content
stringlengths 1
15.9M
|
|---|
\section{Introduction}
The integrable spin chain with an inverse square interaction has
received extensive studies in these years.
One of the famous model is the Haldane--Shastry (HS)
spin chain~\cite{Hal88,Sha88}.
The su($m$) HS spin chain is integrable, and
contrary to the ordinary Heisenberg spin chain it has the
$\mathcal{Y}(s\ell_m)$ Yangian symmetry
even in a finite chain~\cite{HHTBP92}.
While the quasi-particle (spinon)
for the Haldane--Shastry has a square dispersion relation,
there exists another integrable spin chain whose quasi-particle
has a linear dispersion relation.
This model, which we call the Polychronakos spin chain,
was originally introduced in
Ref.~\citen{Poly93a},
and thanks to a linear dispersion relation we can compute the
partition function~\cite{Frah93,Poly94a}.
What is remarkable is that,
similar to the case of HS model,
the Polychronakos spin chain also possesses the
$\mathcal{Y}(s\ell_m)$ Yangian symmetry~\cite{Hikam94d}.
{}From the viewpoint of the conformal field theory, it was
realized~\cite{BPS94,Schou94}
that the Yangian symmetry can be embedded into the level-1 WZNW
theory, and that the first Yangian invariant operator is the Virasoro
generator $L_0$.
Indeed, this supports the fact that
the energy spectrum for the
Polychronakos spin chain is equally spaced.
As the Polychronakos spin chain has the exact Yangian symmetry
in the case of finite $N$ spins,
the character for the level-1 WZNW theory can be given in a large $N$
limit of the partition function of the Polychronakos spin
chain~\cite{Hikam94d}.
In this sense, the partition function of spin chain with finite number
of lattice sites
is viewed as the \emph{restricted} character formula.
In a computation of such partition function,
the Yangian invariant bases which are called the
``motifs''~\cite{HHTBP92,Hal94} play an important role.
These motifs span the Fock space of the Yangian invariant system, and
the degeneracy for each motif can be given from the
representation theory of the Yangian algebra.
The representation for the motif was given for su(2) case in the
original paper~\cite{HHTBP92}, and
for su($m$) case it was established by one of the
author~\cite{Hikam94d,Hikam94e},
where the main observation is that the partition function for
finite-site Polychronakos spin chain can be identified with the
Rogers--Szeg{\"o} (RS)
polynomial~\cite{Andre76} at some special values.
In this paper we consider the su($m|n$) supersymmetric extension of the
Polychronakos spin chain (SP model), whose Hamiltonian is written as
\begin{equation}
\label{Hamilton_P}
\mathcal{H}^{(m|n)}
=
\sum_{1 \leq i< j \leq N}
\frac{
1 - P_{i j}
}{
( z_i - z_j )^2
} .
\end{equation}
Here $z_i$ (for $i=1,2,\dots,N$)
are zeros of the $N$-th Hermite polynomial.
The supersymmetric spin operator $P_{i j}$ is written as~\cite{Hal94}
\begin{equation}
P_{i j}
=
\sum_{\alpha , \beta = 1}^{n+m}
c_{i, \alpha}^\dagger \, c_{j, \beta}^\dagger \,
c_{i , \beta} \, c_{j, \alpha} ,
\end{equation}
Here the creation--annihilation operators, $c_{i,\alpha}^\dagger$ and
$c_{i, \alpha}$, are
\begin{equation*}
c_{i, \alpha}
:
\begin{cases}
\text{bosonic} & \text{for $\alpha= 1, \dots, m$},
\\[2mm]
\text{fermionic}
& \text{for $\alpha= m+1, \dots, m+n$ ,}
\end{cases}
\end{equation*}
and we have a constraint
\begin{equation*}
\sum_{\alpha=1}^{m+n} c_{i, \alpha}^\dagger \, c_{i, \alpha} = 1 .
\end{equation*}
Note that the spin operator $P_{ij}$ is a permutation operator,
satisfying
\begin{align*}
P_{ij} \, P_{jk}
&= P_{jk} \, P_{ki} = P_{ki} \, P_{ij} ,
&
P_{ij}^{~2}
& = 1 .
\end{align*}
As was shown in Ref.~\citen{Poly94a}, the su($m$) Polychronakos spin
chain is
a static limit of the spin Calogero model confined in the harmonic
potential.
We can then compute the partition function of the su($m$)
Polychronakos spin model by factoring out the
dynamical degree of freedom from the spin Calogero model.
By using a similar approach for the case of SP
model~\eqref{Hamilton_P},
one can obtain
the corresponding partition function
$\mathcal{Z}_N^{(m|n)}(q)
= \Tr q^{\displaystyle \mathcal{H}^{(m|n)}}$
as~\cite{BiruUjinWada99a}
\begin{equation}
\label{partition}
\mathcal{Z}_N^{(m|n)}(q)
=
\sum_{
\substack{
\sum_{i=1}^m a_i + \sum_{j=1}^n b_j =N
\\
a_i \geq 0 , \quad b_j \geq 0
}
}
\frac{
(q ; q)_N
}{
\displaystyle
\prod_{i=1}^m ( q ; q)_{a_i} \cdot
\prod_{j=1}^n ( q ; q)_{b_j}
}
\cdot
q^{\displaystyle
\frac{1}{2}
\sum_{j=1}^n b_j \, ( b_j - 1)
} .
\end{equation}
See \S~\ref{sec:pre} for definitions.
We note that the partition function~\eqref{partition} has the duality
for $m, n \neq 0$,
\begin{equation*}
\mathcal{Z}_N^{(m|n)}(q)
=
q^{\displaystyle
\frac{N \, (N-1)}{2}
} \,
\mathcal{Z}_N^{(n|m)}(q^{-1}) .
\end{equation*}
We remark that in the pure fermion case, \emph{i.e.}, when $m=0$,
we need to multiply prefactor in~\eqref{partition}
for taking into account the non-zero ground state energy of the
related spin Calogero model~\cite{Poly94a,BiruUjinWada99a}, and
in the following we always suppose $m \neq 0$.
A main purpose of the present Article
is to introduce motifs as eigenstates
of~\eqref{Hamilton_P}, and to give representation for the motifs.
We can naturally define the
supersymmetric RS polynomial from the partition
function~\eqref{partition}, and
based on the recursion relation for those polynomials
we can compute the degeneracy of motifs and the distribution function
for the quasi-particles.
This Article is organized as follows.
In \S~\ref{sec:pre} we explain notations used in this paper.
We review some important properties for the Schur polynomials
following Refs.~\citen{Macdo95,WFult97Book}.
In \S~\ref{sec:RSpoly} we introduce a supersymmetric analogue of the
Rogers--Szeg{\"o} polynomial.
This polynomial
would reproduce the partition
function~\eqref{partition} at some special values, and we study the
recursion relation for
these polynomials.
In \S~\ref{sec:representation}
we introduce motif as eigenstates of the
SP model, and
we give the representation for
motif using the supersymmetric skew Schur polynomials.
In \S~\ref{sec:distribute} we calculate the distribution function for
quasi-particles and
the central charge by
use of the recursion relation for the restricted partition function.
We also discuss a relationship with the character formula in
\S~\ref{sec:WZW}.
The last section is devoted to discussions and concluding remarks.
\section{Preliminaries}
\label{sec:pre}
\subsection{Notation}
We denote a $q$-polynomial as
\begin{equation}
(t ; q)_N
=
\prod_{i=1}^N
( 1 - t \, q^{i-1} ) ,
\end{equation}
for $N>0$, and we set $(t;q)_0 = 1$.
For this polynomial, we have an identity~\cite{Andre76},
\begin{equation}
\label{help_Eq}
\frac{1}{
(t ; q)_\infty
}
=
\sum_{N=0}^\infty
\frac{
t^N
}{
(q ; q)_N
} .
\end{equation}
A partition $\lambda$
(see Refs.~\citen{Macdo95,WFult97Book} for detail)
is given by a sequence of weakly decreasing
positive integers, and we often write it as
$\lambda = [ \lambda_1 , \lambda_2 ,\dots, \lambda_m]$.
The conjugate diagram $\Tilde{\lambda}$ is given by flipping a diagram
$\lambda$ over its main diagonal (from upper left to lower right).
A skew diagram $\lambda/\mu$
is obtained by removing a smaller Young diagram
$\mu=[\mu_1, \mu_2, \dots]$ from a
larger one $\lambda$
that contains it.
Hereafter, we often use the skew Young diagrams
$\langle m_1 , m_2 , \dots, m_r \rangle$,
which denote the border strip of $r$-columns such that the length of
the $i$-th column is $m_i$;
\begin{equation}
\label{border_strip}
\langle m_1, m_2, \dots , m_r \rangle =
\begin{array}{*{4}{|p{\arraycolsep}}|ccc}
\cline{4-4}
\multicolumn{3}{c|}{} & & \uparrow & & \\
\cline{4-4}
\multicolumn{3}{c|}{} & & & & \\
\cline{4-4}
\multicolumn{3}{c|}{} & \vdots & m_1 & & \\
\cline{3-4}
\multicolumn{2}{c|}{} & & & \downarrow & \uparrow & \\
\cline{3-4}
\multicolumn{2}{c|}{} &
\vdots & \multicolumn{1}{|c}{} & & m_2 & \\
\cline{2-3}
\multicolumn{1}{c|}{} & & &\multicolumn{1}{|c}{} & & \downarrow
& \uparrow \\
\cline{2-3}
\multicolumn{1}{c|}{} & \vdots & \multicolumn{2}{|c}{} & & & m_3 \\
\cline{1-2} & & \multicolumn{2}{|c}{} & & & \downarrow\\
\cline{1-2}
\vdots & \multicolumn{6}{|c}{} \\
\end{array}
\end{equation}
\subsection{Schur Polynomial}
For each partition $\lambda$,
we define the su($m$) Schur polynomial
$s_\lambda(x) = s_\lambda(x_1, \dots, x_m)$
by the Jacobi--Trudi
formula~\cite{Macdo95,WFult97Book};
\begin{equation}
s_\lambda(x)
=
\frac{
\det
\left(
(x_j)^{\lambda_i + m - i}
\right)_{1 \leq i , j \leq m}
}{
\det
\left(
(x_j)^{m - i}
\right)_{1 \leq i , j \leq m }
} .
\end{equation}
Note that $s_\lambda(x)=0$ when $\length(\lambda) > m$.
For our brevity, we define
the elementary and complete symmetric polynomials $e_N(x)$ and
$h_N(x)$ respectively as
\begin{align}
e_N(x)
& = s_{[1^N]}(x) ,
&
h_N(x)
& = s_{[N]}(x) .
\label{element_complete}
\end{align}
The polynomial $h_N(x)$ is the sum of all distinct monomials of degree
$N$ in the variables $x$, while $e_N(x)$ is the sum of all monomials
$x_{i_1} \dots x_{i_\ell}$ for all strictly increasing sequences
$1 \leq i_1 < \dots < i_\ell \leq m$.
We have the generating functions for polynomials $e_N(x)$ and
$h_N(x)$ as
\begin{gather}
\prod_{i=1}^m ( 1 + t \, x_i)
=
\sum_{N=0}^m e_N(x) \, t^N ,
\label{generate_element}
\\[2mm]
\prod_{i=1}^m
\frac{1}{
1 - t \, x_i
}
=
\sum_{N=0}^\infty h_N(x) \, t^N .
\label{generate_complete}
\end{gather}
We note that the Schur polynomials $s_\lambda(x)$ can be also given
as~\cite{WFult97Book}
\begin{align}
\label{SchurDet}
s_\lambda (x)
& =
\det
\left(
h_{\lambda_i + j - i} (x)
\right)_{1 \leq i , j \leq \length(\lambda)}
\nonumber \\
& =
\det
\left(
e_{\Tilde{\lambda}_i + j - i} ( x)
\right)_{1 \leq i , j \leq \length(\Tilde{\lambda})} .
\end{align}
The above definitions for the Schur polynomials can be generalized
straightforwardly to the
skew Young diagram $\lambda/\mu$.
We define the skew Schur polynomials by~\cite{Macdo95,WFult97Book}
\begin{align}
s_{\lambda/\mu} (x)
& =
\det
\left(
h_{\lambda_i - \mu_j + j - i}(x)
\right)_{1 \leq i , j \leq \length(\lambda)}
\nonumber \\
& =
\det
\left(
e_{\Tilde{\lambda}_i - \Tilde{\mu}_j + j - i} (x)
\right)_{1 \leq i , j \leq \length(\Tilde{\lambda})} .
\label{skewSchur}
\end{align}
where $\Tilde{\lambda}$ and $\Tilde{\mu}$ denote the conjugate
partitions.
\subsection{Supersymmetric Schur Polynomial}
The su($m|n$) supersymmetric analogue of the Schur polynomials can be
defined as
follows.
As an extension of~\eqref{SchurDet},
we define the supersymmetric Schur polynomials
(sometimes called as the hook Schur polynomial, or bisymmetric
polynomial)
$S_\lambda(x,y) = S_{\lambda}(x_1, \dots, x_m , y_1 , \dots, y_n)$
as
(see, e.g., Ref.~\citen{WFult97Book})
\begin{equation}
S_\lambda ( x , y)
=
\det
\left(
c_{\lambda_i + j - i}
\right)_{
1 \leq i, j \leq \length(\lambda)
} ,
\end{equation}
where $c_N = c_N(x,y)$ is given by
\begin{equation}
\label{defineCN}
\frac{\displaystyle
\prod_{j=1}^n ( 1 + t \, y_j )
}{
\displaystyle
\prod_{i=1}^m ( 1 - t \, x_i)
}
=
\sum_{N=0}^\infty
c_N \, t^N .
\end{equation}
The coefficients $c_N$ correspond to a supersymmetric analogue of the
complete symmetric polynomial $h_N(x)$~\eqref{generate_complete}.
One easily sees a duality;
\begin{equation*}
S_\lambda(x_1, \dots, x_m, y_1, \dots , y_n)
=
S_{\Tilde{\lambda}}(y_1, \dots, y_n , x_1 , \dots, x_m) ,
\end{equation*}
where $\Tilde{\lambda}$ is the conjugate partition.
We can also define the supersymmetric analogue
of the elementary symmetric polynomials.
For our later convention, we set the supersymmetric elementary
polynomials as
\begin{equation}
\label{susyElement}
E_N(x,y)
=
S_{[1^N]}(x,y) .
\end{equation}
Based on identities~\eqref{generate_element},~\eqref{generate_complete},
and~\eqref{defineCN},
the supersymmetric elementary function is
decomposed~\cite{BalanIBars82a,BarMorRue83a} as
\begin{equation*}
E_N(x,y)
=
\sum_{k=0}^N
e_{k}(x) \cdot h_{N-k}(y) .
\end{equation*}
By using this relation it is easy to see that
$E_N(x,y) \neq 0$ for arbitrary $N$.
We note that the generating function for the polynomials $E_N(x,y)$ is
given by
\begin{equation}
\label{generateSuperE}
\frac{\displaystyle
\prod_{i=1}^m ( 1 - t \, x_i )
}{
\displaystyle
\prod_{j=1}^n ( 1 + t \, y_j)
}
=
\sum_{N=0}^\infty (-)^N \, E_N(x,y) \, t^N .
\end{equation}
\section{Recursion Relation for the Partition Function}
\label{sec:RSpoly}
As the Rogers--Szeg{\"o} polynomial reproduces the partition function
for the su($m$) Polychronakos model at some special values~\cite{Hikam94d},
it is natural to study a supersymmetric analogue of the RS polynomials
which is related with the partition function~\eqref{partition}.
In this section
we give the recursion relations for such supersymmetric RS
polynomials in terms
of the Young diagrams.
\subsection{Supersymmetric Rogers--Szeg{\"o} Polynomials}
We define
polynomials
$H_N^{(m|n)}(x,y) = H_N^{(m|n)}(x_1, \dots, x_m, y_1, \dots, y_n ; q)$
as
\begin{equation}
\label{polyH}
H_N^{(m|n)}(x,y)
=
\sum_{
\substack{
\sum_{i=1}^m a_i + \sum_{j=1}^n b_j = N
\\
a_i \geq 0 , \quad b_j \geq 0
}
}
\frac{
(q;q)_N
}{
\displaystyle
\prod_{i=1}^m (q ; q)_{a_i} \cdot
\prod_{j=1}^n (q^{-1} ; q^{-1} )_{b_j}
}
\cdot
x_1^{a_1} \dots x_m^{a_m} \cdot
\left(
- \frac{y_1}{q}
\right)^{b_1} \dots
\left(
- \frac{y_n}{q}
\right)^{b_n} ,
\end{equation}
which we call the su($m|n$) supersymmetric Rogers--Szeg{\"o} (SRS)
polynomial.
It is easy to see that the polynomial $H_N^{(m|n)}(x,y)$ gives the
partition function~\eqref{partition}
for the SP model~\eqref{Hamilton_P},
\begin{equation}
\mathcal{Z}_N^{(m|n)}(q)
=
H_N^{(m|n)}(x = 1 , y=1) .
\end{equation}
Next we try to derive the recursion relation for the SRS
polynomials~\eqref{polyH}.
For this purpose, it is useful to give a generating function of the
polynomial~\eqref{polyH}.
So we introduce a function
$G^{(m|n)}(t) = G^{(m|n)}(t ; x_1, \dots, x_m , y_1, \dots, y_n ;q )$
as
\begin{equation}
\label{generate_Function}
G^{(m|n)}(t)
=
\frac{1}
{
\displaystyle
\prod_{i=1}^m ( t \, x_i ; q)_\infty
\cdot
\prod_{j=1}^n
( - t \, y_j \, q^{-1} ; q^{-1} )_\infty
} .
\end{equation}
We see that,
by use of an identity~\eqref{help_Eq},
$G^{(m|n)}(t)$
is a generating function for the
SRS polynomial~\eqref{polyH};
\begin{equation}
\label{generateH}
G^{(m|n)}(t)
=
\sum_{N=0}^\infty
\frac{
H_N^{(m|n)}(x,y)
}{
(q ; q)_N
} \cdot
t^N .
\end{equation}
By definition~\eqref{generate_Function},
the function $G^{(m|n)}(t)$ satisfies
$q$-difference equation,
\begin{equation}
\label{differenceG}
\left(
\prod_{j=1}^n
(1 + t \, y_j)
\right) \cdot
G^{(m|n)} (q \, t)
=
\left(
\prod_{i=1}^m
(1 - t \, x_i)
\right) \cdot
G^{(m|n)} (t) .
\end{equation}
Substituting~\eqref{generateH} to above equation and using the
elementary symmetric polynomials~\eqref{generate_element},
we see that the SRS polynomial satisfies the following
recursion relation for any ($\max(n,m) +1$)-consecutive polynomials;
\begin{equation}
\label{FirstRecur}
H_N^{(m|n)}(x,y)
=
\sum_{k=1}^{\max(n,m)}
(-)^{k-1}
\frac{
(q ;q )_{N-1}
}{
(q ; q)_{N-k}
} \cdot
\left(
e_k(x)
- (-)^k \, q^{N-k} \, e_k(y)
\right) \cdot
H_{N-k}^{(m|n)}(x,y) ,
\end{equation}
where
we have set
$H_k^{(m|n)}(x,y)=0$ for $k<0$.
However, by using the expression~\eqref{generateSuperE} which contains
supersymmetric elementary polynomials, we can rewrite the
$q$-difference equation~\eqref{differenceG} as
\begin{equation*}
G^{(m|n)} (q \, t)
=
\left(
\sum_{N=0}^\infty (-)^N \, E_N(x,y) \, t^N
\right) \cdot
G^{(m|n)} (t) .
\end{equation*}
Substituting~\eqref{generateH} to above equation,
we obtain another type of recursion relation for the SRS polynomials
as
\begin{equation}
\label{recurH2}
H_N^{(m|n)}(x,y)
=
\sum_{k=1}^N
(-)^{k+1}
\frac{
(q ; q)_{N-1}
}{
(q ; q)_{N-k}
} \cdot
E_k(x,y) \cdot
H_{N-k}^{(m|n)}(x,y) .
\end{equation}
Notice that,
since
the elementary polynomials $E_k(x,y)$ do not vanish for
arbitrary $k$,
the polynomials $H_N^{(m|n)}(x,y)$ now depend on
every lower degree polynomials
$H_j^{(m|n)}(x,y)$ for $j<N$
contrary to the first type recursion relation~\eqref{FirstRecur}.
In the case of $n=0$, both~\eqref{FirstRecur} and~\eqref{recurH2}
reduce to the recursion relation for the Rogers--Szeg{\"o}
polynomial~\cite{Hikam94d,Hikam94e}.
\subsection{Supersymmetric Skew Schur Polynomial}
We define another $q$-polynomials
$F_N^{(m|n)}(x, y)
=
F_N^{(m|n)}(x_1, \dots, x_m , y_1, \dots , y_n ; q)
$
following Ref.~\citen{KiriKuniNaka96a};
\begin{equation}
\label{defineF}
F_N^{(m|n)}(x,y)
=
\sum_{r=1}^N
\sum_{
\substack{
m_1 + \dots + m_r =N
\\
m_r \geq 1
}}
q^{
\displaystyle
\frac{N(N+1)}{2}
-
\sum_{i=1}^r
\left(
m_1 + \dots + m_i
\right)
}
\times
S_{\langle m_1 , \dots, m_r \rangle}(x,y) ,
\end{equation}
where $S_{\langle m_1 , \dots , m_r \rangle}(x,y)$ is the
supersymmetric skew Schur polynomial for the skew Young diagram
$\langle m_1 , \dots , m_r \rangle$~\eqref{border_strip}.
Being motivated from the expression of the
skew Schur polynomials~\eqref{skewSchur} in non-supersymmetric case,
we define the above mentioned supersymmetric Schur polynomials
through the supersymmetric elementary
polynomials~\eqref{susyElement}
as
\begin{equation}
\label{Det_express}
S_{\langle m_1 , \dots, m_r \rangle}(x,y)
=
\begin{vmatrix}
E_{m_r} & E_{m_r + m_{r-1}} & \dots & \dots & E_{m_r +
\dots + m_1}
\\
1 & E_{m_{r-1}} & E_{m_{r-1} + m_{r-2}}& \dots & E_{m_{r-1} +
\dots + m_1}
\\
0 & 1 & E_{m_{r-2}}& \dots & \vdots
\\
\vdots & \ddots & \ddots & \ddots& \vdots
\\
0 & \dots & 0 & 1 & E_{m_1}
\end{vmatrix} .
\end{equation}
Remark that, since $E_N(x,y) \neq 0$ for arbitrary $N$,
$S_{\langle m_1, \dots , m_r \rangle}(x,y)$ will be non-trivial for
arbitrary set of $m_1$, $m_2, \dots, m_r$.
Expanding the determinant in~\eqref{Det_express}
along the first row, we get a recursion relation for the
supersymmetric Schur polynomials as
\begin{equation*}
S_{\langle m_1, \dots m_r \rangle}(x,y)
=
\sum_{i=1}^r
(-)^{i+1} \,
E_{m_r + \dots + m_{r-i+1}}(x,y) \cdot
S_{\langle m_1, \dots, m_{r-i} \rangle}(x,y) .
\end{equation*}
By substituting the above relation to the
expression~\eqref{defineF} and
following Appendix~A in Ref~\citen{KiriKuniNaka96a},
we find
that the polynomial
$F_N^{(m|n)}(x,y)$ satisfies the recursion relation,
\begin{equation}
F_N^{(m|n)}(x,y)
=
\sum_{k=1}^N
(-)^{k+1}
\frac{
(q ; q)_{N-1}
}{
(q ; q)_{N-k}
} \cdot
E_k(x,y) \cdot
F_{N-k}^{(m|n)}(x,y) ,
\end{equation}
which is exactly same with the recurrence relation for the polynomials
$H_N^{(m|n)}(x,y)$~\eqref{recurH2}.
We can easily recognize that the initial conditions are same for two
sets of polynomials, $H_N^{(m|n)}(x,y)$ and $F_N^{(m|n)}(x,y)$,
\emph{e.g.},
\begin{align}
F_0^{(m|n)}(x,y)
& = 1 ,
\nonumber
\\[2mm]
F_1^{(m|n)}(x,y)
& = S_{[1]}(x,y) ,
\nonumber
\\[2mm]
F_2^{(m|n)}(x,y)
& =
q \, S_{[1^2]}(x,y) + S_{[2]}(x,y) ,
\label{poly_example}
\\[2mm]
H_0^{(m|n)}(x,y)
& = 1 ,
\nonumber
\\[2mm]
H_1^{(m|n)}(x,y)
& =
\sum_i x_i + \sum_j y_j ,
\nonumber
\\[2mm]
H_2^{(m|n)}(x,y)
& =
\sum_i x_i^{~2} + q \, \sum_j y_j^{~2}
+ (1+q)
\Bigl(
\sum_{i_1 < i_2} x_{i_1} \, x_{i_2}
+ \sum_{i,j} x_i \, y_j
+ \sum_{j_1 < j_2} y_{j_1} \, y_{j_2}
\Bigr) .
\nonumber
\end{align}
As a result, we conclude that
\begin{equation}
\label{FequalH}
F_N^{(m|n)} (x,y)
= H_N^{(m|n)}(x,y) ,
\end{equation}
and that the polynomial $F_N^{(m|n)}(x=1,y=1)$ also gives the
partition function $\mathcal{Z}^{(m|n)}(q)$ for the SP model.
\section{Motif: Eigenstates of SP Model}
\label{sec:representation}
We study eigenstates for the supersymmetric
su($m|n$)
Polychronakos model~\eqref{Hamilton_P}.
Due to the Yangian symmetry of the system,
we can use the motif~\cite{HHTBP92,Hal94} which spans the Fock
space of the Yangian invariant spin systems.
The motif $d$ for ($N+1$)-site spin chain is given by an
$N$ sequence of
0's and 1's;
$
d= (d_1, d_2, \dots , d_{N})
$
with
$d_j \in \{ 0, 1 \}$.
A value $d_j=1$
(resp. $d_j=0$)
denotes that the $j$-th energy level is occupied
(resp. empty) by the quasi-particle.
The energy of the SP model is given by
\begin{equation}
\label{energy_motif}
E(d) = \sum_{j=1}^{N} j \, d_j .
\end{equation}
We remark that the energy of the Yangian invariant Haldane--Shastry
type spin chain~\cite{Hal88,Sha88,HHTBP92} is given by
$E_{\text{HS}}
=\sum_j j \, ( N-j) \, d_j
$.
We define $g_N(d)$ as a degeneracy of the motif $d=(d_1, \dots,d_N)$
for ($N+1$)-site spin chain.
We note that $g_N(d)$ generally depends on $x$ and $y$, and that
the degeneracy for each motif is given by setting
$x_i=y_j=1$.
Then the partition function~\eqref{partition} is written as
\begin{equation}
\label{partition_motif}
\mathcal{Z}_{N+1}(q)
=
\sum_d
q^{\displaystyle
E(d)
}
\,
g_N(d) .
\end{equation}
In the su($m$) case, we have a selection rule for motifs
such that $m$-consecutive 1's are forbidden.
On the contrary, in the su($m|n$)
supersymmetric case we do not have such selection
rules;
every sequence $d$ is permitted generally~\cite{Hal94,BasuMa99a}.
{}From the first few polynomials~\eqref{poly_example} we can read
as
\begin{align*}
g_0(~)
& = S_{[1]}(x,y) = e_1(x) + e_1(y) ,
\\
g_1(0)
& =
S_{[2]}(x,y) ,
\\
g_1(1)
& =
S_{[1^2]}(x,y) .
\end{align*}
Below we give two methods to obtain the degeneracy for the general
motifs.
\subsection{Recursion Relation for Motif}
We study the representation of motif based on the first recursion
relation~\eqref{FirstRecur}.
Using~\eqref{partition_motif}, we can translate the recursion
relation~\eqref{FirstRecur} for the SRS polynomials to that for motifs.
We give the several examples as follows.
By putting some special $m$ and $n$ to
the first recurrence relation~\eqref{FirstRecur} and translating it in
terms of motif,
we get
the following equations;
\begin{enumerate}
\def(\Alph{enumi}).{(\Alph{enumi}).}
\item su($2$);
\begin{align}
g_N(d_1 , \dots, d_N)
& =
\delta_{d_N, 0} \cdot e_1(x)
\cdot g_{N-1}(d_1, \dots, d_{N-1})
\nonumber \\
& \qquad
-
\left(
\delta_{d_N , 0} \cdot \delta_{d_{N-1} , 0}
-
\delta_{d_N , 1} \cdot \delta_{d_{N-1} , 0}
\right) \cdot e_2(x) \cdot
g_{N-2}(d_1,\dots,d_{N-2}) .
\end{align}
\item su($2|1$);
\begin{align}
g_N(d_1, \dots, d_N)
& =
\left(
\delta_{d_N,0} \, e_1(x) + \delta_{d_N , 1} e_1(y)
\right) \cdot g_{N-1}(d_1, \dots, d_{N-1})
\nonumber \\
& \qquad
-
\left(
\delta_{d_N , 0} \delta_{d_{N-1} , 0}
-
\delta_{d_N , 1} \delta_{d_{N-1} , 0}
\right) \, e_2(x) \cdot
g_{N-2}(d_1,\dots, d_{N-2} ) .
\end{align}
\item su($2|2$);
\begin{align}
g_N (d_1, \dots, d_N)
& =
\left(
\delta_{d_N,0} \, e_1(x) + \delta_{d_N, 1} \, e_1(y)
\right) \cdot
g_{N-1}(d_1, \dots, d_{N-1})
\nonumber \\
& \qquad
-
\Bigl(
\bigl(
\delta_{d_N , 0} \, \delta_{d_{N-1},0}
-
\delta_{d_N,1} \, \delta_{d_{N-1} , 0}
\bigr) \, e_2(x)
\nonumber \\
& \qquad \qquad
-
\bigl(
\delta_{d_N , 0} \, \delta_{d_{N-1},1}
-
\delta_{d_N,1} \, \delta_{d_{N-1} , 1}
\bigr) \, e_2(y)
\Bigr) \cdot
g_{N-2}(d_1, \dots, d_{N-2}).
\end{align}
\end{enumerate}
See that
the recurrence equation for
the su($2|2$) case reduces to those for su($2|1$) and su($2$) cases
by setting $e_2(y)=0$ and $e_1(y)=e_2(y)=0$, respectively.
By using these recurrence equations for the motifs, we can compute
degeneracy for each motif by setting $x_i = y_j = 1$.
We give some examples for $N=4$ below.
\begin{enumerate}
\def(\Alph{enumi}).{(\Alph{enumi}).}
\item su($2$) and su($1|1$);
\bigskip
\begin{center}
\begin{tabular}{cccc}
motif & energy & degeneracy for su(2) & degeneracy for su($1|1$) \\
\hline \hline
(000) & 0 & 5 & 2\\
(100) & 1 & 3 & 2\\
(010) & 2 & 4 & 2\\
(001) & 3 & 3 & 2\\
(101) & 4 & 1 & 2\\
(110) & 3 & $-$ & 2 \\
(011) & 5 & $-$ & 2 \\
(111) & 6 & $-$ & 2 \\
\hline \hline
& \text{total} & $2^4$ & $2^4$
\end{tabular}
\end{center}
\item su(3) and su($2|1$);
\bigskip
\begin{center}
\begin{tabular}{cccc}
motif & energy & degeneracy for su($3$) & degeneracy for su($2|1$) \\
\hline \hline
(000) & 0 & 15 &9 \\
(100) & 1 & 15& 12 \\
(010) & 2 & 21 & 16 \\
(001) & 3 & 15& 12 \\
(101) & 4 & 9 & 12 \\
(110) & 3 & 3 & 8 \\
(011) & 5 & 3 & 8 \\
(111) & 6 & $-$ & 4 \\
\hline \hline
& \text{total} & $3^4$ & $3^4$
\end{tabular}
\end{center}
\end{enumerate}
\subsection{Motif and Skew Young Diagram}
We have shown that the representation of motif is given recursively by
use of the first recursion relation~\eqref{FirstRecur}.
In this subsection, we shall show that,
by use of the skew Young
diagram and the polynomials $F_N^{(m|n)}(x,y)$, we can directly give
the representation for motif.
We can then give the decomposition rule for each motif, and clarify
the hyper-multiplet structure in the spectrum of the SP model.
{}From a correspondence between a power of $q$ in~\eqref{defineF} and
the energy for motif $d$ in~\eqref{partition_motif}, we can define a
map from motif
$d$
to the skew Young diagram
$\langle m_1 ,\dots, m_r \rangle$~\cite{KiriKuniNaka96a}.
A rule for translation is:
we read a motif $d=(d_1, d_2, \dots)$
from the left, and
we add a box under (resp. left) the box when we encounter `$d_j=1$'
(resp. `$d_j=0$').
One sees that there is a one-to-one correspondence between motifs and
skew Young diagrams.
As we have stressed before, the supersymmetric skew Schur
polynomials
$S_{\langle m_1, \dots, m_r \rangle}(x,y)$ defined in~\eqref{Det_express} do
not vanish for arbitrary set of $m_1, \dots, m_r$,
and
this fact proves that
there is no selection rule
for motif $d$ while the $m$-consecutive 1's are forbidden in the
non-supersymmetric
su($m$) case.
We give examples up to $N=4$ below.
We have decomposed following a rule of the supersymmetric Young
diagrams.
\bigskip
\begin{center}
\begin{tabular}{cll}
motif & skew Young diagram & decomposition \\
\hline \hline
(~) & $ \langle 1 \rangle$ & $[1]$ \\
&& \\
(0) & $\langle 1,1 \rangle$ & $[2]$ \\
(1) & $\langle 2 \rangle$ & $[1^2]$ \\
&& \\
(11) & $\langle 3 \rangle$ & $[1^3]$ \\
(01) & $\langle 1,2 \rangle$ & $[2,1]$ \\
(10) & $\langle 2,1 \rangle$ & $[2,1]$ \\
(00) & $\langle 1,1,1 \rangle$ & $[3]$ \\
&& \\
(111) & $\langle 4 \rangle$ & $[1^4]$ \\
(110) & $ \langle 3 , 1 \rangle$ & $[2,1^2]$ \\
(101) & $ \langle 2 , 2 \rangle$ & $[2^2] \oplus [2, 1^2]$ \\
(011) & $ \langle 1 , 3 \rangle$ & $[2,1^2]$ \\
(100) & $ \langle 2 , 1 ,1 \rangle$ & $[3,1]$ \\
(010) & $ \langle 1 , 2 , 1 \rangle$ & $[3,1] \oplus [2^2]$ \\
(001) & $ \langle 1 , 1, 2 \rangle$ & $[3,1]$ \\
(000) & $ \langle 1, 1, 1, 1 \rangle$ & $[4]$
\end{tabular}
\end{center}
\section{Distribution Function}
\label{sec:distribute}
In the preceding sections, we have shown that the motifs, which are
the eigenstates of the SP model,
satisfy some recursion relation.
As the motifs are composed of the ``quasi-particles'', we
shall consider their distribution function.
Once the recursion relation for the partition function is given, it is
straightforward to compute the distribution function by use of the
asymptotic form of the partition function.
This method was originally introduced in
Refs.~\citen{Hikam95b,Hikam97e}
for a study of the exclusion
statistics~\cite{Hal91b}.
Namely in the SP model the energy dispersion relation is
linear, so we can regard $q^k$ as
\begin{equation}
V_k = \exp \left( - \beta ( \varepsilon_k - \mu) \right) ,
\end{equation}
in which $\beta$, $\mu$, and $\varepsilon_k$ denote the inverse of
temperature, the chemical potential, and the $k$-th energy
respectively.
Correspondingly
the polynomial $H_N^{(m|n)}(x,y)$
can be viewed as the \emph{restricted} partition function $\varphi_N$,
in which particles can occupy up to $N$-th energy level
$\varepsilon_N$
(here dispersion relation for energy $\varepsilon_k$ is arbitrary).
By assuming an asymptotic form of the restricted partition
function
\begin{equation}
\label{Jost}
\varphi_k \sim w^{-k},
\end{equation}
we find that the occupation of state $\varepsilon_k$ is given by
\begin{equation}
\label{distribution}
\langle n_{\text{av}} \rangle
\simeq
\frac{1}{k} \cdot
\frac{\partial}{\partial ( \beta \, \mu)} \log \varphi_k
=
-
\frac{\partial}{\partial ( \beta \, \mu)} \log w ,
\end{equation}
where the spectral parameter $w = w(V)$
is a function of $V=\exp \left( - \beta \, (\varepsilon - \mu) \right)$,
and it can be determined from the recursion
relation for the restricted partition function.
In the same way, we can compute the specific heat from the partition
function, and we get the central charge $c$ of the theory
as~\cite{FrenkSzene93a,Schou97a,BouwkSchou98a,BerkoBMcCo98a}
\begin{equation}
\label{central_charge}
c
=
- \frac{6}{\pi^2}
\int_0^1 \mathrm{d}V \, \frac{1}{V} \log w(V) ,
\end{equation}
which is often rewritten by use of the Rogers' dilogarithm function
(see, \emph{e.g.}, Ref.~\citen{Kiril} and references therein).
In this section we demonstrate for some simple supersymmetric cases.
\begin{enumerate}
\def(\Alph{enumi}).{(\Alph{enumi}).}
\item
su($1|1$);
The recursion relation for the su($1|1$) SRS polynomials
is read off as
\begin{equation*}
H_N^{(1|1)}(x,y)
=
\left(
x + q^{N-1} \, y
\right) \, H_{N-1}^{(1|1)}(x,y) ,
\end{equation*}
which shows that the restricted partition function satisfies
\begin{equation}
\varphi_N
=
\left( 1 + V_{N-1} \right) \, \varphi_{N-1} .
\end{equation}
We see using~\eqref{Jost} that the spectral parameter is given by
\begin{equation}
w = ( 1+ V)^{-1} .
\end{equation}
We thus obtain the central charge~\eqref{central_charge} as
\begin{equation*}
c=\frac{1}{2} ,
\end{equation*}
which denotes that the theory coincides with the free fermion
theory.
This fact is simply realized by explicitly computing the character
in \S~\ref{sec:WZW}.
\item
su($m$);
Based on a recursion relation for the RS polynomial, we see that the
restricted partition function $\varphi_k$ satisfies
\begin{equation}
\varphi_N
=
\sum_{i=1}^m (-)^{i-1} \,
{}_m \mathrm{C}_i \, ( 1- V_{N-1}) \, (1 - V_{N-2}) \,
( 1- V_{N-i+1}) \, \varphi_{N-i} ,
\end{equation}
where ${}_m \mathrm{C}_i$ denote binomial coefficients.
By assuming $V_N$ does not change drastically for $N$,
we find~\cite{Hikam97e} that the spectral parameter $w$~\eqref{Jost}
satisfies
the functional equation,
\begin{equation}
V =
\left(
1 - ( 1 - V) \, w
\right)^m .
\end{equation}
We also see
that the central
charge~\eqref{central_charge} is computed
as
\begin{equation*}
c = m-1 ,
\end{equation*}
which is known as the central charge for the level-1 su($m$) WZNW
theory.
\item
su($2|1$);
The recursion relation for the partition function for the su($2|1$)
SP model is given as
\begin{equation*}
H_N^{(2|1)}(x,y) =
\left( x_1 + x_2 + q^{N-1} \, y \right) \cdot
H_{N-1}^{(2|1)}(x,y)
-
(1 - q^{N-1}) \cdot x_1 \, x_2 \cdot
H_{N-2}^{(2|1)}(x,y) ,
\end{equation*}
which, in terms of the restricted partition function, is rewritten as
\begin{equation}
\varphi_N
= ( 2 + V_{N-1} ) \cdot \varphi_{N-1} - ( 1 - V_{N-1}) \cdot
\varphi_{N-2} .
\end{equation}
By using an asymptotic form~\eqref{Jost}, we see that the spectral
parameter satisfies the functional equation,
\begin{equation*}
(1-V) \, w^2 - (2+V) \, w + 1 = 0 ,
\end{equation*}
which gives
\begin{equation}
\langle n_{\text{av}}^{\mathrm{su}(2|1)} \rangle
\simeq
-
\frac{\partial}{\partial ( \beta \, \mu)}
\log
\left(
\frac{1}{2 \, (1 - V)}
\left(
2 + V - \sqrt{ V^2 + 8 \, V}
\right)
\right).
\end{equation}
We obtain the central charge from~\eqref{central_charge} as
\begin{equation*}
c = \frac{3}{2}= 1+ \frac{1}{2} ,
\end{equation*}
which shows that the theory has one boson and one fermion.
\item
su($2|2$);
We have the recursion relation for the restricted partition function as
\begin{equation}
\varphi_N
=
2 \, ( 1 + V_{N-1} ) \cdot \varphi_{N-1}
- ( 1 - V_{N-1} ) \, ( 1 - V_{N-2}) \cdot \varphi_{N-2} ,
\end{equation}
which gives the functional equation for the spectral
parameter~\eqref{Jost} as
\begin{equation*}
(1-V)^2 \, w^2 - 2 \, (1+V) \, w + 1 = 0 ,
\end{equation*}
The average number of particle and the central charge are respectively
given as
\begin{gather}
\langle n_{\text{av}}^{\mathrm{su}(2|2)} \rangle
\simeq
2 \,
\frac{\partial}{\partial ( \beta \, \mu)}
\log \left( 1 + \sqrt{V} \right)
=
2 \, \langle n^{\mathrm{su}(2)} \rangle
,
\\[2mm]
c = 2 = 1+ 2 \times \frac{1}{2} . \nonumber
\end{gather}
This result is consistent with the fact that the theory includes one
boson and two fermions.
\item su($1|2$);
The recursion relation for the restricted partition function is
given by
\begin{equation}
\varphi_N
=
( 1+ 2 \, V_{N-1} ) \, \varphi_{N-1}
+
( 1 - V_{N-1} ) \, V_{N-2} \, \varphi_{N-2} .
\end{equation}
Correspondingly we get the functional equation for the spectral
parameter~\eqref{Jost},
\begin{equation*}
V \, (V-1) \, w^2 - ( 1 + 2 \, V) \, w + 1 = 0 .
\end{equation*}
One sees a duality:
$w^{\mathrm{su}(1|2)}(V) = V^{-1} \cdot w^{\mathrm{su}(2|1)}(V^{-1})$.
The distribution function~\eqref{distribution}
and the central charge~\eqref{central_charge}
are respectively computed as
follows;
\begin{gather}
\langle n_{\text{av}}^{\mathrm{su}(1|2)} \rangle
\simeq
-
\frac{\partial}{\partial ( \beta \, \mu)}
\log
\left(
\frac{
1 + 2 \, V - \sqrt{1 + 8 \, V}
}{
2 \, V \, (V-1)
}
\right) ,
\\[2mm]
c = 1 = 2 \times \frac{1}{2} ,
\nonumber
\end{gather}
which shows that
the su($1|2$) theory has two independent fermions.
We also find a duality for the occupation of states as
\begin{equation}
\langle n_{\text{av}}^{\mathrm{su}(1|2)}(V) \rangle
+
\langle n_{\text{av}}^{\mathrm{su}(2|1)}(V^{-1}) \rangle
=1 .
\end{equation}
\end{enumerate}
In Fig.~\ref{fig:su2}
we plot an average occupation of states at a temperature $T>0$
for su($m|n$) cases ($m\leq 2$, $n=0,1,2$)
and for the free fermion case.
We see that
\begin{align*}
\left.
\langle n_{\text{av}} \rangle
\right|_{ \varepsilon \to - \infty}
& =
\begin{cases}
\frac{1}{2} ,
& \text{for su($2$)},
\\[2mm]
1 ,
& \text{for others},
\end{cases}
&
\left.
\langle n_{\text{av}} \rangle
\right|_{\varepsilon \to \infty}
& \simeq
\begin{cases}
\sqrt{V}, & \text{for su($2|2$)},
\\[2mm]
\frac{1}{\sqrt{2}}\sqrt{V},
& \text{for su($2|1$)},
\\[2mm]
3 \, V ,
&
\text{for su($1|2$)},
\end{cases}
\end{align*}
and that
\begin{align*}
\left.
\langle n_{\text{av}} \rangle
\right|_{\varepsilon = \mu}
& =
\begin{cases}
\frac{1}{2} , & \text{for su($2|2$)},
\\[2mm]
\frac{4}{9} ,
& \text{for su($2|1$)},
\\[2mm]
\frac{5}{9},
&
\text{for su($1|2$)}.
\end{cases}
\end{align*}
\begin{figure}[htbp]
\begin{center}
\begin{psfrags}
\psfrag{MMM}{$\varepsilon - \mu$}
\psfrag{NNN}{$\langle n_{\text{av}} \rangle$}
\psfig{file=su2DF.eps}
\end{psfrags}
\caption{Average number of particle is plotted for free fermion
and su($m|n$) cases ($m,n \leq 2$).
A bold line is the distribution function for the free fermion,
$\displaystyle \langle n_{\text{av}} \rangle = \frac{V}{1+V}$,
and solid lines denote those for su(2) and su($2|2$).
A dotted and a dashed line respectively show the distribution
function
for su($2|1$) and su($1|2$) cases.
The energy is given in units $\beta^{-1}$.}
\label{fig:su2}
\end{center}
\end{figure}
To close this section, we observe that the central
charge~\eqref{central_charge} for the su($m|n$) SRS polynomials
($m\neq 1$) will be given generally as
\begin{equation}
\label{center_general}
c = m - 1 + \frac{n}{2} .
\end{equation}
\section{Character Formula}
\label{sec:WZW}
The SP model has the supersymmetric Yangian
symmetry, and it is known that the first conserved operator for the
Yangian operator is the Virasoro generator $L_0$.
We can thus conclude that the character
$\ch^{(m|n)}_{\Lambda}(q)$ for the
su$(m|n)_1$ WZNW model with the highest weight $\Lambda$
is related to the SRS
polynomial as
\begin{equation}
\label{chara_form}
\ch^{(m|n)}_\Lambda(q)
=
\lim_{
\substack{N\to \infty
\\
\text{condition}}
}
H_N^{(m|n)}(x=1, y=1 ; q ) .
\end{equation}
We do not know how to treat the formula~\eqref{chara_form} for general
cases, and
below we give some examples.
\begin{enumerate}
\def(\Alph{enumi}).{(\Alph{enumi}).}
\item
su($1|1$);
It is easy to realize that the su($1|1$)
SRS polynomial is given
as the mere $q$-product,
\begin{equation}
H_N^{(1|1)}(x,y ; q )
=
\prod_{i=1}^N
\left( x + q^{i-1} \, y \right) .
\end{equation}
We then obtain the character formula
as
\begin{align}
\ch^{(1|1)}_\Lambda (q)
& =
\lim_{N\to \infty}
H_N^{(1|1)}(x=1,y=1;q)
\nonumber \\
& =
\prod_{i=0}^\infty
\left( 1 + q^i \right) ,
\end{align}
which denotes that the su$(1|1)$ theory
coincides with the free spinless fermion theory as was demonstrated in
Ref.~\citen{Hal94}.
\item su($m$)~\cite{Hikam94d};
We have
\begin{equation}
\ch_{\Lambda_j}^{(m)} (q)
=
\lim_{
\substack{N \to \infty \\
N \equiv j \mod m}
}
q^{\displaystyle
\frac{m-1}{2 \, m}
N^2
}
\cdot
H_N^{(m|0)}(x=1, y=1; q^{-1}) ,
\end{equation}
which coincides with the character formula for su$(m)_1$ WZNW theory.
\item
su($1|n$);
Based on a numerical computation by use of
\textsc{Mathematica}
we suggest that
\begin{align}
\ch^{(1|n)}_\Lambda (q)
& =
\lim_{N\to \infty}
H_N^{(1|n)}(x_i=1,y=1;q)
\nonumber \\
& =
\prod_{i=0}^\infty
\left(
1 + q^i
\right)^n .
\end{align}
Indeed we can prove this identity from the definition~\eqref{polyH} as
follows~\cite{Warnaar};
\begin{align*}
\ch^{(1|n)}_\Lambda (q)
& =
\lim_{N \to \infty}
\sum_{b_1, \dots, b_n \geq 0}
\frac{(q;q)_N}{
(q;q)_{b_1} \dots
(q;q)_{b_n} \,
(q;q)_{N-b_1- \dots - b_n}
} \,
q^{\displaystyle \frac{1}{2} \sum_{j=1}^n b_j \, (b_j - 1)}
\\
& =
\sum_{b_1, \dots, b_n \geq 0}
\frac{1}{
(q;q)_{b_1} \dots
(q;q)_{b_n}
} \,
q^{\displaystyle \frac{1}{2} \sum_{j=1}^n b_j \, (b_j - 1)}
\\
& =
\left(
\sum_{b=0}^\infty
\frac{1}{(q;q)_b} \,
q^{\frac{1}{2} b \, (b-1)}
\right)^n
\\
& =
\prod_{i=0}^\infty ( 1 + q^i )^n .
\end{align*}
This identity also supports that the theory is written by $n$
fermions~\eqref{center_general}.
\end{enumerate}
\section{Concluding Remarks}
We have studied eigenstates of the
supersymmetric extension of the
Polychronakos spin chain, which is the Yangian invariant system.
As it is known that the Yangian symmetry is realized in the level-1
WZNW theory~\cite{BPS94,Schou94},
the partition function can be regarded as the restricted character
formula.
We have introduced the supersymmetric Rogers--Szeg{\"o} polynomials,
and
have shown that these polynomials give the partition function for the
SP model.
The SRS polynomials satisfy two types of recursion
relations,~\eqref{FirstRecur} and~\eqref{recurH2};
the first one is useful to find out the number of degenerate
multiplets in each motif, while the second one gives us the
decomposition rule or hyper-multiplet structure of each motif by the
skew Young diagrams.
The first recursion relation has another aspect.
The spectrum~\eqref{energy_motif} shows that
the motif denote the quasi-particle excitation of the SP model.
In fact the recursion relation denotes the exclusion statistics
of the quasi-particles, and
we have computed the distribution function following
Ref.~\citen{Hikam97e}.
One possible application of our results is for the edge states of the
quantum Hall effect.
As is well known from studies on the quantum Hall effect, the theory
of the edge
state of the Laughrin state is related to the Calogero--Sutherland
model, and the quasi-particles have exclusion statistics.
Our supersymmetric theories should also have close connections with the
edge states.
Especially the $c=3/2$ su($2|1$) theory gives quasi-hole excitations
for the Pfaffian wave function~\cite{MoRe91}, and the
$c=2$ su($2|2$) theory is for the Haldane--Rezayi
states~\cite{HaldReza89}.
We hope to discuss in detail in a future issue.
It has been found earlier that, the partition function
of the vertex model, associated with the vector representation of
$U_q(\Hat{s\ell}_m)$ quantized affine algebra, exactly coincides
with that of the $su(m)$ Polychronakos model~\cite{KiriKuniNaka96a}.
So, as
a future study, it might be interesting to investigate the partition function
of the vertex model associated with the vector representation of
$U_q(\hat{s\ell}_{m|n})$ algebra and examine whether such partition
function also coincides with the polynomial
partition function~\eqref{polyH}
of $su(m|n)$ SP
model. Moreover, in analogy with the non-supersymmetric
case~\cite{KiriKuniNaka96a},
the spectral decomposition of the above mentioned vertex model
should be intimately connected with the supersymmetric skew
Young diagrams and associated Schur functions which have been
studied by us.
Such intriguing relations between the
vertex model with nearest-neighbor type interaction and Polychronakos
spin chain with long-ranged interaction may enlighten our understanding
of the common integrable structure behind these models.
Another
interesting problem might be to study the supersymmetric version of
one-row Macdonald polynomial.
As is well known,
the one-row Macdonald polynomial can be viewed as
a deformation of RS polynomial.
Thus one can obtain a realization of
one-row Macdonald polynomial by deforming the spinon representation
associated with $su(m)$ RS polynomial~\cite{Hikam96c}.
In analogy with this
non-supersymmetric case, it should be possible to find out a spinon
representation for our supersymmetric RS polynomial~\eqref{polyH}.
Furthermore,
by deforming such spinon representation in an appropriate way,
one may obtain some novel realization of supersymmetric one-row
Macdonald polynomial.
\section*{Acknowledgement}
The authors would like to thank Y.~Komori for communications at
the early stage of this work.
One of the authors (BBM) thanks to Japan Society for the Promotion of
Science for a fellowship (JSPS-P97047) which supported this work.
\newpage
|
\section*{Introduction}
Shell effects manifest themselves in terms of the existence of magic
numbers in nuclei. Inclusion of spin-orbit coupling in the shell model
is known to account for the magic numbers$^{1)}$.
Shell effects are also found to appear more subtly
e.g. as anomalous kink in the isotope shifts of Pb nuclei.
It was shown that in the relativistic mean-field (RMF) theory,
this kink appears naturally due to the inherent Dirac-Lorentz
structure of nucleons$^{2)}$. By including an
appropriate isospin dependence of the spin-orbit potential in the
density dependent Skyrme approach, the anomalous kink in the isotope
shifts in Pb nuclei could be produced$^{3)}$.
The shell gaps at the magic numbers are known to produce important effects.
Consequences due to the shell gaps are broadly termed as shell effects.
The clearest manifestation of the shell effects can be seen in
the pronounced dip in neutron and proton separation energies at
the magic numbers in figures by Wapstra and Audi$^{4)}$ on
the experimental separation energies all over the periodic table.
Recently, shell effects have become focus of much attention$^{5,6)}$.
This is due to the reason that the shell effects in nuclei near the drip
lines constitute an important ingredient to understand r-process abundances
and heavy nucleosynthesis$^{7,8)}$.
In this talk, I focus upon 2-neutron separation energies to
demonstrate that the shell efffects show a significant dependence
upon the compressibility of nuclear matter. Using empirical data
on the 2-neutron separation energies and thus implicitly the empirical
shell effects, I will show that these data can be used to constrain
the compressibility of nuclear matter. As the focus of this conference
shows, the compression modulus K is an important fundamental property
of the nuclear matter. It represents a cardinal point on the behaviour of
equation of state (EOS) of the nuclear matter.
The compressibility of the nuclear matter has received a significant attention
in the last decade and various approaches$^{9-11)}$ have been employed
to extract the compression modulus. However, an unambiguous
determination of the compression modulus of the nuclear matter has remained
difficult due to a lesser sensitivity of the giant monopole resonance
data to the compressibility$^{12,13)}$.
\section*{The Density-Dependent Skyrme Theory}
Here I employ the density-dependent Skyrme theory$^{14)}$ to examine
the shell effects. The Skyrme approach has generally been found to
be very successful in providing ground-state properties of nuclei.
The energy density functional used in the Skyrme approach is the
standard one as given in Ref.$^{15)}$. The corresponding energy
per nucleon for the symmetric nuclear matter (with the Coulomb force
switched off) is given by
\begin{equation}
(E/A)_\infty = k\rho^{2/3}(1 + \beta\rho) + {3\over8} t_0\rho
+ {1\over 16}t_3\rho^{1+\alpha}
\end{equation}
where $k = 75$ MeV.fm$^2$. The constant $\beta$ is given in terms of the
constants $t_1$ and $t_2$ and $x_2$ of the Skyrme force by
\begin{equation}
\beta = {2m\over\hbar^2}{1\over4}\Big[{1\over4}(3t_1 + 5t_2) +
t_2x_2\Big].
\end{equation}
The constant $\beta$ is also related to the effective mass $m^*$ and
the saturation density $\rho_0$ by
\begin{equation}
m/m^* = 1 +\beta\rho_0
\end{equation}
The incompressibility (or the compression modulus) of the infinite
nuclear matter is given as the curvature of the EOS curve and
can be written as
\begin{eqnarray}
K &= &9\rho_0^2{d^2(E/A)(\rho)\over d\rho^2}\Big |_{\rho_0} \\
\nonumber
&= &-2k\rho_0^{2/3} + 10k\beta\rho_0^{5/3} +
{t_3\over 16}\alpha(\alpha + 1)\rho_0^{1 + \alpha}
\label{eqn}
\end{eqnarray}
The parameters $t_0$, $t_3$ and $\alpha$ are usually obtained from the
nuclear matter properties. The other parameters $t_1$, $t_2$ and
various $x$ parameters are obtained from fits to properties
of finite nuclei. The strength W$_0$ responsible for the spin-orbit
interaction is obtained by reproducing the spin-orbit splittings in
nuclei such as $^{16}$O and $^{40}$Ca.
\section*{The Skyrme Forces}
In order to show the effect of the incompressibility of nuclear matter
on shell effects and consequently also on ground-state properties of nuclei,
I have constructed a series of zero-range Skyrme forces. Experimental data
on ground-state binding energies and charge radii of key nuclei such
as $^{16}$O, $^{40}$Ca, $^{90}$Zr, $^{116}$Sn, $^{124}$Sn and $^{208}$Pb
are taken into account in the least-square minimization. The ground-state
properties in the Skyrme theory are calculated using the Hartree-Fock method.
With a view to vary K over a large range, the correlation between
K and the saturation density is implicitly taken into account.
This correlation has been summarized for Skyrme type of
forces in Fig. 4 of Blaizot$^{9)}$. Accordingly, there exists an
inverse correlation between the saturation density (Fermi momentum)
and the compression modulus. Keeping this in mind, I have varied the
saturation density $\rho_0$ over the range $0.140 - 170 fm^{-3}$
in steps of $.005 fm^{-3}$ as shown in Table 1.
The effective mass is fixed at 0.79. This value is
required for being consistent with giant quadrupole resonance
energies of heavy nuclei$^{15)}$. Eq. (3) then provides
the value of $\beta$ which can be used in eq. (2) to connect
the coefficients $t_1$, $t_2$ and $x_2$. In order to do a systematic
variation of the nuclear compressibility, I have kept the saturation
binding energy fixed at a value of $-16.0$ MeV. This value is consistent
with most nuclear mass models$^{16)}$ and is close to physically
acceptable values.
The coefficients $\alpha$, $t_0$ and $t_3$ are determined by the
eqs. (1), (4) and the saturation condition. However, we have
allowed $t_0$ and $t_3$ to vary so that a fit to finite nuclei
provides an incompressibility that is consistent with the inverse
correlation. On the other hand, in an exhaustive computational
exercise it is found that the violation of the above correlation
results in bad fits for ground-state binding energies of key nuclei.
I have fixed $x_1$ and $x_2$ at zero for convenience. Given a value
of $\beta$, the coeffients $t_1$ and $t_2$ in eq. (2) are obtained
from the fits to the ground-state binding energies of finite nuclei.
Thus, I have obtained various Skyrme parameter sets with
incompressibility values K = 200, 220, 249, 270, 305, 327, 360 and
393 MeV, respectively. These forces encompass a broad range of physically
plausible values of K. The nuclear matter properties of these forces
are shown in Table 1. It can be seen that the force with K = 270 MeV
was obtained for the sake of interpolation. Table 2 shows the total
binding energies of some key nuclei obtained with the HF+BCS approach
using the various Skyrme forces. It can be seen that all of these forces
reproduce the total binding energies of the key nuclei from $^{16}$O to
$^{208}$Pb very well.
\section*{Two-neutron Separation Energies and the Compressibility}
I have selected the chain of Ni isotopes in order to probe the shell
effects. As most Ni isotopes are known to be spherical and the experimental
binding energies are known over a large range of Ni isotopes,
the chain of Ni isotopes serves as an ideal test bench to probe the
shell effects. Even-mass Ni isotopes from A=52 to A=70 are considered.
This includes the neutron magic number N=28 (A=56) where we intend
to investigate the shell effects due to the major shell closure.
With a view to obtain the ground-state properties of nuclei,
we have performed spherical Skyrme Hartree-Fock calculations in
coordinate space. Herein pairing is included using the BCS formalism
with constant pairing gaps. Since nuclei under focus such as $^{56}$Ni
and $^{58}$Ni are not far away from the stability line, the BCS
scheme suffices to be a suitable mechanism for the pairing.
Ground-state binding energies and $rms$ charge and neutron radii for
the Ni isotopes are calculated using the various Skyrme forces.
The binding energies of nuclei are used to obtain the 2-neutron
separation energies as
\begin{equation}
S_{2n}(Z,N) = B (Z, N) - B (Z, N-2),
\end{equation}
where B represents the total binding energy of a nucleus.
As mentioned earlier, effects due to the shell gaps affect various
nuclear properties besides the particle separation energies.
First, the calculated charge radii of Ni isotopes are shown in Fig. 1.
The charge radii show an increase with the incompressibility.
Evidently, forces with the incompressibility about 200-250 MeV
show a little dispersion in the values of the charge radii. However,
above $K \sim 300$ MeV, there is a clear increase in the charge radii as
a function of K and forces with a large incompressibility
yield large values of charge radii for any given isotope.
Inevitably, a large value of K hinders synthesis of nuclei with a
radial extension which is diminuted (compressed) contrary to
that with a lower values of K. The empirical $rms$ charge radii of
$^{58}$Ni, $^{60}$Ni, $^{62}$Ni and $^{64}$Ni are taken from the
compilation of Ref.$^{17)}$ by solid circles. The values are
3.776, 3.815, 3.846 and 3.868 fm, respectively, obtained
by folding the proton density distributions with the finite
size 0.80 fm of protons. These results derive from experiments on
muonic atoms. The $rms$ charge radii deduced from the precision
measurements in muonic atoms (taken from the recent compilation of
Ref.$^{18)}$ are 3.776, 3.813, 3.842 and 3.860 fm, respectively,
which are almost identical to the previous compilation$^{17)}$.
The empirical charge radii shown in Fig. 1 encompass the theoretical
curves between K=270-327 MeV. Whereas the charge radius for $^{58}$Ni
points towards K=270 MeV, that for the heavier Ni isotopes
is in the vicinity of K=327 MeV. Since the charge radii deduced from
various methods have been obtained with a significant precision,
these data could, in principle, be used to constrain the incompressibility.
However, as there is still some model dependence in extraction of the
charge radii, I would like to withhold any conclusions from this figure.
On the other hand, the general trend of the experimental data points
to a higher value of the compression modulus.
Figure 2 shows the corresponding $rms$ neutron radii. The neutron
radii show a monotonous increase with the mass number, with the
exception of a slight kink at the magic number N=28. The change in the
$r_n$ values with K is similar to that for the charge radii, i.e.
for lower values of K, there is a very little change in $r_n$ with K.
A significant change in the neutron radii, however, can be seen with large
K values. The empirical $rms$ neutron radii obtained from 800-MeV
polarized-proton scattering experiment$^{17,19)}$ for the
isotopes $^{58}$Ni and $^{64}$Ni are shown in the figure by the solid
points. The values lie between the curves for K=300 and K=327 MeV.
The 2-neutron separation energy $S_{2n}$ for the Ni isotopes obtained
with the various Skyrme forces are shown in Fig. 3. For the sake of
clarity of the presentation, I have selected a set of the Skyrme
forces in the figure. The dramatic fall in the $S_{2n}$ curves is seen
conspicuously for the nucleus just above the N=28 magic number.
Such a kink in the $S_{2n}$ values signifies the presence of the shell
effects which arise from closure of a major shell. All the Skyrme
forces produce such a kink in Fig. 3. However, the slope of the kink
between A=56 (N=28) and A=58 (N=30) changes from one force to the
other (as a function of the incompressibility $K$).
The difference in the $S_{2n}$ values of $^{56}$Ni and $^{58}$Ni
can be taken as a measure of the shell effects. The force
with the lowest incompressibility ($K=200$ MeV) gives this
difference as 6.59 MeV with a slope which is minimum amongst
all the curves. This implies that the shell effects due to this
force are the weakest one. As the K value increases, the corresponding
difference shows a smooth increase and the ensuing steepness increases
gradually. For the force with the highest incompressibility ($K=393$ MeV),
the difference amounts to 9.46 MeV. This difference is higher than the
experimental difference$^{20)}$ of 8.37 MeV. It implies that the shell
effects show a strong dependence on the compression modulus K.
Notwithstanding this correlation, the difference in the experimental values
of $S_{2n}$ can serve as a calibration for K. Consequently, the experimental
difference and the slope is closest to the curve for K=270 MeV.
Thus, on the basis of the empirical $S_{2n}$ values, I infer that the
incompressibility of nuclear matter lies in the neighbourhood
of $K \sim 270$ MeV.
\section*{Shell Effects at the Neutron Drip-Line}
Shell effects near the drip lines have been a matter of discussion
of late$^{5,6)}$. It has been shown earlier that within the
non-relativistic approaches of the Skyrme type, the shell effects near
the drip lines and in particular near the neutron drip line are
quenched$^{5)}$. This is contrary to that shown in the RMF theory
by Sharma et al.$^{6)}$, where the shell effects were observed to
remain strong. Before such a debate is resolved, it is important to
settle the issue of the shell effects near the stability line.
Fig. 4 shows the 2-neutron separation energies for the Ni isotopes
using the force SkP within the Hartree-Fock Bogolieubov approach$^{21)}$.
The force SkP has been found to be successful in reproducing the
ground-state properties of nuclei. The comparison of the SkP results
with the experimental $S_{2n}$ values shows that empirically there
exist strong shell effects than predicted by the Skyrme force SkP.
The compression modulus of the force
SkP is about 200 MeV. Thus, the behaviour of the force SkP about the
weak shell effects is consistent with the results from forces with
low value of the incompressibility in Fig. 3. This suggests again
that the shell effects in Ni are commensurate to a higher value of
the compression modulus of the nuclear matter as inferred above.
How the shell effects behave along the neutron drip line as a function of
the compression modulus can be visualized in Fig. 5. Here I have chosen
the chain of Zr isotopes. This chain includes nuclei on both the sides of
the magic number N=82 (A=122). The total binding energies obtained in the
Hartree-Fock calculations with the various Skyrme forces are shown. For
nuclei from A=116 (N=76) to A=122 (N=82), the total binding energy shows a
steep increase in the value with all the forces. The forces with lower
K show a stronger binding in general as compared to those with higher
K, which is as expected. For nuclei above A=122, there is a striking
difference in the way the binding energies progress with mass number.
For the force with K = 200 MeV, the binding energy of nuclei heavier
than $^{122}$Zr shows an increase in the value with the mass number,
implying that a further sequential addition of a pair of neutrons to
the N=82 core does contribute to the binding energy. As the K value
increases, the binding energy contribution from neutrons above N=82
starts diminishing. For K values 300 MeV and above, a stagnation
is observed in the binding energies. Such a behaviour implies
that the shell effects become stronger as the incompressibility
increases. This is again an indication that the shell effects are
strongly correlated to the incompressibility of the nuclear matter.
Consequently, the incompressibility inferred from the ground-state
data along the stability line suggests that the shell effects near
the neutron drip line are strong. This is consistent with the
strong shell effects about the neutron drip line as concluded in
the RMF theory$^{6)}$.
For nuclei near the drip lines, the HFB approach is considered to
be more suitable than the HF+BCS one. However, the effects of the
continuum are expected to be significant more for nuclei very
near the drip line and in particular the observables which are
most affected by coupling to the continuum are the $rms$ radii.
The total binding energies are known to show a little difference
in the HF+BCS and HFB approaches. The main character of the shell
effects is not altered by the inclusion of the continuum,
as has been pointed out in Ref.$^{8)}$ using the Skyrme force SIII.
However, it will be interesting to see how a judicious choice of a
force compatible with the experimental data as discussed above, will
affect the r-process nucleosynthesis.
\section*{Summary and Conclusions}
In the present work, I have shown that there exists a correlation between
the shell effects and the compressibility of nuclear matter. The latter
is shown to influence the shell effects significantly. Consequently,
the 2-neutron separation energies show a strong dependence on
the compression modulus. The ensuing correlation then provides a
calibration for the compression modulus of nuclear matter. It is shown
that using this correlation, the empirical data on the 2-neutron
separation energies and thus implicitly the shell effects can be used to
constrain the compressibility of nuclear matter. This procedure allows
me to conclude that the incompressibility K should be within 270-300 MeV.
This conclusion is also consistent with the experimental data on
charge radii. A natural consequence of the present study is that the
shell effects near the neutron drip line are predicted to be strong.
This is in conformity with the strong shell effects observed at
the neutron drip line in the RMF theory.
\section*{References}
\re
1) M.G. Mayer and J.H.D. Jensen, {\it Elementary Theory of
Nuclear Shell Structure} (Wiley, New York, 1955).
\re
2) M.M. Sharma, G.A. Lalazissis and P. Ring, Phys. Lett.
{\bf 317}, 9 (1993).
\re
3) M.M. Sharma, G.A. Lalazissis, J. K\"onig,
and P. Ring, Phys. Rev. Lett. {\bf 74} 3744 (1994).
\re
4) C. Borcea, G. Audi, A.H. Wapstra and P. Favaron,
Nucl. Phys. {\bf A565} 158 (1993).
\re
5) J. Dobaczewski et al., Phys. Rev. Lett. {\bf 72} 981 (1994);
ibid {\bf 73} 1869 (1994).
\re
6) M.M. Sharma, G.A. Lalazissis, W. Hillebrandt and P. Ring,
Phys. Rev. Lett. {\bf 72} 1431 (1994); ibid {\bf 73} 1870 (1994).
\re
7) K.-L. Kratz et al., Astrophys. J. {\bf 403} 216 (1993).
\re
8) B. Chen et al., Phys. Lett. {\bf B355} 37 (1995).
\re
9) J.P. Blaizot, Phys. Rep. {\bf 64} 171 (1980).
\re
10) M.M. Sharma et al., Phys. Rev. {\bf C38} 2562 (1988).
\re
11) M.V. Stoitsov, P. Ring and M.M. Sharma, Phys. Rev.
{\bf C50} 1445 (1994).
\re
12) J.M. Pearson, Phys. Lett. {\bf B271} 12 (1991).
\re
13) S. Shlomo and D.H. Youngblood, Phys. Rev. {\bf C47} 529 (1993).
\re
14) D. Vautherin and D.M. Brink, Phys. Rev. {\bf C7} 296 (1973).
\re
15) M. Brack, C. Guet and H.B. Hakansson, Phys. Rep. {\bf 123}
275 (1985).
\re
16) P. M\"oller, J.R. Nix, W.D. Myers and W.J. Swiatecki,
At. Data Nucl. Data Tables {\bf 59} 185 (1995).
\re
17) C.J. Batty, E. Friedmann, H.J. Gils and H. Rebel,
Adv. Nucl. Phys. {\bf 19} 1 (1989).
\re
18) G. Fricke et al., Atomic Data and Nuclear Data Tables
{\bf 60} 178 (1995).
\re
19) L. Ray, Phys. Rev. {\bf C19} 1855 (1979).
\re
20) A.H. Wapstra and G. Audi, Nucl. Phys. {\bf A565} 1 (1993).
\re
21) J. Dobaczewski et al., Nucl. Phys. {\bf A422} 103 (1984) and private
communication (1994).
\newpage
\leftline{\Large {\bf Figure Captions}}
\parindent = 2 true cm
\begin{description}
\item[Fig. 1.] {Charge radii for Ni isotopes with various Skyrme
forces. The experimental values for a few nuclei from
Refs.$^{17,18)}$ are shown by solid dots.}
\item[Fig. 2.] {Neutron radii for Ni isotopes with various Skyrme
forces. The experimental values for $^{58}$Ni and $^{64}$Ni
from Refs.$^{19)}$ are shown by solid dots.}
\item[Fig. 3.] {Two-neutron separation energy S$_{2n}$ for Ni isotopes
obtained with various Skyrme forces. Comparison with the experimental
data from Ref.$^{20)}$ is also made.}
\item[Fig. 4.] {Total binding energy of Zr isotopes near the neutron
drip line calculated with the Skyrme forces.}
\end{description}
\begin{table}
\begin{center}
\caption{The nuclear matter properties of the various Skyrme
interactions. The saturation energy $E/A$ has been fixed at $-$16.0 MeV
and the effective mass m* has been kept at 0.79.}
\bigskip
\begin{tabular}{c c c c c c c}
\hline
& Force && $\rho_0$ ($fm^{-3}$) & K (MeV) & $a_{sym}$ (MeV) & \\
\hline
& I && 0.140 & 393 & 38.0 & \\
& II && 0.145 & 360 & 32.4 &\\
& III && 0.150 & 327 & 32.5 &\\
& IV && 0.155 & 305 & 32.3 & \\
& V && 0.160 & 249 & 32.5 & \\
& VI && 0.165 & 220 & 32.0 &\\
& VII && 0.170 & 200 & 31.8 & \\
\hline
\end{tabular}
\end{center}
\end{table}
\begin{table}
\begin{center}
\caption{The total binding energy (MeV) of nuclei obtained from
the HF+BCS calculations with the Skyrme interactions. The empirical
values (exp.) are shown for comparison.}
\bigskip
\begin{tabular}{c c c c c c c c c c c c c}
\hline
& && && & & & K (MeV) & & & & \\
&Nucleus && exp. && 393 & 360 & 327 & 305 & 249 & 220 & 200 &\\
\hline
&~~$^{16}$O && $-$127.6 && $-$127.5 & $-$127.5 & $-$125.5 & $-$127.5 &
$-$127.5 & $-$127.6 & $-$127.7 &\\
&~~$^{40}$Ca && $-$342.0 && $-$342.5 & $-$342.4 & $-$342.2 & $-$342.0 &
$-$342.4 & $-$342.5 & $-$342.4 &\\
&~~$^{90}$Zr && $-$783.9 && $-$786.6 & $-$786.0 & $-$785.6 & $-$785.2 &
$-$785.3 & $-$784.6 & $-$784.3 &\\
&~~$^{116}$Sn && $-$988.7 && $-$989.2 & $-$988.1 & $-$987.0 & $-$986.1 &
$-$986.1 & $-$985.5 & $-$985.4 &\\
&~~$^{120}$Sn && $-$1020.5 && $-$1021.3 & $-$1020.1 & $-$1019.6 & $-$1019.3
& $-$1019.1 & $-$1018.9 & $-$1018.9 & \\
&~~$^{124}$Sn && $-$1050.0 && $-$1051.0 & $-$1049.7 & $-$1050.0 & $-$1050.5 &
$-$1049.9 & $-$1050.1 & $-$1050.3 & \\
&~~$^{208}$Pb && $-$1636.5 && $-$1636.5 & $-$1635.7 & $-$1636.0 & $-$1636.5 &
$-$1636.2 & $-$1637.2 & $-$1637.0 & \\
\hline
\end{tabular}
\end{center}
\end{table}
\end{document}
|
\section{Introduction}
Computations of pion production in nucleon-nucleon
($NN$) collisions near threshold, allow a confrontation of
our understanding of $NN$ interactions with data
in a kinematic region for which
chiral symmetry, and therefore quantum chromodynamics
QCD, could be very important.
The reaction
$pp \rightarrow pp\pi^0$ near threshold has attracted much
attention in the years since the first IUCF data appeared
\cite{M90} and has exposed serious disagreements with earlier
theoretical calculations \cite{KR66,SSY69,MS91,N92}.
The existence of many conflicting models claiming to explain
this discrepancy \cite{LR92,HGM94,HO95,Han+95}
calls for a principle to organize the several
potentially significant mechanisms of pion production.
Chiral Perturbation Theory ($\chi$PT) has been applied to
mesonic \cite{W79,GL84}, one-baryon \cite{JM91,BKM95}, and
nuclear \cite{W90,OK92,ORV94,V94,vKolckn,KSW,SKS98,monster}
processes where typical momenta of the order of the pion mass, $m_\pi$,
allow a systematic expansion of observables in powers of $m_\pi/M_{QCD}$,
where $M_{QCD}\sim 1$ GeV.
Cohen {\it et al} \cite{bira} have adapted the power counting
and applied $\chi$PT to
near-threshold pion production, where momenta are of order $\sqrt{m_\pi m_N}$,
$m_N$ being the nucleon mass.
They estimated leading and next-to-leading contributions,
the latter including important short-range contributions,
related to the isoscalar components of the
potential (and possibly described by
$\sigma$ and $\omega$ meson exchanges) \cite{LR92}.
Subsequently, van Kolck {\it et al} \cite{KMR96} showed that
next-to-next-to-leading contributions ({\it e.g.}, from $\rho-\omega$
meson exchange) are also relevant;
data could then be explained within the very large theoretical uncertainties
associated with $S$-wave pion rescattering and the short-range structure of
the nuclear force.
Other $\chi$PT-inspired calculations
have also stressed the importance of understanding
rescattering \cite{SC95}
and the effect of loops \cite{SC95,Han+98,moalem,cagadadoulf,newsc}.
Since these $\chi$PT-inspired calculations include
a larger number of contributions than other, model calculations,
one concludes that a large theoretical uncertainty
plagues all calculations performed to date.
Given such uncertainties, it is natural to examine other channels using
the same techniques.
Here we are going to discuss $NN\rightarrow d\pi$ and $\rightarrow pn\pi$,
which traditionally have been considered better understood
than $pp \rightarrow pp\pi^0$.
We consider energies near
threshold where the calculation simplifies,
because the pion emerges mostly in an $S$ wave.
We will show that an understanding of these channels
is still in the future.
We adopt here the conventional nuclear approach
of grouping all $NN$ interactions generated by mesons of
small momentum in a
potential,
while the contributions associated with energies comparable
to $m_\pi$ are accounted for in a kernel to be evaluated
between wavefunctions generated by the potential.
Splitting the problem this way, one should still strive to
calculate wavefunctions and kernels from the same theory or model,
otherwise ambiguities arise from off-shell extrapolations
(or equivalently, from nucleon-field redefinitions).
$\chi$PT is the only known tool for performing this task,
and at the same time is consistent with QCD, because
its symmetries are treated correctly.
$\chi$PT is also unique in that it offers the possibility
of doing systematic calculations: an expansion in momenta
provides a power counting to organize the calculation even
though coupling constants are not small.
$\chi$PT separates interactions in long-range effects
calculated explicitly with pion exchange
and short-range effects accounted for by contact
interactions with an increasing number of derivatives.
Parameters not constrained by chiral symmetry depend
on details of QCD dynamics, and are at the present
unknown functions of QCD parameters. Predictive power
is not lost, however, because at any given order
in the power counting only
a finite number of unknown parameters appear; after they
are fitted to a finite set of data, all else can be predicted
at that order. These predictions are called
``low-energy theorems''.
Since the Lagrangian of $\chi$PT is the most general
one consistent with QCD symmetries, $\chi$PT is a generalization
of current algebra.
Unfortunately, the only $NN$ potential derived
in $\chi$PT \cite{OK92} and fitted to
low-energy phase shifts \cite{ORV94} produces poor results
near the pion production threshold.
Attempts to remedy the situation are in progress \cite{SKS98,monster}.
For the time being we will rely on modern, ``realistic''
phenomenological potentials which fit $NN$ data very well.
By considering more than one of those, we can estimate
the otherwise uncontrolled error stemming
from our inconsistent use of potential and kernel.
We are going to see that, in contrast to neutral
pion production, the error is small in the channel considered here.
Because such realistic potentials reproduce low-energy phase shifts
with identical long-range tails, they must contain the equivalent
to the leading order in the chiral expansion.
Although our approach should be considered phenomenological,
our leading-order result will be an approximation to a low-energy
theorem.
At sub-leading orders in the expansion, an apparent ambiguity arises,
concerning the correct treatment of the energy transferred
in pion rescattering. This
can be seen in the conflicting results of estimates of the same
kernel
\cite{bira,SC95}. This issue is under study \cite{usandthem};
here we will limit ourselves to the most natural prescription
that the transferred energy is $\simeq m_\pi/2$.
We will concentrate most of our efforts on the kernel.
We will discuss a
reasonable power counting for pion production that
generalizes for $\pi d$ and $\pi pn$ channels the discussion of
Ref. \cite{bira}. In leading order, the new ingredient here is
pion rescattering using the Weinberg-Tomozawa (WT)
term that dominates isospin-dependent $\pi N$ scattering.
This term does not contribute to the reaction
$pp\rightarrow pp\pi^0$.
In the latter, the leading non-vanishing order consists
only of an impulse term (IA), in which a single pion
is emitted from a nucleon, and a similar contribution
from the delta ($\Delta$).
This leading order underpredicts the
experimental data by a factor of
approximately 5, due to
two cancellations not incorporated in our power counting:
{\it (i)} among different regions in coordinate space for each term
evaluated between initial and final wavefunctions;
and {\it (ii)} between the total impulse and delta contributions.
Oversight {\it (i)} results from our present inability
to treat the potential and the kernel on the same footing.
Oversight {\it (ii)} is somewhat accidental, but actually expected
from the fact that, in energy, the pion threshold sits
midway between the elastic threshold and the delta pole.
The sensitivity of the delta contribution to the realistic
potential used is the main source of dependence
on the $NN$ potential of the final result.
As a consequence of the accidentally small leading order,
effects which are usually negligible acquire
prominence in neutral pion production in the $pp$ reaction.
One effect is isospin-independent
pion rescattering.
$\chi$PT is critically necessary to assess the size of this contribution.
First, in principle $\chi$PT allows one to determine from
$\pi N$ data not only the momentum- and energy-independent
$2\pi N^\dagger N$ vertex, but also terms which are quadratic in
energy and momentum. This is important in view
of the fact that in pion production
the virtual pion has energy of order $m_\pi/2$
and the combination of parameters
of relevance is thus different from the combination that
appears at the $\pi N$ scattering threshold.
There is by now a number of consistent
estimates of the relevant parameters to third order
\cite{BKM95,ulfandall}. An estimate
of the uncertainty of this contribution to pion production
can be made by using also a lower-order determination \cite{sub}.
Second, $\chi$PT is the only way to account for rescattering as a component
of a Feynman diagram without destroying chiral symmetry.
There have been attempts to estimate rescattering by
simply connecting a nucleon line to
a $\pi N$ amplitude with an arbitrary off-shell
extension \cite{HGM94,HO95}. By considering field redefinitions
in the most general chiral Lagrangian, it is easy to show
that the off-shell ambiguity in the pion leg is
equivalent to a set of short-range $\pi (N^\dagger N)^2$ interactions,
and that an inconsistent treatment of both
effects leads to violation of chiral symmetry in a way
that is contradictory with QCD \cite{friar}.
Using $\chi$PT, it was found \cite{bira,SC95} that rescattering interferes
destructively with the leading-order effects in
$pp\rightarrow pp\pi^0$. This further
interference makes agreement with data more difficult.
Although the magnitude of the effect
is still being assessed \cite{usandthem}, it is certain
that the uncertainty stemming from different determinations
of the $\chi$PT parameters is large \cite{KMR96}.
This disagreement between theoretical evaluations and cross-section
data for $pp\rightarrow pp\pi^0$ can be largely removed
if $\sigma$, $\rho$ and other
heavy mesons are included \cite{KMR96}.
When first suggested \cite{LR92},
it looked as if $pp\rightarrow pp\pi^0$ was a clear signal
of these otherwise elusive mechanisms.
Among the first corrections in our power counting,
one finds two-pion-exchange
loop graphs and $\pi (N^\dagger N)^2$ counterterms (that behave
properly under chiral symmetry).
A full $\chi$PT calculation requires the calculation of these
loops, and although some steps have been taken in this
direction \cite{moalem,newsc},
it is a herculean task that remains to be completed.
Even then, it will still require that the counterterms
be fitted to some pion production data (say, right at threshold)
so that other data (say, the energy dependence close to threshold,
or other channels)
be predicted.
In the case of $pp\rightarrow pp\pi^0$, even
the most sophisticated phenomenological models
have to recourse to
such a fit of a short-range counterterm
\cite{hanhart}.
In any case, after all this one would then be interested
in determining whether such counterterms are of natural size,
and whether they can be further understood as the result of
heavy-meson exchange.
In view of all other uncertainties, the authors of Ref. \cite{bira}
took the point of view that
an estimate of this class of sub-leading contributions
could more easily be made by modeling them
with meson exchange in Z-graphs, following Ref. \cite{LR92}.
It turns out that the counterterms so produced are of natural
size, which lends them credence, but that they are not
sufficient to achieve agreement with data.
In order to study convergence, further meson exchanges
(chiefly $\rho-\omega$) that contribute to higher-order
counterterms were considered, and shown to be smaller but
still relevant \cite{KMR96}.
The conclusion of Ref.\cite{KMR96}
was that it is possible to describe the $pp\rightarrow pp\pi^0$ reaction
consistently with QCD, reasonable meson exchanges, and realistic
potentials, but only within a very large theoretical uncertainty.
It is our intention to assess here our theoretical understanding
of the
$N N \rightarrow d \pi, \rightarrow pn \pi$ reactions by
analyzing the effect of the same microscopic mechanisms.
These reactions have been studied for some time.
Conservation of parity, angular momentum, and isospin
constrains the possible channels for these reactions.
In the case of unbound final nucleons, sufficiently close
to threshold the strong $S$-wave two-nucleon interaction
implies that the most important channels should
be those in which the final nucleons
have relative orbital angular momentum $L_{NN}=0$.
Some of these partial waves are listed in Table~\ref{Tab.01}.
\begin{table}
\caption{Partial-wave amplitudes for the $NN\rightarrow NN\pi$
reactions
that can contribute to the isotropic cross section right at
threshold with an $L_{NN}=0$ final state.
$L_{\pi(NN)}$ denotes the pion angular momentum.}
\begin{tabular}{lccc}
Element & $NN$ Initial state & $NN$ final state & $L_{\pi(NN)}$ \\ \tableline
$a_0$ & $^1S_0$ & $^3S_1$ & 1 \\
$a_1$ & $^3P_1$ & $^3S_1$ & 0 \\
$a_2$ & $^1D_2$ & $^3S_1$ & 1 \\
$b_0$ & $^3P_0$ & $^1S_0$ & 0
\end{tabular}
\label{Tab.01}
\end{table}
Near threshold we expect the pion, too, to be in an $S$-wave,
and the cross sections to be nearly isotropic, with \cite{MR55}
\begin{equation}
\frac{d\sigma}{d\Omega} (pp\rightarrow pn \pi^+)=
\frac{1}{4}\left(|a_1|^2+\frac{1}{3}|b_0|^2\right)
+\ldots,
\label{difcrosspn}
\end{equation}
\begin{equation}
\frac{d\sigma}{d\Omega} (pp\rightarrow d \pi^+)=
\frac{1}{4} |a_1'|^2+ \ldots,
\label{difcrossd}
\end{equation}
\begin{equation}
\frac{d\sigma}{d\Omega} (pp\rightarrow pp \pi^0)=
\frac{1}{12} |b_0|^2 +\ldots
\label{difcrosspp}
\end{equation}
Here $a_1$ and $a_1'$ have quite different magnitudes due to
the different wavefunctions; in the case of the deuteron final state
the effect of the $D$ wave is also important, and it is incorporated
in the following.
The cross section for $np\rightarrow d \pi^0$ is related to
that of $pp\rightarrow pn \pi^+$ by isospin considerations.
There is a relative factor
2 from $pp$ being pure
isospin $I=1$ and $np$ having $I=1$ and $I=0$ in equal
probabilities; and there are Coulomb effects.
The total cross-section for the deuteron channels is function of
$\eta=q/m_\pi$ where $q$ is the maximum pion momentum.
It is well-known \cite{GW54,R54} that sufficiently close to threshold
the cross-section for $n p \rightarrow d \pi^0 $ is
\begin{equation}
\sigma_{n p}= \frac{\alpha}{2}\eta + {\cal O}(\eta^3),
\label{snp}
\end{equation}
\noindent
while for $p p \rightarrow d \pi^+ $ it is
\begin{equation}
\sigma_{p p}= \alpha C_0^2 \eta + {\cal O}(\eta^3).
\label{spp}
\end{equation}
\noindent
Here $\alpha$ is a constant which depends on the strong
interactions that produce the pion
and $C_0(\eta)$ is
a Coulomb factor arising mostly from the pion-deuteron electromagnetic
interaction.
If the experimental data for
$p p \rightarrow d \pi^+ $ are corrected for
these electromagnetic effects, then both reactions
near threshold can be viewed
as a determination of the single constant $\alpha$.
The threshold behavior of the reaction $pp\rightarrow pn\pi^+$
was estimated by Gell-Mann and Watson \cite{GW54}, who predicted an energy
dependence $\sigma\propto\eta^{4}$ under the approximation
of zero-range nuclear forces.
Early theoretical analyses of the reaction
$p p \rightarrow d \pi^+$, like in ours,
split it into a kernel ---where
the two-nucleon system emits the pion--- and
effects of initial- and final-state
interactions ---taken into account by using
deuteron and scattering wave
functions that satisfy the Schr\"odinger
equation with a given $NN$ potential.
The numerical results of
the early analyses \cite{KR66} were that
this reaction is dominated by the Weinberg-Tomozawa (WT) term
(Fig.~\ref{Fig.1}b).
The main competitor process
was thought to be the impulse term (IA)
(Fig.~\ref{Fig.1}a). The early
estimate of this IA gave a small
contribution \cite{KR66} due to a
cancellation in the matrix elements between
the $S$ and $D$ waves of the deuteron final
state. It was believed that the WT
term alone could account for
the essential features of the data.
\begin{figure}
\epsfxsize=13.0cm
\epsfysize=4.0cm
\centerline{\epsffile{fig01.ps}}
\vspace{.5cm}
\caption{ Various contributions to the $p p \rightarrow
d \pi^+, pn\pi^+$ reactions. A single (double)
solid line stands for a nucleon (delta) and a single
(double) dashed line represents a pion (sigma, omega,
rho); $\Psi_i$ ($\Psi_f$) is the wave function for the
initial (final) state.
\label{Fig.1}}
\end{figure}
With the advent of accelerators capable of
producing intense high-quality beams of
protons with GeV energies and
very precise detecting systems, good and
accurate data for $np\rightarrow d \pi^0$ \cite{triumf},
$p p \rightarrow d \pi^+$ \cite{cosy,iucf},
and $p p \rightarrow pn \pi^+$ \cite{steve1,steve2}
near threshold became available.
For the deuteron channels, the new data confirm the threshold behavior
(\ref{snp}) \cite{triumf}, and
show that
the theoretical calculations started by Koltun and Reitan were
in the right direction.
More recent calculations of the WT and IA mechanisms
\cite{niskanen,Han+98},
however, indicated that their strength might not be sufficient
to explain the new data.
As for the unbound final state, the
recent measurements in the Indiana Cooler \cite{steve2} give
\begin{equation}
|a_1|^2\propto \eta^{3.2}.
\label{apot}
\end{equation}
A calculation \cite{unpharry}
using IA, on-shell pion rescattering,
and $\sigma,\omega$ Z-graphs finds good agreement with data in all
$pp\pi^0$, $pn\pi^+$, and $d\pi^+$ channels with a soft $\pi N$
form factor.
Our motivation here is to examine what a chiral power counting
suggests for the pion production in the $pp\rightarrow d\pi^+$
and $pp\rightarrow pn\pi^+$ reactions near threshold.
For the $d \pi^+$ reaction, we will compute $\sigma_{pp}$
but will disregard the Coulomb interaction in both the initial
state ---which is not so important due to the not-so-small initial
energy--- and final state. This means we do not calculate $C_0(\eta)$,
and can only compare with Coulomb-corrected $pp$ data.
We will see that $\sigma_{pp}$ is indeed approximately linear with $\eta$
at threshold, so our result can be considered a calculation of $\alpha$ in
Eq.~(\ref{spp}). Alternatively, it is a calculation of $2 \sigma_{np}$.
For the $pn\pi^+$ final state, we will also disregard the Coulomb
interaction and see that our cross section has an energy dependence
similar to the data.
Moreover, we
will show that the WT term is indeed the dominant one, not only for the
$^3S_1$ $pn\pi^+$ final state but also for the deuteron channel,
which is consistent with the early analyses for
this final state.
The IA term is smaller than WT because of cancellations
between different regions in coordinate space, and
between the $S$ and $D$ waves of the deuteron final
state. The same cancellations affect the contribution from
an explicit $\Delta$ in the intermediate state
(Fig.~\ref{Fig.1}d), more so in the $d \pi^+$ channel than in the
$pn \pi^+ $ channel. The relative smallness of this term here
reduces the effects of using different $NN$ potentials
compared to the $pp\rightarrow pp\pi^0$ reaction.
However the delta contribution is affected by a number of
uncertainties. We include also the same sub-leading terms that proved
important in the $pp\rightarrow pp\pi^0$ reaction: isospin-independent
rescattering (ST, Fig.~\ref{Fig.1}b),
and heavy meson exchange simulating short-range
mechanisms ($\sigma,\omega$ and $\rho-\omega$, Fig.~\ref{Fig.1}c,e).
The sub-leading Galilean correction to the WT term (GC)
is included as well. These contributions are all relatively small.
We will see that, if we neglect the uncertain
$\Delta$ contributions,
each of the the cross-sections is underestimated by a factor of $\sim 2$.
In Section II we discuss the power counting and the chiral Lagrangian.
In Section III the kernel is obtained.
In Section IV the calculation of the cross-sections is outlined;
some technical details are relegated to an Appendix.
Section V describes our input and discusses our results.
An outlook is presented in Section VI.
\section{Implementing $\chi$PT}
Near threshold for pion production the total energy of the two colliding
nucleons is of order
$2 m_N + m_\pi$,
so that the center-of-mass initial kinetic energy of each nucleon is
$m_\pi/2$. This energy is smaller than the nucleon mass,
so we can use a non-relativistic framework.
The nonrelativistic kinetic
energy formula holds; the mass $2m_N$ plays no dynamical
role,
and the typical momentum of real and virtual
particles involved in this process is
$p_{\rm typ} \sim \sqrt{m_N m_\pi}$.
This requires
some adaptation of the usual effective theory ideas
which have been developed for momenta typical
of most nuclear systems, $Q \sim m_{\pi}$.
Because the mass difference between the delta isobar and the nucleon,
$\delta= m_\Delta - m_N$, is numerically of order of the typical
excitation energy we are interested in, $m_\pi$,
the $\Delta$ must be included explicitly as a
degree of freedom in the Lagrangian.
On the other hand, $\sqrt{m_N m_\pi}$
is smaller than the characteristic mass scale of QCD,
$M_{QCD}\sim 1$GeV, at least in the chiral limit $m_\pi\rightarrow 0$,
so that the contribution of other states (the Roper, the $\rho$ meson,
{\it etc.}) can be buried in short-range interactions.
We thus seek a theory of non-relativistic nucleons and deltas
interacting with pions that is consistent with the symmetries of QCD.
$\chi$PT is implemented via the most general
Lagrangian involving these degrees of freedom aided by power-counting
arguments. The seminal idea was contained
in a paper by Weinberg \cite{W79}. This
idea was developed systematically for
interactions of mesons \cite{GL84} and
for interactions of mesons with a baryon
\cite{JM91,BKM95}. The generalization of
these techniques to describe properties
of more than one baryon was also due to
Weinberg \cite{W90} and was carried out
in detail in Ref. \cite{OK92,ORV94,V94};
see also Refs. \cite{vKolckn,KSW,monster}.
When dealing with typical momenta of order
$Q \sim m_{\pi}$, the usual power counting
suggests that we order terms in the chiral
Lagrangian according to the ``index''
$\Delta=d+\frac{f}{2}-2$,
where $d$ is the sum of the number of derivatives,
the number of powers of $m_{\pi}$,
and the number of powers of $\delta$;
and $f$ is the number of fermion field operators.
In the following we will consider interactions
with $\Delta$ up to 4.
The Lagrangian with $\Delta= 0$
for each interaction \cite{W79,GL84,JM91,BKM95,W90,ORV94,V94} is
\newcommand{\mbox{\boldmath $\pi$}}{\mbox{\boldmath $\pi$}}
\newcommand{\mbox{\boldmath $\tau$}}{\mbox{\boldmath $\tau$}}
\newcommand{\mbox{\boldmath $T$}}{\mbox{\boldmath $T$}}
\begin{eqnarray}
{\cal L}^{(0)} & = &
\frac{1}{2}(\dot{\mbox{\boldmath $\pi$}}^{2}-(\vec{\nabla}\mbox{\boldmath $\pi$})^{2})
-\frac{1}{2}m_{\pi}^{2}\mbox{\boldmath $\pi$}^{2} \nonumber \\
& & +N^{\dagger}[i\partial_{0}-\frac{1}{4 f_{\pi}^{2}} \mbox{\boldmath $\tau$} \cdot
(\mbox{\boldmath $\pi$}\times\dot{\mbox{\boldmath $\pi$}})]N +\frac{g_{A}}{2 f_{\pi}}
N^{\dagger}(\mbox{\boldmath $\tau$}\cdot\vec{\sigma}\cdot\vec{\nabla}\mbox{\boldmath $\pi$})N
\nonumber \\
& & +\Delta^{\dagger}[i\partial_{0}- \delta]\Delta
+\frac{h_{A}}{2 f_{\pi}}[N^{\dagger}(\mbox{\boldmath $T$}\cdot
\vec{S}\cdot\vec{\nabla}\mbox{\boldmath $\pi$})\Delta +h.c.] +\cdots \, ,
\label{la0}
\end{eqnarray}
\noindent
where $f_{\pi}=93$ MeV is the pion decay constant,
$\delta = m_{\Delta}-m_{N}$ is the isobar-nucleon mass difference,
$g_{A}$ is the axial-vector coupling of the nucleon, $h_{A}$
is the~$\Delta N \pi$ coupling, and $\vec{S}$ and $\mbox{\boldmath $T$}$ are the
transition spin and isospin matrices, normalized such that
\begin{eqnarray}
S_{i}S^{+}_{j} & = & \frac{1}{3} (2\delta_{ij} -
i\varepsilon_{ijk} \sigma_{k}) \label{S1} \\
T_{a}T^{+}_{b} & = & \frac{1}{3} (2\delta_{ab}
- i\varepsilon_{abc} \tau_{c}). \label{S2}
\end{eqnarray}
\noindent
Notice that we defined the fields $N$ and $\Delta$ in
such a way that there is no factor of $\exp (-i m_N t)$ in their
time evolution. Hence $m_N$ does not appear explicitly at this index,
corresponding to static baryons.
We also wrote ${\cal L}^{(0)}$ in the rest frame
of the baryons, which is the natural choice. (Galilean invariance
will be ensured by including terms with additional derivatives.)
Chiral symmetry determines the coefficient of the so-called
Weinberg-Tomozawa term
($N^{\dagger}\mbox{\boldmath $\tau$} \cdot(\mbox{\boldmath $\pi$}\times\dot{\mbox{\boldmath $\pi$}})N$)
but not of the single-pion interactions ($g_A, h_A$).
The Lagrangian with $\Delta=1$ is
\cite{JM91,BKM95,ORV94,V94},
\begin{eqnarray}
{\cal L}&^{(1)}&
=\frac{1}{2m_{N}}[N^{\dagger}\vec{\nabla}^{2}N
+\frac{1}{4 f_{\pi}^{2}}(iN^{\dagger}\mbox{\boldmath $\tau$}\cdot
(\mbox{\boldmath $\pi$}\times\vec{\nabla}\mbox{\boldmath $\pi$})\cdot\vec{\nabla}N + h.c.)]
\nonumber \\
& & +\frac{1}{f_{\pi}^{2}}N^{\dagger}[(c_2 +c_3 - \frac{g_A ^2}{8 m_{N}})
\dot{\mbox{\boldmath $\pi$}}^{2} -c_3 (\vec{\nabla}\mbox{\boldmath $\pi$})^{2}
-2c_1 m_{\pi}^{2} \mbox{\boldmath $\pi$}^{2} -
\frac{1}{2} (c_4 + \frac{1}{4m_{N}})
\varepsilon_{ijk} \varepsilon_{abc} \sigma_{k} \tau_{c}
\partial_{i}\pi_{a}\partial_{j}\pi_{b}]N \nonumber \\
& & +\frac{\delta m_{N}}{2} N^{\dagger}[\tau_{3}-\frac{1}{2 f_{\pi}^{2}}
\pi_3 \mbox{\boldmath $\pi$}\cdot\mbox{\boldmath $\tau$}]N
+\frac{1}{2m_{N}}\Delta^{\dagger}[\vec{\nabla}^{2} +\cdots]\Delta
\nonumber \\
& & -\frac{g_{A}}{4 m_{N} f_{\pi}}[iN^{\dagger}\mbox{\boldmath $\tau$}\cdot\dot{\mbox{\boldmath $\pi$}}
\vec{\sigma}\cdot\vec{\nabla}N + h.c.]
-\frac{h_{A}}{
2 m_{N} f_{\pi}}[
iN^{\dagger}\mbox{\boldmath $T$}\cdot\dot{\mbox{\boldmath $\pi$}}\vec{S}\cdot\vec{\nabla}
\Delta + h.c.] \nonumber \\
& & -\frac{d_1}{f_{\pi}}
N^{\dagger}(\mbox{\boldmath $\tau$}\cdot\vec{\sigma}\cdot\vec{\nabla}\mbox{\boldmath $\pi$})N\,
N^{\dagger}N
-\frac{d_2}{2 f_{\pi}} \varepsilon_{ijk} \varepsilon_{abc}
\partial_{i}\pi_{a}
N^{\dagger}\sigma_{j}\tau_{b}N\, N^{\dagger}\sigma_{k}\tau_{c}N
+\cdots \, , \label{la1}
\end{eqnarray}
\noindent
where the $c_{i}$'s are coefficients of ${\cal O}(1/M)$,
$\delta m_{N} \sim m_d -m_u$ is the quark mass difference contribution to
the neutron-proton mass difference,
and the $d_{i}$'s are coefficients of ${\cal O}(1/f_\pi^2 M)$.
These seven numbers are not fixed by chiral symmetry, but it is important
to point out that Galilean invariance requires that the other coefficients
explicitly shown above be related to those appearing in ${\cal L}^{(0)}$.
This in particular
fixes the strength of
the single-pion interactions in terms of the
lowest-order coefficients $g_A$ and $h_A$, and of the common mass $m_{N}$.
The Lagrangian with $\Delta=2$ is
\begin{eqnarray}
{\cal L}^{(2)} & = & \frac{d_1^{\prime}+e_1}{2 m_N f_{\pi}}
[iN^{\dagger}\mbox{\boldmath $\tau$}\cdot\dot{\mbox{\boldmath $\pi$}}
\vec{\sigma}\cdot\vec{\nabla}N\, N^{\dagger}N + h.c.]
\nonumber \\
& & -\frac{e_1}{2 m_N f_{\pi}}
[iN^{\dagger}\mbox{\boldmath $\tau$}\cdot\dot{\mbox{\boldmath $\pi$}}\vec{\sigma}N\,
\cdot N^{\dagger}\vec{\nabla}N + h.c.] \nonumber \\
& & +\frac{e_2}{2 m_N f_{\pi}}
[N^{\dagger}\mbox{\boldmath $\tau$}\cdot\dot{\mbox{\boldmath $\pi$}}\vec{\sigma}\times
\vec{\nabla}N\,
\cdot N^{\dagger}\vec{\sigma}N + h.c.]
+\cdots \, , \label{la2}
\end{eqnarray}
\noindent
where the $e_i$'s are other coefficients of ${\cal O}(1/f_\pi^2 M)$.
Among the Lagrangians with higher indices, we find
\begin{equation}
{\cal L}^{(4)} =
\frac{g}{2 m_N f_{\pi}}
[iN^{\dagger}\mbox{\boldmath $\tau$}\cdot\dot{\mbox{\boldmath $\pi$}}
\vec{\sigma}\cdot\vec{\nabla}\vec{\nabla}N \cdot N^{\dagger}\vec{\nabla}N
+ h.c.]
+\cdots \, , \label{la3}
\end{equation}
\noindent
where $g$ is
a coefficient of
${\cal O}(1/f_\pi^2 M^3)$.
The two nucleons in the $N N \rightarrow NN \pi$ reaction
can interact repeatedly by the exchange of mesons of momenta
$Q\sim m_\pi$ before and after the emission of the pion.
We account for this through the iteration of a potential,
which produces initial and final wavefunctions that
differ from the free ones.
The emission of the pion, on the other hand,
involves the larger momentum $p_{\rm typ} \sim \sqrt{m_N m_\pi}$.
The sub-diagrams that involve such typical momentum
form the kernel of ``irreducible diagrams'',
which is evaluated between wavefunctions.
The unusually high momentum in the kernel
requires modification in the usual power counting.
While in the usual power counting energy and momenta are counted
as equal, here energies are $\sim m_\pi$ but momenta
$\sim \sqrt{m_N m_\pi}$. This changes the usual correspondence
between index and order.
The leading nucleon propagator, for example, includes both the
static term in (\ref{la0}) and the kinetic term in (\ref{la1})
at the same order in power counting for pion production,
contrary to the situation in
$\chi$PT applied to one-baryon systems. The nucleon
propagator then is $\sim 1/m_\pi$.
Relativistic corrections $p^4/8m_N^3 + \ldots$ are relatively smaller by
$\sim m_\pi/m_N$ and can be considered higher-order insertions.
Note that this is completely consistent with
our decomposition of the full amplitude into a kernel
and wavefunctions obtained from a Schr\"odinger equation,
and,
contrary to what is stated in Ref. \cite{cagadadoulf},
does {\it not} imply the need of a relativistic framework.
The delta propagator differs from the nucleon
by the presence of the mass difference $\delta\sim 2m_\pi$.
As pointed out first in Ref. \cite{bira},
the delta propagator is then actually $\sim -1/m_\pi$ and tends to
interfere destructively with the nucleon contributions to the kernel.
The pion propagator, on the other hand,
is $\sim 1/m_\pi m_N$.
In interactions, each time derivative is associated with a factor
of $m_\pi$, while a space derivative a factor of $\sqrt{m_N m_\pi}$.
Finally, a loop brings a $(m_N m_\pi)^{3/2}m_\pi/(4\pi)^2$.
A detailed analyses of various contributions
can found in Ref. \cite{bira}.
The new elements here are those associated with
the
isospin-dependent WT
pion rescattering.
The order of the WT
contribution is evaluated as follows. The
term proportional to $\partial_0\bbox{\pi}$
of Eq.~(\ref{la0}) yields an
explicit factor of $m_\pi/f_\pi^2$; the
pion-nucleon interaction provides a factor
of $\sqrt{m_\pi\,m_N}/f_\pi$, and the
pion propagator $(m_\pi\,m_N)^{-1}$. The
total net result is
of order ${\cal O}(f_{\pi}^{-3} \sqrt{m_\pi /M})$,
which is the same order of the impulse approximation
and the delta contribution \cite{bira}.
Therefore due to power counting arguments,
we expect that contribution of WT, IA, and $\Delta$ terms to have the same
importance; they constitute our leading order.
We will also include
other terms which are of order
${m_\pi /M}$ or higher relative to the leading one:
the isospin-independent pion rescattering
(SG),
the Galilean correction to the WT
term (GC), and contact terms, modeled by
the heavy meson exchange ($\sigma$, $\omega$ and $\rho-\omega$).
\section{The Kernel}
We now obtain the explicit forms of the various
contributions by evaluating the most important
irreducible diagrams in momentum space.
Our notation is as follows:
$\omega_q^2 = \vec{q}^{\: 2} + m_{\pi}^2$
is the energy of the (on-shell) pion produced
with momentum $\vec{q}$ in the center of mass;
$\vec{p}$ ($\vec{p'}$)
is the center-of-mass momentum of the
incoming (outgoing) nucleon labeled
``1'' (those of nucleon ``2'' are opposite);
$\vec{k}= \vec{p} -\vec{p'}$
($k^0= m_\pi/2$)
is the momentum (energy) transferred;
$\omega_k^2 = \vec{k}^2 + m_{\pi}^2$;
$\vec{P}= \vec{p} +\vec{p'}$;
$\vec{\sigma}^{(i)}$
is the spin of proton $i$;
$\vec{\Sigma}_a= \vec{\sigma}^{(1)}\tau_a^{(1)} -
\vec{\sigma}^{(2)}\tau_a^{(2)}$
where $a$ is the isospin of emitted pion;
and $T(\vec{k})\equiv \vec{\sigma}^{(1)}\cdot\vec{k} \;
\vec{\sigma}^{(2)}\cdot\vec{k}$.
We define the $T$-matrix in terms of the
$S$-matrix via $S=1+iT$.
According to the previous discussion, we
expect the leading contributions
to arise {}from the diagrams in
Figs.~\ref{Fig.1}a,b, and d.
In the case of the Weinberg-Tomozawa
diagram (Fig.~\ref{Fig.1}b), we get
\begin{equation}
T^{WT}_a=-\frac{g_A}{4f_\pi^3}\,\epsilon_{abc}\tau_b^{(1)}\tau_c^{(2)}\,
\frac{\omega_q+k^0}{w_k^2-(k^0)^2}\;\vec{S}\cdot\vec{k}\; .
\label{WT}
\end{equation}
\noindent
The Galilean correction to WT contribution is smaller by a factor of
$m_\pi/M$ and is given by
\begin{equation}
T^{GC}_a=-\frac{g_A}{8m_Nf_\pi^3}\,\epsilon_{abc}\tau_b^{(1)}\tau_c^{(2)}\,
\frac{1}{w_k^2-(k^0)^2}\;\vec{k}\cdot\vec{P}\;\vec{S}\cdot\vec{k}\; .
\label{GC}
\end{equation}
The impulse term (Fig.~\ref{Fig.1}a)
is discussed in detail in \cite{bira}.
The principle of irreducibility says
that we have to redraw this
diagram. Since the outgoing pion
carries an energy of the order of
the pion mass, the energies of the
$NN$ intermediate state before and
after pion emission differ by
$\sim m_\pi$. Therefore both of the
intermediate states cannot simultaneously be
within $\sim m_\pi^2/m_N$ of being on-shell:
at least one intermediate state, before
or after emission, is off shell by $\sim m_\pi$.
This single, relatively-high-momentum
($\sim \sqrt{m_{\pi} m_N}$) pion exchange
must therefore be included in the
irreducible class of operators for
our process (unlike the usual case).
All other initial- and final-state
interactions will be considered
reducible and included in the wave
functions.
Therefore, in the case of pion exchange with a
nucleon in the intermediate state
we get
\begin{equation}
T^{IA}_a=\frac{ig_A^3}{8 m_{N} f_{\pi}^3}\; \frac{1}{\omega_k^2-(k_0)^2}
\left[\vec{\Sigma}_a\cdot\vec{p'}\; T(\vec{k})-
T(\vec{k})\, \vec{\Sigma}_a\cdot\vec{p}\;\right]\, ,
\label{IA}
\end{equation}
\noindent which is listed for comparative
purposes only. We will actually calculate
the impulse approximation directly
from Eq.(6) in Ref.~\cite{bira}, in the same
fashion as done by
Koltun and Reitan \cite{KR66}.
Recoil corrections are expected to be smaller
by a factor of $m_{\pi}/M$. Since the impulse contribution will prove
to be small, we can ignore the recoil contributions in this first approach.
The $\Delta$ contribution (Fig.~\ref{Fig.1}d)
to the kernel is given by
\begin{eqnarray}
T^{\Delta}_a & = &
\frac{-ig_A h_A^2}{18 m_{N} f_{\pi}^3}\; \frac{1}{\omega_k^2-(k^0)^2}
\frac{\omega_q}{\delta^2-\omega_q^2} \times \nonumber \\
& & \left[\left(\vec{k}^2 \omega_q - \vec{k}\cdot\vec{P}\;\delta\right)
\vec{\Sigma}_a\cdot \vec{k}
+\frac{i}{2} \omega_q \left(\tau_a^{(1)}\;\vec{\sigma}^{(1)}
\cdot\vec{k}\;\vec{\sigma}^{(2)}\cdot(\vec{P}\times\vec{k})
-\tau_a^{(2)}\;\vec{\sigma}^{(1)}\cdot(\vec{P}\times\vec{k})
\;\vec{\sigma}^{(2)}\cdot\vec{k}\right)\right]\nonumber \\
& & +i\epsilon_{abc}\tau_b^{(1)}\tau_c^{(2)}\left[\left(\omega_q\,\vec{k}
\cdot\vec{P}-\delta\,\vec{k}^2\right)\vec{S}\cdot\vec{k}-
\frac{i}{4}\delta\left(\vec{\sigma}^{(1)}
\cdot\vec{k}\;\vec{\sigma}^{(2)}\cdot(\vec{P}\times\vec{k})
+\vec{\sigma}^{(1)}\cdot(\vec{P}\times\vec{k})
\;\vec{\sigma}^{(2)}\cdot\vec{k}\right)\right].
\label{Delta}
\end{eqnarray}
\noindent
Results similar to Eqs. (\ref{IA})
and (\ref{Delta}) follow for shorter-range
terms where the two nucleons exchange a heavier meson rather
than a pion.
In the case of a nucleon intermediate state,
such a contribution is automatically included in the potential.
In any reasonable model, the contributions from a delta
intermediate state
turn out to be smaller than those
in diagram of Fig.~\ref{Fig.1}d. For example,
they could arise {}from $a_1$ exchange, but
then the relatively high $a_1$ mass suppresses
this contribution;
the contribution from the $\rho$ which is formally
of higher-order is likely to be more important.
Since, as we are going to see, the delta contribution
from pion exchange is not large,
we will not go into such detailed analysis for the purpose of estimating
the effect of the $\Delta$: we use
Eq.(\ref{Delta}). We will further discuss the uncertainties related to
the delta contribution below.
There are other corrections of order $m_{\pi}/M$
compared to the leading terms.
Fig.~\ref{Fig.1}b represents also isospin-independent rescattering:
\begin{eqnarray}
T^{ST}_a&=&i\frac{g_A}{f_{\pi}^3}\, \frac{1}{\omega_k^2 -(k^0)^2}
\left\{\left[\left(c_2 +c_3- \frac{g_A^2}{8 m_N}\right) k^0
\omega_q -2c_1 m_{\pi}^2\right]\vec{\Sigma}_a\cdot\vec{k}
\right..\nonumber \\
& -& \left.\frac{\delta m_N}{8}\left[\delta_{3a}\vec{\tau}^{(1)}\cdot
\vec{\tau}^{(2)}\,\left(\vec{\sigma}^{(1)}-\vec{\sigma}^{(2)}\right)
+\vec{\sigma}^{(1)}\tau_3^{(1)}\tau_a^{(2)}-
\vec{\sigma}^{(2)}\tau_a^{(1)}\tau_3^{(2)}\right]\cdot\vec{k}\right\}.
\label{ST}
\end{eqnarray}
The short-range mechanisms provided by the $\Delta=2, 3, 4$ Lagrangians
involve several unknown constants.
Chiral symmetry tells us nothing about the strength of these coefficients.
We can use data to determine some of them.
Alternatively,
we can use a model to determine these coefficients and then try to explain
the experimental results. Here we use the mechanism first proposed
by Lee and Riska \cite{LR92} and by Horowitz {\it et al} \cite{HGM94},
where the short-range interaction is supposed to originate {}from
Z-graphs with $\sigma$ and $\omega$ exchanges, as shown in Fig.~\ref{Fig.1}c.
In this case,
\begin{eqnarray}
T^{\sigma, \omega}_a & = & - \frac{i\, g_A}{4 f_{\pi} m_N^2}\; \omega_q
\left[\left(\frac{g_\sigma^2}{\vec{k}^2 + m_{\sigma}^2}+
\frac{g_\omega^2}{\vec{k}^2 + m_{\omega}^2}\right)
\vec{\Sigma}_a\cdot\vec{P}\right. \nonumber\\
& &\left. - i\,\frac{g_\omega^2 (1+C_\omega)}{\vec{k}^2 + m_{\omega}^2}
\;\vec{\sigma}^{(1)}\times\vec{\sigma}^{(2)}\cdot\vec{k}
\, \left(\tau_a^{(1)}+\tau_a^{(2)}\right)\right] ,
\label{sigom}
\end{eqnarray}
where $m_\sigma$ ($m_\omega$) and $g_\sigma$ ($g_\omega$) are
the mass and the vector coupling to nucleons of the
$\sigma$ ($\omega$) meson,
and $C_\omega$ denotes the ratio of tensor to vector coupling for the
$\omega$ meson.
In the case of $pp\rightarrow pp \pi^0$,
the contribution due to $\rho-\omega$ exchange (Fig.~\ref{Fig.1}e)
is not negligible
\cite{KMR96}, so we also include it here in order to get an estimate
of the convergence of our expansion. This contribution leads to
\begin{eqnarray}
T^{\rho, \omega}_a & = & - \frac{i\, g_\rho g_\omega g_{\pi\rho\omega}}
{4 m_N^2}\; \frac{\omega_q}{m_\omega}
\left(\frac{g_\rho^2}{\vec{k}^2 + m_{\rho}^2}\cdot
\frac{g_\omega^2}{\vec{k}^2 + m_{\omega}^2}\right) \nonumber \\
& & \left\{\left(2+C_\rho+C_\omega\right)\left(\vec{k}^2
\vec{\Sigma}_a\cdot\vec{P}-\vec{k}\cdot\vec{P}\vec{\Sigma}_a
\cdot\vec{k}\right)\right. - \nonumber\\
& &\left. -i\,\left(1+C_\rho\right)\left(1+C_\omega\right)
\vec{k}^2\,\vec{\sigma}^{(1)}\times\vec{\sigma}^{(2)}\cdot\vec{k}
\, \left(\tau_a^{(1)}+\tau_a^{(2)}\right)\right\},
\label{rhoom}
\end{eqnarray}
\noindent
where $g_{\pi\rho\omega}$ is the $\pi \rho \omega$ coupling \cite{KMR96} and
the other coefficients have the same meaning as in the previous equation.
At momenta much smaller than the meson masses,
these short-range contributions are indeed
contact interactions.
\section{Cross-section}
\subsection{The $pp\rightarrow d\pi^+$ reaction}
We are concerned with evaluating the matrix elements of the above
operators between the initial $^3P_1$ and the deuteron final
wave functions.
To evaluate the influence of the potential in the amplitudes,
we use Reid93 \cite{Reid93}
and Argonne V18 \cite{v18} potentials which, for a given $pp$ channel,
are local potentials. Thus we evaluate the operators between coordinate
space initial ($i$) and final ($f$)
wave functions expressed by
\begin{equation}
\langle \vec{r}|i \rangle =\mp{\sqrt{2}\over pr} i \, u_{1,1}(r)e^{i
\delta_{1,1}} \left(\sqrt{2\pi}\;\sqrt{3}\right)\, |^3P_1\rangle \;,
\end{equation}
\noindent
where $-$ ($+$) is for third-component of the angular momentum
$M_J=+1$ ($-1$),
and
\begin{equation}
\langle \vec{r}|f \rangle ={1\over r} \left[u(r)\;|^3S_1\rangle
+w(r)\;|^3D_1\rangle \right]\;,
\end{equation}
\noindent where the deuteron wave functions are normalized as
\begin{equation}
\int_0^\infty dr\left[u^2(r)+w^2(r)\right]=1\;.
\end{equation}
Here and in the following
the spin angular part is expressed, as usual, as
\begin{equation}
| ^{2S+1}L_J \rangle = \sum_{m_L\,m_S}\langle L\,m_L\,S\,m_S|J\,m_J\rangle
Y_L^{m_L}(\hat{\bf r})\chi_S^{m_S}\;,
\end{equation}
\noindent
where $Y_L$ is the spherical harmonic function.
We convert the operators of Eqs.~(\ref{WT}-\ref{rhoom}) to
configuration space by inverting the Fourier transforms.
The resulting operators can then be used in configuration-space matrix
elements.
We define the matrix
elements of the operators of Eqs.~(\ref{WT}-\ref{rhoom}) as
\begin{eqnarray}
{\cal M}^{X} & = & \langle f|T^{X}|i\rangle \nonumber \\
& = & \mp{\sqrt{12\pi}\;i\over p}
e^{i\delta_{1,1}}
\int_0^\infty dr\left[\frac{u(r)}{\sqrt{m_\pi}}\,H^{X}_S(r)+
\frac{w(r)}{\sqrt{2m_\pi}}\,H^{X}_D(r)\right]\,u_{1,1}(r),
\label{Mdef}
\end{eqnarray}
\noindent
where $X$ represents $WT$, $IA$, {\it etc.}
$H^{X}_{S,D}(r)$ is the corresponding operator,
obtained using the matrix elements given in the Appendix.
To compare results before the cross-section evaluation, we define
dimensionless amplitudes $J^X$, where $X=WT,IA,\Delta,$ {\it etc.}
via
\begin{equation}
{\cal M}^{X} = \mp \,\sqrt{12\pi}\;i\;e^{i\delta_{1,1}}
\frac{g_A}{f_\pi^3}\;\cdot\frac{m_\pi}{4\pi}\, \sqrt{\frac{2}{3}}
\; J^{X}.
\label{Mdef2}
\end{equation}
We will plot $J^{X}$ in terms of $\eta$ to evaluate the energy dependence
of the amplitude,
and $dJ^X/dr$ in terms of $r$ to study the $r$ dependence of the integrand.
The contribution from the Weinberg-Tomozawa term is given by
\begin{equation}
H^{WT}_S(r)=H^{WT}_D(r)={g_A \over 4 f^3_\pi}\frac{3 m_\pi}{4\pi}
\sqrt{\frac{2}{3}}\;{(1+\tilde{m}_\pi r)\over r^2} e^{-\tilde{m}_\pi r},
\label{wt}
\end{equation}
\noindent where $\tilde{m}_\pi=\sqrt{\frac{3}{4}}m_\pi$.
The Galilean correction to the WT term is given by
\begin{eqnarray}
H^{GC}_S(r)&=&{g_A \over 4 f^3_\pi}\frac{3 m_\pi}{4\pi}\sqrt{\frac{2}{3}}\;
\frac{m_\pi}{4m_N}\left[A_{GC}(r)+B_{GC}(r)+C_{GC}(r)\right],
\label{gcs}\\
H^{GC}_D(r)&=&{g_A \over 4 f^3_\pi}\frac{3 m_\pi}{4\pi} \sqrt{\frac{2}{3}}\;
\frac{m_\pi}{4m_N}\left[A_{GC}(r)+B_{GC}(r)+D_{GC}(r) \right],
\label{gcd}
\end{eqnarray}
\noindent where
\begin{eqnarray}
A_{GC}(r)&=&{(1+\tilde{m}_\pi r)\over r^2} e^{-\tilde{m}_\pi r}, \\
B_{GC}(r)&=&-2 \frac{ e^{-\tilde{m}_\pi r}}{r}\left(1+\frac{2}{\tilde{m}_\pi r}
+\frac{2}{(\tilde{m}_\pi r)^2}\right)\frac{\partial}{\partial r}, \\
C_{GC}(r)&=&2 \frac{ e^{-\tilde{m}_\pi r}}{r^2}\left(1+\frac{4}{\tilde{m}_\pi r}
+\frac{4}{(\tilde{m}_\pi r)^2}\right), \\
D_{GC}(r)&=&2 \frac{ e^{-\tilde{m}_\pi r}}{r^2}\left(1+\frac{1}{\tilde{m}_\pi r}
+\frac{1}{(\tilde{m}_\pi r)^2}\right).
\end{eqnarray}
The impulse approximation is given by
\begin{eqnarray}
H^{IA}_S(r)&=&{g_A \over f_\pi}\frac{m_\pi}{m_N}\sqrt{\frac{1}{3}}\;
\left[\frac{1}{r}+\frac{\partial}{\partial r}\right],
\label{ias}\\
H^{IA}_D(r)&=&{g_A \over f_\pi}\frac{m_\pi}{m_N}\sqrt{\frac{1}{3}}\;
\left[-\frac{2}{r}+\frac{\partial}{\partial r}\right].
\label{iad}
\end{eqnarray}
The $\Delta$ resonance contribution
be written as
\begin{eqnarray}
H^{\Delta}_S(r)&=&\frac{1}{9m_N}\left({g_A \over f_\pi}\right)^3\left(
\frac{h_A}{g_A}\right)^2\frac{ m_\pi \tilde{m}_\pi^2}{(m_\pi-\delta)}
\sqrt{\frac{2}{3}}\;\frac{1}{4\pi}\left[A_{\Delta}(r)+B_{\Delta}(r)+
C_{\Delta}(r)\right], \label{dels}\\
H^{\Delta}_D(r)&=&\frac{1}{9m_N}\left({g_A \over f_\pi}\right)^3\left(
\frac{h_A}{g_A}\right)^2\frac{ m_\pi \tilde{m}_\pi^2}{(m_\pi-\delta)}
\sqrt{\frac{2}{3}}\;\frac{1}{4\pi}\left[A_{\Delta}(r)+B_{\Delta}(r)+
D_{\Delta}(r)\right],
\label{deld}
\end{eqnarray}
\noindent
where
\begin{eqnarray}
A_{\Delta}(r)&=&2{(1+\tilde{m}_\pi r)\over r^2} e^{-\tilde{m}_\pi r}, \\
B_{\Delta}(r)&=&-2\frac{ e^{-\tilde{m}_\pi r}}{r}
\left(1+\frac{1}{\tilde{m}_\pi r}+\frac{1}{(\tilde{m}_\pi r)^2}\right)
\frac{\partial}{\partial r}\;, \\
C_{\Delta}(r)&=&2 \frac{ e^{-\tilde{m}_\pi r}}{r^2}
\left(1+\frac{2}{\tilde{m}_\pi r}+\frac{2}{(\tilde{m}_\pi r)^2}\right), \\
D_{\Delta}(r)&=& \frac{ e^{-\tilde{m}_\pi r}}{r^2}
\left(-1+\frac{1}{\tilde{m}_\pi r}+\frac{1}{(\tilde{m}_\pi r)^2}\right).
\end{eqnarray}
The isospin-independent seagull term is given by
\begin{equation}
H^{SG}_S(r)=H^{SG}_D(r)={g_A \over f^3_\pi}\frac{m_\pi^2}{4\pi}
\sqrt{\frac{2}{3}}F_{SG}\;{(1+\tilde{m}_\pi r)\over r^2} e^{-\tilde{m}_\pi r},
\label{sg}
\end{equation}
\noindent where
\begin{equation}
F_{SG}=4c_1+{\delta m_N\over 4 m_\pi^2}-\left(c_2+c_3-{g_A^2\over8m_N}\right).
\end{equation}
We also include the effects of the $\sigma$ and $\omega$ exchange and
$\rho-\omega$ term.
The result for the $\sigma$ and $\omega$ exchange is
\begin{equation}
H^{\sigma, \omega}_S(r)=\frac{g_A}{f_\pi} {m_\pi \over\,m_N^2}
\sqrt{\frac{2}{3}}
\left\{ \left[f_\sigma(r)+ f_\omega(r)\right]
\left({\partial\over\partial r}+{1\over r}\right)
+ \frac{1}{2}\frac{\partial}{\partial r}\left[
f_\sigma(r)+ f_\omega(r)\right]\right\}\, ,
\label{sig-omS}
\end{equation}
\noindent and
\begin{equation}
H^{\sigma, \omega}_D(r)=\frac{g_A}{f_\pi} {m_\pi \over\,m_N^2}
\sqrt{\frac{2}{3}}
\left\{ \left[f_\sigma(r)+ f_\omega(r)\right]
\left({\partial\over\partial r}-{2\over r}\right)
+ \frac{1}{2}\frac{\partial}{\partial r}\left[
f_\sigma(r)+ f_\omega(r)\right]\right\}\, ,
\label{sig-omD}
\end{equation}
\noindent
where the function $f_h(r)$ accounts for exchange of the meson $h$ between
nucleons,
\begin{equation}
f_h(r)={g_h^2\over 4\pi}{e^{-m_h r}\over r}.
\end {equation}
\noindent
We follow Ref.\cite{HGM94} by including the effects of form factors
in these heavy-meson contributions, as
given in the Bonn potential \cite{machleidt}.
For completeness, we repeat that monopole form factors are used
at each meson vertex according to the replacement
\begin{equation}
g_h\to g_h{\Lambda_h^2-m_h^2\over\Lambda_h^2-k_\mu k^\mu},
\label{formf}
\end{equation}
where $k_\mu$ is the transferred momentum and $\Lambda_h$ is the
cutoff mass.
With $M$ denoting the common $\rho-\omega$ mass (780 MeV),
$\rho-\omega$ exchange yields
\begin{eqnarray}
H^{\rho-\omega}_S(r)&=&\frac{g_\rho\,g_\omega\,g_{\pi\rho\omega}}{4m_N^2}
{m_\pi \over\,4\pi} \sqrt{\frac{2}{3}}
\frac{2(2+C_\rho)}{M(\Lambda^2-M^2)}\times\nonumber \\
& & \left\{[A_{\rho-\omega}(r; M, \Lambda)
-A_{\rho-\omega}(r; \Lambda, M)]
\left({\partial\over\partial r}+{1\over r}\right)\right.
\nonumber \\
& & \left. \;\;
-[B_{\rho-\omega}(r; M, \Lambda)
-B_{\rho-\omega}(r; \Lambda, M)]
\left({\partial\over\partial r}-{1\over r}\right)\right\},
\label{roS}
\end{eqnarray}
where
\begin{eqnarray}
A_{\rho-\omega}(r; m_1, m_2) &=&
m_1 \, \frac{e^{-m_1r}}{m_1r}\left[ 4m_1^2\left(\frac{1}{m_1r}+
\frac{1}{(m_1r)^2}\right)-(m_2^2-m_1^2)m_1r
+3m_1^2+m_2^2\right],\nonumber \\
B_{\rho-\omega}(r; m_1, m_2) &=&
m_1 \, \frac{e^{-m_1r}}{m_1r}\left[ 4m_1^2\left(1+\frac{3}{m_1r}+
\frac{3}{(m_1r)^2}\right)-(m_2^2-m_1^2)(1+m_1r)\right].
\end{eqnarray}
$H^{\rho-\omega}_D(r)$
can be obtained making the replacement
\[
\left(\frac{\partial}{\partial r}\pm\frac{1}{r}\right)\longrightarrow
\left(\frac{\partial}{\partial r}\pm\frac{2}{r}\right)
\]
above.
The final steps consist of computing the total matrix element ${\cal M}$,
\begin{equation}
{\cal M}={\cal M}^{WT}+{\cal M}^{GC}+{\cal M}^{IA}+{\cal M}^{ST}+{\cal M}^{\Delta}
+{\cal M}^{\sigma,\omega}+{\cal M}^{\rho-\omega},
\label{calm}
\end{equation}
squaring it, and integrating over the available phase space.
We find
\begin{equation}
\sigma={1\over 16\pi}\;\frac{m_\pi}{p}\;E_d\,\omega_q\,\eta|{\cal M}|^2,
\label{cross}
\end{equation}
where $p$ is the magnitude of the center-of-mass 3-momentum,
\[
p=m_\pi\cdot \sqrt{\frac{m_N}{m_\pi}\cdot\sqrt{1+\eta^2}+
\frac{\eta^2\;m_N}{2M_d}+\frac{(M_d-2m_N)m_N}{m_\pi^2}},
\]
and $E_d$ is energy of the produced deuteron
of mass $M_d$, $E_d=\sqrt{M_d^2+q^2}$.
\subsection{The $pp\rightarrow pn\pi^+$ reaction}
As we did in the deuteron case, we evaluate the matrix elements
for the unbound final state using the same operators of
Eqs.~(\ref{WT})-(\ref{rhoom}). We will consider here just the absolute
threshold limit where $L_{\pi(NN)}=0$. According to selection rules,
we have 2 channels, as stated in Tab.~\ref{Tab.01}:
$^3P_1\rightarrow {^3S_1}$ ($T_i=1,\,T_f=0$) and
$^3P_0\rightarrow {^1S_0}$ ($T_i=1,\,T_f=1$). Again,
to evaluate the influence of the potential in the amplitudes,
we use Reid93 \cite{Reid93}
and Argonne V18 \cite{v18} potentials which, for a given $pp$ or $pn$
channel, are local potentials. Thus we evaluate the operators between
coordinate space initial ($i$) and final ($f$)
wave functions for the two channels, expressed by
\begin{itemize}
\item $^3P_1\rightarrow$ $^3S_1$ channel:
\begin{eqnarray}
\langle \vec{r}|i \rangle &=& \mp{\sqrt{2}\over pr} i \, u_{1,1}(r)e^{i
\delta_{1,1}} \left(\sqrt{2\pi}\;\sqrt{3}\right)\, |^3P_1\rangle \;,
\\ \nonumber
\langle \vec{r}|f \rangle &=& {1\over p'r} i \, u_{0,1}(r)e^{i
\delta_{0,1}} \sqrt{4\pi}\, |^3S_1\rangle \;,
\end{eqnarray}
\item $^3P_0\rightarrow$ $^1S_0$ channel:
\begin{eqnarray}
\langle \vec{r}|i \rangle &=& {\sqrt{2}\over pr} i \, u_{1,0}(r)e^{i
\delta_{1,0}} \sqrt{4\pi}\, |^3P_0\rangle \;,
\\ \nonumber
\langle \vec{r}|f \rangle &=& {1\over p'r} i \, u_{0,0}(r)e^{i
\delta_{0,0}} \sqrt{4\pi}\, |^1S_0\rangle \;,
\end{eqnarray}
\end{itemize}
\noindent
where $-$ ($+$) is for third-component of the angular momentum
$M_J=+1$ ($-1$).
We convert the operators of Eqs.~(\ref{WT}-\ref{rhoom}) to
configuration space by inverting the Fourier transforms.
The resulting operators can then be used in configuration-space matrix
elements.
We define the matrix
elements of the operators of Eqs.~(\ref{WT}-\ref{rhoom}) as
\begin{itemize}
\item $^3P_1\rightarrow$ $^3S_1$ channel
\begin{equation}
{\cal M}^{X} = \langle ^3S_1 | T^X | ^3P_1 \rangle =
\mp \,4\pi\,\sqrt{3}\;i\;e^{i(\delta_{0,1}+\delta_{1,1})}
\frac{g_A}{f_\pi^3}\;\cdot\frac{1}{4\pi}\, \sqrt{\frac{2}{3}}
\; J^{X}.
\label{Mdef3p1}
\end{equation}
\item $^3P_0\rightarrow$ $^1S_0$ channel
\begin{equation}
{\cal M}^{Y} = \langle ^1S_0 | T^Y | ^3P_0 \rangle =
4\pi\,\sqrt{2}\;i\;e^{i(\delta_{1,0}+\delta_{0,0})}
\frac{g_A}{f_\pi^3}\;\cdot\frac{1}{4\pi}
\; J^{Y}.
\label{Mdef3p0}
\end{equation}
\end{itemize}
\noindent
with
\begin{eqnarray}
J^X&=&\frac{m_\pi}{p\,p'}\int_0^\infty dr\,u_{0,1}\,H^X(r)\,u_{1,1}(r)
\\ [0.3cm]
J^Y&=&\frac{m_\pi}{p\,p'}\int_0^\infty dr\,u_{0,0}\,H^Y(r)\,u_{1,0}(r)
\end{eqnarray}
\noindent
where $X$ and $Y$ represents $\Delta$, $IA$, etc.,
$J^X$ and $J^Y$ are dimensionless integrals,
$H^{X}(r)$ and $H^Y(r)$ are the corresponding operators,
obtained using the matrix elements given in the Appendix, which are the
same used in the deuteron final state (for the
$^3P_1\rightarrow$ $^3S_1$ channel) and in the $pp\rightarrow pp\pi^0$
\cite{bira} (for the $^3P_0\rightarrow$ $^1S_0$ channel).
We will plot $J^{X}$ and $J^Y$ in terms of $p'$ to evaluate the energy
dependence of the amplitude,
and $dJ^X/dr$ and $dJ^Y/dr$ in terms of $r$ to study the $r$ dependence of
the integrand.
For the $^3P_1\rightarrow$ $^3S_1$ channel, the coordinate space
expressions for the amplitudes are pretty much
the same of the $S$-wave deuteron amplitudes. For instance, the
Weinberg-Tomozawa term gives
\begin{equation}
H^{WT}(r)=\frac{3}{4}\;{(1+\tilde{m}_\pi r)\over r^2} e^{-\tilde{m}_\pi r},
\label{wtpn}
\end{equation}
\noindent
where $\tilde{m}_\pi=\sqrt{\frac{3}{4}}m_\pi$.
For the $^3P_0\rightarrow ^1S_0$ channel, the expressions follow closely the
work of Cohen {\it et al} \cite{bira},
the isospin matrix elements being the only
difference. As we did in the deuteron case, we include the effects of
form factors when dealing with heavy meson contributions.
The final steps consist of computing the total matrix element ${\cal M}$
for the two channels, using Eq.~(\ref{calm}) for the $^3S_1$ final state
and the same equation but without WT and the GC terms for the $^1S_0$
final state, since in this state the isovector contributions are zero.
The cross section is obtained by squaring the total amplitude, and
integrating over the available phase space. We find
\begin{equation}
\sigma=\sum_{spins}\,\frac{1}{v}\int_0^{p_{max}'}\;dp'\;
\frac{{p'}^2\,q}{(2\pi)^3}\;|{\cal M}|^2\;\frac{m_n}{2m_n+w(q)}
\label{crosspn}
\end{equation}
where $v$ is the laboratory velocity of the incident proton, $q$ is
the pion 3-momentum, $w(q)=\sqrt{q^2+m_\pi^2}$,
$p$ is the magnitude of the center-of-mass initial 3-momentum and
${p}_{max}'=\sqrt{p^2-m_N m_\pi}$. The $\Sigma_{spins}$ indicates that
(a) a sum over final spin states and (b) an average over
initial spin states must be made, which result in factors of $3/4$
for the $^3S_1$
and $1/4$ for the $^1S_0$ final states.
\section{Input Parameters and Results}
The various amplitudes considered in the last section depend on
several parameters that we can determine {}from other processes.
The WT, GC to WT, and the impulse-approximation operators depend on
the pion mass, $m_\pi = 138$ MeV \cite{pdb}, and on
\begin{equation}
\frac{g_A}{f_{\pi}}=\frac{g_{\pi NN}}{m_N};
\end{equation}
\noindent
we use the value of $g_{\pi NN}$ appropriate for each potential.
The $\Delta$ operator of Eq.(\ref{Delta}) further depends
on the $\Delta-N$ mass splitting $\delta = 294$ MeV \cite{pdb} and
on the $\pi N \Delta$
coupling constant, $h_A$. This has been fixed {}from $P$-wave
$\pi N$ scattering (see, e.g., Ref.\cite{weise}),
\begin{equation}
\frac{h_A}{g_A} \simeq 2.1.
\end{equation}
The seagull operator of Eq.(\ref{ST}) depends on four parameters
$c_{1,2,3}$ and $\delta m_{N}$.
The $c_{i}$'s can be obtained by fitting $S$-wave $\pi N$ scattering.
In Ref.\cite{BKM95} they were found to be
\begin{equation}
\left[4c_1+\frac{\delta m_N}{4m_\pi^2}-\left(c_2+c_3
-\frac{g_A^2}{8m_N}\right)\right]=-\frac{2.31}{2m_N}\; ,
\end{equation}
\noindent
from the $\sigma$-term, the isospin-even scattering length,
and the axial polarizability, to ${\cal O}(Q^3)$.
We refer to this as ``SeaI''.
A different determination from an ${\cal O}(Q^2)$ fit to $\pi N$
sub-threshold
parameters \cite{sub} gives $-0.29/2m_N$ instead.
We refer to this as ``SeaII''.
Newer determinations \cite{ulfandall} give values closer to the more
negative value, but we use both values in order to estimate
the importance of this contribution.
Note that the analysis of Ref.\cite{BKM95} does not include
the isobar explicitly.
Since the inclusion of the $\pi N \Delta$ interaction
only affects $S$ waves at one order higher than the $c_{i}$'s,
the above values can still be used to estimate the effect of
$S$-wave rescattering.
The parameter, $\delta m_{N}$, can in principle also
be determined {}from S-wave $\pi N$ scattering, but would require
a careful analysis of other isospin-violating effects. Chiral
symmetry relates it to the
strong interaction contribution to the nucleon mass splitting, which
is also difficult to determine directly.
Estimates of the electromagnetic contribution $\bar{\delta} m_{N}$
are more reliable, $\bar{\delta} m_{N}\sim -1.5$ MeV \cite{jerry},
and give $\delta m_{N} \sim 3$ MeV. To be definite, we use
\begin{equation}
\delta m_{N}=3 \, \mbox{MeV}.
\label{deltmn}
\end{equation}
Finally, the $\sigma, \omega$ and the $\rho-\omega$ operators involve
$g_{h}$, $\Lambda_{h}$, $m_{h}$, and $C_\omega$, parameters listed in
Table A.3 of Ref. \cite{machleidt}. They also involve $g_{\pi\rho\omega}$,
discussed in Ref. \cite{KMR96}.
These heavy-meson contributions
correspond to chiral Lagrangian coefficients of
natural size:
$d_1^{\prime} + e_1 \approx -1.5 (1/f_{\pi}^2 M)$,
$e_1 \approx -2 (1/f_{\pi}^2 M)$,
$e_2 \approx 2 (1/f_{\pi}^2 M)$,
and $g \approx 4 (1/f_{\pi}^2 M^3)$.
\subsection{The $pp\rightarrow d\pi^+$ reaction}
The relative sizes of the various contributions
to the matrix element $J$ of this reaction as function of $\eta$
are shown in Fig. \ref{Fig.2} for the Reid93 potential and
in Fig. \ref{Fig.3} for the AV18 potential. There is a very strong similarity
between these two sets of results.
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig02.ps}}
\vspace{-4.5cm}
\caption{Matrix elements $J$ as function of $\eta$
for the various contributions to $pp\rightarrow d\pi^+$,
calculated with wavefunctions
from the Reid93 potential.}
\label{Fig.2}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig03.ps}}
\vspace{-4.5cm}
\caption{Matrix elements $J$ as function of $\eta$
for the various contributions to $pp\rightarrow d\pi^+$,
calculated with wavefunctions
from the Argonne V18 potential.}
\label{Fig.3}
\end{figure}
Our leading order comprises WT, IA and $\Delta$. The first
noticeable observation is that WT is by far the largest contribution.
This is a consequence of cancellations in IA and $\Delta$
which were not anticipated by the power counting.
In Figs.~\ref{Fig.4}, \ref{Fig.5}, and \ref{Fig.6}
we can see typical integrands for the three contributions
at $\eta=0.3$, in the case of the Argonne V18 potential.
While for WT the contributions from the $S$ and $D$ deuteron waves
add and are dominated by the region around $r=1.5$ fm,
for IA and $\Delta$ the $S$ and $D$ waves tend to interfere destructively,
and contributions from different $r$ regions approximately
cancel.
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig04.ps}}
\vspace{-4.5cm}
\caption{Typical integrands ($\eta=0.3$) of the Weinberg-Tomozawa
contribution to $pp\rightarrow d\pi^+$ as function of the radial coordinate
$r$ for deuteron $S$ and $D$ waves of the Argonne V18 potential.
The $S$ wave is given by the dashed line, the $D$ wave by the dot-dashed
line and the sum by the solid line.}
\label{Fig.4}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig05.ps}}
\vspace{-4.5cm}
\caption{Typical integrands ($\eta=0.3$) of the impulse
contribution to $pp\rightarrow d\pi^+$ as function of the radial coordinate
$r$ for deuteron $S$ and $D$ waves of the Argonne V18 potential.
Lines are as in Fig. 4.}
\label{Fig.5}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig06.ps}}
\vspace{-4.5cm}
\caption{Typical integrands ($\eta=0.3$) of the delta
contribution to $pp\rightarrow d\pi^+$ as function of the radial coordinate
$r$ for deuteron $S$ and $D$ waves of the Argonne V18 potential.
Lines are as in Fig. 4.}
\label{Fig.6}
\end{figure}
Our formally sub-leading contributions are all
considerably smaller than WT. This suggests that
the theoretical control of this reaction is greater
than for $pp\rightarrow pp\pi^0$.
Moreover, sub-leading contributions come with
different signs, partially cancelling.
In fact, the $\sigma$,$\omega$ and GC
terms come out with similar size
---as predicted by the power counting--- but with opposite signs;
there is an almost complete
accidental cancellation between them.
$\rho-\omega$ exchange is of higher order and indeed very small.
As a result, when the small contribution from
the combination of seagull parameters ``SeaII''
is considered, the sum of all sub-leading
contributions is small. The sum is somewhat larger
and of opposite sign to leading order
when ``SeaI'' is used.
In Fig. \ref{Fig.10} we compare our
leading and sub-leading results for the $pp$ cross-section
(without Coulomb) using the Reid93 potential
with data from Refs. \cite{triumf,cosy,iucf}.
We see that our curves are approximately constant, as expected.
Recent data points for $\eta<0.1$ are not all consistent, but they cluster
around an average $\alpha$ of about 180 to 190 $\mu$b.
The value of $\alpha$ in leading order (WT+IA+$\Delta$)
is a factor of about 1.5 below data.
The destructive interference with next-order contributions
increases the disagreement by an amount depending
on the value of $\pi N$ isoscalar rescattering term.
The change is bigger when ``SeaI'' is used. This can be
seen in Fig. \ref{Fig.10} as ``SumI'' and ``SumII''.
Because of the cancellations among sub-leading terms,
the result exhibits comparable dependence
on the short-range contributions,
as can be seen in Fig. \ref{Fig.10}
as ``SumII without $\sigma$, $\rho$, $\omega$''.
We summarize our results in Fig. \ref{Fig.11} where we show
the sum of all the contributions we considered
for the two sets of seagull parameters, and for the two
potentials.
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig10.ps}}
\vspace{-4.5cm}
\caption{Reduced cross-section $\sigma/\eta$
for $pp\rightarrow d\pi^+$ as function of $\eta$
calculated with the Reid93 potential
in leading order ($WT+IA+\Delta$), in leading plus sub-leading order
with two sets of parameters (SumI and SumII),
and in leading plus sub-leading order
without heavy-meson exchange (SumII without $\rho$, $\sigma$, $\omega$), and
compared with data from TRIUMF (circles) \protect\cite{triumf},
COSY (diamonds) \protect\cite{cosy} and IUCF (squares) \protect\cite{iucf}.}
\label{Fig.10}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig11.ps}}
\vspace{-4.5cm}
\caption{Reduced cross-section $\sigma/\eta$ for $pp\rightarrow d\pi^+$
as function of $\eta$
for the sum of all the contributions considered
with two sets parameters (SumI, SumII) for
the Reid93 (solid line) and Argonne V18 (dashed line) potentials,
compared with data from TRIUMF (circles) \protect\cite{triumf},
COSY (diamonds) \protect\cite{cosy} and IUCF (squares) \protect\cite{iucf}.}
\label{Fig.11}
\end{figure}
\subsection{The $pp\rightarrow pn\pi^+$ reaction}
The relative sizes of the various contributions
to the matrix element $J$ of this reaction as function of $p'$
are shown for the $^3S_1$ final state
in Fig. \ref{Fig.12} for the Reid93 potential and
in Fig. \ref{Fig.13} for the AV18 potential;
and for the $^1S_0$ final state
in Fig. \ref{Fig.14} for the Reid93 potential and
in Fig. \ref{Fig.15} for the AV18 potential.
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig12.ps}}
\vspace{-4.5cm}
\caption{Matrix elements $J$ as function of $p'$
for the various contributions to $pp\rightarrow pn\pi^+$,
calculated with wavefunctions
from the Reid93 potential: $^3S_1$ final state.}
\label{Fig.12}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig13.ps}}
\vspace{-4.5cm}
\caption{Matrix elements $J$ as function of $p'$
for the various contributions to $pp\rightarrow pn\pi^+$,
calculated with wavefunctions
from the Argonne V18 potential: $^3S_1$ final state.}
\label{Fig.13}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig14.ps}}
\vspace{-4.5cm}
\caption{Matrix elements $J$ as function of $p'$
for the various contributions to $pp\rightarrow pn\pi^+$,
calculated with wavefunctions
from the Reid93 potential: $^1S_0$ final state.}
\label{Fig.14}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig15.ps}}
\vspace{-4.5cm}
\caption{Matrix elements $J$ as function of $p'$
for the various contributions to $pp\rightarrow pn\pi^+$,
calculated with wavefunctions
from the Argonne V18 potential: $^1S_0$ final state.}
\label{Fig.15}
\end{figure}
Here again the WT contribution is the largest;
since it contributes only to the $^3S_1$ final state,
this channel is dominant.
Most of the other contributions are much smaller and tend to
cancel to some extent.
The exception is the $\Delta$,
which has a significant destructive interference with WT
in the $^3S_1$ state.
The IA contribution is small due to the same type of
cancellation observed before among different regions in coordinate
space.
In Fig. \ref{Fig.19} we compare our
results for the different final states
using the Reid93 potential
with data from Ref. \cite{steve1,steve2}.
In Fig. \ref{Fig.20} we summarize our results for
the two potentials considered and compare them to the same data.
We see that the theory produces a correct shape for the
$\eta$ dependence but fails in magnitude by a factor of $\sim 5$.
Use of ``SeaI'' further worsens the results.
Once again the differences between the two potentials are minimal.
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig19.ps}}
\vspace{-4.5cm}
\caption{Reduced cross-section $\sigma/\eta^2$
for $pp\rightarrow pn\pi^+$ as function of $\eta$
calculated with the Reid93 potential:
$^3S_1$ final state (dashed line),
$^1S_0$ final state (dash-dotted line),
their sum (solid line),
and data from IUCF \protect\cite{steve1,steve2}.}
\label{Fig.19}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig20.ps}}
\vspace{-4.5cm}
\caption{Reduced cross-section $\sigma/\eta^2$
for $pp\rightarrow pn\pi^+$ as function of $\eta$
for the sum of all the contributions:
Reid93 potential (dotted line),
Argonne V18 potential (solid line),
and data from IUCF \protect\cite{steve1,steve2}.}
\label{Fig.20}
\end{figure}
\subsection{Discussion}
The $d\pi^+$ final state has been considered before in the literature.
Ref. \cite{KR66} has found that the impulse term was very small due to
the strong cancellation between the $S$ and $D$ waves in the matrix element
shown above. Mostly due to the WT,
$\alpha$ of 146 to 160 $\mu$b was
obtained
using the older, higher value of the $\pi NN$ coupling constant;
using the more recent value,
we get 124 to 136 $\mu$b.
A similar analysis \cite{horowitz}
included also
form factors at the $\pi\pi NN$ vertices, which led to a decrease
of approximately 20\% in the cross section; with the more modern
coupling constants, the overall result was near 100 $\mu$b.
Recently, Ref. \cite{hanhart}
re-analyzed this reaction using a covariant
approach. The result for $\alpha$ using only the WT term was
again near 100 $\mu$b, but the amplitude contains what we refer to as the
Galilean correction to the WT term,
which has an opposite sign.
Our results for the WT term alone are in numerical
agreement with these works, since we find that it gives
an $\alpha$ of about 100 $\mu$b.
Other contributions tend to worsen the description of the data.
Particularly damaging are the contributions of the rescattering type.
Here, as for $pp\rightarrow pp\pi^0$, the $S$-wave seagulls
tend to interfere destructively with the leading mechanism.
As in the case of the $pp\rightarrow pp\pi^0$ reaction,
the $\Delta$ contribution shows appreciable dependence
on the potential used. But unlike that reaction, here
it generates relatively small dependence on the potential in the final result,
as consequence of the large relative size of WT.
The $\Delta$ contribution is the one that presents
the largest change when we compare $pn$ and $d$ final states.
In Fig. \ref{Fig.21} we compare the radial distributions
of the delta contribution for the $^3S_1$ $pn$ final state at
$p'=0.18$ fm$^{-1}$
and for the deuteron final state (same as in Fig. \ref{Fig.6}).
{}From this we can see how this contribution has different
signs in the two channels, being small and constructive with
WT for $d$ ($J^{WT}_d=0.130$ and $J^{\Delta}_d=0.015$; $\eta=0.3$)
and larger and destructive for $pn$
($J^{WT}_{pn}=0.260$ and $J^{\Delta}_{pn}=-0.11$; $\eta=0.3$; $p'=0.18$)
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig21.ps}}
\vspace{-4.5cm}
\caption{Typical integrands ($\eta=0.3$) of the delta
contribution as function of the radial coordinate
$r$ for the $^3S_1$ $pn$ final state at $p'=0.18$ (solid line)
compared to the deuteron final state (dashed line).}
\label{Fig.21}
\end{figure}
Note, however, that the $\Delta$ contribution is subject to large
uncertainties.
First, the $\pi N\Delta$ coupling constant is not well determined
and appears squared; although we do not know whether this coupling should
be bigger or smaller than the value used here, it could
contribute to a decrease of the $\Delta$ effect.
Second, the $\pi N\Delta$ form factor seems to be much
softer than the corresponding $\pi NN$ form factor;
our neglecting both enhances the $\Delta$ contribution
relative to the WT term.
Third, we have neglected the $\Delta-N$ mass difference
in energies; since the $\Delta$ mass would appear in the denominator,
the delta contribution would decrease by a factor of
$\sim 2m_N/(m_N+m_\Delta)$.
Fourth, we have neglected the kinetic energy
of the delta, account of which would further decrease
its amplitude.
Fifth, it is known that there can be some cancellation
between the pion- and rho-exchange delta terms; including
the latter would also diminish the delta contribution.
Assuming each of these gives a 10\% decrease, we could very well be
overestimating the $\Delta$ contribution by 50\% or more.
If we neglect this contribution altogether,
we obtain the results in
Fig. \ref{Fig.22} for the cross section in the $d$ channel
and in Fig. \ref{Fig.23} for the cross section in the $pn$ channel.
This brings theory to underestimate both sets of data
by a common factor of $\simeq 2$.
This implies that there must be further corrections to the amplitude of
about 50\%.
This is not unlike the $pp\rightarrow pp\pi^0$ reaction considered
in Ref. \cite{KMR96},
where theory tends to fail by a similar factor.
The case for failure of theory there is less clear-cut, however,
because of the lack of a large contribution as for the WT
term here. As a consequence, the usually small effect
of other mechanisms is enhanced and the result is dominated
by shorter-range dynamics; more sensitivity
to the potential and seagull terms surfaces,
and it is possible to find a combination of parameters that
includes data \cite{KMR96}.
No such gimmicks work here.
For example, we find that
heavy-meson exchange ---hailed as solution in the
$pp\rightarrow pp\pi^0$ reaction---
does not help much in $\rightarrow d\pi^+$ and $\rightarrow pn\pi^+$,
in agreement with the findings in Refs. \cite{niskanen,unpharry}.
The dependence on the $\Delta$ contribution (which is a particular
type of rescattering) suggests that we need better
control over longer-range contributions, such as $\pi N$ rescattering
and two-pion exchange.
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig22.ps}}
\vspace{-4.5cm}
\caption{Reduced cross-section $\sigma/\eta$ for $pp\rightarrow d\pi^+$
as function of $\eta$. The graph show the sum of all the contributions,
excluding the $\Delta$: Reid93 potential (dotted line),
Argonne V18 potential (solid line),
and data from TRIUMF (circles) \protect\cite{triumf},
COSY (diamonds) \protect\cite{cosy} and IUCF (squares)
\protect\cite{iucf}.}
\label{Fig.22}
\end{figure}
\begin{figure}
\vspace{-4.5cm}
\epsfxsize=13.0cm
\centerline{\epsffile{fig23.ps}}
\vspace{-4.5cm}
\caption{Reduced cross-section $\sigma/\eta^2$ for $pp\rightarrow pn\pi^+$
as function of $\eta$. The lines shown the
sum of all the contributions, excluding the $\Delta$:
Reid93 potential (dotted line),
Argonne V18 potential (solid line),
and data from IUCF \protect\cite{steve1,steve2}.}
\label{Fig.23}
\end{figure}
\section{Conclusion}
We have calculated the cross section near threshold for the
reactions $pp \rightarrow d \pi^+, \rightarrow pn \pi^+$,
using a chiral power counting to order interactions.
The interactions included contain most of the interactions
found in the literature. In particular, they include
mechanisms used in various models of the
$pp\rightarrow pp\pi^0$ reaction.
In contrast to the latter,
results here depend relatively little on the wavefunction
employed and on short-range interactions.
The Weinberg-Tomozawa term dominates, but the delta contribution
can be important, with large uncertainty.
Our computed results typically
fall a factor of about two below the data. Our kernels vary as the
cube of a generic meson-nucleon coupling constant, so such discrepancy
can be parametrized as a 12\% deficiency in the coupling constants. The
pion-nucleon
coupling constant is known to higher precision than that, but not to much
higher precision.
Thus the discrepancy we find might not be a very serious problem.
Our main conclusion
is that a relatively long-range mechanism ---
such as $\pi N$ rescattering and/or two-pion exchange--- is needed
for the description of these reactions.
Accordingly, we suggest that advance in understanding pion production
in $NN$ collisions must follow
not from the study of $pp\rightarrow pp\pi^0$ by itself
---as has been the trend of theoretical study to date---
but from focus on an understanding of long-range effects
that afflict all channels.
\vspace{2cm}
{\large \bf Acknowledgments}
We thank C. Hanhart for a helpful discussion.
This research was supported in part by the U.S. DOE
under grant DE-FG03-97Er41014 and NSF under grant PHY 94-20470.
The work of C.A.dR. was supported by the Brazilian FAPESP
under contract numbers 97/05817-0 and 97/6209-4. He thanks
the Nuclear Theory group at the University of Washington
for hospitality during the initial stages of this work.
\vspace{1.5cm}
{\Large \bf Appendix}
\vspace{.5cm}
The calculation of the cross-section in Section IV relies
on the following matrix elements.
For isospin,
\begin{eqnarray}
&&\langle 00\,|\;i\epsilon_{abc}\;\tau_b^{(1)}\tau_c^{(2)}\,|11\rangle =
-2 \;,\\
&&\langle 00\,|\vec{A}\,\tau_a^{(1)}-\vec{B}\,\tau_a^{(2)}\,|11\rangle =
\vec{A}+\vec{B} \;,\\
&&\langle 00\,|\vec{A}\,\tau_3^{(1)}\tau_a^{(2)}-\vec{B}\,\tau_a^{(1)}
\tau_3^{(2)}\,|11\rangle = -\left(\vec{A}+\vec{B}\right) \;,\\
&&\langle 00\,|\,\delta_{3a}\,\vec\tau^{(1)}\cdot\vec\tau^{(2)}\,|11\rangle=0 \;, \\
&&\langle 00\,|\frac{1}{2}\left[\tau^{(1)}_a + \tau^{(2)}_a\right]\,|11\rangle=0\;,
\end{eqnarray}
\noindent
where $\vec{A}$ and $\vec{B}$ are spin operators or scalar products.
The index $a$ represents the isospin of the emerging pion,
$\langle 00\,|$ is the isospin
state of the deuteron and $|11\rangle$ the $pp$ isospin initial state.
For spin,
\begin{eqnarray}
&&\langle ^3S_1|\vec{S}\cdot \hat r |^3P_1\rangle = -\sqrt{\frac{2}{3}}\;, \\
&&\langle ^3D_1|\vec{S}\cdot \hat r |^3P_1\rangle = -\sqrt{\frac{1}{3}}\;, \\
&&\langle ^3S_1|i \vec{S}\cdot\vec{p}\,|^3P_1\rangle = -
\sqrt{\frac{2}{3}}\left({\partial\over\partial r}+{2\over r}\right)\; ,\\
&&\langle ^3D_1|i \vec{S}\cdot\vec{p}\,|^3P_1\rangle = -
\sqrt{\frac{1}{3}}\left({\partial\over\partial r}-{1\over r}\right)\; ,\\
&&\langle ^3S_1|\vec{\sigma}^{(1)}\cdot\vec{\sigma}^{(2)}\; i \vec{S}\cdot\vec{p}\,|^3P_1\rangle = -
\sqrt{\frac{2}{3}}\left({\partial\over\partial r}+{2\over r}\right)\; ,\\
&&\langle ^3D_1|\vec{\sigma}^{(1)}\cdot\vec{\sigma}^{(2)}\;i \vec{S}\cdot\vec{p}\,|^3P_1\rangle = -
\sqrt{\frac{1}{3}}\left({\partial\over\partial r}-{1\over r}\right)\; ,\\
&&\langle ^3S_1|\hat{S}_{12}\,i \vec{S}\cdot\vec{p}\,|^3P_1\rangle = - 2
\sqrt{\frac{2}{3}}\left({\partial\over\partial r}-{1\over r}\right)\; ,\\
&&\langle ^3D_1|\hat{S}_{12}\,i \vec{S}\cdot\vec{p}\,|^3P_1\rangle = - 2
\sqrt{\frac{1}{3}}\left({\partial\over\partial r}+{5\over r}\right)\; .
\end{eqnarray}
\noindent
where $\hat{S}_{12}$ is the usual tensor operator.
|
\section{Introduction}
Perovskite manganites exhibiting a variety of exotic electronic properties
\cite{Millis} that include a spectacular decrease of electrical resistance
in a magnetic field \cite{VonHel}, the so-called colossal magnetoresistance
(CMR)\ \cite{Jin}, have attracted wide attention in recent years. A
significant feature of the electronic structure of the ferromagnetic CMR
manganites, revealed by recent experiments \cite{Park,Soulen} and theory
\cite{pickett}, is that the charge carriers are almost completely
spin-polarized at the Fermi level E$_{F}$. These materials are half-metallic
ferromagnets, where the majority spin states near E$_{F}$ are delocalized
and the minority spin channel is effectively localized. Since half-metallic
ferromagnetism and magnetoresistance (MR), especially at low fields, seem to
be intimately related to each other \cite{Hwang} - the latter arising from
the former- there is an intense search for half-metallic magnets which could
be candidate materials for the realization of MR applications. While several
double-perovskite oxides of the kind A$_{2}$BB'O$_{6}$ (A being an alkaline
earth or rare earth ion and B, B' being $d$-transition metal ions) have been
theoretically predicted to be half-metallic antiferromagnets \cite{pickett2}
, one such material, Sr$_{2}$FeMoO$_{6}$ \cite{nakagawa}, has recently \cite
{Kobayashi,Kim} been shown to be a half-metallic ferrimagnet exhibiting a
significant tunneling-type magnetoresistance at room temperature. More
recently, Asano et al. \cite{asano} have shown that it is possible to have
either positive or negative MR in thin films of Sr$_{2}$FeMoO$_{6}$ grown by
pulsed laser deposition.
Ba$_{2}$FeReO$_{6}$ \cite{sleight} and Ca$_{2}$FeReO$_{6}$ are
double-perovskites whose structure and properties are quite similar to those
of Sr$_{2}$FeMoO$_{6}$. In both materials a valence degeneracy between the
B-site cation occurs, giving rise to the observed metallic and magnetic
properties ; in the SFMO case, the valence-degeneracy is between Fe$^{3+}$+Mo
$^{5+}\rightleftharpoons $ Fe$^{2+}$+Mo$^{6+}$ states, while in the Ba$_{2}$
FeReO$_{6}$ or Ca$_{2}$FeReO$_{6}$ \ case, the degenerate oxidation states
are Fe$^{3+}$+Re$^{5+}$ $\rightleftharpoons \ $Fe$^{2+}$+Re$^{6+}$. A major
difference between the Mo and Re oxides however, is that Mo$^{5+}$ is $
4d^{1} $, whereas Re$^{5+}$ is $5d^{2}$. This would mean that the conducting
charge carrier density of the Re compound would be twice as much as in the
compound Mo, while the localized spins centered on Fe remain the same in
both oxides. We believe that this difference could have an influence on the
magnetotransport behavior, especially in view of the recent report \cite
{majumdar} that the bulk low field MR in ferromagnetic metals is mainly
determined by the charge carrier density. In view of the foregoing, we
consider it important to investigate the magnetic and transport properties
of A$_{2}$FeReO$_{6}$ (A=Ba and Ca) and our results are reported in this
communication.
\section{Results and discussion}
Polycrystalline samples of A$_{2}$FeReO$_{6}$ (with\ A=Ba or Ca) were
synthesized by standard solid state methods. First, a precursor oxide of the
composition A$_{2}$ReO$_{5.5}$ (A=Ba or Ca) was prepared by reacting
stoichiometric amounts of ACO$_{3}$ and Re$_{2}$O$_{7}$ at 1000 ${^{\circ }}$
C in air for 2 h. Second, this resultant oxide was mixed with required
quantities of Fe$_{2}$O$_{3}$ and Fe powder to obtain the desired
composition A$_{2}$FeReO$_{6}$. Finally, pellets of this mixture were heated
in an evacuated sealed silica tube at 910 ${{}^{\circ }}$C during 4 days,
followed by another treatment at 960 ${{}^{\circ }}$C during the same time,
with intermediate grinding. X-Ray powder diffraction (XRD) patterns were
taken using a conventional diffractometer with Cu K$\alpha $ radiation ($
\lambda =$1.5406 \AA ). The XRD patterns of final compounds are shown in
Fig.1a and Fig.1b. Ba$_{2}$FeReO$_{6}$ is indexed on the basic of a cubic
cell ($Fm3m$) with $a\approx $8.054(1) \AA , similar to Ba$_{2}$YRuO$_{6}$
\cite{Battle}.\ Ca$_{2}$FeReO$_{6}$ is distorted to a monoclinic symmetry ($
P21/n$) with $a\approx $5.396(2) \AA , $b\approx $5.522(2) \AA , $c\approx $
7.688(1) \AA\ and $\beta =90.4{^{\circ }}${, similar to those of La}$_{{2}}${
CuIrO}$_{{6}}${\ and Nd}$_{{2}}${MgTiO}$_{{6}}${\ \cite{Anderson}.} These
results indicate the formation of the double-perovskite (Fig.2) for both
compounds, as reported earlier \cite{sleight} and consistent with recent
results \cite{Kobayashi,Kim}. One also notes that no impurity phase can be
detected in the XRD indicating a clean single phase in each case.
\begin{figure}
\centerline{
\psfig{figure=fig1.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{X-Ray powder Diffraction of double-perovskite (a): Ba$_{2}$FeReO$_6$
,(b): Ca$_{2}$FeReO$_{6}$.}
\end{figure}
\begin{figure}
\centerline{
\psfig{figure=fig2.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{Idealized structure of the ordered double-perovskite.}
\end{figure}
Resistivity ($\rho $) was measured between 5 K and 300 K, on bars with the
approximate dimensions of (1x3x8) mm$^{3}$, using the standard four-probe
method. The temperature (T) dependence of $\rho $ is shown in Fig.3 at
various magnetic fields of 0, 0.2 and 5 T for Ba$_{2}$FeReO$_{6}$. The
resistivity gradually decreases when the temperature decreases suggesting a
metallic behavior below 300 K. The zero-field resistivity was also measured
up to 385 K (inset of Fig.3) but no change in the metallic behavior was
observed in this region. In contrast, Ca$_{2}$FeReO$_{6}$ shows a
semiconducting behavior from 5 K to 300 K (Fig.4). While Ba$_{2}$FeReO$_{6}$
exhibits magnetoresistance at 5 K, the resistivity of Ca$_{2}$FeReO$_{6}$
remains unchanged, even under an applied magnetic field of 8 T. We find,
that the latter resistivity does not fit with an activation energy law ($
\rho (T)\propto e^{\frac{E_{a}}{kT}}$). As often found in the manganites\cite
{Urushibara}, the resistivity at room temperature is large (50 m$\Omega $.cm
for Ba$_{2}$FeReO$_{6}$ and 20 m$\Omega $.cm for Ca$_{2}$FeReO$_{6}$ ), well
above the Mott limit, but perhaps reflecting intergrain resistance in these
polycrystalline samples. A more detailed understanding of the transport of
these materials will require single crystals or oriented films.
To investigate the MR\ of the Ba$_{2}$FeReO$_{6}$, we present the MR at
different temperatures in Fig.5. As shown in Fig.3 , the Ba$_{2}$FeReO$_{6}$
compound exhibits negative MR, with MR defined as $MR(T,H)=\left[
R(H)-R(0)\right] /R(H)$. The MR is equal to 3 \% at 10 K in a 2000 Oe field.
This MR under low field at 10K is smaller than that found in Sr$_{2}$FeMoO$
_{6}$ (10 \%)\cite{Kobayashi}. However, the zero field $\rho (T)$ behavior
of Sr$_{2}$FeMoO$_{6}$ is rather different at low temperature than Ba$_{2}$
FeReO$_{6}$ -\ in Sr$_{2}$FeMoO$_{6}$\ it tends to increase slightly at 10
K. This behavior might be due to the preparation procedure of the sample
which dramatically affects the $\rho (T)$ \cite{Kobayashi}, most probably
this is a result of the difference in the structure of these compounds.
Indeed, in the series Ba$_{2}$FeReO$_{6}$, Sr$_{2}$FeReO$_{6}$ \cite{Abe}
and Ca$_{2}$FeReO$_{6}$, the crystal symmetry decreases from cubic to
tetragonal to orthorhombic (or distorted monoclinic). This is clearly a
manifestation of the decreasing Re-O-Fe bond angle from 180 degrees.
Accordingly, the conduction band width would be expected to decrease as we
go from Ba to Sr to Ca. Thus, it is not surprising that Ba$_{2}$FeReO$_{6}$
is metallic and Ca$_{2}$FeReO$_{6}$ is not. A similar conclusion occurs when
comparing Ba$_{2}$FeMoO$_{6}$ and Sr$_{2}$FeMoO$_{6}$ \cite
{Kobayashi,Maignan} ; Ba$_{2}$FeMoO$_{6}$ is metallic (and cubic) whereas Sr$
_{2}$FeMoO$_{6}$, whose structure is tetragonal, has a resistivity which
increases when T decreases \cite{Kobayashi}. The MR strongly increases at
low field with a slower increase at higher field. This effect occurs mainly
at low temperature, since at room temperature the MR\ is very small. The
features are characteristic of intergrain magnetoresistance \cite{Kim}. At
low temperature (Fig.5), a small hysteretic behavior also appears but thus
is not of relevance to the issues discussed in this paper.
\begin{figure}
\centerline{
\psfig{figure=fig3.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{Resistivity vs temperature under different magnetic fields for Ba$
_{2}$FeReO$_{6}$. The inset shows the zero field dependence of the
resistivity from 5-385 K.}
\end{figure}
\begin{figure}
\centerline{
\psfig{figure=fig4.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{Resistivity vs temperature under zero field for Ca$_{2}$ FeReO$_{6}$
.}
\end{figure}
Magnetization measurements (M)\ were made with a SQUID (MPMS Quantum Design)
magnetometer. These DC measurements were carried out with increasing
temperature after the sample was zero field cooled (ZFC). The temperature
dependence of M for a magnetic field of 10 Oe is shown in Fig.6 for Ba$_{2}$
FeReO$_{6}$. Like several other double-perovskite compounds \cite{Abe,Longo}
, A$_{2}$FeReO$_{6}$ (A=Ba and Ca) exhibit ferrimagnetic behavior due to an
antiferromagnetic superexchange interaction between spins of Fe$^{3+}$ ($
S=5/2$) and Re$^{5+}$ ($S=1$). The ferrimagnetic Curie temperature ($T_{c}$)
of Ba$_{2}$FeReO$_{6}$ was determined as 315 K from the temperature
dependence of the magnetization shown in Fig.6. For Ba$_{2}$FeReO$_{6}$ ,
the low temperature saturation magnetization value, taken under a magnetic
field of 2 T, is 24.9 emu/g, which is close to the expected value based on Fe
$^{3+}$ and Re$^{5+}$ moments (27.3 emu/g). The Ca$_{2}$FeReO$_{6}$ compound
exhibits a higher $T_{c}$, above 385 K (not determined). For the Ca$_{2}$
FeReO$_{6}$ compound, the low temperature saturation magnetization value,
measured under a field of 5 T, is calculated to be 40 emu/g which is a
little higher than the experimental value (30 emu/g) but in agreement with
a previous report \cite{Sleight2}. This is probably due to the fact that the
full saturation magnetization was not reached even with a 5T magnetic field.
The Curie temperature of Ba$_{2}$FeReO$_{6}$ is lower than the reported
value of T$_{c} \approx $410 K for Sr$_{2}$FeReO$_{6}$ \cite{Abe,Longo} but
close to the value of Ba$_{2}$FeMoO$_{6}$ ($T_{c}$=340 K) \cite{Maignan},
whereas the $T_{c}$ of Ca$_{2}$FeReO$_{6}$ seems to be higher ($>$ 385 K).
This means that these differences are likely caused by the larger ionic
radius of Ba$^{2+}$ (1.47 \AA\ versus 1.31 \AA\ for Sr$^{2+}$ and 1.18 \AA\
for Ca$^{2+}$ \cite{Shannon}) which leads to a larger Re-O-Fe (or Mo-O-Fe)
bond length and therefore a smaller exchange and lower Curie temperature.
\begin{figure}
\centerline{
\psfig{figure=fig5.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{Field dependence of the normalized MR at different temperatures for
Ba$_{2}$FeReO$_{6}$.}
\end{figure}
\begin{figure}
\centerline{
\psfig{figure=fig6.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{Magnetization vs temperature under a magnetic field of 10 Oe for Ba$
_{2}$FeReO$_{6}$.}
\end{figure}
The specific heat was measured by relaxation calorimetry in the temperature
range 2-16 K. The specific heat data for Ba$_{2}$FeReO$_{6}$ and Ca$_{2}$
FeReO$_{6}$ are shown in Fig. 7. In our analysis of the low temperature
specific heat we include lattice ($C_{latt}$), metallic ($C_{el}$) and
hyperfine ($C_{hyp}$) contributions. When the temperature decreases below 3
K, the specific heat increases due to the hyperfine contribution $
C_{hyp}=A/T^{2}$. Since the nuclear spin of Fe$^{56}$ is zero, the hyperfine
contribution in our samples arises from the Re$^{186}$ nuclear spin $I=1$.
The experimental values of $A$ are found to be 135$\pm $4 mJ-Kmole and 180$
\pm $5 mJ-K/mole for Ba$_{2}$FeReO$_{6}$ and Ca$_{2}$FeReO$_{6}$
respectively. In addition, Ba$_{2}$FeReO$_{6}$ has the expected metallic
contribution, $\gamma T$, with $\gamma =23.1\pm 0.2$ mJ/mole-K$^{2}$. Using $
\gamma =\pi ^{2}k_{B}^{2}N(E_{F})/3$, we find the density of states at the
Fermi energy $N(E_{F})$ to be 6.1$\times 10^{24}$ eV$^{-1}$mole$^{-1}$. This
value is larger than the $N(E_{F})$ obtained from the band structure
calculation for Sr$_{2}$FeMoO$_{6}$ (1.2$\times 10^{24}$ eV$^{-1}$mole$^{-1}$
)\cite{Kobayashi}, probably because Re$^{5+}$ has two electrons on the $d$
orbital ($5d^{2}$), while Mo$^{5+}$ has only one ($4d^{1}$).
To achieve a good fit for our Ba$_{2}$FeReO$_{6}$ specific heat data two
lattice terms are required: $C_{latt}=\beta _{3}T^{3}+\beta _{5}T^{5}$,
where $\beta _{3}=0.438\pm 0.003$ mJ/mole-K$^{4}$ and $\beta _{5}=5.9\times
10^{-4}$ mJ/mole-K$^{6}$. Since $\Theta _{D}=(12\pi ^{4}pR/5\beta
_{3})^{1/3} $, where p=10 is the number of atoms per formula unit, we find
the Debye temperature $\Theta _{D}$ to be 354 K, which is similar to $\Theta
_{D}$ of other perovskites \cite{Hamilton}.
The best fit for Ca$_{2}$FeReO$_{6}$ does not include $\gamma T$ and $\beta
_{5}T^{5}$ terms, but requires an additional term $C^{\prime }\approx \alpha
T^{a}$, where $a$ is close to 2. The absence of the charge carrier term $
\gamma T$ is consistent with the insulating resistivity of Ca$_{2}$FeReO$
_{6} $. The $C^{\prime }\approx \alpha T^{a}$ term, or more precisely
\begin{figure}
\centerline{
\psfig{figure=fig7.eps,width=7.0cm,height=8.5cm,clip=}
}
\caption{Specific heat of Ba$_{2}$FeReO$_{6}$ and Ca$_{2}$FeReO$_{6}$
plotted as $C/T$ vs $T^{2}$. Lines are the best fit to the form $
C=A/T^{2}+\gamma T+\beta _{3}T^{3}+\beta _{5}T^{5}$ for Ba$_{2}$FeReO$_{6}$
and $C=A/T^{2}+\gamma T+\beta _{3}T^{3}+C^{\prime }$ for Ca$_{2}$FeReO$_{6}$
(see text for values parameters and definition of ${C}^{\prime}$).}
\end{figure}
$C^{\prime }=C^{\prime }(\Delta ,B,T)$, which is the specific heat of
excitations with a dispersion relation $\epsilon =\Delta +Bk^{2}$ of
non-magnetic origin, was also found in the charge-ordered perovskite
manganites \cite{Vera}. In the case of insulating Ca$_{2}$FeReO$_{6}$ a
charge of 3+ on the Fe site and a charge of 5+ on the Re site create a
situation similar to the Mn$^{3+}$-Mn$^{4+}$ charge ordering in manganites.
The fit to our Ca$_{2}$FeReO$_{6}$ specific heat data gives $\Delta $ = 7 K
and $B$ = 22 meV-\AA\ similar in magnitude to the parameters found in the
manganites. We believe that the reasons for the presence of the $C^{\prime
}(\Delta ,B,T)$ term in Ca$_{2}$FeReO$_{6}$ are similar to those in La$
_{0.5} $Ca$_{0.5}$MnO$_{3}$ \cite{Vera}. The lattice contribution $\beta
_{3}T^{3}$ ($\beta _{3}=0.25\pm 0.006$ mJ/mole-K$^{4}$) in Ca$_{2}$FeReO$_{6}
$ is smaller than in Ba$_{2}$FeReO$_{6} $, which can be explained by the
different crystal structure (cubic for Ba$_{2}$FeReO$_{6}$ and distorted
monoclinic for Ca$_{2}$FeReO$_{6}$) and smaller Ca mass.
A ferrimagnet has a magnetic contribution to the specific heat, $
C_{mag}=\delta T^{3/2}$, which is similar to that of a ferromagnetic one.
However, this contribution, can not be resolved from the specific heat data
alone, since our data can be fit well without the magnetic term. Since Ca$
_{2}$FeReO$_{6}$ and Ba$_{2}$FeReO$_{6}$ have a high Curie temperature, and
hence, strong exchange interaction (J), the magnetic term should be small
(since $\delta \propto \frac{1}{J^{3/2}}$).
\section{Conclusion}
In summary, we have investigated the transport, thermal and magnetic
properties of two polycrystalline double-perovskites Ba$_{2}$FeReO$_{6}$ and
Ca$_{2}$FeReO$_{6}$. Ba$_{2}$FeReO$_{6}$ displays a metallic behavior below
385 K whereas Ca$_{2}$FeREO$_{6}$ is insulating below this temperature. The
specific heat of Ba$_{2}$FeReO$_{6}$ gives a low temperature metallic
contribution with an electron density of states at the Fermi level close to
the band structure value. Insulating Ca$_{2}$FeReO$_{6}$ has no metallic
term in the specific heat but rather an extra contribution most likely
caused by charge ordering of Fe, Re. The Ba$_{2}$FeReO$_{6}$ compound
exhibits a negative MR at 10K, smaller than the analogous compound Sr$_{2}$
FeMoO$_{6}$. Magnetic measurements indicate a ferrimagnetic behavior,with a $
T_{c}=315K$ for Ba$_{2}$FeReO$_{6}$ and above 385 K for Ca$_{2}$FeReO$_{6}$
. These data have been explained and compared with the analogous compounds Sr
$_{2}$Fe(Re,Mo)O$_{6}$.
\acknowledgments{\ Partial support of NSF-MRSEC at University of Maryland is
acknowledged. The work at Bangalore was supported by the Department of
Science and Technology, Government of India. K.R. thanks the Council of
Scientific and Industrial Research, New Delhi, for the award of a fellowship.
}
|
\section{Introduction}
Undoubtedly, we can say that this decade is a further, permanent progress
in experimental high energy particle physics. Let's modestly mention
LEP achievements \cite{lep} as well as top discovery \cite{top}. However,
not less
impressive results in low energy physics have been obtained. Especially much
has been done in neutral current physics, where data coming from both
deep inelastic neutrino-hadron, neutrino-electron scattering as well as
electron-hadron interactions have been enriched lately by exquisitely
precise measurements
of parity nonconservation (PNC) in heavy atoms, such as cesium \cite{wood}
and thalium \cite{thal}. The CCFR collaboration data on quark-Z boson
couplings has also improved \cite{ccfr}.
This kind of experiments is a nice (and not too
expensive regarding high energy collider physics) tool to
probe the standard electroweak model (SM)
and its parameters. Moreover, searches
beyond the standard model physics using low-energy data
complement quite well the efforts made at high energy colliders.
To visualize this statement we use in this work, as a representative
for `new physics', the classical Manifest Left-Right Symmetric (MLRS)
model. Its principle advantage over the SM is space inversion invariance at
high energies, implied not only by the gauge group but also by a discrete
symmetry (replacement of the left by the right fields and {\it vice versa}).
As a consequence the
left and right couplings $g_L$, $g_R$ are equal,
$g_L=g_R=g$, and the Yukawa matrices in the quark and lepton sectors are
hermitian. A minimal Higgs sector with a bidoublet $\Phi$, and
left $\Delta_L$ and right $\Delta_R$
triplets is adopted \cite{lr,gun}, with the additional assumption that
only $\Phi$ and $\Delta_R$ have non-vanishing VEVs\footnote{
$$<\Phi>=\frac{1}{\sqrt{2}} \left( \matrix{\kappa_1 & 0 \cr
0 & \kappa_2} \right),\;
<\Delta_R>=\frac{1}{\sqrt{2}} \left( \matrix{0 & 0 \cr
v_R & 0} \right),\;<\Delta_L>=0.$$
}.
In this model we have four non-standard parameters, namely, additional
gauge boson masses $M_{W_2}$, $M_{Z_2}$ and mixing angles in both charged
and neutral gauge sectors ($\zeta$, $\phi$). We use them to
parametrize the mentioned low-energy neutral data.
These parameters are not
independent of each other, as they are functions
of the VEVs of $\Phi$ and $\Delta_{R}$. Based on this fact
we end up with two independent factors
$\gamma=M_{Z_1}^2/M_{Z_2}^2$ and $\epsilon=2 \kappa_1
\kappa_2/(\kappa_1^2+\kappa_2^2)$.
This way, exploiting the full strength of the MLRS enabled us
to obtain quite impressive limits on the $M_{Z_2}$
mass (much above 1 TeV).
The results are better
than those from direct searches
at high energy hadron colliders and comparable to those extracted from
the LEPI data.
All numerics are done with the CERN code MINUIT \cite{min}.
\section{$M_{Z_2}$ mass and low-energy neutral current experiments}
The low energy processes' momentum transfer being much smaller than the
intermediate gauge boson masses, contact four-fermion Lagrangians
can be effectively used.
Four-fermion
neutrino-hadron $(\nu N)$,
neutrino-electron $(\nu e)$ and parity-violating
electron-hadron $(eN)$ interactions can be written in the conventional form as
follows \cite{98}
\begin{eqnarray}
L^{\nu N} & = &- \frac{G_F}{\sqrt{2}} \bar{\nu} \gamma^{\mu}(1-\gamma_5)
\nu \sum_{i=u,d} [ \epsilon_L(i) \bar{q}_i \gamma_{\mu}(1-\gamma_5)q_i
+ \epsilon_R(i) \bar{q}_i \gamma_{\mu}(1+\gamma_5)q_i], \nonumber \\
&& \\
L^{\nu e} & = &- \frac{G_F}{\sqrt{2}} \bar{\nu} \gamma^{\mu}(1-\gamma_5)
\nu \bar{e}_i \gamma_{\mu}(g_V^{\nu e}-g_A^{\nu e} \gamma_5)e, \\
L^{\nu N} & = & \frac{G_F}{\sqrt{2}}
\sum_{i=u,d} [ C_{1i} \bar{e} \gamma_{\mu}\gamma_5 e
\bar{q}_i \gamma_{\mu} q_i + C_{2i} \bar{e} \gamma_{\mu} e
\bar{q}_i \gamma_{\mu}\gamma_5 q_i].
\end{eqnarray}
Here,
$\epsilon_{L,R}(i),g_{V,A}^{\nu e}, C_{ij}$ are model-dependent coefficients.
It is usual to consider only pure left-handed currents in $L^{\nu e}$
and $L^{\nu N}$. In the SM they can be derived by comparison with the neutral
current Lagrangian ( $ T^L_{3i} $ and $Q_i$
are the weak isospin of fermion i and its charge, respectively)
\begin{equation}
L^{SM}_{NC}=\frac{g}{2 \cos{\Theta_W}} \sum_{u,d,\nu,e}
\bar{\Psi_i} \gamma^{\mu} \left( g_V^i-g_A^i \gamma_5 \right) \Psi_i Z_{\mu}
\end{equation}
with
\begin{eqnarray}
g_V^i & \equiv & T^L_{3i}-2Q_i \sin^2{\Theta_W}, \\
g_A^i & \equiv & T^L_{3i}.
\end{eqnarray}
Using The SM definition
$\frac{G_F}{\sqrt{2}}=
\frac{\pi \alpha}{2 \sin^2{\Theta_W} (1-\Delta r) M_{W_1}^2}$, we get
\begin{eqnarray}
\epsilon_{L}^{SM}(i) & = & \rho_{\nu N} ( T_{3i}-Q_i \kappa_{\nu N}
\sin^2{\Theta_W})+\lambda_{iL}, \\
\epsilon_{R}^{SM}(i) & = & \rho_{\nu N} ( -Q_i \kappa_{\nu N}
\sin^2{\Theta_W})+\lambda_{iR}, \\
C_{1i}^{SM} & = & \rho_{eq}' ( -T_{3i}+2Q_i \kappa_{eq}'
\sin^2{\Theta_W})+\lambda_{1i}, \\
C_{2i}^{SM} & = & \rho_{eq} ( -1/2+2 \kappa_{eq}
\sin^2{\Theta_W}) (2 T_{3i})+\lambda_{2i}, \\
{(g_V^{\nu e})}^{SM} & = & \rho_{\nu e} ( -1/2+ \kappa_{\nu e}
\sin^2{\Theta_W}), \\
{(g_A^{\nu e})}^{SM} & = & \rho_{\nu e} ( -1/2).
\end{eqnarray}
The $\rho,\lambda$ and $\kappa$ factors include the effects
of one-loop radiative corrections to the low energy processes
\cite{98}. At tree level $\rho=\kappa=1$ and $\lambda=0$.
Now let us procede to the left-right symmetric model.
The masses of the gauge bosons $(M_{Z_{1,2}},M_{W_{1,2}})$ and the
mixing angles
$\zeta, \phi$ in charged and neutral gauge sectors are the following
($\kappa_+=\sqrt{\kappa_1^2+\kappa_2^2}$) \cite{pol,nasze}
\begin{eqnarray}
M^2_{W_{1,2}} &=& \frac{g^2}{4} \left[
\kappa^2_+ +v_R^2 \mp \sqrt{v_R^4+4 \kappa^2_1\kappa^2_2} \right], \\
M^2_{Z_{1,2}} &=& \frac{1}{4} \left\{ \left[
g^2 \kappa^2_+ +2v_R^2 \left( g^2+g'^2 \right) \right] \right. \nonumber \\
& \mp & \left. \sqrt{
\left[ g^2 \kappa^2_+ +2v_R^2 \left( g^2+g'^2 \right) \right]^2- 4g^2
\left( g^2+2g'^2 \right) \kappa^2_+ v_R^2 } \right\}, \\
\tan{2\xi}&=&-\frac{2\kappa_1 \kappa_2}{v_R^2}, \\
\sin{2\phi}&=&
-\frac{g^2 \kappa^2_+ \sqrt{\cos{2\Theta_W}}}{2\cos^2{\Theta_W}
\left( M^2_{Z_2}-M^2_{Z_1} \right)}.
\end{eqnarray}
Obviously these are functions of the three VEVs $v_R$, $\kappa_1,\kappa_2$.
Certainly, $M_{W(Z)_1} < M_{W(Z)_2}$, so $\kappa_+^2,\kappa_1 \kappa_2
<< v_R^2$ and
we expand the above formulas leaving terms up to $O\left(
(\frac{\kappa_+}{v_R})^2,
(\frac{\kappa_1 \kappa_2}{v_R^2}) \right)$
(we will comment on this approximation at the end of the Chapter).
The result can be cast in the form
$\left( \beta=\frac{M_{W_1}^2}{M_{W_2}^2} \right) $
\cite{pol,nasze,mas}
\begin{eqnarray}
\gamma \equiv \frac{M_{Z_1}^2}{M_{Z_2}^2} & = & \frac{\cos{2 \Theta_W}}
{2 \cos^4{\Theta_W}} \beta , \\
\zeta & = & -\epsilon \beta, \\
\phi & = & -\frac{(\cos{2\Theta_W})^{3/2}}{2 \cos^4{\Theta_W}} \beta , \\
\mbox{\rm and} && \nonumber \\
\rho_{LR} \equiv \frac{M_{W_1}^2}{M_{Z_1}^2 \cos^2{\Theta_W}} & = &
1+\left[-\epsilon^2+\frac{1}{2} (1-\tan^2{\Theta_W})^2 \right] \beta ,
\end{eqnarray}
where
\begin{equation}
\epsilon = \frac{2 \kappa_1 \kappa_2}{\kappa_1^2+\kappa_2^2}, \;\;\;
0 \leq \epsilon \leq 1.
\end{equation}
In the MLRS model the neutral current interaction for any fermions can
be written
\begin{equation}
L_{NC}=\frac{e}{2\sin{\Theta_W}\cos{\Theta_W}}
\sum\limits_{i=up,\;down,l,\nu }\sum\limits_{j=1,2} \bar{\psi}_i\gamma^\mu
\left[
A^{ji}_L \Omega_L^{i}P_L + A^{ji}_R \Omega_R^{i} P_R
\right]
\psi_i Z_{j\mu}.
\end{equation}
The couplings $A^{1,2;\;i}_{L,R}$ are given by
\begin{eqnarray}
A_L^{1i} & = & \cos{\phi}\;g^i_L+\sin{\phi}\;g_L^{\prime i},\\
A_R^{1i} & = & \cos{\phi}\;g^i_R+\sin{\phi}\;g_R^{\prime i},\\
A_L^{2i} & = & \sin{\phi}\;g^i_L-\cos{\phi}\;g_L^{\prime i},\\
A_R^{2i} & = & \sin{\phi}\;g^i_R-\cos{\phi}\;g_R^{\prime i},
\end{eqnarray}
where
\begin{eqnarray}
g^i_L & = & 2 T^L_{3i}-2Q_i\sin^2\Theta_W , \\
g_L^{\prime i} & = & \frac{2\sin^2\Theta_W }{\sqrt{\cos 2\Theta_W }}
\left( Q_i-T^L_{3i}\right), \\
g^i_R & = & -2 Q_i \sin^2 \Theta_W , \\
g_R^{\prime i} & = & \frac{2}{\sqrt{\cos 2\Theta_W }}
\left( Q_i \sin^2 \Theta_W - T^R_{3i}\cos^2 \Theta_W \right).
\end{eqnarray}
$\Omega_{L,R}$ are analogous to Cabbibo-Kobayashi mixing matrices in the
charged sector and are the identity matrices for charged fermions
$$\Omega_{L,R}^{i}=I\;\;\; \mbox{\rm for}\;\; i=u,d,l.$$
To have a link with the model independent Lagrangians (Eq.(1-3))
we now assume that only
pure left-handed neutrinos play a role in neutral low energy
physics\footnote{If it was not true then we would certainly have had an
indication on the long standing Dirac-Majorana neutrino nature problem
\cite{zral}.}.
Then $\Omega_L^{\nu} \simeq I, \Omega_R^{\nu} \simeq 0$.
We can now, quite analogously to the SM case, find low energy
LR model coefficients
\begin{eqnarray}
\epsilon_{L,R}^{LR}(i) & = & \Lambda ( A_L^{1 \nu}
A_{L,R}^{1i}+ \gamma A_L^{2\nu}A_{L,R}^{2i} ), \\
C_{1i}^{LR} & = & \Lambda(g_A^{1l} g_V^{1i} + \gamma g_A^{2l} g_V^{2i} ), \\
C_{2i}^{LR} & = & \Lambda(g_V^{1l} g_A^{1i} + \gamma g_V^{2l} g_A^{2i} ), \\
{(g_V^{\nu e})}^{LR} & = & \Lambda(A^{1\nu}_L g_V^{1l}
+ \gamma A^{2 \nu}_L g_V^{2l} ), \\
{(g_A^{\nu e})}^{LR} & = & \Lambda(A^{1\nu}_L g_A^{1l}
+ \gamma A^{2 \nu}_L g_A^{2l} ),
\end{eqnarray}
where
\begin{eqnarray}
g_{V,A}^i & = & \frac{1}{2} \left( A_L^{i \nu} \pm A_R^{i \nu} \right), \\
\Lambda &=& \frac{\rho_{LR}}{ \left( \cos^2{\zeta}+\beta \sin^2{\zeta}
\right)} .
\end{eqnarray}
The $\Lambda$ factor is connected with the L-R definition of the $G_F$
constant. If we assume the only natural situation when right-handed neutrinos
are too heavy to be directly produced in the muon decay and the light are
left-handed (negligible right-handed admixture) then
the $G_F$ definition follows \cite{cz}
\begin{equation}
\frac{G_F}{\sqrt{2}}=
\frac{\pi \alpha}{2 \sin^2{\Theta_W} M_{W_1}^2 (1-\Delta r)}
\left( \cos^2{\zeta}+\beta \sin^2{\zeta} \right) .
\end{equation}
The parameters of Eqs. (31-35) are
`bare', reproducing the 'bare' SM couplings for
$\phi=\gamma=0$.
We improve them by adding the SM corrections
(Eqs.(7)-(12)), so the data is fitted with
\begin{eqnarray}
\epsilon_{L,R}^{LR}(i) & = & \Lambda \left[ A_L^{1 \nu} \left
( \cos{\phi}
\epsilon_{L,R}^{SM}(i) + \frac{1}{2}\sin{\phi}
g'^{i}_{L,R} \right) + \frac{1}{2} \gamma A_L^{2 \nu} A_{L,R}^{2i} \right], \\
C_{1i}^{LR} & = & \Lambda \left[ \cos{2 \phi}- \sin{2 \phi}
\frac{\sin^2{\Theta_W}}{\sqrt{\cos{2 \Theta_W}}} \right]
\left[ C_{1i}^{SM}- \gamma \left( -T_{3i}+2Q_i \sin^2{\Theta_W} \right)
\right], \\
C_{2i}^{LR} & = & \Lambda \left[ \cos{2 \phi}- \sin{2 \phi}
\frac{\sin^2{\Theta_W}}{\sqrt{\cos{2 \Theta_W}}} \right]
\left[ C_{2i}^{SM}- \gamma \left( -\frac{1}{2}+2 \sin^2{\Theta_W} \right)
(2T_{3i}) \right] \nonumber \\
&& \\
{(g_V^{\nu e})}^{LR} & = & \Lambda \left[
A_L^{1 \nu} \left( \cos{\phi}-\frac{\sin{\phi}}{\sqrt{\cos{2\Theta_W}}}
\right)
{(g_V^{\nu e})}^{SM} \right. \nonumber \\
&+& \left. \gamma A_L^{2 \nu} \left( \sin{\phi}+\frac{\cos{\phi}}
{\sqrt{\cos{2\Theta_W}}} \right) \left( -\frac{1}{2}+2 \sin^2{\Theta_W}
\right) \right] ,\\
{(g_A^{\nu e})}^{LR} & = & \Lambda
\left[ A_L^{1 \nu} ( \cos{\phi}+\sin{\phi}\sqrt{\cos{2\Theta_W}})
{(g_A^{\nu e})}^{SM} \right. \nonumber \\
&+& \left. \gamma A_L^{2 \nu} \left( \sin{\phi}-\cos{\phi}
\sqrt{\cos{2\Theta_W}} \right) \left( -\frac{1}{2} \right) \right].
\end{eqnarray}
In Table 1 we show the 1998 data \cite{98} for all the couplings
that are used.
As the Standard Model one-loop corrections to these
theoretical formulas (Eqs.(7)-(12)) have been calculated in the
$\overline{MS}$ scheme, we take the value of $\sin^2{\Theta_W}$ in the same
scheme
$\sin^2{\Theta_W} \equiv \hat{s}_Z^2 =0.23124 \pm 0.00017$ \cite{98}.
In Fig.1 we show the 90 \% C.L. allowed region for
$\gamma-\phi$ parameters. The dotted line shows the results for the
data from Table 1. The solid line follows from supplementing the
previous with an
additional parameter which measures the neutral to charged current cross
section ratio in neutrino scattering off nuclei
and is given by the CCFR collaboration \cite{ccfr}.
This parameter in the frame of our model should be defined in the following
way
\begin{equation}
\kappa^2= 1.7897 g_L^2+1.1479 g_R^2-0.0916 \delta_L^2-0.0782 \delta_R^2
\end{equation}
where
\begin{eqnarray}
g_{L,R}^2& = & \left( \epsilon_{L,R}^{LR}(u) \right)^2+ \left(
\epsilon_{L,R}^{LR}(d) \right)^2, \\
\delta_{L,R}^2& = & \left( \epsilon_{L,R}^{LR}(u) \right)^2- \left(
\epsilon_{L,R}^{LR}(d) \right)^2.
\end{eqnarray}
The CCFR collaboration has found it to be
\begin{equation}
\kappa^2=0.5820 \pm 0.0041.
\end{equation}
We can see that $\kappa^2$ does not change the predictions for the $\gamma$
parameter. Using Eq.(17) we get $M_{Z_2} \geq$ 410 GeV. This result is
not substantially different from other analyses \cite{pol,chay}.
\begin{table}[h]
\centering
\begin{tabular}{c c c}
\cline{1-3}
& \ Experimental Value & \ Correlations \\
\cline{1-3}
$\epsilon_L(u)$ & $ 0.328 \pm 0.0016$ & \\
$\epsilon_L(d)$ & $ -0.440 \pm 0.011$ & non- \\
$\epsilon_R(u)$ & $ -0.179 \pm 0.0013$ & Gaussian \\
$\epsilon_R(d)$ & $ -0.027 \matrix{ +0.077 \cr -0.048}$ & \\
\cline{1-3}
$g_L^2$ & $0.3009 \pm 0.0028$ & \\
$g_R^2$ & $0.0328 \pm 0.003$ & \\
$\Theta_L$ & $2.50 \pm 0.035$ & small \\
$\Theta_R $ & $4.56 \matrix{ +0.42 \cr -0.27}$ & \\
\cline{1-3}
$g_V^{\nu e}$ & $-0.041 \pm 0.015$ & $-0.04$ \\
$g_A^{\nu e}$ & $-0.507 \pm 0.014$ & \\
\cline{1-3}
$C_{1u}$ & $-0.216 \pm 0.046$ & $-0.997$ $-0.78$ \\
$C_{1d}$ & $0.361 \pm 0.041$ & 0.78 \\
$C_{2u}-\frac{1}{2}C_{2d}$ & $-0.03 \pm 0.12$ & \\
\cline{1-3}
\end{tabular}
\caption{\footnotesize{
Data used for the neutral data analysis \cite{98}. Appropriate formulas are
given in the text. $g_{L,R}^2=\epsilon^2_{L,R}(u)-\epsilon^2_{L,R}(d)$,
$\tan{\Theta_{L,R}}=\frac{\epsilon_{L,R}(u)}{\epsilon_{L,R}(d)}$}}
\end{table}
Until now the left-right observables
$\beta,\gamma,\zeta,\phi$ have been treated as independent, i.e.
we do not take into account the relations (13)-(16)
which reflect the fact that the VEVs link them to one another.
When we use these relations the situation changes substantially (Fig2).
We have chosen as independent two phenomenologically handful
parameters: $\epsilon$ and $\gamma$.
Different $\epsilon$'s describe left-right models with different bidoublet
VEVs $\kappa_1,\kappa_2$. We know from
phenomenological considerations that the reduction of FCNC favors
left-right models with
$\epsilon \simeq 0$ \cite{gun} (ellipses denoted with (b)).
Then (see Eq.(18)) there is no $W_L-W_R$ mixing.
However, to make this possibility open we also
show the results for left-right models with $\epsilon \simeq 1$ (
ellipses denoted with (a)).
The dotted ellipses are obtained when all of the data from Table 1 is taken
into account.
The solid ones correspond to the inclusion of the
$\kappa^2$ Eq.(47) parameter.
Fig.2 shows that the MLRS relations among the
fitted parameters $\beta,\gamma,\phi,\zeta$, and the CCFR data
make it possible to shrink considerably the allowed
space for the $\gamma$ factor. From Eqs. (17)-(20) it is possible to find
limits on the rest of the
left-right parameters:
\begin{eqnarray*}
\left. \matrix{
M_{Z_2} \geq 1475\; \mbox{\rm GeV} \cr
M_{W_2} \geq 875\; \mbox{\rm GeV} \cr
| \phi | \leq 0.0028\; \mbox{\rm rad}} \right\}
&& \mbox{\rm for models without }\; W_L-W_R\; \mbox{\rm mixing}, \\
&& \\
\left. \matrix{
M_{Z_2} \geq 1205\; \mbox{\rm GeV} \cr
M_{W_2} \geq 715\; \mbox{\rm GeV} \cr
| \zeta | \leq 0.013\; \mbox{\rm rad} \cr
| \phi | \leq 0.0042\; \mbox{\rm rad}
} \right\}
&& \mbox{\rm for models with possible}
W_L-W_R \; \mbox{\rm mixing} \; (\epsilon \neq 0)
\end{eqnarray*}
These results are comparable with previous LEPI analyses ($M_{Z_2}
\geq 0.8 \div 1.5$ TeV) \cite{lep1}
and better than that which follow from direct searches for additional
gauge bosons in hadron
colliders ($M_{Z_2} \geq 630$ GeV) \cite{hc}.
Finally, let us comment on the approximation made
in Eqs.(17)-(20). Our fitted observables (Eq.(36)-(41))
are functions of $\beta,\gamma$
(so $M_{W_2},M_{Z_2}$) and mixing angles $\zeta,\phi$.
Taking into account exact formulas (13)-(16) for these quantities we
have checked that, at 90 \% C.L., quantities of the form
$(\kappa_+/v_R)^2$ and $\kappa_1 \kappa_2/v_R^2$ do not exceed
at the worst case 0.04 so, when neglecting squares of them,
relations (17)-(20) are quite reliable.
\section{Conclusions}
We point out the importance of examining non-standard
models using relations
among physical parameters such as masses and mixings that follow
from the Higgs sector.
In our analysis we used the most up to date low energy experimental data,
including the CCFR $\kappa^2$.
Thanks to these we obtained
new limits on {left-right symmetric model } parameters that are
comparable with those from high energy physics.
\begin{ack}
This work was supported by Polish Committee for Scientific Research under
Grants Nos. 2P03B08414 and 2P03B04215.
J.G. appreciates also the support of the Humboldt-Stiftung.
\end{ack}
|
\section{}'.
Observation of mesoscopic quantum phenomena in magnetism
has remained a challenging problem. The first striking demonstration of
quantum tunneling and quantum phase interference was found on
Mn$_{12}$ acetate and Fe$_8$, molecular clusters having a spin ground state $S = 10$
\cite{Sessoli93, Friedman96, Sangregorio97, WW_RS99}. Several models and theories have been proposed to explain
in detail the experimental results, published during the last five years by several authors
\cite{theories}, but there is not yet satisfactory agreement between theory and
experiments concerning mainly the relaxation rate and the resonance line width
\cite{Friedman98}. This letter is intended to report more accurate measurements which
should help to find a satisfactory explanation of the observed quantum phenomena.
Several authors have pointed out that in the Mn$_{12}$ carboxylate family
different isomeric forms give rise to different relaxation rates \cite{Sun98}.
This has also been observed in Mn$_{12}$ acetate \cite{Aubin97}.
We found that a minor species Mn$_{12}$ac(2) \cite{remark1}, randomly
distributed in the crystal of the major species Mn$_{12}$ac(1), exhibits a faster
relaxation rate which becomes temperature independent below 0.3 K. Even if this
second species has been only partially characterised \cite{remark1} we can exploit
it as a local probe providing unique information on the tunnelling process. We
used a recently developed method \cite{Ohm98,Wernsdorfer99} for measuring the
intrinsic line width broadening due to local fluctuating fields and found strong evidence
for the influence of nuclear spins on resonance tunnelling.
In the first part of this letter, we focus on the low temperature and low field limit
which is particularly interesting because phonon-mediated relaxation is astronomically
long and can therefore be neglected. In this limit, only the two lowest levels with quantum numbers
$M_z$~=$\pm$10 are involved. They are coupled by a tunnel matrix element $\Delta/2$
where $\Delta$ is the tunnel splitting which is estimated to be about 10$^{-10}$~K for
Mn$_{12}$ \cite{Prok98}. In an ideal system, resonant tunnelling requires that the
magnetic field (local to the giant spin) is smaller than the field associated with the tunnel
splitting $\Delta$ which means a field smaller than 10$^{-9}$~T for the Mn$_{12}$ac
clusters. This fact would make it very difficult to observe the tunnelling of isolated giant
spins at constant applied field. This dilemma is solved by invoking the dynamics of dipolar interaction between
molecules and nuclear spins \cite{Prok98}. The tunnelling scenario can be summarised as
follows: the rapidly fluctuating hyperfine field brings molecules into resonance. The
dipolar field of tunnelled spins can lift the degeneracy and remove from resonance a large
number of neighbouring spins. However, a gradual adjustment of the dipolar fields
across the sample (up to 0.03 T in Mn$_{12}$ac), caused by tunnelling relaxation,
brings other molecules into resonance and allows continuous relaxation. Therefore, one
expects a fast relaxation at short times, and slow logarithmic relaxation at long times.
\begin{figure}[t]
\centerline{\epsfxsize=6 cm \epsfbox{fig1.eps}}
\caption{Magnetic hysteresis curves of the minor species Mn$_{12}$ac(2) \cite{remark1} at several temperature. Before the measurement, the major species Mn$_{12}$ac(1) was
demagnetised \cite{remark2}. Resonant tunnelling is evidenced by four steps. Notice that the loops are about temperature independent below 0.6 K. $M_S$ is the magnetisation of
saturation of the whole crystal.}
\label{fig1}
\end{figure}
Recently, we developed a method \cite{Ohm98,Wernsdorfer99} for measuring
the intrinsic line width broadening due to local fluctuating fields of the nuclear spins. It is
based on the general idea that the short time relaxation rate is directly connected to the
number of molecules which are in resonance at a given longitudinal applied field $H$.
The Prokof'ev - Stamp theory \cite{Prok98} predicts that the magnetisation should
relax at short times with a square-root time dependence:
\begin{equation}
M(H,t)=M_{\rm in}+(M_{\rm eq}(H)-M_{\rm in})\sqrt{\Gamma_{\rm
sqrt}(H)t} \label{eq1}
\end{equation}
Here $M_{\rm in}$ is the initial magnetisation at time $t$~= 0 (i.e. after a
rapid field change), and $M_{\rm eq}(H)$ is the equilibrium
magnetisation. The rate function $\Gamma_{\rm sqrt}(H)$ is
proportional to the normalised distribution $P(H)$ of molecules which are in resonance at
the applied field H:
\begin{equation}
\Gamma_{\rm sqrt}(H)\sim\frac{\Delta^2}{\hbar}P(H)
\label{eq2}
\end{equation}
where $\hbar$ is Planck's constant. Thus, the measurements of $\Gamma_{\rm
sqrt}(H)$, as a function of $H$, should give direct access to the distribution $P(H)$.
Our measuring procedure is as follows. Starting from a well defined magnetisation state
\cite{remark2}, we apply a magnetic field $H$ in order to measure the short-time square
root relaxation behaviour. By using eq. (1), we get the rate function $\Gamma_{\rm
sqrt}(H)$ at the field $H$. Then, starting again from the same well defined magnetisation
state, we measure $\Gamma_{\rm sqrt}(H)$ at another field $H$, yielding the field
dependence of $\Gamma_{\rm sqrt}(H)$ which is proportional to the dipolar distribution
$P(H)$ (eq. (2)).
\begin{figure}[t]
\centerline{\epsfxsize=6 cm \epsfbox{fig2.eps}}
\caption{Typical relaxation measurements of the minor species Mn$_{12}$ac(2) \cite{remark1} measured at H = 0.39 T and for several temperatures. For each curve, the major species
Mn$_{12}$ac(1) was demagnetised \cite{remark2} and Mn$_{12}$ac(2) was saturated in a field of -1.4 T. The $\Gamma_{\rm sqrt}$ relaxation rate was approximately found in the range from $0.014 > M/M_S > 0.01$, i.e. in the short time region. $M_S$ is the magnetisation of saturation of the whole crystal.}
\label{fig2}
\end{figure}
This technique can be used for following the time evolution of molecular states in
the sample during a tunnelling relaxation \cite{Wernsdorfer99}. Starting from a well
defined magnetisation state \cite{remark2}, and after applying a field $H_{dig}$, we let
the sample relax for a time $t_{dig}$, called 'digging field and digging time',
respectively. During the digging time, a small fraction of the molecular spins tunnel and
reverse the direction of their magnetisation. Finally, we apply a field $H$ to measure the
short time relaxation in order to get $\Gamma_{\rm sqrt}(H)$ (eq. (1)). The entire
procedure is then repeated to probe the distribution at other fields $H$ yielding
$\Gamma_{\rm sqrt}(H,H_{dig},t_{dig})$ which is proportional to the number of spins
which are still free for tunnelling. With this procedure one obtains the distribution
$P(H,H_{dig},t_{dig})$, which we call the 'tunnelling distribution'.
We used this new technique, which we call 'hole digging' method \cite{remark3},
for studying Fe$_8$ molecular clusters \cite{Wernsdorfer99} and found that tunnelling
causes rapid transitions of molecules near $H_{dig}$, thereby "digging a hole" in
$P(H,H_{dig},t_{dig})$ around $H_{dig}$, and also pushing other molecules away
from resonance. The hole widens and moves with time, in a way depending on sample
shape; the width dramatically depends on thermal annealing of the magnetisation of the
sample. For small initial magnetisation \cite{remark2}, the hole width shows an intrinsic
broadening which may be due to nuclear spins \cite{Wernsdorfer99}.
\begin{figure}[t]
\centerline{\epsfxsize=6 cm \epsfbox{fig3a.eps}}
\centerline{\epsfxsize=6 cm \epsfbox{fig3b.eps}}
\caption{(a): Quantum hole digging in an initial dipolar distribution of the minor species
Mn$_{12}$ac (2). (b): Difference between the initial dipolar distribution $\Gamma_{\rm init}$ and quantum tunnel distribution $\Gamma_{\rm dig}$ of Fig 3(a). The continuos lines are Gaussian functions fitted to the
data yielding the hole line width $\sigma$.}
\label{fig3}
\end{figure}
For Mn$_{12}$ac(1) \cite{remark1}, T $<$ 1.5K and small applied fields, relaxation
measurements are very time consuming because the pure quantum relaxation rate between
$M_z$~=$\pm$10 is of the order of years or longer \cite{Wernsdorfer96}. However, we
found that one can use the minor species Mn$_{12}$ac(2) \cite{remark1} which showed
to have a much faster tunnelling rate. Furthermore, it has the advantage of being diluted
over the entire crystals with a concentration of 1 to 2 percent, thus the internal dipolar
fields hardly change during relaxation of Mn$_{12}$ac(2). As Mn$_{12}$ac(2)
experiences the same environment (concerning mainly hyperfine fields) as the major
species Mn$_{12}$ac(1), we propose to use Mn$_{12}$ac(2) as a local probe of any
fluctuating field acting on the giant spins of Mn$_{12}$ac molecular clusters.
Below about 1.5 K, we found that the magnetisation of Mn$_{12}$ac(2) can be
reversed in an applied field smaller than 2 T whereas that of Mn$_{12}$ac(1) hardly
reverses because of the very small tunnelling rate. Fig. 1 presents a typical hysteresis loop
measurements of Mn$_{12}$ac(2) which is almost temperature independent below 0.6 K
\cite{remark4}. These loops are strongly field sweeping rate dependent and show
quantum tunnelling resonances at about equidistant fields of $\Delta$H $\approx$ 0.39 T in comparison to 0.45 T for Mn$_{12}$ac(1).
Fig. 2 presents typical relaxation measurements of the minor species
Mn$_{12}$ac(2). For each curve, the major species Mn$_{12}$ac(1) was demagnetised
\cite{remark2} and Mn$_{12}$ac(2) was saturated in a field of -1.4 T. Approximate square
root relaxation was found in the range from $0.014 > M/M_S > 0.01$, where $M_S$ is the
magnetisation of saturation of the entire crystal. The fact that the relaxation is not exactly
proportional to $\sqrt{t}$ in the short time region, is irrelevant for the discussion of this letter \cite{remark5}.
Fig. 3a presents tunnelling distributions for Mn$_{12}$ac(2) for digging times
between t$_0$ = 0 and 128 s. Note the depletion ("hole digging") around the digging field $H_{dig}$ = 0.39 T. This hole-digging arises because only spins in resonance can
tunnel. Although the hole is narrow, it is still several orders of magnitude larger than the
field associated with the tunnel splitting $\Delta$. The hole could be fitted to a Gaussian
function yielding the line width $\sigma$ (see Fig. 3b) which we studied as a function of
temperature and digging time (fig. 4). We defined an intrinsic line width $\sigma_0$ by a linear
extrapolation of the curves to $t_{dig}$ = 0. For temperatures between 0.04 and 0.3K, $\sigma_0$ $\approx$ 12 mT. For T $>$ 0.3 K, $\sigma_0$ increase rapidly.
The physical origin of the line width $\sigma_0$ is tentatively assigned to the fluctuating
hyperfine fields. A simple calculation of the random hyperfine field distribution was made
by Hartmann-Boutron {\it et al.} \cite{Hartmann96}, who evaluated the maximum nuclear field
operating on the lowest $M$ = 10 levels of Mn$_{12}$. Using the same approach it is
possible to calculate the whole spectrum of hyperfine levels. Assuming for the hyperfine
coupling constants the values a(MnIII)= 6.9 mT and a(MnIV)= 8.5 mT, in agreement
with currently accepted values for these ions \cite{Zheng96}, we calculate a Gaussian
distribution of fields with a width of ca. 16 mT, in good agreement with the above
reported experimental result. A detailed calculation of the random hyperfine field
distribution can be found in Ref. \cite{Prok98}.
\begin{figure}[t]
\centerline{\epsfxsize=5.5 cm \epsfbox{fig4.eps}}
\caption{Hole line width $\sigma$, obtained by the measurements of quantum hole digging like in Fig. 3 (a) and (b), as a function of digging time for several temperatures. The intrinsic
line width $\sigma_0$ is obtained by an linear extrapolation to $t_{dig}$ = 0 s.}
\label{fig4}
\end{figure}
We applied also the 'hole digging' method at temperatures between 1.5 and 4
K. At these temperatures, the relaxation rates of the minor species are very fast and can
therefore be neglected. As pointed out by several groups, the relaxation of the major
species Mn$_{12}$ac(1) is non-exponential at temperatures below 4 K but, nevertheless,
we approximately adjusted an exponential law for the short time relaxation regime (1 $-$
100s) in order to yield a relaxation rate $\Gamma$. We emphasise that the Prokof'ev
Stamp theory \cite{Prok98} cannot be applied in the higher temperature regime because it
neglects thermal activation to higher energy levels. However, the main idea of the 'hole
digging' method should still hold, i.e. it should answer the question of whether the line
width is homogeneously or inhomogeneously broadened. A typical result of the 'hole
digging' experiment at 2 K is presented in Fig. 5 which shows that it is impossible to dig
a hole in the $\Gamma(H)$ dependence suggesting that the line width is homogeneously
broadened, as first suggested by Friedman {\it et al.} \cite{Friedman98}. This finding is also in good
agreement with a recent calculation of Leuenberger and Loss \cite{Leuenberger99}, see also \cite{Fort98}
which is based on thermally assisted spin tunnelling induced by quadratic anisotropy and
weak transverse magnetic fields. Their model is minimal in the sense that it is sufficient to
explain the measurements without including hyperfine fields. Indeed, our measurements
show that the inhomogeneous hyperfine broadening of about 12 mT is small compared to
the homogeneous broadening of about 30 mT (see fig. 5) which might be due to spin-
phonon coupling \cite{remark6}.
In conclusion, this letter is intended to report more accurate measurements which
should help to find a satisfactory explanation of the observed quantum phenomena in
molecular clusters. We used the recently developed 'hole digging' method for measuring
the intrinsic line width broadening due to local fluctuating fields and found strong
evidence for the influence of nuclear spins in the tunnel process at low temperature. At
higher temperatures, spin-phonon coupling seems to dominate the resonant quantum
tunnel transitions which leads to homogeneously broadened line widths.
\bigskip\centerline{***}\medskip
A. Caneschi is acknowledged for providing the Mn$_{12}$ac samples. We are
deeply indebted to P. Stamp for many fruitful discussions. We thank B. Barbara, A. Benoit, E. Bonet Orozco, D. Mailly, and P. Pannetier for the help of constructing our micro-SQUID magnetometer.
\begin{figure}[t]
\centerline{\epsfxsize=5.5 cm \epsfbox{fig5.eps}}
\caption{Quantum hole digging in an initial dipolar distribution of the major species
Mn$_{12}$ac (1), at T = 2 K. The relaxation rate (found by an exponential short time fit)
is plotted versus field $H$ after a digging relaxation at $H_{dig}$ = 0.925 T. After $t_{dig}$ = 10, 80 and 320 s, the reversed fraction of magnetisation is 0.042, 0.214, 0.573 of $M_S$,
respectively. The shift of the maximum of $\Gamma(H)$ is due to the change of the
internal dipolar fields.}
\label{fig5}
\end{figure}
\vskip-12pt
|
\section{Introduction.}
\label{Sec:intro}
In present paper we study projective modules over non-commutative tori.
(We always consider finitely generated
projective modules.)
Our main goal is to describe all modules that admit
constant curvature connections. It is well known that constant
curvature connections correspond to maximally supersymmetric BPS
fields \cite {cds} ; this means that we give conditions for existence of
1/2 BPS states.
The main results of the paper are formulated in the following theorems.
\begin{thm}
\label{thm:M1}
Let ${{\cal A}_{\theta}}$ be a non-commutative torus.
Then for every projective ${{\cal A}_{\theta}}$ module $E$ with a constant
curvature connection corresponding element of the group
$K_0({{\cal A}_{\theta}})$ is a generalized quadratic exponent.
Conversely,if $\mu$ is a positive generalized quadratic exponent in
$K_0({{\cal A}_{\theta}})$ then there exists such a projective module $E$ with constant
curvature connection that $[E]=\mu$.
(Here $[E]$ stands for the K-theory class of $E$.The definition of
generalized quadratic exponent will be given later. )
\end{thm}
\begin{thm}
\label{thm2}
Let ${{\cal A}_{\theta}}$ be an irrational non-commutative torus. In this case projective modules over ${{\cal A}_{\theta}}$ which admit
constant curvature connection are in one-to-one
correspondence with positive generalized quadratic exponents in $K_0({{\cal A}_{\theta}})$.
\end{thm}
This theorem is an immediate consequence of Theorem 1 and of the
following
very strong result by M. Rieffel (see \cite{Rf1}):
for irrational non-commutative torus ${{\cal A}_{\theta}}$ projective modules $E$ and $F$
are isomorphic if and only if the classes $[E],[F]\in K_0({{\cal A}_{\theta}})$ are
equal.
Our main results were formulated and partially proved in \cite {K-S},
Appendix D.
It is assumed in \cite {K-S} that every linear combination of entries of
the matrix
$\theta$ is irrational. It is proved that in this case a projective module
can
be transformed into a free module by means of complete Morita equivalence
iff corresponding $K$-theory class is a generalized quadratic exponent.
This statement can be used to prove
Theorem 1.1 in the conditions of Appendix D of \cite {K-S}).
The paper is organized as follows. In the introduction we remind
the main notions and results we need and explain how we plan to prove the
main theorem.
In the section 2 we introduce the notion of generalized quadratic
exponent and we study its properties.
Section 3 is about integral generalized quadratic exponents
and finite dimensional representations of rational non-commutative tori.
In section 4 we present a proof of main results.
Let us remind the definition of a
non-commutative torus (see \cite{R5} for more
details). Let $L$ be a lattice ${\bf Z\!\!Z}^n$ in a vector space $V^*={\bf I\!R}^n$.
Let $\theta$ be real valued skew-symmetric bilinear form on $R^n$.
We will think about $\theta$ as a two-form, that is an element
of $\Lambda^2 V$. Non-commutative torus ${{\cal A}_{\theta}}$ is the universal
${\bf C}^*$-algebra generated by unitary operators $U_\alpha$, $\alpha\in L$
obeying relations
\begin{equation}
\label{eq:rel}
U_\alpha U_\beta=e^{\pi i \theta(\alpha, \beta)} U_{\alpha+\beta}.
\end{equation}
Any element from ${{\cal A}_{\theta}}$ can be represented uniquely by a sum
$a=\sum_{{\bf \alpha}\in L} c_{\alpha}U_{\alpha}$,
where $c_{\alpha}$ are complex numbers. Assigning to every $a\in{{\cal A}_{\theta}}$ the
coefficient $c_0$ in the representation above we obtain a canonical trace
$\tau$ on ${{\cal A}_{\theta}}$.
Let $\{e_i\}$ be a basis of $L$. One can say that ${{\cal A}_{\theta}}$ is the universal
${\bf C}^*$-algebra generated by unitary operators $U_1,\cdots,U_n$
obeying the relations
\begin{equation}
\label{eq:relother}
U_i U_j=e^{2\pi i \theta(e_i, e_j)} U_j U_i.
\end{equation}
To check that these two definitions are equivalent one should take
$U_i=U_{e_i}$.
The transformations
$\delta _k U_{e_k}=U_{e_k}$, $1\leq k\leq n$ ,
$\delta _l U_{e_k}=0$, $k\not=l,1\leq k,l\leq n$ can be
regarded as generators of Abelian Lie algebra $L_{\theta}$ of
infinitesimal automorphisms on ${\cal A}_{\theta}$.
We can naturally identify ${L_{\theta}}$ with $V$.
Let us remind the definition of a connection in a ${\cal
A}_{\theta}$-module following \cite{Con1} (we do not need the general
notion of connection \cite{Con2}). First we need the notion of a
smooth part of a projective module.
Any element from ${{\cal A}_{\theta}}$ can be considered as a (generalized) function
on the n-dimensional torus whose Fourier coefficients are $c_{\alpha}$
(see above). The space of smooth functions on $T^n$ forms a subalgebra
of ${{\cal A}_{\theta}}$. We denote it by ${{{\cal A}_{\theta}}}^{smooth}$ and call it the smooth part of
${{\cal A}_{\theta}}$. If $E$ is a projective ${{\cal A}_{\theta}}$ module one can
define its smooth part $E^{smooth}$ in a similar manner (see \cite{Rf1}).
A connection on projective module $E$
can be defined as follows:
{\it ${{\cal A}_{\theta}}$ connection on a right ${{\cal A}_{\theta}}$ module $E$
is a linear map $\nabla :L_{\theta}
\rightarrow \mbox{\rm End}_{{\bf C}}E$,
satisfying the condition
$${\nabla}_{\delta}(ea)=({\nabla}_{\delta}e)a+e(\delta(a)),$$
where $e\in E^{smooth}$, $a\in{{\cal A}_{\theta}}^{smooth}$, and $\delta\in L_{\theta}$.
The curvature $F_{\mu,\nu}=[{\nabla}_{\mu},{\nabla}_{\nu}]$ of connection ${\nabla}$ is
considered as a two-form on ${L_{\theta}}$ with values in $\mbox{\rm End}_{{{\cal A}_{\theta}}}E$.
(Here $\mbox{\rm End}_{{{\cal A}_{\theta}}}E$ stands for the space of
endomorphisms of ${{\cal A}_{\theta}}^{smooth}$-module $E^{smooth}$ and $\mbox{\rm
End}_{{\bf C}}E$
denotes the space of ${\bf C}$-linear
endomorphisms of $E^{smooth}$.)}
We always consider Hermitian modules and Hermitian connections.
This means that if $E$ is a right ${{\cal A}_{\theta}}$ module it is equipped with
${{\cal A}_{\theta}}$ valued Hermitian inner product $\langle\cdot,\cdot\rangle$
(for the detailed list of properties see \cite{Bl}) ; all connections
that we will consider should be compatible with this inner product.
If $E$ is endowed with a ${{\cal A}_{\theta}}$-connection, then one can define a
Chern character
\begin{equation}
\label{eqchern}
\mbox{\rm ch}(E)=\sum_{k=0}\frac{\hat{\tau}(F^k)}{(2\pi i)^k k!}=
\hat{\tau}(e^{\frac{F}{2\pi i}}),
\end{equation}
where $F$ is a curvature of a connection on $E$, and $\hat{\tau}$ is the
canonical trace on $\hat{A}=\mbox{\rm End}_{{{\cal A}_{\theta}}}(E)$
(we use that
${{\cal A}_{\theta}}$ is equipped with a canonical trace $\tau=c_0$). One can consider
$\mbox{\rm ch}(E)$ as an element in the Grassmann algebra
$\Lambda^\cdot({L_{\theta}}^*)=\Lambda^\cdot (V^*)$.
We have a lattice $L$ in $V^*$. Thus we can talk about integral
elements in $\Lambda^\cdot (V^*)$ which are just the elements of
$\Lambda^\cdot L$.
In the commutative case ${\rm ch}(E)$ is integral.
In non-commutative case this is wrong, but there exists an integral
element $\mu(E)\in \Lambda^\cdot (V^*)$ related to
$\mbox{\rm ch}(E)$ by the
formula (see \cite{Ell}, \cite{Rf1})
\begin{equation}
\label{eqrelchmu}
\mbox{\rm ch}(E)=e^{\iota(\theta)}\mu(E),
\end{equation}
Here ${\iota(\theta)}$ stands for the operation of contraction with
${\theta}$ considered as an element of $\Lambda^2 V$.
In particular, formula (\ref{eqrelchmu}) means that
$e^{-\iota(\theta)}\mbox{\rm ch}(E)$ is an integral element
of $\Lambda^\cdot (V^*)$. The group $\Lambda^{even} L$ can be naturally
identified with the group $K_0({{\cal A}_{\theta}})$. Moreover
$\mu(E)$ is the class of the module $E$ in the $K_0({{\cal A}_{\theta}})$ group (see
\cite{Ell}).
Let us remind that the element $\mu\in K_0({{\cal A}_{\theta}})$ is called positive
if $(e^{\iota(\theta)}\mu)_{(0)}>0$ (the zero component is positive).
A well known theorem of M. Rieffel (see \cite{Rf1}) says that if $\theta$
is irrational then every positive element of $\mu$ is represented by a
projective module over ${{\cal A}_{\theta}}$.
Let $E$ be a projective ${{\cal A}_{\theta}}$ module with a
constant curvature connection ${\nabla}$. Denote by $F$ the curvature of ${\nabla}$.
Then since $F$ is a 2-form with values in ${\bf C}$ we obtain that
$\hat{\tau}(F^k)=\hat{\tau}(1) F^k$. The number $\hat{\tau}(1)$ is
called the dimension of the module $E$ and we denote it by $d_E$.
Then the formula \ref{eqchern} becomes
\begin{equation}
\label{eq:ch}
ch(E)=d_E e^{\frac{F}{2\pi i}}.
\end{equation}
We see that in this case $ch(E)$ is a quadratic exponent (i.e. an
expression of the form $Ce^a$ where $C$ is a constant and $a\in \Lambda
^2(V^+)$). It follows from (4) and from this fact that $\mu (E)$ is a
generalized quadratic exponent (i.e. a limit of quadratic exponents). This
gives a proof of the first statement of Theorem \ref{thm:M1}. The proof of
the second statement of this theorem is based on the study of
generalized quadratic
exponents in sections 2 and 3. In section 3 we will
study integral generalized quadratic exponents keeping in mind that
$K_0({{\cal A}_{\theta}})$ is exactly
the integral lattice in ${\Lambda^{even}(V)}$, where $V={L_{\theta}}^*$. We will prove some
auxiliary technical results saying that something is rational
or integral which we will use in our construction of the module
in section 4.\\
In section 4 we will construct a desired module together
with constant curvature connection in four steps.
First we find explicitly the curvature $F$ as a 2-form on ${L_{\theta}}$.
Second, we construct some spaces of functions on ${\bf I\!R}^p\times {\bf Z\!\!Z}^{n-2p}$
with action of generators of
some non-commutative torus ${\cal A}_{\widetilde{\theta}}$ and with constant
curvature connection having the curvature form $F$.
We do not construct an ${\cal A}_{\widetilde{\theta}}$ module at this step.
Moreover, we even will not specify what space of functions we will
take.
Third, we will check that our construction in the previous step
is just a particular case of Rieffel's construction in \cite{Rf1}
where he constructs ${\cal A}_{\widetilde{\theta}}$
projective modules. So we can construct the desired module $\widetilde{E}$ over
${\cal A}_{\widetilde{\theta}}$ using Rieffel's construction.
Fourth, we will see that $\tau=\theta-\widetilde{\theta}$ is a rational element
of $\Lambda^2 {L_{\theta}}=\Lambda^2 V^*$.
Also, we will use Rieffel's explicit calculation
of $[\widetilde{E}]$ (of the class of $\widetilde{E}$ in
$K_0({\cal A}_{\widetilde{\theta}})\subset {\Lambda^{even}(V)}$ )to find a simple relation
between $[\widetilde{E}]$ and $\mu$.
Finally, we show that we can construct
a projective module $E$ with constant curvature
connection over ${{\cal A}_{\theta}}$ such that $[E]=\mu$ by taking $E$ to be a tensor
product of $\widetilde{E}$ by some finite dimensional module $M$ over ${{\cal A}_{\tau}}$.
\section{Generalized quadratic exponents.}
\label{sec:gen}
In this section we introduce generalized quadratic exponents
and study their properties.
Let $V$ be a finite dimensional vector space over ${\bf I\!R}$. Let $V^*$ be
a dual space. Then the space $V\oplus V^*$ has a natural symmetric
bilinear product given by
\[
\langle (x_1,y_1),(x_2,y_2)\rangle=y_2(x_1)+y_1(x_2),
\]
where $x_1,x_2\in V$ and $y_1,y_2\in V^*$. Consider a Clifford algebra
${\mbox{\rm Cl}(V\oplus V^*)}$. It naturally acts on the vector space ${\Lambda^{\cdot}(V)}$; we denote this action
by ${\rho}$.
Note, that there is a natural inclusion $i$ of $V\oplus V^*$ into ${\mbox{\rm Cl}(V\oplus V^*)}$.
\begin{defi}
\label{defi:gqe}
An element $q\in {\Lambda^{\cdot}(V)}$ is called a generalized quadratic exponent
if there exists a maximal isotropic subspace $U\subset V\oplus V^*$
such that for any $x\in U$ we have ${\rho}(x)q=0$.
\end{defi}
If the projection of $U$ onto $V^*$ is bijective we can represent $U$ as a
graph of a linear operator $a: V^*\rightarrow V$. The operator $a$ is
antisymmetric; it can be considered as an element of $\Lambda ^2(V)$. The
element $q$ can be represented in the form const$\cdot e^a$, i.e. it is a
quadratic exponent. The set of maximal isotropic subspaces we just
considered is
dense in the set of all maximal isotropic subspaces; this means that
quadratic exponents are dense in the set of all generalized quadratic
exponents.
In the next proposition we will describe all possible generalized quadratic
exponents.
\begin{prop}
\label{prop:gqe}
Let $q\in{\Lambda^{\cdot}(V)}$ be a generalized quadratic exponent. Then there exists
a subspace $W\subset V$, a non-degenerate element $\widetilde{q}_1\in \Lambda^2(V/W)$
and nonzero element $w\in \Lambda^{{\mbox{\rm \tiny dim}} W}W$ such that
\[
q=w\wedge e^{q_1}.
\]
where $q_1\in \Lambda^2(V)$ is any preimage of $\widetilde{q}_1\in \Lambda^2(V/W)$
under the natural projection from $\Lambda^2(V)$ onto $\Lambda^2(V/W)$.
\end{prop}
{\bf Proof:} Let $U$ be the maximal isotropic subspace corresponding to $q$.
It is easy to see that we can choose a basis $\{\xi_i\}$
of $V$ and a dual basis
$\{\eta_i\}$ of $V^*$ such that $U$ is spanned by the vectors
$\eta_1-\sum_{i=1}^j a_{1,i} \xi_i,...,
\eta_j-\sum_{i=1}^{j} a_{j,i} \xi_i,
\xi_{j+1},...,\xi_{{\mbox{\rm \tiny dim}} W}$.
Thus, $q$ satisfies the following system of equations:
\[
\begin{array}{l}
\frac{\partial q}{\partial \xi_1}-(\sum_{i=1}^j a_{1,i} \xi_i)\wedge q=0\\
\cdot\\
\cdot\\
\cdot\\
\frac{\partial q}{\partial \xi_j}-(\sum_{i=1}^j a_{j,i} \xi_i)\wedge
q=0\\[12pt]
\xi_{j+1}\wedge q=0\\
\cdot\\
\cdot\\
\cdot\\
\xi_{{\mbox{\rm \tiny dim}} W}\wedge q=0
\end{array}
\]
The partial derivatives in this system are understood as left derivatives
in the sense of superalgebra.
It is easy to see that any solution of this system is of the form
\[
C\cdot\xi_{j+1}\wedge\cdots\wedge\xi_{{\mbox{\rm \tiny dim}} W}\wedge
e^{\sum_{k=1}^j\sum_{l=1}^j a_{k,l} \xi_k\wedge \xi_l},
\]
where C is a constant. The proposition follows easily from the
above formula. $W$ is the subspace spanned by $\xi_{j+1},\cdots,\xi_{{\mbox{\rm \tiny dim}}
W}$ and
$\widetilde{q}_1$ is the projection of
${\sum_{k=1}^j\sum_{l=1}^j a_{k,l} \xi_k\wedge \xi_l}$.\\
{\hspace*{3in} Q.E.D.}
${\Lambda^{\cdot}(V)}$ is a graded vector space. If $q\in{\Lambda^{\cdot}(V)}$ let us denote by
$q_{(i)}\in \Lambda^i V$ the projection $q$ on $\Lambda^i V$.
\begin{cor}
\label{cor:const}
Let $q$ be a generalized quadratic exponent. If $q_{(0)}$ is
not zero then there is a non-degenerate element $a\in\Lambda^2 V$
and a nonzero real number $C$ such that
\[
q=C e^{a}.
\]
\end{cor}
{\bf Proof:} Immediately follows from Proposition \ref{prop:gqe}.\\
{\hspace*{3in} Q.E.D.}
Let $b\in \Lambda^2 (V^*)$. Then $b$ acts naturally on ${\Lambda^{\cdot}(V)}$. If we choose
basis $\{\xi_i\}$ in $V$ then we can write the action of $b$ as
$\sum_{k,l} b_{k,l}\frac{\partial}{\partial \xi_k \partial \xi_l}$.
Another way of thinking is to think about $b$ as an element of ${\mbox{\rm Cl}(V\oplus V^*)}$.
We have a canonical map from $\Lambda^\cdot (V^*)$ to ${\mbox{\rm Cl}(V\oplus V^*)}$ since
$V^*$ is isotropic subspace in $V\oplus V^*$. Then the action
of $b$ is simply ${\rho}(b)$.
\begin{prop}
\label{prop:closure}
Let $q$ be a generalized quadratic exponent and $b$ any element
in $\Lambda^2 (V^*)$. Then $e^{{\rho}(b)}q$ is a generalized quadratic
exponent.
\end{prop}
{\bf Proof:}
We will reduce the proposition to the case where $b$ is decomposable. Since,
${\rho}(b)=\sum_{k,l} b_{k,l}\frac{\partial}{\partial \xi_k \partial \xi_l}$
in some basis $\{\xi_i\}$ of $V$ and the operators
$\frac{\partial}{\partial \xi_k \partial \xi_l}$ commute it is enough
to prove the proposition in the case when
${\rho}(b)=c\frac{\partial}{\partial \xi_k \partial \xi_l}$, where c is a real
number. In this case
$e^{{\rho}(b)}=1+{\rho}(b)$.
Our goal is to show that there exists a subspace $\widetilde{W}\subset V\oplus V^*$
such that for any $x\in\widetilde{W}$ we have ${\rho}(x)(q+{\rho}(b)(q))=0$.
Let $W$ be a subspace of $V\oplus V^*$ such that if $x\in W$ then ${\rho}(x)q=0$.
We can choose a basis $\{v_1+w_1,\cdots,v_k+w_k,v_{k+1},\cdots,v_{{\mbox{\rm \tiny dim}} W}\}$
of $W$, where $v_i\in V^*$ and $w_i\in V$, and the vectors $\{w_i\}$ are
linearly independent.
A simple calculation shows that
\[
{\rho}(v_l)(q+{\rho}(b)(q))=0+{\rho}(v_l){\rho}(b)(q)={\rho}(b){\rho}(v_l)(q)=0,
\]
for $l>k$. Also, we can easily see that for $l<k+1$
\[
\begin{array}{l}
{\rho}(v_l+w_l)(q+{\rho}(b)(q))={\rho}(v_l+w_l){\rho}(b)(q)=\\[12pt]
[{\rho}(v_l+w_l),{\rho}(b)](q)+{\rho}(b)
{\rho}(v_l+w_l)(q)
=[{\rho}(w_l),{\rho}(b)](q)={\rho}(\iota(w_l)b)q,
\end{array}
\]
where $\iota(w_l)$ is plugging the vector $w_l$ in the 2-form $b$.
$\iota(w_l)b$ is an element of $V^*$. Since $b^2=0$ we see that
$0=\iota(w_l)(b^2)=2(\iota(w_l)b)b$. Thus,
\[
{\rho}(\iota(w_l)b)q={\rho}(\iota(w_l)b+(\iota(w_l)b)b)q={\rho}(\iota(w_l)b)(q+{\rho}(q)).
\]
Therefore, we see that
\[
{\rho}([v_l-\iota(w_l)b]+w_l)(q+{\rho}(b)q)=0.
\]
Denote by $\widetilde{W}$ the subspace of $V\oplus V^*$ spanned by
the vectors
$[v_1-\iota(w_1)b]+w_1,\cdots,[v_k-\iota(w_k)b]+w_k,v_{k+1},...,
v_{{\mbox{\rm \tiny dim}} W}$. It is easy to check that $\widetilde{W}$ is a maximal isotropic
subspace of $V\oplus V^*$ and we showed that ${\rho}(x)(q+{\rho}(b)q)=0$
for any $x\in \widetilde{W}$. Thus, $q+{\rho}(b)q$ is a generalized quadratic exponent.\\
{\hspace*{3in} Q.E.D.}
\section{Integral generalized quadratic exponents.}
\label{sec:int}
In this section we study integral generalized
quadratic exponents and we prove a couple of auxiliary propositions
that we will use in our construction.
Let $V$ be a finite dimensional vector space and let $L$ be a lattice
in it. Denote by $n$ the dimension $V$.
Then, $V\cong {\bf I\!R}^n$ and $L\cong {\bf Z\!\!Z}^n$. We denote the dual lattice to $L$
by $L^*$. Obviously $L^*\subset V^*$. We call an element of ${\Lambda^{\cdot}(V)}$
integral if it lies in $\Lambda^{\cdot}L$.
Define $U_\mu$ a subspace
of $V^*$ as follows:
\[
U_\mu=\left\{ x\in V^*\,\mid \, \iota(x)\mu=0\right\}.
\]
Denote by $W_\mu\subseteq V$ the orthogonal complement to $U_\mu$.
\begin{prop}
\label{prop:intgqe}
Let $\mu\in {\Lambda^{\cdot}(V)}$ be an integral generalized quadratic exponent.
Then $L_\mu=L\cap W_\mu$ is a lattice in $W_\mu$. We can identify
$\Lambda ^{{\mbox{\rm \tiny dim}} W_\mu} L_\mu$ with ${\bf Z\!\!Z}$ (the isomorphism is not canonical
but it is specified up to a sign). Let $\alpha\in \Lambda ^{{\mbox{\rm \tiny dim}} W_\mu}
L_\mu$ be
a volume form (an
element that corresponds to $1$ under the isomorphism with ${\bf Z\!\!Z}$).
Then $\mu_{({\mbox{\rm \tiny dim}} W_\mu)}=N \alpha$, where $N$ is a nonzero integer.
\end{prop}
{\bf Proof:} Let $\mu\in {\Lambda^{\cdot}(V)}$ be an integral generalized quadratic exponent.
Let $k$ be the largest integer such that $\mu_{(k)}\neq 0$ and
for all $l>k$ we have $\mu_{(l)}=0$.
{}From Proposition \ref{prop:gqe} easily follows that
$U_\mu=\left\{ x\in V^*\,\mid \, \iota(x)\mu=0\right\}=
\left\{ x\in V^*\,\mid \, \iota(x)\mu_{(k)}=0\right\}$.
Moreover, from Proposition \ref{prop:gqe} follows that $\mu_{(k)}$
is a decomposable element of ${\Lambda^{\cdot}(V)}$ and that
$k={\mbox{\rm dim}} V-{\mbox{\rm dim}} U_\mu={\mbox{\rm dim}} W_\mu$. Since $\mu$ is integral $\mu_{(k)}$
is also integral. Thus, the subspace $U_\mu$ is spanned by $U_\mu\cap L^*$.
Therefore, $U_\mu\cap L^*$ is a lattice in $U_\mu$. This immediately
implies that $L_\mu=W_\mu\cap L$ is a lattice in $W_\mu$ since $W_\mu$ is
the orthogonal complement to $U_\mu$.
{}From the above discussion it is easy to see that
$\Lambda ^{{\mbox{\rm \tiny dim}} W_\mu} (W_\mu\cap L)=\Lambda ^{{\mbox{\rm \tiny dim}} W_\mu} L_\mu\cong {\bf Z\!\!Z}$.
Since, by the definition $\alpha$ corresponds to $\pm 1$ under such an
isomorphism and $\mu_{({\mbox{\rm \tiny dim}} W_\mu)}=\mu_{(k)}$ is an integral element
we obtain that $\mu_{({\mbox{\rm \tiny dim}} W_\mu)}=N \alpha$ for some integer $N$.\\
{\hspace*{3in} Q.E.D.}
Under the conditions in the above proposition we can easily find
a complement $\widetilde{L}_\mu$ ($\widetilde{L}_\mu\cong {\bf Z\!\!Z}^{n-{\mbox{\rm \tiny dim}} W_\mu}$)
to $L_\mu$ in $L$. It is not unique but we do
not care about that. Let $Y_\mu$ be the subspace of $V$ spanned by
$\widetilde{L}_\mu$. Then it is obvious that $V=W_\mu\oplus Y_\mu$ and
$L=L_\mu\oplus \widetilde{L}_\mu$.
Next results will be used in the construction in section
\ref{sec:constr}. Since, they do not use any theory of
non-commutative tori we state them here. But they need some explanation
concerning their origin.
Let $\mu\in{\Lambda^{even}(V)}$ be an integral generalized
quadratic exponent which will be an element of $K_0$ representing
a projective module of ${{\cal A}_{\theta}}$.
We can think about $\theta$ as an element of $\Lambda^2 V^*$.
If there exists a projective module $E$ over ${{\cal A}_{\theta}}$ with constant
curvature
connection such that $[E]=\mu$ then by the result of G. Elliott (see
\cite{Ell})
$d e^{\frac{F}{2\pi i}}=e^{\iota(\theta)}\mu$, where $d$ is the dimension
of the module. $\widetilde{\theta}$ which satisfies the conditions
of the lemma below will be constructed in section \ref{sec:constr}.
\begin{lem}
\label{lem:ration}
Let $\mu\in{\Lambda^{even}(V)}$ be an integral generalized
quadratic exponent.
Let us assume that we fixed the isomorphism between
$\Lambda ^{{\mbox{\rm \tiny dim}} W_\mu} L_\mu$ and ${\bf Z\!\!Z}$ (see Proposition \ref{prop:intgqe})
so that
$\mu_{({\mbox{\rm \tiny dim}} W_\mu)}=N\alpha$ with $N$ being natural number (here
$\alpha\in \Lambda ^{{\mbox{\rm \tiny dim}} W_\mu} L_\mu$ corresponds to $1$ in ${\bf Z\!\!Z}$).
Let $\theta$ and $\widetilde{\theta}$ be elements of $\Lambda^2 V^*$ such that
$\theta-\widetilde{\theta}$ is zero on $V\otimes Y_\mu$ (that is if $X_\mu\subset V^*$
is the orthogonal complement to $Y_\mu$ then
$\theta-\widetilde{\theta}\in \Lambda^2 X_\mu$). Assume that
\begin{equation}
\label{eq1}
e^{\iota(\theta)}\mu=c e^{\iota(\widetilde{\theta})}\alpha,
\end{equation}
where $c$ is a real number. Then $c=N$, that is
\begin{equation}
\label{eq2}
e^{\iota(\theta)}\mu=N e^{\iota(\widetilde{\theta})}\alpha,
\end{equation}
and $N (\theta-\widetilde{\theta})$ is an integral element of $\Lambda^2 X_\mu$.
\end{lem}
{\bf Proof:} Let us denote ${\mbox{\rm dim}} W_\mu$ by $k$. Then formula
\ref{eq1} implies that
\[
\mu_{(k)}=(e^{\iota(\theta)}\mu)_{(k)}=c (e^{\iota(\widetilde{\theta})}\alpha)_{(k)}=
c \alpha_{(k)}.
\]
Thus, $c=N$ and we proved formula \ref{eq2}. From formula \ref{eq2}
it easily follows that
\begin{equation}
\label{eq3}
\mu=N e^{\iota(\widetilde{\theta}-\theta)}\alpha.
\end{equation}
This means that $\mu_{k-2}=N \iota(\widetilde{\theta}-\theta)\alpha$.
$\mu_{k-2}$ is an integral element.
$\widetilde{\theta}-\theta$ is in $\Lambda^2 X_\mu$ which is dual to $W_\mu$
and $\alpha$ is a nonzero element of $\Lambda^{{\mbox{\rm \tiny dim}} W_\mu}W_\mu$.
Thus, $N(\widetilde{\theta}-\theta)$ is an integral element of $\Lambda^2 X_\mu$.\\
{\hspace*{3in} Q.E.D.}
\subsection{Modules over the rational non-commutative tori.}
\label{subsec:ration}
Let us assume that conditions of Lemma \ref{lem:ration} are satisfied.
We denote ${\theta}-\widetilde{\theta}$ by $\tau$. Let us remind the definition
of ${{\cal A}_{\tau}}$. ${{\cal A}_{\tau}}$ is a universal ${\bf C}^*$ algebra having unitary
generators $U_\beta$, where $\beta\in L$, obeying the relations
\[
U_{\beta_1}U_{\beta_2}=e^{\pi i \tau(\beta_1,\beta_2)} U_{\beta_1 +\beta_2}.
\]
We can reformulate this definition in a slightly different way.
Let $\beta_1,\cdots,\beta_n$ (where $n={\mbox{\rm dim}} V$)
be a basis of a free ${\bf Z\!\!Z}$ module $L$.
${{\cal A}_{\tau}}$ is
a universal ${\bf C}^*$ algebra having unitary
generators $U_i$, $1\leq i\leq n$, obeying the relations
\[
U_i U_j=e^{2\pi i \tau(\beta_i,\beta_j)} U_j U_i.
\]
It is obvious that the two definitions are equivalent.
\begin{prop}
\label{prop:taumod}
Under the conditions of Lemma \ref{lem:ration} there exists
an $N$ dimensional module $M$ over ${{\cal A}_{\tau}}$.
\end{prop}
{\bf Proof:} Since $N\tau$ is integral form and we have a freedom
in choosing a basis $\{\beta_i\}$ of $L$, we can choose it
so that
\[
\begin{array}{l}
N\tau(\beta_{2i-1},\beta_{2i})=q_1q_2\cdots q_i,\\[12pt]
\tau(\beta_k,\beta_l)=0~~~{\rm unless}~~~k=2i-1~{\rm and}~l=2i~~
{\rm or}~~k=2i~{\rm and}~l=2i-1,
\end{array}
\]
where $q_1,q_2,\cdots$ are integers (see \cite{Igusa})
and moreover the basis $\{\beta_i\}$ respects the decomposition of $L$ into
$L_\mu\oplus \widetilde{L_\mu}$.
In this basis the algebra ${{\cal A}_{\tau}}$ is generated by unitary
generators $U_i$ obeying the relations
\begin{equation}
\label{eq4}
U_{2i-1}U_{2i}=e^{2\pi i \frac{q_1\cdots q_i}{N}}U_{2i}U_{2i-1}
\end{equation}
( all other generators commute). Note that it may happen
that there exists an integer $m$ such that if $i>m$ then all $q_i$ are
zero.
So we see that our algebra ${{\cal A}_{\tau}}$ is a tensor product of algebras
${{\cal A}_{\tau}}_i$, where ${{\cal A}_{\tau}}_i$ is generated by two unitary generators
$U_{2i-1}$ and $U_{2i}$ obeying relations \ref{eq4} or
${{\cal A}_{\tau}}_i$ is generated by only one unitary generator (this is the case
when $i>m$, in particular if $\beta_i\in \widetilde{L_\mu}$).
Thus it is enough to show that we can construct finite dimensional
modules $M_i$ over ${{\cal A}_{\tau}}_i$ such that $({\mbox{\rm dim}} M_1)({\mbox{\rm dim}} M_2)\cdots$
divides $N$. Really, in such case we can take $M$ to be the direct sum
of $M_1\otimes M_2\otimes\cdots$ taken $\frac{N}{({\mbox{\rm \tiny dim}} M_1)({\mbox{\rm \tiny dim}} M_2)\cdots}$
times.
If ${{\cal A}_{\tau}}_i$ is generated by one unitary generator then it has a 1 dimensional
module over it, $U_i$ acts by 1. We choose $M_i$ to be this module
in this case.
If ${{\cal A}_{\tau}}_i$ is generated by two unitary generators
$U_{2i-1}$ and $U_{2i}$ obeying relations \ref{eq4} then it has
a module of dimension $\frac{N}{{\rm GCD}(N,q_1\cdots q_i)}$, where
GCD stands for greatest common divisor. We choose $M_i$ to be a module of
the dimension $\frac{N}{{\rm GCD}(N,q_1\cdots q_i)}$.
Thus, it is enough to show that
$\frac{N^m}{{\rm GCD}(N,q_1)\cdots{\rm GCD}(N,q_1\cdots q_m)}$ divides $N$,
where $m$-the number of tori
${{\cal A}_{\tau}}_i$ generated by two generators. Therefore it is enough to prove that
$\frac{{\rm GCD}(N,q_1)\cdots{\rm GCD}(N,q_1\cdots q_m)}{N^{m-1}}$ is
an integer.
\begin{lem}
\label{lem:division}
\[
\frac{{\rm GCD}(N,q_1)\cdots{\rm GCD}(N,q_1\cdots q_m)}{N^{m-1}}
\]
is an integer if
\[
\frac{q_1^2q_2}{N},\frac{q_1^3q_2^2q_3}{N^2},\cdots,
\frac{q_1^{m}q_2^{m-1}\cdots q_m}{N^{m-1}}
\]
are integers.
\end{lem}
{\bf Proof:} Denote by $a_i$ the
$\frac{{\rm GCD}(N,q_1\cdots q_i)}{{\rm GCD}(N,q_1\cdots q_{i-1})}$ and
$b_i=q_i/a_i$. Then we can write
$q_1^{m}q_2^{m-1}\cdots q_m=(a_1^{m}a_2^{m-1}\cdots a_m)
(b_1^{m}b_2^{m-1}\cdots b_m)$. We will prove by induction that
$ \frac{a_1^{k}a_2^{k-1}\cdots a_k}{N^{k-1}}$ are integers.
The initial case $k=1$ is obvious.
For $k>1$ we have
\[
\frac{a_1^{k}a_2^{k-1}\cdots a_k}{N^{k-1}}=
\frac{a_1^{k-1}a_2^{k-2}\cdots a_{k-1}}{N^{k-2}} \left(\frac{a_1\cdots a_k}{N}
\right).
\]
$\frac{a_1^{k-1}a_2^{k-2}\cdots a_{k-1}}{N^{k-2}}$ is an integer by induction
hypothesis and we also know that $\frac{N}{a_1\cdots a_k}$ are relatively prime
with $b_1,\cdots b_k$. But on the other hand
\[
\frac{q_1^{k}q_2^{k-1}\cdots q_k}{N^{k-1}}=
\frac{a_1^{k-1}a_2^{k-2}\cdots a_{k-1}}{N^{k-2}}
\left(\frac{a_1\cdots a_k}{N}\right)
\left(b_1^{k}b_2^{k-1}\cdots b_k\right).
\]
Thus, $\frac{a_1^{k-1}a_2^{k-2}\cdots a_{k-1}}{N^{k-2}}
\left(\frac{a_1\cdots a_k}{N}\right)$ is an integer.\\
{\hspace*{3in} Q.E.D.}
To prove the proposition it is enough to show that
\[
\frac{q_1^2q_2}{N},\frac{q_1^3q_2^2q_3}{N^2},\cdots,
\frac{q_1^{m}q_2^{m-1}\cdots q_m}{N^{m-1}}
\]
are all integers. Let $\{\gamma_i\}$ be a basis of $V^*$ dual
to $\{\beta_i\}$. We know that $\mu=N e^{\iota(\tau)}\alpha$
is an integral element of ${\Lambda^{even}(V)}$. Denote by $k$ the dimension
of $W_\mu$ as before.
Then, $\alpha=\pm \beta_1\wedge\beta_2\wedge\cdots\wedge\beta_k$
and the numbers
\[
\begin{array}{l}
\langle \gamma_3\wedge\cdots\wedge\gamma_k,N\iota(\tau)\alpha\rangle,
\langle \gamma_5\wedge\cdots\wedge\gamma_k,
N\frac{(\iota(\tau))^2}{2}\alpha\rangle,\\[12pt]
\cdots,
\langle \gamma_{2m+1}\wedge\cdots\wedge\gamma_k,
N\frac{(\iota(\tau))^m}{m!}\alpha\rangle
\end{array}
\]
are integers. A straightforward calculation shows that
\[
\langle \gamma_{2j+1}\wedge\cdots\wedge\gamma_k,
yN\frac{(\iota(\tau))^j}{j!}\alpha\rangle=
\pm
\frac{q_1^{j}q_2^{j-1}\cdots q_j}{N^{j-1}}.
\]
Thus all numbers $\frac{q_1^{j}q_2^{j-1}\cdots q_j}{N^{j-1}}$ are
integers.\\
{\hspace*{3in} Q.E.D.}
\section{Proof of Theorem \ref{thm:M1}.}
\label{sec:proof}
First, let us show that if we have a projective
${{\cal A}_{\theta}}$-module with constant curvature connection
then the corresponding class $\mu=[E]\in K_0({{\cal A}_{\theta}})$ is a positive
generalized quadratic exponent.
Really, we know from formula \ref{eqrelchmu} that
\[
\mu=[E]=e^{-\iota(\theta)}ch(E)=e^{\iota(-\theta)}ch(E).
\]
Also, from formula \ref{eq:ch} we see that
\[
ch(E)=d_E e^{\frac{F}{2\pi i}}.
\]
Thus $ch(E)$ is a generalized quadratic exponent since $F$ is an element
of $\Lambda^2 V$ (recall that $V={L_{\theta}}^*$).
>From Proposition \ref{prop:closure} it immediately follows that
$\mu$ is a generalized quadratic exponent since $-\theta\in \Lambda^2 (V^*)$.
Therefore we proved that $\mu=[E]$ is a generalized quadratic exponent.
It is a positive element of $K_0({{\cal A}_{\theta}})$ because it represents a genuine
${{\cal A}_{\theta}}$ module. Thus we proved the statement of Theorem \ref{thm:M1} in one
direction.
This was the easy part. The hard part is to prove the second half, that is
to show that if $\mu\in K_0({{\cal A}_{\theta}})$ is a positive generalized quadratic exponent
then there exists a projective module $E$ with a
constant curvature connection which represents the class $\mu$, {\it i.e.},
$\mu=[E]$. In the next subsection we present an explicit construction
of such a module.
\subsection{Construction of ${{\cal A}_{\theta}}$-module $E$ with constant curvature
connection.}
\label{sec:constr}
In this section we will construct explicitly a ${{\cal A}_{\theta}}$-module $E$ with a
constant
curvature connection over representing $\mu\in K_0({{\cal A}_{\theta}})$.
Assuming that such a module exists we see that
$ch(E)=e^{\iota(\theta)}\mu$ is a generalized quadratic exponent
(follows from Proposition \ref{prop:closure} and the fact that
$\mu$ is a generalized quadratic exponent). Moreover,
$ch(E)_{(0)}>0$ therefore from Corollary \ref{cor:const} follows that
\[
ch(E)=e^{\iota(\theta)}\mu=d_E e^{\frac{F}{2\pi i}},
\]
where $F$ is the curvature form, and $d_E$ is the dimension of the module
$E$. Thus, reversing the previous arguments it is obvious that it is enough
to construct a projective ${{\cal A}_{\theta}}$- module
with constant curvature connection satisfying the following
properties\\
a) the curvature form is $F$;\\
b) the dimension of the module is $d_E$. \\
In Section \ref{sec:int} we defined a subspace $W_\mu\subseteq V={L_{\theta}}^*$
associated
with the generalized quadratic exponent $\mu$. Since $\mu\in K_0({{\cal A}_{\theta}})$ we
see that $\mu$ is integral. Thus, $L_\mu=L\cap W_\mu$ is the integral lattice
in $W_\mu$ by Proposition \ref{prop:intgqe}. As in section \ref{sec:int}
we denote by $k={\mbox{\rm dim}} W_\mu$ and we choose a complement $\widetilde{L_\mu}$ to
$L_\mu$ in $L$. Denote by $Y_\mu$ the span of $\widetilde{L_\mu}$ in $V$.
It is obvious that ${{L_{\theta}}}^*=V=W_\mu\oplus Y_\mu$. Thus, we have a
natural decomposition ${L_{\theta}}=V^*=W^*_\mu\oplus Y^*_\mu$. Note
that the space $Y^*_\mu=U_\mu$ was defined in section \ref{sec:int}.
Since $\mu_{(k)}=d_E (e^{\frac{F}{2\pi i}})_{(k)}$
$k$ is even integer, that is $k=2p$, $p\in {\bf Z\!\!Z}$. Denote by
$q$ the rank of the free abelian group $\widetilde{L_\mu}$.
We have $q=n-2p$, where $n={\mbox{\rm dim}} {L_{\theta}}$
the dimension of ${{\cal A}_{\theta}}$.
Since $F^p$ is nonzero it follows that $F|_{W_\mu^*}$
($F$ restricted to $W_\mu^*$) is a non-degenerate 2-form.
\subsubsection{Construction of operators ${\nabla}_x$ for $x\in{L_{\theta}}$.}
Let ${{\cal H}eis}$ be a Heisenberg algebra generated
by the operators ${\nabla}_x$, for $x\in W_\mu^*$ which satisfy the relation
\[
[{\nabla}_x,{\nabla}_y]=F(x,y),
\]
where $x,y\in W_\mu^*$. The algebra ${{\cal H}eis}$ has a unique irreducible
representation
which can be realized in the space of square integrable functions on
${\bf I\!R}^p$. Moreover, the action of ${\nabla}_x$ is given by an operator
\[
({\nabla}_x(f))(z)=2\pi i \langle \phi(x), z\rangle f(z) +
\sum_i \psi_i(x)\frac{\partial f(z)}{\partial z_i},
\]
where $\phi: W_\mu\rightarrow ({\bf I\!R}^p)^* $ is some linear map, and
$\psi_i: W_\mu\rightarrow {\bf I\!R}$ are some linear functions on $W_\mu^*$.
In particular, we see that these operators preserve the space
of Schwartz functions on ${\bf I\!R}^p$.
The above construction provides us with the action of the operators
${\nabla}_x$ for $x\in W_\mu^*$ only. First we will extend the above construction
to obtain an action
of all operators ${\nabla}_x$, $x\in {L_{\theta}}=W_\mu^*\oplus Y_\mu^*$. Then, we
will obtain an action of some non-commutative torus ${{\cal A}_{\widetilde{\theta}}}$ so that
${\nabla}$ becomes an ${{\cal A}_{\widetilde{\theta}}}$ connection.
We extend the space from the space of Schwartz functions on ${\bf I\!R}^p$
to the space of Schwartz functions on
${\bf I\!R}^p\times \widetilde{L_\mu}={\bf I\!R}^p\times {\bf Z\!\!Z}^q$. Denote it by $H$.
If $x\in {L_{\theta}}=W_\mu^*\oplus Y_\mu^*$ we denote by $x_W$ the projection
of $x$ on $W^*_\mu$ and by $x_Y$ the projection of $x$ on $Y^*_\mu$
(obviously $x=x_W+x_Y$).
We define the action of ${\nabla}_x$ on an element $f(z,a)\in H$, where $z\in{\bf I\!R}^p$
and $a\in \widetilde{L_\mu}$, as follows
\begin{equation}
\label{eq:actn}
({\nabla}_x(f))(z,a)=({\nabla}_{x_W}(f))(z,a)+2\pi i \langle x_Y,a\rangle f(z,a),
\end{equation}
where the action of ${\nabla}_{x_W}$ is the same as above (only along $z$'s).
Notice that the operators ${\nabla}_x$, $x\in{L_{\theta}}$ satisfy the commutation relations
\[
[{\nabla}_x,{\nabla}_y]=[{\nabla}_{x_W},{\nabla}_{y_W}]=F(x_W,y_W)=F(x,y),~~~x,y\in{L_{\theta}},
\]
since ${\nabla}_{x_Y}$ (recall
that $({\nabla}_{x_Y}(f))(z,a)=2\pi i \langle x_Y,a\rangle f(z,a)$)
commutes with ${\nabla}_{y}$ for any $y\in{L_{\theta}}$.
Thus, we constructed operators
${\nabla}_x$, $x\in{L_{\theta}}$ which satisfy the desired commutation relations.
Next, we will construct operators acting on the space $H$ which
generate a non-commutative torus ${{\cal A}_{\widetilde{\theta}}}$ such that
\begin{equation}
\label{cond}
\begin{array}{l}
{\rm \bf first}~~~{\nabla}~~{\rm is~an}~~{{\cal A}_{\widetilde{\theta}}}~~{\rm connection}\\
{\rm \bf second}~~~\theta-\widetilde{\theta}~~{\rm is~an~element~of}~~
\Lambda^2 W^*_\mu.
\end{array}
\end{equation}
\subsubsection{Construction of operators satisfying conditions
\ref{cond}.}
Let us choose a basis $\beta_1,\cdots,\beta_{2p}$ of $L_\mu$ and
a basis
$\beta_{2p+1},\cdots,\beta_{2p+q}$ of $\widetilde{L_\mu}$.
We will construct operators $V_i$, $1\leq i\leq 2p+q$
acting on $H$ which generate a non-commutative torus ${{\cal A}_{\widetilde{\theta}}}$ which
satisfies conditions \ref{cond}.
\begin{lem}
\label{lem:oper1}
For $1\leq i\leq 2p$ there exists an operator ${\widetilde{V_i}}$ acting on $H$
such that\\
$
{\rm first}~~~({\widetilde{V_i}}(f))(z,a)=
e^{2\pi i \chi_i(z)}f(z+y_i,a),~{\it where}~z\in {\bf I\!R}^p,~~
a\in \widetilde{L_\mu},
$\\
\hspace*{0.5in}
for some $y_i\in {\bf I\!R}^p$ and some linear function $\chi_i\in({\bf I\!R}^p)^*$;\\
{\rm second}
$[{\nabla}_x,{\widetilde{V_i}}]=2\pi i \langle x,\beta_i \rangle {\widetilde{V_i}}$, for
$x\in {L_{\theta}}$.
\end{lem}
{\bf Proof:}
Let us introduce an operator $W(y,\chi)$, where
$y\in {\bf I\!R}^p$ and $\chi\in ({\bf I\!R}^p)^*$
\[
(W(y,\chi)f)(z,a)=e^{2\pi i \chi(z)}f(z+y,a).
\]
A straightforward calculation shows that
$[{\nabla}_x,W(y,\chi)]W(y,\chi)^{-1}$ is an operator of multiplication by a
real number and moreover we obtain a non-degenerate pairing
between the spaces $W_\mu^*$ and ${\bf I\!R}^p\oplus({\bf I\!R}^p)^*$. Thus an choosing
appropriate element $y_i\in {\bf I\!R}^p$ and $\chi_i\in ({\bf I\!R}^p)^*$
we can put ${\widetilde{V_i}}=W(y_i,\chi_i)$.\\
{\hspace*{3in} Q.E.D.}
We define the operators $\widetilde{V_i}$ for $1\leq i\leq 2p$ as in the above lemma.
We define the operators $\widetilde{V_i}$ for $2p+1\leq i\leq 2p+q$ acting on $H$
by the formula
\[
(\widetilde{V_i}f)(z,a)=f(z,a-\beta_i).
\]
\begin{lem}
\label{oper2}
For any $1\leq i \leq n=2p+q$, and any $x\in{L_{\theta}}$ we have
\begin{equation}
\label{eq:comrel}
[{\nabla}_x,\widetilde{V_i}]=2\pi i \langle x,\beta_i\rangle \widetilde{V_i}.
\end{equation}
\end{lem}
{\bf Proof:} For $i\leq 2p$ formula \ref{eq:comrel} follows from
Lemma \ref{lem:oper1}. For $i>2p$ formula \ref{eq:comrel} follows
from an easy straightforward calculation.\\
{\hspace*{3in} Q.E.D.}
It is easy to check that the operators $\widetilde{V_i}$ are generators
of some non-commutative torus. Moreover, these operators satisfy
the first condition in (\ref{cond}) but they do not satisfy the
second condition. To remedy this we will modify the operators $\widetilde{V_i}$
replacing them with operators
$V_i=e^{2\pi i l_i(\cdot)}\widetilde{V_i}$.
If $l\in Y_\mu^*$ then the operator $e^{2\pi i l(\cdot)}\widetilde{V_i}$ acts
on $H$ by the formula
\[
(e^{2\pi i l(\cdot)}\widetilde{V_i}(f))(z,a)=e^{2\pi i l(a)}(\widetilde{V_i}(f))(z,a).
\]
Moreover, we have
\[
[{\nabla}_x,e^{2\pi i l(\cdot)}\widetilde{V_i}]= 2\pi i \langle x,\beta_i\rangle
e^{2\pi i l(\cdot)}\widetilde{V_i}
\]
which follows from an easy straightforward calculation (since the operator
$e^{2\pi i l(\cdot)}$ commutes with the operators ${\nabla}_x$, $x\in{L_{\theta}}$).
\begin{prop}
\label{mainconst}
For $1\leq i\leq 2p+q$ there exists a linear function
$l_i\in Y_\mu^*$ on $Y_\mu$ such that\\
if we define
$V_i=e^{2\pi i l_i(\cdot)}\widetilde{V_i}$
then
\begin{equation}
\label{eq:tettild}
V_i V_j=e^{2\pi i \widetilde{\theta}_{ij}}V_j V_i,
\end{equation}
and $\theta-\widetilde{\theta}$ is an element of $\Lambda^2 W^*_\mu$
\end{prop}
{\bf Proof:} First, it is easy to see that there exists a 2-form
$\sigma\in \Lambda^2 {L_{\theta}}$ such that
\[
\widetilde{V_i} \widetilde{V_j}=e^{2\pi i \sigma_{ij}}\widetilde{V_j} \widetilde{V_i}.
\]
An easy calculation shows that the operator $e^{2\pi i l(\cdot)}$
commutes with the operators $\widetilde{V_i}$ for $i\leq 2p$. If $i>2p$ then
we have
\[
(\widetilde{V_i}\circ e^{2\pi i l(\cdot)})= e^{-2\pi i l(\beta_i)}
(e^{2\pi i l(\cdot)}\circ\widetilde{V_i}).
\]
This gives us that\\
if $i,j\leq 2p$ then
\begin{equation}
\label{eq:ijleqtp}
V_i V_j=e^{2\pi i \sigma_{ij}}V_j V_i;
\end{equation}
if $i\leq 2p$ and $j>2p$ then
\begin{equation}
\label{eq:ileqtpjgtp}
V_i V_j=e^{2\pi i (\sigma_{ij}+l_i(\beta_j))}V_j V_i;
\end{equation}
and if $i,j>2p$ then
\begin{equation}
\label{eq:ijgtp}
V_i V_j=e^{2\pi i (\sigma_{ij}+l_i(\beta_j)-l_j(\beta_i))}V_j V_i.
\end{equation}
For $1\leq i\leq 2p$ we define $l_i\in Y_\mu^*$ by the formula
\[
l_i(\beta_j)=\theta(\beta_i,\beta_j)-\sigma_{ij}
\]
on the basis $\{\beta_j\}$, $j>2p$ of $Y_\mu$.
For $2p<i\leq2p+q$ we define $l_i\in Y_\mu^*$ by the formula
\[
l_i(\beta_j)=\frac{1}{2}(\theta(\beta_i,\beta_j)-\sigma_{ij})
\]
on the basis $\{\beta_j\}$, $j>2p$ of $Y_\mu$.
Equations \ref{eq:ijleqtp}, \ref{eq:ileqtpjgtp}, and \ref{eq:ijgtp}
show that
\[
V_i V_j=e^{2\pi i \theta(\beta_i,\beta_j)}V_j V_i
\]
if either $i$ or $j$ greater then $2p$ and
\[
V_i V_j=e^{2\pi i \sigma_{ij}}V_j V_i
\]
if $i,j\leq 2p$.
Thus we constructed the linear functions $l_i\in Y_\mu^*$ such that
the conditions \ref{eq:tettild} are satisfied.\\
{\hspace*{3in} Q.E.D.}
We define the operators $V_i$ as in the above Lemma. We easily
see that the operators $V_i$ generate a non-commutative torus ${{\cal A}_{\widetilde{\theta}}}$,
where $\widetilde{\theta}(\beta_i,\beta_j)=\sigma_{ij}$ if both $i,j\leq 2p$
and $\widetilde{\theta}(\beta_i,\beta_j)={\theta}(\beta_i,\beta_j)$ otherwise.
Thus, the operators $V_i$ satisfy the condition \ref{cond}.
\subsubsection{Construction of a projective ${{\cal A}_{\widetilde{\theta}}}$
module.}
Now we will identify our construction with the construction given in
\cite{Rf1}.
Let $G$ be a central extension of the abelian group
${\bf I\!R}^p\times \widetilde{L_\mu}\times ({\bf I\!R}^p)^*\times (\widetilde{Y_\mu}/L_\mu^*)$
given by the natural pairing between ${\bf I\!R}^p\times \widetilde{L_\mu}$ and
$({\bf I\!R}^p)^*\times (\widetilde{Y_\mu}/L_\mu^*)$. We see that $G$ is a
Heisenberg group and it acts naturally on $H$.
We denote this representation by
$\rho$.
Moreover, for each $V_i$ there
exists a unique element $g_i\in G$ such that $\rho(g_i)=V_i$.
One can easily recognize the construction of elementary modules
over non-commutative tori in M. Rieffel's paper \cite{Rf1}.
Thus,
choosing an appropriate space of functions on ${\bf I\!R}^p\times \widetilde{L_\mu}$
we get a projective ${{\cal A}_{\widetilde{\theta}}}$module $\widetilde{E}$ with constant
curvature connection ${\nabla}$
such that\\
{\it first}, the curvature of ${\nabla}$ is $F$;\\
{\it second}, $\theta-\widetilde{\theta}$ is an element of $\Lambda^2 W_\mu^*$.
Next, we would like to find explicitly the class $[\widetilde{E}]$ in $K_0({{\cal A}_{\widetilde{\theta}}})$.
Note, that in our construction of module $\widetilde{E}$ we canonically
identified the space ${L_{\theta}}$ with the space $L_{\widetilde{\theta}}$. Thus, we can
think about $[\widetilde{E}]$ as an integral element of
$\Lambda^{even}{L_{\theta}}^*=\Lambda^{even}V$. To find the
class $[\widetilde{E}]$ we would have to do some calculations. Fortunately,
they were already done by M. Rieffel in \cite{Rf1}. So, we will apply his
results to our case.
Let us remind that in paper \cite{Rf1} M. Rieffel introduced a linear
map $\widetilde{T}:L_{\widetilde{\theta}}\rightarrow {\bf I\!R}^p\times {\bf I\!R}^q\times({\bf I\!R}^p)^*$.
In our notation $L_{\widetilde{\theta}}$ is canonically identified with
${L_{\theta}}=V=W_\mu\oplus Y_\mu$ and ${\bf I\!R}^q$ with $Y_\mu$. Thus, in our terms
we have a linear map
$\widetilde{T}:W_\mu\oplus Y_\mu\rightarrow {\bf I\!R}^p\times Y_\mu\times({\bf I\!R}^p)^*
={\bf I\!R}^p\times({\bf I\!R}^p)^*\times Y_\mu$. It is easy to see from the
explicit construction of the operators $V_i$
that $\widetilde{T}$ maps $W_\mu$ to ${\bf I\!R}^p\times({\bf I\!R}^p)^*$,
and $Y_\mu$ to $Y_\mu$. Moreover the restriction
of $\widetilde{T}$ on $Y_\mu$ is the identity map.
M. Rieffel found in \cite{Rf1} that
\begin{equation}
\label{eq:class}
[\widetilde{E}]=d\prod_{j=1}^p\bar{Y}_j\wedge\bar{Y}_{j+p},
\end{equation}
where $d={\rm det}(\widetilde{T})$ and
\begin{equation}
\label{eq:Y}
\bar{Y}_j=
\begin{array}{l}
\widetilde{T}^{-1}(\bar{e}_j)~~~{\rm for}~~~1\leq j\leq p\\[10pt]
\widetilde{T}^{-1}(e_{j-p})~~~{\rm for}~~~p+1\leq j\leq 2p,
\end{array}
\end{equation}
where $\{e_j\}$ is a basis of ${\bf I\!R}^p$ and $\{\bar{e}_j\}$ is the dual basis
of $({\bf I\!R}^p)^*$.
Since $\widetilde{T}$ is the identity map on $Y_\mu$ we see that
\[
{\rm det}(\widetilde{T}^{-1})=\pm\frac{\prod_{j=1}^p\bar{Y}_j\wedge\bar{Y}_{j+p}}
{\alpha},
\]
where $\alpha$ is the volume form on $W_\mu$ (see Proposition \ref{prop:gqe}
for the definition of $\alpha$). Note, we put a $\pm$ sign because we
do not want to specify precisely how to pick a volume form. The lattice
$L_\mu$ specifies the volume form upto a sign. Later it will be easy to
make the right choice of the sign so that everything would agree with
M. Rieffel's paper \cite{Rf1}.
We get
\[
d={\rm det}(\widetilde{T})=1/{\rm det}(\widetilde{T}^{-1})= \pm\frac{\alpha}
{\prod_{j=1}^p\bar{Y}_j\wedge\bar{Y}_{j+p}}.
\]
Thus we obtain that
\begin{equation}
\label{eq:classEt}
[\widetilde{E}]=\pm\alpha.
\end{equation}
\subsubsection{Construction of a projective ${{\cal A}_{\theta}}$
module $E$.}
{}From the above results and Proposition \ref{prop:intgqe} we see that
if we make the right choice of the sign (so that $[\widetilde{E}]=\alpha$) then
\begin{equation}
\label{defN}
N=\frac{\mu_{(2p)}}{\alpha}
\end{equation}
is a positive integer. Moreover, we have
\[
d_{\widetilde {E}} e^{\frac{F}{2\pi i}}=e^{\iota(\theta)}\mu
\]
and
\[
d e^{\frac{F}{2\pi i}}=e^{\iota(\widetilde{\theta})}\alpha.
\]
Therefore,
\begin{equation}
\label{equal}
e^{\iota(\theta)}\mu=\left(\frac{d_{\widetilde
E}}{d}\right)e^{\iota(\widetilde{\theta})}\alpha.
\end{equation}
{}From equations \ref{equal} and \ref{defN} we see that
$\frac{d_{\widetilde E}}{d}=N$.
One can easily check that the conditions of Lemma \ref{lem:ration}
are satisfied. Therefore
$N(\theta-\widetilde{\theta})$ is an integral element of $(W_\mu)^*=X_\mu$.
{}From Proposition \ref{prop:taumod} follows that there exists
$N$-dimensional
module $M$ over ${\cal A}_{\theta-\widetilde{\theta}}$.
Denote by
\begin{equation}
\label{E}
E=\widetilde{E}\otimes M.
\end{equation}
{}From Proposition 5.4 and Theorem 5.6 in M. Rieffel's paper
\cite{Rf1} it follows
that $E$ is a projective module over ${{\cal A}_{\theta}}$ (since
$\theta=\widetilde{\theta}+(\theta-\widetilde{\theta})$) with constant curvature
connection with the curvature
given by formula
\[
\Omega=F\otimes {\rm Id}_M=F
\]
and
\[
{\rm ch}(E)={\mbox{\rm \tiny dim}}(M){\rm ch}(\widetilde{E}).
\]
Therefore we see that
\[
{\rm ch}(E)=N{\rm ch}(\widetilde{E})=
N e^{\iota(\widetilde{\theta})}\alpha=
N \left(\frac{d}{d_E}\right)e^{\iota(\theta)}\mu=
e^{\iota(\theta)}\mu.
\]
Thus we constructed a projective ${{\cal A}_{\theta}}$-module $E$
with constant curvature connection
such that
$[E]=\mu$. This finishes the proof of Theorem \ref{thm:M1}\\
{\hspace*{3in} Q.E.D.}
{\bf Acknowledgements.} We would like to express our deep gratitude to
Dmitry Fuchs, Anatoly Konechny and Marc Rieffel for
numerous fruitfull discussions.
\newpage
|
\section{Introduction}
The so-called mean field equation of the Onsager-Joyce-Montgomery (OJM)
theory$^{1,2,3}$ for the equilibrium vorticity distributions of a
two-dimensional inviscid and incompressible fluid (and guiding center line
charge model in plasmas) were derived more than twenty years ago but the
volume of research concerning these equations continues unabated. Recently,
the mean field thermodynamic limit for the planar vortex model has been
shown to exist by Caglioti et al $^4$, and independently by Kiessling$^5$,
and the existence of negative temperature states in this model was
demonstrated by Eyink and Spohn $^6$. Furthermore, these authors established
the maximum entropy principle$^7$ for this theory. Onsager$^1$ predicted
that the negative temperature equilibrium states exhibit a clustering of the
point vortices into coherent structures. Exact solutions of the OJM mean
field equations obtained by Chen et al $^8$ confirm the existence of such
coherent states in inviscid turbulent flows.
A OJM type equilibrium statistical theory for nearly parallel thin vortex
filaments was very recently put forth by Lions and Majda$^9$. They
established the mean field thermodynamic limit and the maximum entropy
principle for their problem. The mean field equation that Lions and Majda
obtained, is radically different from other mean field theories for vortex
dynamics in the sense that their equations are time-dependent. Nonetheless,
in a special singular limit of perfectly parallel vortex filaments, their
mean field equation reduces to the sinh-Poisson mean field equation of the
planar point vortex problem.
Concerning the relationship between the OJM and other statistical theories,
Majda and Holen$^{10}$ showed that both the OJM theory and Kraichnan's
truncated spectral theory$^{11}$ are statistically sharp with respect to the
recent infinite constraints theory of Miller$^{12,13}$ and Robert$^{14}$,
i.e., the few constraints and infinite constraints theory agree for low
energies. Chorin$^{15}$ has earlier indicated that a few constraints is
sufficient for a reasonable theory of statistical equilibrium in these
models. He also demonstrated numerically that the Miller-Robert theory is
valid only for moderate temperatures and has no Kosterlitz-Thouless
phase-transition except at zero temperature$^{16}$. Turkington and Whittaker$%
^{17}$ gave a numerical algorithm for the Miller-Robert theory. Recently,
Turkington$^{18}$ critiqued the Miller-Robert theory and presented a few
constraints equilibrium statistical theory for 2-D coherent structures,
where the equality constraints of the Miller-Robert theory are replaced by
inequalities. Turkington$^{18}$ argued that the finite ultra-violet cutoff
implicit in the Miller-Robert theory does not reflect the true inviscid
vortex dynamics as represented in the two dimensional Euler equations. In
their preprint$^{19}$, Boucher, Ellis, and Turkington presented rigorous
large deviation results and proved the existence of the mean field
thermodynamic limit for Turkington's theory.
Yet, several issues remain unresolved ---in spite of the above recent works,
it is still not completely clear where the OJM theory stands in relation to
other equilibrium statistical theories of two dimensional turbulence$%
^{11,12,13,14}$ especially with regard to their relative efficacy in
modelling real physical problems in fluid mechanics. Moreover, it is still
not clear how the OJM theory is related to the physicists' standard
variational formulations of mean field theory such as in Chapter 6 and 7 of
the text$^{20}$. It is the main aim of this paper to derive equation (\ref
{mft1}) in a new way, thereby demonstrating that the OJM theory is indeed a
mean field theory in the sense of Weiss, Peierls and Feynman$^{20}$ .
Specifically, we will prove that equation (\ref{mft1}) is the limit, as the
number of particles $N\rightarrow \infty ,$ of mean field equations that are
derived using the Bogoliubov-Feyman inequality$^{21}$ (\ref{bog}) , the
Gibbs entropy function and Landau's approximation$^{20}$. Previous
derivations of the OJM theory$^{1,2,3}$ are based on Boltzmann's entropy
function instead of Gibbs' entropy function. We note here the significant
point that our result holds for all initial vorticity distributions $q_o(x).$
A OJM-type theory for point vortex systems on a rotating sphere$^{22}$ has
recently been obtained by Lim$^{23}$, and the corresponding mean field
thermodynamic limit was established by methods similar to those of Kiessling$%
^5.$
\section{Mean field theory}
All the above statistical theories for two dimensional turbulence in
inviscid fluids have a common thread in the maximum entropy principle of
information theory$^7$. They lead to nonlinear elliptic equations, the
so-called mean field equations which have the generic form
\[
\Delta \Psi =F(\Psi ,\gamma )
\]
for the stream function $\Psi $ of the mean field locally averaged vorticity
distributions $\bar{q}(x)=-\Delta \Psi ,$ and some parameters $\gamma .$
Indeed it can be shown that $\bar{q}(x)$ are exact stationary solutions of
the Euler equations. The mean field equation of the OJM theory for a single
component vortex gas is the nonlinear elliptic equation
\begin{equation}
\Delta \Psi =ke^{-\beta \Psi }. \label{mft1}
\end{equation}
Heuristically, this equation has the correct form for a mean field theory in
the sense that the precise interactions between particles has been replaced
by an approximate energy expression where the particles interact with some
\textit{mean }field.
The main tool for the work in this section is the Bogoliubov-Feynman
inequality$^{20}$ for the free energy in the form proven by Feynman$^{21}$:
\begin{equation}
F\leq F_0+\left\langle H_1\right\rangle _0\equiv F_{var}, \label{bog}
\end{equation}
where $F_0$ is the free energy based on the approximate Hamiltonian $H_0$,
where $H=H_0+H_1,$ and the second term on the right is the thermal average
of the remainder $H_1$ in the canonical measure based on $H_0.$ In other
words,
\begin{equation}
\left\langle \cdot \right\rangle _0\equiv \frac{\int_CdV\text{ }(\cdot )%
\text{ }\exp \{-\beta H_0\}}{Z_0}, \label{zero}
\end{equation}
where the partition function $Z_0=\int_CdV$ $\exp \{-\beta H_0\}$ is based
on the approximate Hamiltonian $H_0.$ In the standard mean field theory$%
^{20} $, one choose $H_0$ so that $Z_0$ can be evaluated easily, and yet it
must have certain basic physical properties to make the theory physically
relevant. For negative temperatures, a derivation similar to Feynman's
yields
\begin{equation}
F\geq F_0+\left\langle H_1\right\rangle _0\equiv F_{var}^{-}. \label{bogneg}
\end{equation}
For the OJM theory of point vortices in the plane where the vortices are
identical and have vorticity or charge $\lambda $, we shall choose the
coarse-grained approximate Hamiltonian based on the division of the physical
domain $\Omega \subseteq R^2$ of area $A$ into $M$ equal boxes,
\[
H_0=-\frac 12\sum_{i=1}^M\sum_{j\neq i}^Mn_in_j\lambda ^2\log |\vec{x}_i^0-%
\vec{x}_j^0|.
\]
Here $n_i$ denotes the number of vortices in box $B_i$ which has area $h^2,$
and $\vec{x}_i^0$ is the location of the center of $B_i.$ Furthermore, the
total number of particles is $N$, i.e.,
\[
\sum_{i=1}^Mn_i=N.
\]
Thus, $\vec{x}_i^0$ are no longer dependent on time but depend instead on
the lattice which is implicit in the coarse-grained Hamiltonian $H_0.$ Since
the full Hamiltonian of the point vortex model is
\[
H=-\frac 12\sum_{i=1}^N\sum_{j\neq i}^N\lambda ^2\log |\vec{x}_i-\vec{x}_j|,
\]
the remainder $H_1$ is given by
\[
H_1=H-H_0
\]
\[
=-\frac 12\sum_{i=1}^M\left( \sum_{j=1}^{n_i}\sum_{k\neq j}^{n_i}\lambda
^2\log |\vec{x}_j-\vec{x}_k|\right)
\]
\[
-\frac 12\sum_{i=1}^M\sum_{i^{\prime }\neq 1}^M\left[ \left(
\sum_{j=1}^{n_i}\sum_{k=1}^{n_{i^{\prime }}}\lambda ^2\log |\vec{x}_j-\vec{x}%
_k|\right) -n_in_{i^{\prime }}\lambda ^2\log |\vec{x}_i^0-\vec{x}_{i^{\prime
}}^0|\right]
\]
\medskip\
\[
=-\frac 12\sum_{i=1}^M\left( \sum_{j=1}^{n_i}\sum_{k\neq j}^{n_i}\lambda
^2\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\right)
\]
\[
-\frac 12\sum_{i=1}^M\sum_{i^{\prime }\neq 1}^M\left[ \left(
\sum_{j=1}^{n_i}\sum_{k=1}^{n_{i^{\prime }}}\lambda ^2\log |(\vec{x}_i^0-%
\vec{x}_{i^{\prime }}^0)+(\vec{x}_j^{\prime }-\vec{x}_k^{\prime })|\right)
-n_in_{i^{\prime }}\lambda ^2\log |\vec{x}_i^0-\vec{x}_{i^{\prime
}}^0|\right] ,
\]
where we have used
\[
\vec{x}_j-\vec{x}_k=(\vec{x}_i^0-\vec{x}_i^0)+(\vec{x}_j^{\prime }-\vec{x}%
_k^{\prime })=\vec{x}_j^{\prime }-\vec{x}_k^{\prime },
\]
in the first sum, and
\[
\vec{x}_j-\vec{x}_k=(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)+(\vec{x}_j^{\prime
}-\vec{x}_k^{\prime })
\]
in the second sum. In other words the primed vector $\vec{x}_j^{\prime }$
denotes the difference $\vec{x}_j-\vec{x}_i^0,$ which represents the vector
from the center of $B_i$ to vortex $j$ in it.
From
\begin{eqnarray*}
(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)+(\vec{x}_j^{\prime }-\vec{x}_k^{\prime
}) &=&(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)+(\vec{x}_j^{\prime }-\vec{x}%
_k^{\prime }) \\
&=&(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)\left( 1+\frac{(\vec{x}_j^{\prime }-%
\vec{x}_k^{\prime })}{(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)}\right) ,
\end{eqnarray*}
one derives
\begin{eqnarray*}
\log |(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)+(\vec{x}_j^{\prime }-\vec{x}%
_k^{\prime })| &=&\log |(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)|+\log \left( 1+%
\frac{|(\vec{x}_j^{\prime }-\vec{x}_k^{\prime })|}{|(\vec{x}_i^0-\vec{x}%
_{i^{\prime }}^0)|}\right) \\
&=&\log |(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)|+\frac{|(\vec{x}_j^{\prime }-%
\vec{x}_k^{\prime })|}{|(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)|}+O\left(
\frac{|(\vec{x}_j^{\prime }-\vec{x}_k^{\prime })|}{|(\vec{x}_i^0-\vec{x}%
_{i^{\prime }}^0)|}\right) ^2,
\end{eqnarray*}
using the complex form of the logarithm, a step which we omit here for the
sake of brevity. Substituting this last inequality back into $H_1$ and
keeping only terms of order $|(\vec{x}_j^{\prime }-\vec{x}_k^{\prime })|/|(%
\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)|,$ we get
\[
H_1=-\frac 12\sum_{i=1}^M\left( \sum_{j=1}^{n_i}\sum_{k\neq j}^{n_i}\lambda
^2\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\right) -\frac
12\sum_{i=1}^M\sum_{i^{\prime }\neq 1}^M\left(
\sum_{j=1}^{n_i}\sum_{k=1}^{n_{i^{\prime }}}\lambda ^2\frac{|(\vec{x}%
_j^{\prime }-\vec{x}_k^{\prime })|}{|(\vec{x}_i^0-\vec{x}_{i^{\prime }}^0)|}%
\right) ,
\]
where the first group of terms represent the intra-box interaction energy,
and the second group represents $O(h)$ terms in the inter-box interaction
energy.
To begin, we compute the partition function $Z_0$ which is based on the
approximate Hamiltonian $H_0,$ as follows
\begin{eqnarray*}
Z_0 &=&\sum_sW(s)h^{2N}\exp \left( -\beta H_0\right) \\
&=&\sum_sW(s)h^{2N}\exp \left( \frac \beta 2\sum_{i=1}^M\sum_{j\neq
i}^Mn_in_j\lambda ^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) ,
\end{eqnarray*}
where the summation is taken over all coarse-grained state (or macro-state) $%
s=(n_1,...,n_M)$ such that $\sum_{i=1}^Mn_i=N,$ and the degeneracy of the
macro-state $s$ is given by
\[
W(s)\equiv \frac{N!}{n_1!...n_M!}.
\]
The probability distribution $P_0(s)$ for a macrostate $s,$ is now defined
by:
\begin{eqnarray}
P_0(s) &=&\int_{D^N(s)}\frac{\exp \left( -\beta H_0\right) }{Z_0}d^NA
\label{prob} \\
&=&\frac{W(s)h^{2N}\exp \left( -\beta H_0\right) }{Z_0}, \nonumber
\end{eqnarray}
where $D^N(s)$ denotes the part of phase space $D^N$ which is occupied by
the microstates associated with $s.$ Finally, we compute the free energy $%
F_0 $ as follows:
\begin{eqnarray*}
F_0 &=&-\frac 1\beta \log Z_0 \\
&=&-\frac 1\beta \left[ N\log h^2+\log \left\{ \sum_sW(s)\exp \left( \frac
\beta 2\sum_{i=1}^M\sum_{j\neq i}^Mn_in_j\lambda ^2\log |\vec{x}_i^0-\vec{x}%
_j^0|\right) \right\} \right] .
\end{eqnarray*}
In the next step in this procedure, we will compute $\left\langle
H_1\right\rangle _0$ up to $O(h):-$%
\begin{eqnarray*}
\left\langle H_1\right\rangle _0 &=&\int\limits_{D^N}H_1(\vec{x})\text{ }P_0(%
\vec{x})\text{ }d\vec{x} \\
&=&\int\limits_{D^N}H_1(\vec{x})\frac{\exp \left( -\beta H_0(\vec{x})\right)
}{Z_0}d\vec{x}_1...d\vec{x}_N \\
&=&\int\limits_{D^N}\left[
\begin{array}{c}
-\frac 12\sum\limits_{i=1}^M\left( \sum\limits_{j=1}^{n_i}\sum\limits_{k\neq
j}^{n_i}\lambda ^2\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\right) \\
-\frac 12\sum\limits_{i=1}^M\sum\limits_{i^{\prime }\neq i}^M\left(
\sum\limits_{j=1}^{n_i}\sum\limits_{k=1}^{n_{i^{\prime }}}\lambda ^2\frac{|(%
\vec{x}_j^{\prime }-\vec{x}_k^{\prime })|}{|(\vec{x}_i^0-\vec{x}_{i^{\prime
}}^0)|}\right)
\end{array}
\right] \frac{\exp \left( -\beta H_0(\vec{x})\right) }{Z_0}d\vec{x}_1...d%
\vec{x}_N \\
&=&\int\limits_{D^N}\left[ -\frac 12\sum_{i=1}^M\left(
\sum_{j=1}^{n_i}\sum_{k\neq j}^{n_i}\lambda ^2\log |\vec{x}_j^{\prime }-\vec{%
x}_k^{\prime }|\right) \right] \frac{\exp \left( -\beta H_0(\vec{x})\right)
}{Z_0}d\vec{x}_1...d\vec{x}_N\text{ }+O(h).
\end{eqnarray*}
where $\vec{x}\equiv (\vec{x}_1,...,\vec{x}_N)$ is a microstate$.$
Substituting the above expressions for $Z_0,$ $F_0$ into (\ref{bog}), the
free energy upper bound when the temperature $\beta >0,$ is given by
\[
F_{var}\equiv F_0+\left\langle H_1\right\rangle _0
\]
\[
=
\begin{array}{c}
-\frac 1\beta N\log h^2-\frac 1\beta \log \left\{ \dsum\limits_sW(s)\exp
\left( \frac \beta 2\sum\limits_{i=1}^M\sum\limits_{j\neq i}^Mn_in_j\lambda
^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) \right\} \\
+\dint\limits_{D^N}\left[ -\frac 12\sum\limits_{i=1}^M\left(
\sum\limits_{j=1}^{n_i}\sum\limits_{k\neq j}^{n_i}\lambda ^2\log |\vec{x}%
_j^{\prime }-\vec{x}_k^{\prime }|\right) \right] \frac{\exp \left( -\beta
H_0(\vec{x})\right) }{Z_0}d\vec{x}_1...d\vec{x}_N\text{ }
\end{array}
;
\]
similarly for negative temperatures $\beta <0,$ the free energy lower bound
is given by
\[
F_{var}^{-}\equiv F_0+\left\langle H_1\right\rangle _0
\]
\[
=
\begin{array}{c}
-\frac 1\beta N\log h^2-\frac 1\beta \log \left\{ \dsum\limits_sW(s)\exp
\left( \frac \beta 2\sum\limits_{i=1}^M\sum\limits_{j\neq i}^Mn_in_j\lambda
^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) \right\} \\
+\dint\limits_{D^N}\left[ -\frac 12\sum\limits_{i=1}^M\left(
\sum\limits_{j=1}^{n_i}\sum\limits_{k\neq j}^{n_i}\lambda ^2\log |\vec{x}%
_j^{\prime }-\vec{x}_k^{\prime }|\right) \right] \frac{\exp \left( -\beta
H_0(\vec{x})\right) }{Z_0}d\vec{x}_1...d\vec{x}_N\text{ }
\end{array}
.
\]
Whether the temperture is positive or negative, the integral in the above
equation can be evaluated as follows:
\begin{equation}
\sum_sW(s)\int\limits_{B_1^{n_1}}d^{n_1}\vec{x}^{\prime }\text{ }%
....\int\limits_{B_M^{n_m}}d^{n_M}\vec{x}^{\prime }\text{ }\left[
\begin{array}{c}
-\frac 12\sum\limits_{i=1}^M\left( \sum\limits_{j=1}^{n_i}\sum\limits_{k\neq
j}^{n_i}\lambda ^2\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\right) \\
\times \frac{\exp \left( \frac \beta 2\sum\limits_{i=1}^M\sum\limits_{j\neq
i}^Mn_in_j\lambda ^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) }{Z_0}
\end{array}
\right] \label{four}
\end{equation}
\[
=-\frac 1{2Z_0}\sum_sW(s)\left[
\begin{array}{c}
\exp \left( \frac \beta 2\sum\limits_{i=1}^M\sum\limits_{j\neq
i}^Mn_in_j\lambda ^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) \\
\times \sum\limits_{i=1}^M\dint\limits_{B_i^{n_i}}d^{n_i}\vec{x}^{\prime }%
\text{ }\left( \sum\limits_{j=1}^{n_i}\sum\limits_{k\neq j}^{n_i}\lambda
^2\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\right)
\end{array}
\right]
\]
\[
=-\frac 1{2Z_0}\sum_s\left\{
\begin{array}{c}
W(s)\prod\limits_{i\neq j}^M|\vec{x}_i^0-\vec{x}_j^0|^{\frac{\beta
n_in_j\lambda ^2}2} \\
\times \sum\limits_{i=1}^M\dint\limits_{B_i^{n_i}}d^{n_i}\vec{x}^{\prime }%
\text{ }\left( \sum\limits_{j=1}^{n_i}\sum\limits_{k\neq j}^{n_i}\lambda
^2\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\right)
\end{array}
\right\} .
\]
An application of the Mean Value Theorem implies that the integrand in each
of the integrals in the last line above can be replaced as follows:
\begin{equation}
\sum_{j=1}^{n_i}\sum_{k\neq j}^{n_i}\lambda ^2\log |\vec{x}_j^{\prime }-\vec{%
x}_k^{\prime }|=\frac{\lambda ^2n_i(n_i-1)}2L(n_i), \label{L}
\end{equation}
where $L(n_i)$ is a large negative constant of the order of
\[
\log |\vec{x}_j^{\prime }-\vec{x}_k^{\prime }|\simeq \log \frac h{\sqrt{n_i}%
}.
\]
Therefore, the term
\[
I=\sum\limits_{i=1}^M\dint\limits_{B_i^{n_i}}d^{n_i}\vec{x}^{\prime }\left(
\sum\limits_{j=1}^{n_i}\sum\limits_{k\neq j}^{n_i}\lambda ^2\log |\vec{x}%
_j^{\prime }-\vec{x}_k^{\prime }|\right)
\]
is given by
\[
I(n_i)=\sum_{i=1}^M\frac{\lambda ^2n_i(n_i-1)}2L(n_i)h^{2n_i}.
\]
Substituting this expression in (\ref{four}), we get
\[
\dint\limits_{D^N}\left[ -\frac 12\sum_{i=1}^M\left(
\sum_{j=1}^{n_i}\sum_{k\neq j}^{n_i}\lambda ^2\log |\vec{x}_j^{\prime }-\vec{%
x}_k^{\prime }|\right) \right] \frac{\exp \left( -\beta H_0(\vec{x})\right)
}{Z_0}d\vec{x}_1...d\vec{x}_N
\]
\[
\simeq \frac 1{2Z_0}\sum_s\left\{ W(s)\left( \prod_{i\neq j}^M|\vec{x}_i^0-%
\vec{x}_j^0|^{\frac{\beta n_in_j\lambda ^2}2}\right) \sum_{i=1}^M\frac{%
\lambda ^2n_i(1-n_i)}2L(n_i)h^{2n_i}\right\} .
\]
Thus, both $F_{var}^{-}$ and $F_{var}$ up to order $O(h)$ is given by$:$%
\begin{eqnarray}
F_{var}^{-}=F_{var}=-\frac 1\beta \left[ N\log h^2+\log \left\{
\sum_sW(s)\exp \left( \frac \beta 2\sum_{i=1}^M\sum_{j\neq i}^Mn_in_j\lambda
^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) \right\} \right] && \label{fvarr} \\
-\frac 1{4Z_0}\sum_s\left\{ W(s)\left( \prod_{i\neq j}^M|\vec{x}_i^0-\vec{x}%
_j^0|^{\frac{\beta n_in_j\lambda ^2}2}\right) \sum_{i=1}^M\lambda
^2n_i(n_i-1)L(n_i)h^{2n_i}\right\} +O(h) &&. \nonumber
\end{eqnarray}
\subsection{Bounds}
The strategy of variational mean field theories$^{20}$ at this point is to
introduce some parameters $\gamma $ into the expressions for $F_{var}$
(respectively $F_{var}^{-})$ and minimize $F_{var}\{\gamma \}$ (resp.
maximize $F_{var}^{-}\{\gamma \})$ with respect to $\gamma .$ This yields
the best possible approximation to the full free energy $F$ within the class
determined by the chosen reduced Hamiltonian $H_0.$ Since
\[
F=-\frac 1\beta \log Z,
\]
we now have the best approximation for $Z.$ In our derivation of the mean
field theory for the point vortex model, these parameters are given by the
macrostate $s=\{n_1,...n_M\}$ which are occupation numbers of the boxes $B_i$
in the statistical coarse-graining procedure. We will now use the facts that
the energy $H$ of the system, the entropy $S_0$ and the partition function $%
Z_0$ are all highly focussed at the equilibrium (most probable) macrostate $%
s^{*}$ in the mean field limit of large $N$. This is essentially the Landau
approximation in the Ginsburg-Landau theory for phase transitions$^{20}$. In
particular, we will use the following consequence of the Landau
approximation
\[
P_0(s^{*})=\frac{W(s^{*})\left( \prod\limits_{j\neq i}^M|\vec{x}_i^0-\vec{x}%
_j^0|^{\frac{\beta n_in_j}2\lambda ^2}\right) }{\sum\limits_sW(s)\left(
\prod\limits_{j\neq i}^M|\vec{x}_i^0-\vec{x}_j^0|^{\frac{\beta n_in_j}%
2\lambda ^2}\right) }\simeq 1,
\]
to analyse the expressions (\ref{fvarr}). Thus $F_{var}$ and $F_{var}^{-}$ (%
\ref{fvarr}) become
\begin{eqnarray}
F_{var}^{-}=F_{var}\simeq -\frac 1\beta \left[ N\log h^2+\log \left\{
W(s^{*})\exp \left( \frac \beta 2\sum_{i=1}^M\sum_{j\neq i}^Mn_in_j\lambda
^2\log |\vec{x}_i^0-\vec{x}_j^0|\right) \right\} \right] && \label{fvarr3}
\\
-\frac 14\left\{ \sum_{i=1}^M\lambda ^2n_i(n_i-1)L(n_i)\right\} &&
\nonumber
\end{eqnarray}
for all $\beta \geq 0$ and for a range of values $0>\beta >\beta ^{*},$
where $\beta ^{*}$ is given by the theory for the existence of the mean
field thermodynamic limit $^{4,5}.$
In order to minimize $F_{var}$ and maximize $F_{var}^{-}$ (\ref{fvarr3})
with respect to $s=\{n_1,...,n_m\}$ under the constraint
\[
\sum_{i=1}^Mn_i=N,
\]
we augment $F_{var}(s)$ and $F_{var}^{-}(s)$ by adding the auxillary term
with Lagrange multiplier $\alpha $ to obtain
\begin{eqnarray*}
\tilde{F}_{var}(s) &=&F_{var}+\alpha (\sum_{i=1}^Mn_i-N), \\
\tilde{F}_{var}^{-}(s) &=&F_{var}^{-}+\alpha (\sum_{i=1}^Mn_i-N),
\end{eqnarray*}
Then taking the gradient $\nabla _s\tilde{F}_{var}(s)=0$ (resp. $\nabla _s%
\tilde{F}_{var}^{-}(s)=0)$ yields
\[
\frac 1\beta (\log n_i+1)+\sum_{j\neq i}^M\lambda ^2n_j\log |\vec{x}_i^0-%
\vec{x}_j^0|
\]
\begin{equation}
+\lambda ^2\left[ \frac{n_i(n_i-1)}4L^{\prime }(n_i)+\frac{2n_i-1}4L(n_i)+%
\frac{n_i(n_i-1)}2L(n_i)\right] +\alpha =0\text{ } \label{grad1}
\end{equation}
for $i=1,...,M.$ We note that $\nabla _s\tilde{F}_{var}(s)$ (resp. $\nabla _s%
\tilde{F}_{var}^{-}(s))$ differs from $\nabla _s\tilde{F}_0(s)$ by the
derivatives of the term
\begin{equation}
\sum_{i=1}^M\frac{\lambda ^2n_i(n_i-1)}4L(n_i), \label{self}
\end{equation}
which is associated with the self energies of each of the $M$ boxes $B_i$ in
the coarse-graining procedure. In working with $F_{var}(s^{*})$ (resp. $%
F_{var}^{-}(s))$ instead of $F_0,$ we have put back these self energy terms.
Minimizing $F_{var}$ (resp. maximizing $F_{var}^{-}$ in the case of negative
temperatures) yields a lower bound for $F_{var}$ (resp. an upper bound for $%
F_{var}^{-})$ which provides the best approximation for the free energy $F$
of the point vortex system$^{20}$ within the ansatz of assuming $H_0$ to be
the approximate (mean field) Hamiltonian. Solving for $n_i$ in (\ref{grad1})
yields the occupation numbers
\[
n_i=e^{-\alpha }\exp \left( -\beta \left( \sum_{j\neq i}^M\lambda ^2n_j\log |%
\vec{x}_i^0-\vec{x}_j^0|+\lambda ^2\left[ \frac{n_i(n_i-1)}4L^{\prime }(n_i)+%
\frac{2n_i^2-1}4L(n_i)\right] \right) \right) .
\]
Next we turn to the analysis of the behaviour of this expression as $%
N,M\rightarrow \infty $ while the vortex strength is scaled by $\frac 1N,$
i.e., $\lambda =\frac 1N$ --- it is easy to see that the following limits
are valid:
\begin{eqnarray}
\frac{n_i}N &\rightarrow &\xi (\vec{x}_i^0)\text{ }d^2x, \label{scale1} \\
\frac{e^{-\alpha }}{Nh^2} &\rightarrow &d, \nonumber \\
\frac 1{N^2}\left[ \frac{n_i(n_i-1)}4L^{\prime }(n_i)+\frac{2n_i^2-1}%
4L(n_i)\right] &\rightarrow &E^1(\vec{x}_i^0), \nonumber
\end{eqnarray}
\[
\sum_{j\neq i}^M\frac{n_j}{N^2}\log |\vec{x}_i^0-\vec{x}_j^0|\rightarrow E^0(%
\vec{x}_i^0).
\]
Thus the mean field equations in the planar point vortex theory is given by:
\begin{equation}
\xi (\vec{x})=\Delta \Psi =d\exp \left( -\beta (E^0(\vec{x})+E^1(\vec{x}%
))\right) , \label{mft2}
\end{equation}
for all temperatures, which differs from that in the OJM theory by the self
energy term $E^1(\vec{x}).$
We will now analyse the self energy expression in some detail. As we let $%
N,M\rightarrow \infty $ the vortex strengths have to be scaled by $\frac 1N$
in order that the mean field (non-extensive) thermodynamic limit exists, as
was shown by Caglioti et al$^4$, Kiessling$^5$ and Eyink and Spohn$^6$.
Since $N\leq M\max_i(n_i)$ and $h^2=A/M,$ we have the bound
\begin{eqnarray*}
L(n_i) &=&\frac 12\log \frac{h^2}{n_i}\geq \frac 12\log \left[ h^2\left(
\frac 1{\max_i(n_i)}\right) \right] \\
&=&\frac 12\log A-\frac 12\log M-\frac 12\log [\max_i(n_i)] \\
&=&\frac 12\log A-\frac 12\log [M\max_i(n_i)] \\
&\geq &\frac 12\log A-\frac 12\log N,
\end{eqnarray*}
which implies that for each $i=1,...,M,$
\begin{equation}
|L(n_i)|\leq \frac 12\log N-\frac 12\log A. \label{lb}
\end{equation}
The self energy term in (\ref{mft2}) scales as follows
\begin{eqnarray*}
|E^1(\vec{x})| &\leq &\frac{n_i^2}{N^2}|L^{\prime }(n_i)|+\frac{n_i^2}{N^2}%
|L(n_i)| \\
&\leq &\frac{n_i^2}{N^2}|L^{\prime }(n_i)|+\frac{n_i\max_in_i}N\left( \frac{%
|L(n_i)|}N\right) \\
&\leq &\frac{n_i^2}{N^2}|L^{\prime }(n_i)|+n_i\frac{|L(n_i)|}N,
\end{eqnarray*}
where we have used the fact that $\left( \frac{\max_in_i}N\right) <1.$ The
first term tends to zero as $N\rightarrow \infty $ because
\begin{eqnarray*}
\frac{n_i^2}{N^2}|L^{\prime }(n_i)| &=&\frac{n_i^2}{N^2}\left( \frac
1{2n_i}\right) =\frac{n_i}{2N^2} \\
&\leq &\frac 1{2N}\rightarrow 0.
\end{eqnarray*}
By (\ref{lb}), the second term tends to zero as $N\rightarrow \infty ,$
i.e.,
\[
n_i\frac{|L(n_i)|}N\leq \frac{n_i}N\left( -\frac 12\log A+\frac 12\log
N\right) \rightarrow 0\text{ },
\]
because the number $n_i$ $\sim \frac NM$ of particles in box $B_i$ stays
about the same as the total number $N$ of particles and the number $M$ of
equal boxes in the statistical coarse-graining procedure both tend to $%
\infty .$ We have shown that
\textbf{Theorem}: The mean field equation (\ref{mft2}) of the
Bogoliubov-Feynman-Landau mean field theory for point vortex dynamics tends
to the mean field equation (\ref{mft1}) of the OJM theory in the mean field
thermodynamic (non-extensive) limit of infinite particles $N\rightarrow
\infty ,$ and infinite number of boxes $M\rightarrow \infty $ in the
coarse-graining procedure in the definition of the approximate Hamiltonian $%
H_0.$
\section{Concluding remarks}
The following discussion gives a brief derivation of another formulation for
$F_{var}$ and $F_{var}^{-}$. By definition, the free energy based on the
Hamiltonian $H_0$ is given by
\[
F_0=\left\langle H_0\right\rangle _0-TS_0,
\]
where the expectation operator $\left\langle \cdot \right\rangle _0$ is
defined by (\ref{zero}), and $S_0$ is the Gibbs entropy function based on $%
H_0,$ i.e.,
\[
S_0=-k_B\sum_sP_0(s)\log P_0(s),
\]
with
\[
P_0(s)\equiv \frac{W(s)h^{2N}\exp (-\beta H_0(s))}{Z_0}.
\]
Thus, the upper bound $F_{var}$ (resp. lower bound $F_{var}^{-})$ for the
free energy $F$
\begin{equation}
F_{var}^{-}=F_{var}=\left\langle H_0+H_1\right\rangle _0-TS_0. \label{bog2}
\end{equation}
is exactly equal to the free energy based on the full Hamiltonian $H,$ but
using the probability distribution $P_0.$
To summarize, we have derived the OJM theory by a new method which is based
on the Bogoliubov-Feynman inequality, the Gibbs entropy function and
Landau's approximation, and showed that it is a mean field theory. It is
remarkable that no additional conditions on the initial vorticity
distributions was needed to prove this result. In the case of a two
component vortex gas, our procedure gives the well-known sinh-Poisson
equation$^2$. The analogue of the above theorem for point vortices on a
rotating sphere is presented in Lim$^{23}$. We note that previous
derivations of the OJM theory$^{1,2,3}$ are based on Boltzmann's entropy
function instead of Gibbs' entropy function, which is the fore-runner of the
information-theoretic entropy function.
In another paper$^{24}$, this author will return to the issue of the
indeterminacy of the OJM mean field equations which was raised by Turkington
\cite{Turk}. This issue concerns the fact that a given continuous vorticity
distribution can be represented in a number of different ways by clouds of
point vortices--- for example, one could use two species of vortices with
equal but opposite circulations, or one could just as well choose an
approximation based on three different species of vortices. The mean field
equations ensuing from these distinct representations of the original
continuous vorticity distributions must necessarily differ; this can be
demonstrated equally well within the traditional formulation and the current
derivation of the OJM theory.
\medskip\
\begin{center}
Acknowledgement
\end{center}
The author would like to thank Andy Majda for many useful discussions on
equilibrium statistics and for arranging office space at CIMS in the summer
of 1998, during which period, this paper was completed. He would also like
to thank John Chu for friendly advice over the past few years.
|
\section{Introduction.}
\label{sec:introduction}
Cosmic ray fluxes are about to be measured with unprecedented
precision both by balloon borne detectors and by space instruments.
The various ongoing experiments are also hunting for traces
of antimatter in the cosmic radiation. The BESS collaboration
\cite{bess_1}
plans to push the limit on the $\bar{\rm He}/{\rm He}$ ratio
down to $10^{-8}$ whereas the AMS spectrometer should reach a
sensitivity of $\sim 10^{-9}$ once it is installed on the International
Space Station Alpha (ISSA) \cite{ams}.
The search for antinuclei has profound cosmological implications.
The discovery of a single antihelium or anticarbon would actually
be a smoking gun for the existence of antimatter islands in our
neighborhood. However, light antinuclei, mostly antiprotons but also
antideuterons, are actually produced in our Galaxy as secondaries.
They result from the interaction of high--energy cosmic--ray protons
with the interstellar gas of the Milky Way disk. In a previous analysis,
Chardonnet {{\it et al.}} \cite{chardonnet97} have estimated the flux of
antideuterium {\mbox{$\bar{\rm D}$}} and antihelium $\bar{\rm ^{3}He}$ secondaries.
The {\mbox{$\bar{\rm D}$}} signal is very weak but may marginally be detected by AMS
on board ISSA. The case of antihelium is, at least for the moment, hopeless.
The dark matter of the Milky Way could be made mostly of elementary
particles such as the heavy and neutral species predicted by supersymmetry.
The mutual annihilations of these relics, potentially concealed in the
halo of our Galaxy, would therefore produce an excess in the cosmic radiation
of gamma rays, antiprotons and positrons. In particular, supersymmetric
antiprotons should be abundant at low energy, a region where the flux of
{\mbox{$\bar{\rm p}$}} secondaries is a priori negligible. There is quite an excitement trying
to extract from the observations a possible {\mbox{$\bar{\rm p}$}} exotic component
which would signal the presence of supersymmetric dark matter
in the Galaxy. Unfortunately, it has been recently realized
\cite{bottino98,bergstrom99,bieber99} that a few processes add up
together to flatten out, at low energy, the spectrum of secondary
antiprotons.
Ionisation losses as well as inelastic but non-annihilating scatterings
on the hydrogen atoms of the galactic disk result into the decrease of the
antiproton energy. The low--energy tail of the {\mbox{$\bar{\rm p}$}} spectrum is
replenished by the more abundant population from higher energies.
That effect is further strengthened by solar modulation which also shifts
the energy spectrum towards lower energies.
As a result of these effects, the secondary {\mbox{$\bar{\rm p}$}}'s are much more
abundant at low energy than previously thought. Disentangling
an exotic supersymmetric contribution from the conventional
component of spallation antiprotons may turn out to be a very
difficult task. The antiproton signal of supersymmetric dark matter
is therefore in jeopardy.
Antideuterons, {\it i.e.}, the nuclei of antideuterium, are free from such problems.
As explained in Sect.~\ref{sec:coalescence}, they form when an antiproton
and an antineutron merge together. The two antinucleons must be
at rest with respect to each other in order for fusion to take place
successfully. For kinematic reasons, a spallation reaction creates very
few low--energy particles. Low-energy secondary antideuterons are even
further suppressed. Energy loss mechanisms are also less efficient in shifting
the antideuteron energy spectrum towards low energies. The corresponding
interstellar (IS) flux is derived in Sect.~\ref{sec:secondary}, for energies in
the range extending from 0.1 up to 100 GeV/n. It reaches a maximum of
$2-5 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX}
for a kinetic energy of $\sim$ 4 GeV/n. A dozen of
secondary antideuterons should be collected by the
AMS/ISSA experiment.
On the other hand, supersymmetric {\mbox{$\bar{\rm D}$}}'s are manufactured at rest
with respect to the Galaxy. In neutralino annihilations, antinucleons are
predominantly produced with low energies. This feature is further
enhanced by their subsequent fusion into antideuterons, hence
a fairly flat spectrum for supersymmetric antideuterium nuclei
as shown in Sect.~\ref{sec:susy}.
Below a few GeV/n, secondary antideuterons are quite
suppressed with respect to their supersymmetric partners. That
low--energy suppression is orders of magnitude more effective
for antideuterons than for antiprotons. This makes
cosmic--ray antideuterons a much better probe of supersymmetric
dark matter than antiprotons.
Unfortunately, antideuteron fluxes are quite small with respect
to {\mbox{$\bar{\rm p}$}}'s. We nevertheless show in Sect.~\ref{sec:conclusion}
that a significant portion of the supersymmetric parameter space may
be explored by measuring the cosmic--ray {\mbox{$\bar{\rm D}$}} flux at low energy.
In particular, an AMS/ISSA caliber experiment should reach a
sensitivity of $4.8 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX}
at solar minimum, pushing it down to
$3.2 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX}
at solar maximum, for a modulated energy of 0.24 GeV/n.
\section{Production of antideuterons.}
\label{sec:coalescence}
At this point, our goal is to derive the cross section for the production
of antideuterons. The processes at stake are both the spallation of a cosmic--ray
high--energy proton on an hydrogen atom at rest and the annihilation
of a neutralino pair. The number $d{\cal N}_{\rm X}$ of particles ${\rm X}$
-- antinucleons or antideuterons -- produced in a single reaction and whose
momenta are $\mbox{$\vec{k_{\rm X}}$}$, is related to the differential production cross section
through
\begin{equation}
d{\cal N}_{\rm X} \; = \; {\displaystyle \frac{1}{\mbox{$\sigma_{\rm tot}$}}} \,
{d^{3} \sigma_{\rm X}}(\sqrt{s} , \mbox{$\vec{k_{\rm X}}$})
\;\; ,
\end{equation}
where $\mbox{$\sigma_{\rm tot}$}$ denotes the total cross section for the process under
scrutiny -- spallation reaction or neutralino annihilation. The total
available energy is $\sqrt{s}$. The corresponding differential probability
for the production of ${\rm X}$ is defined as
\begin{equation}
d{\cal N}_{\rm X} \; = \; {\cal F}_{\rm X} (\sqrt{s} , \mbox{$\vec{k_{\rm X}}$}) \, d^{3} \mbox{$\vec{k_{\rm X}}$}
\;\; .
\end{equation}
For each of the processes under concern, the differential probability
for the production of an antiproton or an antineutron may be derived.
The calculation of the probability for the formation of an antideuteron
can now proceed in two steps. We first need to estimate the probability for
the creation of an antiproton--antineutron pair. Then, those antinucleons
merge together to yield an antinucleus of deuterium.
As explained in Ref.~\cite{chardonnet97}, the production of two antinucleons is
assumed to be proportionnal to the square of the production of one of them.
The hypothesis that factorization of the probabilities holds is fairly well
established at high energies. For spallation reactions, however, the bulk
of the antiproton production takes place for an energy $\sqrt{s} \sim 10$ GeV
which turns out to be of the same order of magnitude as the antideuteron mass.
Pure factorization should break in that case as a result of energy conservation.
It needs to be slightly adjusted. We have therefore assumed that the center
of mass energy available for the production of the second antinucleon is reduced
by twice the energy carried away by the first antinucleon
\begin{equation}
{\cal F}_{\mbox{$\bar{\rm p}$} , \mbox{$\bar{\rm n}$}} (\sqrt{s} , \mbox{$\vec{k_{\pbar}}$} , \mbox{$\vec{k_{\nbar}}$}) \; = \;
{\displaystyle \frac{1}{2}} \,
{\cal F}_{\mbox{$\bar{\rm p}$}} (\sqrt{s} , \mbox{$\vec{k_{\pbar}}$}) \,
{\cal F}_{\mbox{$\bar{\rm n}$}} (\sqrt{s} - 2 E_{\mbox{$\bar{\rm p}$}} , \mbox{$\vec{k_{\nbar}}$}) \; + \;
\left( \mbox{$\vec{k_{\pbar}}$} \leftrightarrow \mbox{$\vec{k_{\nbar}}$} \right)
\;\; .
\end{equation}
Once the antiproton and the antineutron are formed, they combine
together to give an antideuteron with probability
\begin{equation}
{\cal F}_{\mbox{$\bar{\rm D}$}} (\sqrt{s} , \mbox{$\vec{k_{\dbar}}$}) \, d^{3} \mbox{$\vec{k_{\dbar}}$} \; = \;
{\displaystyle \int} \, d^{3} \mbox{$\vec{k_{\pbar}}$} \, d^{3} \mbox{$\vec{k_{\nbar}}$} \;
{\cal C}(\mbox{$\vec{k_{\pbar}}$} , \mbox{$\vec{k_{\nbar}}$}) \;
{\cal F}_{\mbox{$\bar{\rm p}$} , \mbox{$\bar{\rm n}$}} \left( \sqrt{s} , \mbox{$\vec{k_{\pbar}}$} , \mbox{$\vec{k_{\nbar}}$} \right) \;\; .
\label{coalescence_1}
\end{equation}
The summation is performed on those antinucleon configurations for which
\begin{equation}
\mbox{$\vec{k_{\pbar}}$} + \mbox{$\vec{k_{\nbar}}$} \; = \; \mbox{$\vec{k_{\dbar}}$} \;\; .
\end{equation}
The coalescence function ${\cal C}(\mbox{$\vec{k_{\pbar}}$} , \mbox{$\vec{k_{\nbar}}$})$ describes the probability
for a $\mbox{$\bar{\rm p}$} - \mbox{$\bar{\rm n}$}$ pair to yield by fusion an antideuteron. That function
depends actually on the difference
$\mbox{$\vec{k_{\pbar}}$} - \mbox{$\vec{k_{\nbar}}$} = 2 \mbox{$\vec{\Delta}$}$ between the antinucleon momenta so that
relation (\ref{coalescence_1}) may be expressed as
\begin{equation}
{\cal F}_{\mbox{$\bar{\rm D}$}} (\sqrt{s} , \mbox{$\vec{k_{\dbar}}$}) \; = \;
{\displaystyle \int} \, d^{3} \mbox{$\vec{\Delta}$} \; {\cal C}(\mbox{$\vec{\Delta}$}) \;
{\cal F}_{\mbox{$\bar{\rm p}$} , \mbox{$\bar{\rm n}$}} \left( \sqrt{s} ,
\mbox{$\vec{k_{\pbar}}$} \, = \, \frac{\mbox{$\vec{k_{\dbar}}$}}{2} + \mbox{$\vec{\Delta}$} ,
\mbox{$\vec{k_{\nbar}}$} \, = \, \frac{\mbox{$\vec{k_{\dbar}}$}}{2} - \mbox{$\vec{\Delta}$} \right) \;\; .
\label{coalescence_2}
\end{equation}
An energy of $\sim 3.7$ GeV is required to form by spallation an
antideuteron whereas the binding energy of the latter is
$B \sim 2.2$ MeV. The coalescence function is therefore strongly
peaked around
$\mbox{$\vec{\Delta}$} = \vec{0}$ and expression (\ref{coalescence_2}) simplifies into
\begin{equation}
{\cal F}_{\mbox{$\bar{\rm D}$}} (\sqrt{s} , \mbox{$\vec{k_{\dbar}}$}) \; \simeq \; \left\{
{\displaystyle \int} \, d^{3} \mbox{$\vec{\Delta}$} \; {\cal C}(\mbox{$\vec{\Delta}$}) \right\} \;
{\cal F}_{\mbox{$\bar{\rm p}$} , \mbox{$\bar{\rm n}$}} \left( \sqrt{s} ,
\mbox{$\vec{k_{\pbar}}$} \, = \, \frac{\mbox{$\vec{k_{\dbar}}$}}{2} , \mbox{$\vec{k_{\nbar}}$} \, = \, \frac{\mbox{$\vec{k_{\dbar}}$}}{2}
\right) \;\; ,
\end{equation}
where the probability for the formation of the $\mbox{$\bar{\rm p}$} - \mbox{$\bar{\rm n}$}$ pair
has been factored out. The term in brackets may be estimated in the
rest frame of the antideuteron through the Lorentz invariant term
\begin{equation}
{\displaystyle \int} \, \frac{E_{\mbox{$\bar{\rm D}$}}}{E_{\mbox{$\bar{\rm p}$}} \, E_{\mbox{$\bar{\rm n}$}}} \,
d^{3} \mbox{$\vec{\Delta}$} \, {\cal C}(\mbox{$\vec{\Delta}$}) \; \simeq \;
\left( \frac{\mbox{$m_{\dbar}$}}{m_{\mbox{$\bar{\rm p}$}} \, m_{\mbox{$\bar{\rm n}$}}} \right) \,
\left(\frac{4}{3} \pi \mbox{$P_{\rm coal}$}^{3} \right) \;\; .
\end{equation}
In that frame, the antinucleons merge together if the momentum of the
corresponding two--body reduced system is less than some critical value
$\mbox{$P_{\rm coal}$}$. That coalescence momentum is the only free parameter of our
factorization and coalescence scheme. As shown in Ref.~\cite{chardonnet97},
the resulting antideuteron production cross section in proton--proton
collisions is well fitted by this simple one--parameter model. A value of
$\mbox{$P_{\rm coal}$} = 58$ MeV has been derived, not too far from what may be naively
expected from the antideuteron binding energy, {\it i.e.},
$\sqrt{\mbox{$m_{\rm p}$} \, B} \sim 46$ MeV.
The differential probability with which an antiproton is produced during
a proton--proton collision is related to the corresponding Lorentz invariant
cross section through
\begin{equation}
{\mbox{$\sigma_{\rm p - p}^{\rm tot}$}} \, E_{\mbox{$\bar{\rm p}$}} \, {\cal F}_{\mbox{$\bar{\rm p}$}} (\sqrt{s} , \mbox{$\vec{k_{\pbar}}$}) \; = \;
\left. E_{\mbox{$\bar{\rm p}$}} \, \frac{d^{3} \sigma}{d^{3} \mbox{$\vec{k_{\pbar}}$}} \right|_{\rm LI}
\;\; .
\end{equation}
The latter is experimentally well known. It is fairly well fitted by the Tan and Ng's
parametrization \cite{TanNg82} which has been used here. Assuming that
the invariance of isospin holds, the antineutron production cross section is
equal to its antiproton counterpart.
The Lorentz invariant cross section for the production of antideuterons
resulting from the impact of a high--energy cosmic--ray proton on a
proton at rest has been derived by Chardonnet {\it et al.} \cite{chardonnet97}
who showed that
\begin{eqnarray}
E_{\mbox{$\bar{\rm D}$}} \,
{\displaystyle \frac{d^3 \sigma_{\mbox{$\bar{\rm D}$}}}{d^{3} \mbox{$\vec{k_{\dbar}}$}}} & = &
\left( {\displaystyle \frac{\mbox{$m_{\dbar}$}}{m_{\mbox{$\bar{\rm p}$}} \, m_{\mbox{$\bar{\rm n}$}}}} \right) \,
\left({\displaystyle \frac{4}{3}} \pi \mbox{$P_{\rm coal}$}^{3} \right) \times
{\displaystyle \frac{1}{2 \mbox{$\sigma_{\rm p - p}^{\rm tot}$}}} \times \nonumber \\
& \times &
\left\{
E_{\mbox{$\bar{\rm p}$}} {\displaystyle \frac{d^{3} \sigma_{\mbox{$\bar{\rm p}$}}}{d^{3} \mbox{$\vec{k_{\pbar}}$}}}
\left( \sqrt{s} , \mbox{$\vec{k_{\pbar}}$} \right) \,
E_{\mbox{$\bar{\rm n}$}} {\displaystyle \frac{d^{3} \sigma_{\mbox{$\bar{\rm n}$}}}{d^{3} \mbox{$\vec{k_{\nbar}}$}}}
\left( \sqrt{s} - 2 E_{\mbox{$\bar{\rm p}$}} , \mbox{$\vec{k_{\nbar}}$} \right)
\; + \; \left( \mbox{$\vec{k_{\pbar}}$} \leftrightarrow \mbox{$\vec{k_{\nbar}}$} \right)
\right\}
\;\; .
\label{LI_dbar_production}
\end{eqnarray}
The corresponding differential cross section obtains from the summation, in
the galactic frame, of the Lorentz invariant production cross section
(\ref{LI_dbar_production})
\begin{equation}
{\displaystyle \frac{d \sigma_{\rm p H \to \mbox{$\bar{\rm D}$}}}{d \mbox{$E_{\dbar}$}}}
\left\{ \mbox{$E_{\rm p}$} \to \mbox{$E_{\dbar}$} \right\} \; = \;
2 \pi \; k_{\mbox{$\bar{\rm D}$}} \;
{\displaystyle \int_{0}^{\theta_{\rm max}}} \,
\left. E_{\mbox{$\bar{\rm D}$}} \, \frac{d^{3} \sigma}{d^{3} \mbox{$\vec{k_{\dbar}}$}} \right|_{\rm LI} \;
d \left( - \cos \theta \right) \;\; .
\label{integral_production_lab}
\end{equation}
In that frame, $\theta$ denotes the angle between the momenta of the incident
proton and of the produced antideuteron. It is integrated up to a maximal value
$\theta_{\rm max}$ set by the requirement that, in the center of mass
frame of the reaction, the antideuteron energy $E^{*}_{\mbox{$\bar{\rm D}$}}$ cannot exceed
the bound
\begin{equation}
E^{*}_{\mbox{$\bar{\rm D}$} \, {\rm max}} \; = \;
{\displaystyle \frac{s \, - \, 16 {\mbox{$m_{\rm p}$}}^{2} \, + \, {\mbox{$m_{\dbar}$}}^{2}}{2 \sqrt{s}} }
\;\; .
\end{equation}
The integral (\ref{integral_production_lab}) is performed at fixed
antideuteron energy ${\mbox{$E_{\dbar}$}}^{2} = {\mbox{$m_{\dbar}$}}^{2} + {k_{\mbox{$\bar{\rm D}$}}}^{2}$.
In the case of a neutralino annihilation, the differential multiplicity
for antiproton production may be expressed as
\begin{equation}
{\displaystyle \frac{dN_{\mbox{$\bar{\rm p}$}}}{d \mbox{$E_{\pbar}$}}} \; = \;
{\displaystyle \sum_{\rm F , h}} \, B_{\rm \chi h}^{\rm (F)} \,
{\displaystyle \frac{dN_{\mbox{$\bar{\rm p}$}}^{\rm h}}{d \mbox{$E_{\pbar}$}}} \;\; .
\end{equation}
The annihilation proceeds, through the various final states F, towards
the quark or the gluon h with the branching ratio $B_{\rm \chi h}^{\rm (F)}$.
Quarks or gluons may be directly produced when a neutralino pair annihilates.
They may alternatively result from the intermediate production of a Higgs or
gauge boson as well as of a top quark. Each quark or gluon h generates in turn
a jet whose subsequent fragmentation and hadronization yields the antiproton
energy spectrum ${dN_{\mbox{$\bar{\rm p}$}}^{\rm h}} / {d \mbox{$E_{\pbar}$}}$.
Because neutralinos are at rest with respect to each other, the probability
to form, say, an antiproton with momentum $\mbox{$\vec{k_{\pbar}}$}$ is essentially isotropic
\begin{equation}
{\displaystyle \frac{dN_{\mbox{$\bar{\rm p}$}}}{d \mbox{$E_{\pbar}$}}}(\chi + \chi \to \mbox{$\bar{\rm p}$} + \ldots)
\; = \;
4 \pi \, k_{\mbox{$\bar{\rm p}$}} \, \mbox{$E_{\pbar}$} \, {\cal F}_{\mbox{$\bar{\rm p}$}}(\sqrt{s} = 2 \mbox{$m_{\chi}$} , \mbox{$E_{\pbar}$})
\;\; .
\end{equation}
Applying the factorization--coalescence scheme discussed above leads
to the antideuteron differential multiplicity
\begin{equation}
{\displaystyle \frac{dN_{\mbox{$\bar{\rm D}$}}}{d \mbox{$E_{\dbar}$}}}\; = \;
\left( {\displaystyle \frac{4 \, \mbox{$P_{\rm coal}$}^{3}}{3 \, k_{\mbox{$\bar{\rm D}$}}}} \right)
\,
\left( {\displaystyle \frac{\mbox{$m_{\dbar}$}}{m_{\mbox{$\bar{\rm p}$}} \, m_{\mbox{$\bar{\rm n}$}}}} \right)
\;
{\displaystyle \sum_{\rm F , h}} \, B_{\rm \chi h}^{\rm (F)} \,
\left\{
{\displaystyle \frac{dN_{\mbox{$\bar{\rm p}$}}^{\rm h}}{d \mbox{$E_{\pbar}$}}}
\left( \mbox{$E_{\pbar}$} = \mbox{$E_{\dbar}$} / 2 \right)
\right\}^{2} \;\; .
\label{dNdbar_on_dEdbar_susy}
\end{equation}
It may be expressed as a sum, extending over the various quarks and
gluons h as well as over the different annihilation channels F, of the
square of the antiproton differential multiplicity. That sum is weighted
by the relevant branching ratios. The antineutron and antiproton differential
distributions have been assumed to be identical.
The hypothesis that factorization holds is certainly conservative. The
antinucleons which merge together to create an antideuteron are produced
in the same quark or gluon jet. Their momenta are not isotropically distributed
with respect to each other. They tend to be more aligned than what has been
assumed here, with a larger chance to generate an antideuteron. However,
our analysis is meant to be conservative.
\section{The detection of spallation antideuterons.}
\label{sec:secondary}
\begin{figure*}[htb!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_sec_is.ps}}
}
\caption{
The IS secondary flux of antideuterons, expressed
in units of \DFLUX, is presented as a function of kinetic
energy per nucleon. The solid curve corresponds to the median value
of the cosmic--ray proton spectrum, as derived by Bottino {\it et al.} [4].
The dashed and dotted lines respectively stand for
the maximal and minimal values of the primary proton flux
from which the antideuterons originate.
}
\label{fig:dbar_sec_is}
\end{figure*}
\begin{figure*}[htb!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_sec_solmod.ps}}
}
\caption{
The median IS spectrum of Fig.~\ref{fig:dbar_sec_is} (solid curve)
has been modulated at solar maximum (dashed line) and minimum
(dotted line).
}
\label{fig:dbar_sec_solmod}
\end{figure*}
As suggested by Parker, the propagation of cosmic--rays inside the Galaxy is
strongly affected by their scattering on the irregularities of magnetic fields.
This results into a diffusive transport. In the following, we will assume an
isotropic diffusion with an empirical value for the diffusion coefficient.
Our Galaxy can be reasonably well modelled by a thin disk of atomic and
molecular hydrogen, with radius $R \sim 20$ kpc and thickness $\sim$
200 pc. This gaseous ridge is sandwiched between two diffusion regions
which act as confinement domains as a result of the presence of irregular
magnetic fields. They extend vertically up to $\sim 3$ kpc apart from the
central disk. That two-zone diffusion model is in good agreement with the
observed primary and secondary nuclei abundances \cite{webber92}.
Assuming a steady regime, the propagation of cosmic--ray antideuterons
within the Milky Way is accounted for by the diffusion equation
\begin{equation}
- \, K \Delta \psi_{\mbox{$\bar{\rm D}$}} \, + \,
\Gamma_{\mbox{$\bar{\rm D}$}} \, \psi_{\mbox{$\bar{\rm D}$}}\; + \;
\frac{\partial}{\partial E} \left\{ b(E) \psi_{\mbox{$\bar{\rm D}$}}\right\}
\; = \; q_{\mbox{$\bar{\rm D}$}}^{\rm sec} \;\; ,
\label{diffusion_dbar_disk}
\end{equation}
where \mbox{$\psi_{\mbox{$\bar{\rm D}$}}$} is the density of antideuterons per unit of
volume and per unit of energy.
In the left--hand side of relation (\ref{diffusion_dbar_disk}), the first term
describes the diffusion of the particles throughout the galactic magnetic
fields. The coefficient $K$ is derived from measurements of the light element
abundances in cosmic--rays. It is constant at low energies, but beyond a
critical value of ${\cal R}_{0} = 1$ GV, it raises with rigidity $\cal{R}$
like
\begin{equation}
K({\cal{R}}) \; = \; K_{0} \,
\left( 1 + \frac{\cal{R}}{\cal R}_{0} \right)^{0.6} \;\; ,
\label{DIFFUSION_K}
\end{equation}
where $K_{0} = 6 \times 10^{27}$ cm$^{2}$ s$^{-1}$. It is assumed to be
essentially independent of the nature of the species that propagate throughout
the Galaxy.
The second term acounts for the destruction of antideuterons through their
interactions, mostly annihilations, with the interstellar medium.
Antideuterons may also undergo fragmentation if they survive annihilation.
In that case, they are broken apart as most of the cosmic--ray nuclei.
The total collision rate is given by
\begin{equation}
\Gamma_{\mbox{$\bar{\rm D}$}} \; = \;
\sigma_{\mbox{$\bar{\rm D}$} \, {\rm H}} \; v_{\mbox{$\bar{\rm D}$}} \; n_{\rm H} \;\; ,
\label{pbar_collision}
\end{equation}
where $\sigma_{\mbox{$\bar{\rm D}$} \, {\rm H}}$ is the total antideuteron interaction
cross section with protons \cite{sigma_dbar_H}, $v_{\mbox{$\bar{\rm D}$}}$ denotes the
velocity and $n_{\rm H} = 1$ cm\puis{-3} is the average hydrogen density
in the thin matter disk.
The last term in the left--hand side of relation (\ref{diffusion_dbar_disk})
stands for the energy losses undergone by antideuterons as they diffuse in
the galactic ridge. The rate
$b (\mbox{$E_{\dbar}$}) = \dot{\mbox{$E_{\dbar}$}}$
at which the antideuteron energy varies is essentially set by the ionization
losses which the particle undergoes as it travels through interstellar gas.
This mechanism yields the following contribution to the energy loss rate
\begin{equation}
b_{\rm \, ion} (\mbox{$E_{\dbar}$}) \; = \;
- \, 4 \pi \, r_{e}^{2} \; m_{e} c^{2} \; n_{\rm H} \;
{\displaystyle \frac{c}{\beta} } \;
\left\{ \ln \left( \frac{2 \, m_{e} c^{2}}{E_{0}} \right) \, + \,
\ln \left( \beta^{2} \gamma^{2} \right) \, - \, \beta^{2} \right\}.
\end{equation}
In molecular hydrogen, the ionization energy $E_{0}$ has been set equal
to 19.2 eV; here $\gamma = \mbox{$E_{\dbar}$} / \mbox{$m_{\dbar}$}$. The classical radius of the
electron is denoted by $r_{e}$ and the electron mass is $m_{e}$.
In the case of antiprotons, it was realized
\cite{TanNg82,bottino98,bergstrom99} that the dominant energy
loss mechanism is actually their inelastic, but non--annihilating,
interactions with interstellar protons. The latter are excited towards
resonant states and hence absorb part of the antiproton energy. In the {\mbox{$\bar{\rm p}$}}
frame, an incident proton kicks off the antiproton at rest, transfering
some of its kinetic energy. In the case of antideuterons, however,
such a process is no longer possible. In the {\mbox{$\bar{\rm D}$}} frame, the impinging
proton cannot transfer energy without destroying the antideuteron
whose binding energy -- $B \sim 2.2$ MeV -- is much smaller than the
typical kinetic energies at stake. That is why fragmentation generally
dominates the interactions of cosmic--ray nuclei with interstellar matter.
Accordingly, the resulting destruction occurs at fixed energy per nucleon.
In the right--hand side of the diffusion Eq.~(\ref{diffusion_dbar_disk}),
the production rate $q_{\mbox{$\bar{\rm D}$}}^{\rm sec}$ of the spallation antideuterons
involves a convolution over the incident cosmic--ray proton energy
spectrum $\psi_{\rm p}$ of the differential production cross
section (\ref{integral_production_lab})
\begin{equation}
q_{\mbox{$\bar{\rm D}$}}^{\rm disk}(\mbox{$E_{\dbar}$}) \; = \;
{\displaystyle \int_{\mbox{$E_{\dbar}$}}^{+ \infty}} \; d\mbox{$E_{\rm p}$} \;
\psi_{\rm p}(\mbox{$E_{\rm p}$}) \, n_{\rm H} \, v_{\rm p} \;
{\displaystyle \frac{d \sigma_{\rm p H \to \mbox{$\bar{\rm D}$}}}{d \mbox{$E_{\dbar}$}}}
\left\{ \mbox{$E_{\rm p}$} \to \mbox{$E_{\dbar}$} \right\}
\;\; .
\label{sec_source}
\end{equation}
The differential energy distribution $\psi_{\mbox{$\bar{\rm D}$}}$ of secondary
antideuterons is determined by solving Eq.~(\ref{diffusion_dbar_disk}).
We have followed the standard approach which may be found in
Ref.~\cite{berezinskii90}. At the edge of the domain where the cosmic--rays
are confined, the particles escape freely, the diffusion becomes inefficient
and densities vanish. This provides the boundary conditions for solving
Eq.~(\ref{diffusion_dbar_disk}). Then, because the problem is axisymmetric,
the various cosmic--ray distributions may be expanded as series of Bessel
functions of zeroth order. Details may be found in
Refs.~\cite{bottino98,chardonnet96}. The secondary antideuteron
interstellar flux finally obtains from the differential energy spectrum
\begin{equation}
\Phi_{\mbox{$\bar{\rm D}$}}^{\rm sec} \; = \;
\frac{1}{4 \pi} \, \psi_{\mbox{$\bar{\rm D}$}} \, v_{\mbox{$\bar{\rm D}$}}
\;\; .
\end{equation}
The interstellar (IS) flux of spallation antideuterons is presented in
Fig.~\ref{fig:dbar_sec_is} as a function of the kinetic energy per nucleon.
As explained in Bottino {\it et al.} \cite{bottino98}, the IS proton flux is
still uncertain around $\sim$ 20--100 GeV, an energy range that
contributes most to the integral~(\ref{sec_source}). We have borrowed
the parametrization
\begin{equation}
\Phi_{\rm p}^{\rm IS} \; = \; A \, \beta \,
\left(
{\displaystyle \frac{\mbox{$E_{\rm p}$}}{1 \, {\rm GeV}}}
\right)^{- \alpha} \;\; .
\label{proton_flux}
\end{equation}
The median IS proton flux corresponds to a normalization factor of
$A = 15,950$ protons {\DFLUX} with a spectral index of $\alpha = 2.76$.
The normalization factor $A$ has been varied from $12,300$ (minimal)
up to $19,600$ protons {\DFLUX} (maximal). Accordingly, the minimal
and maximal IS proton fluxes respectively correspond to the spectral
indices $\alpha = 2.61$ and $2.89$. In Fig.~\ref{fig:dbar_sec_is}, the solid
curve features the IS secondary antideuterons generated from the
median proton spectrum. The maximal (dashed line) and minimal
(dotted line) distributions delineate the band within which the spallation
antideuteron signal lies. The flux reaches a maximum value
comprised between 2.1 and $4.9 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX} for
a kinetic energy of $\sim$ 4 GeV/n.
The antideuteron spectrum sharply drops below a few GeV/n.
Remember that in the galactic frame, the production threshold is
17 {\mbox{$m_{\rm p}$}}. When a high--energy cosmic--ray proton impinges
on an hydrogen atom at rest, the bulk of the resulting antiprotons and
antineutrons keep moving, with kinetic energies $\sim 10 - 20$ GeV.
For kinematical reasons, the production of antinucleons at rest with
respect to the Galaxy is extremely unprobable. The manufacture of a
low--energy antideuteron is even more unprobable. It actually requires
the creation of both an antiproton and an antineutron at rest. The momenta
need to be aligned in order for fusion to succesfully take place. Low--energy
antideuterons produced as secondaries in the collisions of high--energy
cosmic--rays with the interstellar material are therefore extremely scarce,
with a completely depleted energy spectrum below $\sim$ 1 GeV/n.
Energy losses tend to shift the antideuteron spectrum towards lower
energies with the effect of replenishing the low--energy tail with the
more abundant species which, initially, had a higher energy. This
process tends to slightly soften the strong decrease of the low--energy
antideuteron spectrum. The effect is nevertheless mild. Remember that
in the case of antiprotons, it is actually the inelastic but non--annihilating
interactions which considerably flatten the {\mbox{$\bar{\rm p}$}} distribution. Ionizations
losses are not enough to significantly affect the energy spectrum.
The IS secondary antideuterons are therefore extremely depleted
below $\sim$ 1 GeV/n. The spallation background is negligible
in the region where supersymmetric {\mbox{$\bar{\rm D}$}}'s are expected to be
most abundant. This feature makes the detection of low--energy
antideuterons an interesting signature of the presence of supersymmetric
relics in the Galaxy.
In Fig.~\ref{fig:dbar_sec_solmod}, the median IS {\mbox{$\bar{\rm D}$}} spectrum (solid curve)
has been modulated at solar maximum (dashed line) and solar minimum
(dotted line). We have applied the forced field approximation \cite{perko}
to estimate the effect of the solar wind on the cosmic--ray energies
and fluxes. For the energies at stake, this amounts to simply shift the
IS energy of a nucleus $N$, with charge $Z$ and atomic number $A$, by a
factor of $Z e \Phi$. The solar modulation parameter $\Phi$ has the same
dimensions as a rigidity or an electric potential. The Earth ($\oplus$) and
interstellar (IS) energies, {\em per nucleon}, are therefore related by
\begin{equation}
E_{N}^{\oplus} / A \; = \;
E_{N}^{\rm IS} / A \, - \, \left| Z \right| e \Phi / A \;\; .
\end{equation}
In Perko's approximation, antinuclei are affected in just the same way
as nuclei. Their energy decreases as they penetrate inside the heliomagnetic
field. Once the momenta at the Earth $p_{N}^{\oplus}$ and at the boundaries
of the heliosphere $p_{N}^{\rm IS}$ are determined, the flux modulation
ensues
\begin{equation}
{\displaystyle
\frac{\Phi_{N}^{\oplus} \left( E_{N}^{\oplus} \right)}
{\Phi_{N}^{\rm IS} \left( E_{N}^{\rm IS} \right)}}
\; = \;
\left\{ {\displaystyle \frac{p_{N}^{\oplus}}{p_{N}^{\rm IS}}} \right\}^{2} \,
\;\; .
\end{equation}
Antideuterons undergo an energy loss, {\em per nucleon}, half that of protons
and antiprotons. At solar minimum (maximum) the modulation parameter
$\Phi$ has been set equal to 320 MV (800 MV) \cite{bottino98}. The energy shift
is larger at solar maximum than at solar minimum. Once modulated, the sharply
decreasing IS antideuteron distribution tends therefore to be flatter at solar
maximum as is clear in Fig.~\ref{fig:dbar_sec_solmod}. We estimate that a total of
12--13 secondary antideuterons may be collected by the AMS collaboration
during the space station stage, in the energy range extending up to 100 GeV/n.
These antideuterons correspond to IS energies in excess of $\sim$ 3
GeV/n, a region free from the effects of solar modulation. This result
takes into account the geomagnetic suppression as discussed in
Sect.~\ref{sec:conclusion}.
\section{The supersymmetric antideuteron signal.}
\label{sec:susy}
\begin{figure*}[h!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_prim_is.ps}}
}
\caption{
The IS flux of secondary antideuterons (heavier solid curve)
decreases at low energy whereas the energy spectrum of the
antideuterons from supersymmetric origin tends to flatten.
The four cases of table~\ref{table:susy} are respectively featured
by the solid (a), dotted (b), dashed (c) and dot-dashed (d) curves.
}
\label{fig:dbar_prim_is}
\end{figure*}
\begin{table}[h!]
\[
\begin{array}{|c|c|c|c|c|c|c|c|} \hline
{\rm case} &
m_{\chi} &
P_{g} (\%) &
\Omega_\chi h^{2} &
\Phi_{\mbox{$\bar{\rm p}$}}^{\rm min} \left( 0.24 \; {\rm GeV} \right) &
\Phi_{\mbox{$\bar{\rm D}$}}^{\rm min} \left( 0.24 \; {\rm GeV/n} \right) &
\Phi_{\mbox{$\bar{\rm D}$}}^{\rm max} \left( 0.24 \; {\rm GeV/n} \right) &
N_{\mbox{$\bar{\rm D}$}}^{\rm max}
\\
& & & & & & &
\\
\hline \hline
a & 36.5 & 96.9 & 0.20 & 1.2 \times 10^{-3} & 1.0 \times 10^{-7} &
2.9 \times 10^{-8} & 0.6\\
b & 61.2 & 95.3 & 0.13 & 3.9 \times 10^{-3} & 3.5 \times 10^{-7}&
1.1 \times 10^{-7} & 2.9 \\
c & 90.4 & 53.7 & 0.03 & 1.1 \times 10^{-3} & 1.8 \times 10^{-7} &
6.1 \times 10^{-8} & 2.0 \\
d & 120 & 98.9 & 0.53 & 2.9 \times 10^{-4} & 2.5 \times 10^{-8}&
8.6 \times 10^{-9} & 0.3 \\
\hline \hline
\end{array} \]
\caption{
These four cases illustrate the richness of the supersymmetric
parameter space. There is no obvious correlation between the
antiproton and antideuteron Earth fluxes with the neutralino mass
$ m_{\chi}$. Case (c) is a gaugino-higgsino mixture and still
yields signals comparable to those of case (a), yet a pure gaugino.
Antideuteron fluxes are estimated at both solar minimum and
maximum, for a modulated energy of 0.24 GeV/n. The last column
features the corresponding number of {\mbox{$\bar{\rm D}$}}'s which AMS on board
ISSA can collect below 3 GeV/n.
}
\label{table:susy}
\end{table}
\begin{figure*}[h!]
\centerline{
{\resizebox{0.5\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_prim_solmax.ps}}}
{\resizebox{0.5\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_prim_solmin.ps}}}
}
\caption{
Same as in Fig.~\ref{fig:dbar_prim_is} but modulated at solar
maximum ({\rm left}) and minimum ({\rm right}).
}
\label{fig:dbar_prim_solmod}
\end{figure*}
\begin{figure*}[h!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{susy_over_sec.ps}}
}
\caption{
The supersymmetric--to--secondary IS flux ratio for antiprotons
(lower curves) and antideuterons (upper curves) is presented
as a function of the kinetic energy per nucleon. The supersymmetric
configurations are those reported in table~\ref{table:susy} and
featured in Figs.~\ref{fig:dbar_prim_is} and
\ref{fig:dbar_prim_solmod}. Below a few GeV/n, the flux ratio is
always larger for {\mbox{$\bar{\rm D}$}}'s than for {\mbox{$\bar{\rm p}$}}'s.
For the supersymmetric configurations of table~\ref{table:susy}, the
antiproton signal is swamped into its background. This is not the case for
antideuterons. At low energy, the flux of primaries is several orders of
magnitude above the {\mbox{$\bar{\rm D}$}} background.
}
\label{fig:susy_over_sec}
\end{figure*}
As a theoretical framework, we use the Minimal Supersymmetric extension of the
Standard Model (MSSM) \cite{susy}, which conveniently describes the
supersymmetric phenomenology at the electroweak scale, without too strong
theoretical assumptions. This model has been largely adopted by many
authors for evaluations of the neutralino relic abundance and detection rates
(for reviews, see Refs.~\cite{ICTP,jkg}).
The MSSM is defined at the electroweak scale as a straightforward supersymmetric
extension of the Standard Model. The Higgs sector consists of two Higgs doublets
$H_1$ and $H_2$ and, at the tree level, is fully described by two free parameters,
namely: the ratio of the two vacuum expectation values
$\tan\beta \equiv \langle H_2 \rangle/\langle H_1\rangle$ and the mass of one
of the three neutral physical Higgs fields, which we choose to be the mass $m_A$ of
the neutral pseudoscalar one. Once radiative corrections are introduced, the Higgs
sector depends also on the squark masses through loop diagrams. The radiative
corrections to the neutral and charged Higgs bosons, adopted in the present paper,
are taken from Refs.~\cite{carena,haber}.
The other parameters of the model are defined in the superpotential, which contains
all the Yukawa interactions and the Higgs--mixing term $\mu H_1 H_2$, and in the
soft--breaking Lagrangian, which contains the trilinear and bilinear breaking parameters
and the soft gaugino and scalar mass terms.
In this model, the neutralino is defined as the lowest--mass linear superposition of
photino ($\tilde \gamma$), zino ($\tilde Z$) and the two higgsino states
($\tilde H_1^{\circ}$, $\tilde H_2^{\circ}$)
\begin{equation}
\chi \equiv a_1 \tilde \gamma + a_2 \tilde Z + a_3 \tilde H_1^{\circ}
+ a_4 \tilde H_2^{\circ} \;\; .
\label{eq:neu}
\end{equation}
In order to deal with manageable models, it is necessary to introduce some
assumptions which establish relations among the too many free parameters
at the electroweak scale. We adopt the following usual conditions.
All trilinear parameters are set to zero except those of the third family,
which are unified to a common value $A$.
All squarks and sleptons soft--mass parameters are taken as
degenerate: $m_{\tilde l_i} = m_{\tilde q_i} \equiv m_0$.
The gaugino masses are assumed to unify at $M_{GUT}$, and this implies that
the $U(1)$ and $SU(2)$ gaugino masses are related at the electroweak scale
by $M_1= (5/3) \tan^2 \theta_W M_2$.
When all these conditions are imposed, the supersymmetric parameter space is
completely described by six independent parameters, which we choose to be:
$M_2, \mu, \tan\beta, m_A, m_0, A$. In our analyses, we vary them in the
following ranges:
$20\;\mbox{GeV} \leq M_2 \leq 500\;\mbox{GeV}$;
$20\;\mbox{GeV} \leq |\mu| \leq 500\;\mbox{GeV}$;
$80\;\mbox{GeV} \leq m_A \leq 1000\;\mbox{GeV}$;
$100\;\mbox{GeV} \leq m_0 \leq 1000\;\mbox{GeV}$;
$-3 \leq {\rm A} \leq +3$;
$1 \leq \tan \beta \leq 50$.
The supersymmetric parameter space is constrained by all the experimental limits
achieved at accelerators on supersymmetric and Higgs searches \cite{lep189}. Also
the constraints due to the $b \rightarrow s + \gamma$ process \cite{glenn,barate}
have been taken into account (see Ref.~\cite{noi3} for a discussion of
our implementation of the $b \rightarrow s + \gamma$ constraint and
for the relevant references).
We further require the neutralino to be the Lightest Supersymmetric Particle
(LSP) and the supersymmetric configurations to provide a neutralino relic
abundance in accordance with the cosmological bound
$\Omega_{\chi}h^2 \leq 0.7$ \cite{ICTP}.
For the evaluation of the averaged annihilation cross section
$<\sigma_{\rm ann} v>$, we have followed the procedure outlined in
Ref.~\cite{bffm}. We have considered all the tree--level diagrams which are
responsible for neutralino annihilation and which are relevant to $\mbox{$\bar{\rm p}$}$
production, namely: annihilation into quark--antiquark pairs, into gauge bosons,
into a Higgs boson pair and into a Higgs and a gauge boson. For each final state
we have considered all the relevant Feynman diagrams, which involve the
exchange of Higgs and $Z$ bosons in the s--channel and the exchange of squarks,
neutralinos and charginos in the t and u--channels. Finally, we have included
the one--loop diagrams which produce a two--gluon final state \cite{bu}.
The $\mbox{$\bar{\rm p}$}$ differential distribution
${dN_{\mbox{$\bar{\rm p}$}}} / {d \mbox{$E_{\pbar}$}}$ has been evaluated as discussed in Ref.~\cite{bffm}.
Here we only recall that we have calculated the branching ratios
$B_{\rm \chi h}^{\rm (F)}$ for all annihilation final states F which may
produce $\mbox{$\bar{\rm p}$}$'s. These final states fall into two categories~:
(i) direct production of quarks and gluons and
(ii) generation of quarks through the intermediate production of Higgs bosons,
gauge bosons and $t$ quark.
In order to obtain the distributions ${dN_{\mbox{$\bar{\rm p}$}}^{\rm h}} / {d \mbox{$E_{\pbar}$}}$,
the hadronization of quarks and gluons has been computed by using the Monte
Carlo code Jetset 7.2 \cite{jet}. For the top quark, we have considered it to decay
before hadronization.
The source term for supersymmetric antideuterons
\begin{equation}
q_{\mbox{$\bar{\rm D}$}}^{\rm susy} \left( \chi + \chi \to \mbox{$\bar{\rm D}$} + \ldots \right)
\; = \;
<\sigma_{\rm ann} v> \,
{\displaystyle \frac {dN_{\mbox{$\bar{\rm D}$}}}{d \mbox{$E_{\dbar}$}}} \,
\left\{ {\displaystyle \frac{\rho_{\chi}}{\mbox{$m_{\chi}$}}} \right\}^{2}
\end{equation}
supplements the spallation contribution $q_{\mbox{$\bar{\rm D}$}}^{\rm sec}$ in
the diffusion Eq.~(\ref{diffusion_dbar_disk}). The propagation of
primary antideuterons from the remote regions of the galactic halo
to the Earth has been treated as explained in Ref.~\cite{bottino98}.
The neutralino distribution has been assumed to be spherical, with
radial dependence
\begin{equation}
\rho_{\chi} \; = \;
\rho_{\chi}^{\odot} \,
\left\{
{\displaystyle \frac{a^{2} + r_{\odot}^{2}}{a^{2} + m^{2}}}
\right\} \;\; ,
\end{equation}
where $m^{2} = r^{2} + z^{2}$. The solar system is at a distance
$r_{\odot}$ of 8 kpc from the galactic center. The dark matter
halo has a core radius $a = 3.5$ kpc and its density in the solar
neighborhood is $\rho_{\chi}^{\odot} = 0.4$ GeV cm$^{-3}$
\cite{ICTP}.
In Fig.~\ref{fig:dbar_prim_is}, both primary (supersymmetric) and
secondary (spallation) interstellar antideuterons energy spectra are
presented. The secondary flux (heavier solid line) drops sharply at
low energies as discussed above. The four supersymmetric examples of
table~\ref{table:susy} are respectively featured by the solid
(a), dotted (b), dashed (c) and dot-dashed (d) curves. The corresponding
primary fluxes flatten at low energy where they reach a maximum.
As the secondary {\mbox{$\bar{\rm D}$}} background vanishes, the supersymmetric
signal is the largest. Neutralino annihilations actually take place at rest
in the galactic frame. The fragmentation and subsequent hadronization of
the jets at stake tend to favour the production of low--energy species.
Therefore, the spectrum of supersymmetric antiprotons -- and antineutrons
-- is fairly flat below $\sim$ 1 GeV. For the same reasons, the coalescence of
the primary antideuterons produced in neutralino annihilations
predominantly takes place with the two antinucleons at rest, hence a flat
spectrum at low energy, as is clear in Fig.~\ref{fig:dbar_prim_is}. The fusion
of an antideuteron requires actually that its antinucleon constituents should
be aligned in momentum space. Consequently, secondary antideuterons are
completely depleted below $\sim 1$ GeV while the primary species are mostly
produced in that low--energy regime.
This trend still appears once the energies and fluxes are modulated.
The left and right panels of Fig.~\ref{fig:dbar_prim_solmod} respectively
show the effects of solar modulation at maximum and minimum. The
spallation background somewhat flattens. It is still orders of magnitude below
the supersymmetric signal which clearly exhibits a plateau.
It is difficult to establish a correlation between the {\mbox{$\bar{\rm D}$}} flux
and the neutralino mass. In case (c), for instance, $\mbox{$m_{\chi}$}$ is $\sim$ 3 times
larger as in case (a) and yet the corresponding antideuteron flux is larger.
It is not obvious either that gaugino--like mixtures lead to the largest
{\mbox{$\bar{\rm D}$}} signals. Table~\ref{table:susy} gives a flavor of the complexity
and of the richness of the supersymmetric parameter space.
In Fig.~\ref{fig:susy_over_sec}, the supersymmetric--to--spallation IS flux ratio
for antiprotons (lower curves) and antideuterons (upper curves) are presented
as a function of the kinetic energy per nucleon. In the case of antiprotons, the
primary--to--secondary ratio is much smaller than for antideuterons.
For the configurations of table~\ref{table:susy} presented here, the {\mbox{$\bar{\rm p}$}} primary
flux is at the same level as the spallation background. The supersymmetric
antiproton signal is swamped in the flux of the secondaries. This is not the case for
antideuterons. At low energies, their supersymmetric flux is several orders of
magnitude above background. Antideuterons appear therefore as a much cleaner
probe of the presence of supersymmetric relics in the galactic halo than antiprotons.
The price to pay however is a much smaller flux.
Typical {\mbox{$\bar{\rm D}$}} spectra may reach up to $10^{-6}-10^{-5} \, \DFLUX$. This
corresponds to an antiproton signal of $10^{-2}-10^{-1} \, \DFLUX$, {\it i.e.}, four orders
of magnitude larger. It is therefore crucial to ascertain which portion of the
supersymmetric configurations will be accessible to future experiments
through the detection of low--energy cosmic--ray antideuterons.
\section{Discussion and conclusions.}
\label{sec:conclusion}
In order to be specific, we have estimated the amount of antideuterons
which may be collected by the AMS experiment once it is on board ISSA.
The future space station is scheduled to orbit at 400 km above sea level,
with an inclination of $\alpha = 52^{\circ}$ with respect to the Earth
equator. A revolution takes about 1.5 hours so that ISSA should fly over
the same spot every day. The AMS detector may be pictured as a
cylindrical magnetic field with diameter $D = 110$ cm. At any time, its axis
points towards the local vertical direction. The colatitude of the north
magnetic pole has been set equal to $\Upsilon = 11^{\circ}$. At any given
time $t$ along the orbit, the geomagnetic latitude $\varrho$ of ISSA may be
inferred from
\begin{eqnarray}
\sin \varrho & = &
\sin \Upsilon \,
\cos \left( \Omega_{\rm sid} t \right) \,
\cos \left( \Omega_{\rm orb} t + \varphi \right) +
\\
& + &
\cos \alpha \, \sin \Upsilon \,
\sin \left( \Omega_{\rm sid} t \right) \,
\sin \left( \Omega_{\rm orb} t + \varphi \right) \; + \;
\sin \alpha \, \cos \Upsilon \,
\sin \left( \Omega_{\rm orb} t + \varphi \right) \;\; ,
\nonumber
\end{eqnarray}
where $\Omega_{\rm sid}$ and $\Omega_{\rm orb}$ respectively
denote the angular velocities associated to the sideral rotation of the
Earth and to the orbital motion of the space station. The phase
$\varphi$ depends on the orbital initial conditions and does not affect
the result if a large number of revolutions -- typically 100 -- is considered.
The Earth is shielded from cosmic--rays because its magnetic field
prevents particles from penetrating downwards. At any
given geomagnetic latitude $\varrho$, there exists a rigidity cut-off
${\cal R}_{\rm min}$ below which the cosmic--ray flux is suppressed.
This lower bound depends on the radius $R$ of the orbit through
\begin{equation}
{\cal R}_{\rm min} \; = \;
{\displaystyle \frac{\mu_{\oplus}}{R^{2}}} \,
{\displaystyle \frac{\cos^{4} \varrho}{\varpi^{2}}} \;\; ,
\label{geomagnetic_1}
\end{equation}
where $\mu_{\oplus}$ denotes the Earth magnetic dipole moment
and ${\mu_{\oplus}} / {R_{\oplus}^{2}} \simeq 60$ GV. The term
$\varpi$ stands for
\begin{equation}
\varpi \; = \; 1 \, + \, \sqrt{1 + \cos \theta \, \cos^{3} \varrho} \;\; .
\label{geomagnetic_2}
\end{equation}
It depends on the angle $\theta$ between the cosmic--ray
momentum at the detector and the local east--west line that points
in the ortho--radial direction of an axisymmetric coordinate system.
Notice that because we are interested here in singly charged species, the
rigidity amounts to the momentum $p$.
Once the cosmic--ray energy as well as the geomagnetic latitude are
specified, the solid angle $\Omega_{\rm cut}$ inside which the
direction of the incoming particle lies may be derived from
relations~(\ref{geomagnetic_1}) and (\ref{geomagnetic_2}).
The AMS detector looks upwards within $\sim 27^{\circ}$ around
the vertical. This corresponds to a solid angle of
$\Omega_{\rm det} = 0.68$ sr. Because the apparatus does not point
towards the local east or west, impinging particles may not be seen by
the instrument. The effective solid angle
$\Omega_{\rm eff}$ through which they are potentially detectable
corresponds to the overlap, if any, between $\Omega_{\rm cut}$ and
$\Omega_{\rm det}$. The value of $\Omega_{\rm eff}$ depends on
the cosmic--ray rigidity $p$ as well as on the precise location of the detector
along the orbit.
The detector acceptance may therefore be defined as
\begin{equation}
\aleph \left( p \right) \; = \;
{\displaystyle \frac{\pi}{4}} D^{2} \,
{\displaystyle \int} \Omega_{\rm eff} \left( p , t \right) \, dt \;\;,
\end{equation}
where the time integral runs over the duration $\tau$ of the space
mission. In the case of AMS on board ISSA, $\tau$ is estimated to be
$10^{8}$ s (3 yrs).
Between 100 MeV/n and 100 GeV/n, we infer a total acceptance of
$5.8 \times 10^{9}$ ${\rm m^{2} \; s \; sr \; GeV}$
for antiprotons and of
$6 \times 10^{9}$ ${\rm m^{2} \; s \; sr \; GeV}$ for
antideuterons.
The net number of cosmic--ray species which AMS may collect on
board ISSA is actually a convolution of the detector acceptance with
the relevant differential flux at Earth. For antideuterons, this leads to
\begin{equation}
N_{\mbox{$\bar{\rm D}$}} \; = \;
{\displaystyle \int} \aleph \left( p_{\mbox{$\bar{\rm D}$}}^{\oplus} \right) \,
\Phi_{\mbox{$\bar{\rm D}$}}^{\oplus} \, dT_{\mbox{$\bar{\rm D}$}}^{\oplus} \;\; ,
\label{dbar_convolution}
\end{equation}
where the integral runs on the {\mbox{$\bar{\rm D}$}} modulated energy $T_{\mbox{$\bar{\rm D}$}}^{\oplus}$.
Integrating the secondary flux discussed in Sect.~\ref{sec:secondary}
leads respectively to a total of 12.3 and 13.4 antideuterons, depending on
whether the solar cycle is at maximum or minimum. These spallation
{\mbox{$\bar{\rm D}$}}'s are mostly expected at high energies. As is clear from
Figs.~\ref{fig:dbar_prim_is} and \ref{fig:dbar_prim_solmod}, the secondary
flux drops below the supersymmetric signal below a few GeV/n. The transition
typically takes place for an interstellar energy of 3 GeV/n. Below that value,
the secondary antideuteron signal amounts to a total of only 0.6 (solar
maximum) and 0.8 (solar minimum) nuclei.
Most of the supersymmetric signal is therefore concentrated in a low--energy
band extending from the AMS threshold of 100 MeV/n up to a modulated
energy of 2.6 GeV/n (maximum) or 2.84 GeV/n (minimum) which corresponds
to an upper bound of 3 GeV/n in interstellar space. In this low--energy
region where spallation antideuterons yield a negligible background,
the AMS acceptance is
$2.2 \times 10^{7}$ ${\rm m^{2} \; s \; sr \; GeV}$ for antiprotons and
$5.5 \times 10^{7}$ ${\rm m^{2} \; s \; sr \; GeV}$ for antideuterons.
\begin{figure*}[h!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{ndbar_versus_mchi.ps}}
}
\caption{
The supersymmetric {\mbox{$\bar{\rm D}$}} flux has been integrated over
the range of IS energies extending from 0.1 up to 3 GeV/n.
The resulting yield $N_{\mbox{$\bar{\rm D}$}}$ of antideuterons which AMS
on board ISSA can collect is plotted as a function of the neutralino
mass $\mbox{$m_{\chi}$}$. Modulation has been considered at solar maximum.
}
\label{fig:ndbar_versus_mchi}
\end{figure*}
For each supersymmetric configuration, the {\mbox{$\bar{\rm D}$}} flux has
been integrated over that low--energy range. The resulting yield $N_{\mbox{$\bar{\rm D}$}}$
which AMS may collect on board ISSA is presented as a function of the
neutralino mass $\mbox{$m_{\chi}$}$ in the scatter plot of Fig.~\ref{fig:ndbar_versus_mchi}.
During the AMS mission, the solar cycle will be at maximum. Most of
the configurations are gaugino like (crosses) or mixed combinations of
gaugino and higgsino states (dots). A significant portion of the parameter
space is associated to a signal exceeding one antideuteron -- horizontal
dashed line. In a few cases, AMS may even collect more than a dozen of
low--energy {\mbox{$\bar{\rm D}$}} nuclei. However, when the antideuteron signal exceeds
$\sim$ 20 particles, the associated antiproton flux is larger than what
BESS 95 + 97 \cite{bess_2} has measured.
\begin{figure*}[h!]
\centerline{
{\resizebox{0.5\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_scatter_solmax.ps}}}
{\resizebox{0.5\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_scatter_solmin.ps}}}
}
\caption{
Scatter plots in the plane $m_{\chi}$--$\Phi_{\mbox{$\bar{\rm D}$}}^{\oplus}$. The Earth
antideuteron flux $\Phi_{\mbox{$\bar{\rm D}$}}^{\oplus}$ has been computed at solar
maximum ({\rm left}) and minimum ({\rm right}), for a modulated
energy of 0.24 GeV/n. Configurations lying above the horizontal lines
correspond to the detection of at least one antideuteron in the range
of interstellar energies 0.1 -- 3 GeV, by an experiment of the AMS caliber
on board ISSA.
}
\label{fig:dbar_scatter_solmod}
\end{figure*}
The scatter plot of Fig.~\ref{fig:ndbar_versus_mchi} may be translated
into a limit on the antideuteron flux $\Phi_{\mbox{$\bar{\rm D}$}}^{\oplus}$ at the
Earth. Table~\ref{table:susy} gives a flavor of the relation between that
flux and the yield $N_{\mbox{$\bar{\rm D}$}}$ of low--energy antideuterons. At solar
maximum, a value of $N_{\mbox{$\bar{\rm D}$}} = 1$ translates, on average, into a flux
of $\sim 3.2 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX} for a modulated energy
of 240 MeV/n. The energy spectrum matters of course. For the steep
differential flux of case (a), a value of
$4.8 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX} is necessary in order to achieve a
signal of at least one antideuteron. In case (d) where the spectrum is much
flatter, the same {\mbox{$\bar{\rm D}$}} yield is reached for a flux of only
$2.8 \times 10^{-8}$ {\mbox{$\bar{\rm D}$}} {\DFLUX}. The horizontal dashed lines
of Figs.~\ref{fig:dbar_scatter_solmod} should therefore be understood
as averaged limits. They are nevertheless indicative of the level of
sensitivity which may be reached through the search for low--energy
antideuterons. The left and right panels respectively correspond to a
solar activity taken at maximum and minimum. In these scatter plots,
the {\mbox{$\bar{\rm D}$}} modulated flux is featured as a function of the neutralino
mass $\mbox{$m_{\chi}$}$. The antideuteron energy at the Earth has been set equal
to 240 MeV/n. The flux $\Phi_{\mbox{$\bar{\rm D}$}}^{\oplus}$ is larger at solar
minimum -- when modulation is weaker --than at maximum. The
lower the cosmic--ray energy, the larger that effect. The plateaux of
Figs.~\ref{fig:dbar_prim_solmod} illustrate the flatness of the
supersymmetric {\mbox{$\bar{\rm D}$}} spectra at low energies. These plateaux actually
exhibit a shift by a factor $\sim$ 3 between the left and right panels.
Accordingly, the constellation of supersymmetric configurations
in Figs.~\ref{fig:dbar_scatter_solmod} is shifted upwards, by the same
amount, between solar maximum (left panel) and minimum (right
panel).
At larger energies, the variation of the flux at Earth during
the solar cycle is milder. Above a few GeV/n,
solar modulation has no effect. The number of supersymmetric
antideuterons collected at low energy obtains from the convolution
of Eq.~(\ref{dbar_convolution}). It also varies during the solar cycle,
in a somewhat lesser extent however than the above mentioned plateaux.
Between maximum and minimum, the value of $N_{\mbox{$\bar{\rm D}$}}$ only
varies by a factor of $\sim$ 2, to be compared to a flux increase of
$\sim$ 3. At solar maximum, when AMS/ISSA will be operating,
a signal of one antideuteron translates into a flux sensitivity of
$\sim 3.2 \times 10^{-8}$ antinuclei {\DFLUX}. At minimum, the
same signal would translate into the weaker limit of
$\sim 4.8 \times 10^{-8}$ antideuterons {\DFLUX} and the horizontal
dashed line is shifted upwards by $\sim$ 50\%. The supersymmetric
configurations which an antideuteron search may unravel are
nevertheless more numerous at solar minimum. Between the left and
the right panels, the constellation of representative points is actually
shifted upwards and, relative to the limit of sensitivity, the increase
amounts to a factor $\sim$ 2.
In spite of the low fluxes at stake, the antideuteron channel is sensitive
to a respectable number of supersymmetric configurations.
\begin{figure*}[h!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{ndbar_pbar_scatter.ps}}
}
\caption{
In this scatter plot, the antideuteron yield $N_{\mbox{$\bar{\rm D}$}}$ of
Fig.~\ref{fig:ndbar_versus_mchi} is featured against the supersymmetric
{\mbox{$\bar{\rm p}$}} flux. The antideuteron signal is estimated at solar maximum.
This corresponds to the AMS mission on board the space station. The
{\mbox{$\bar{\rm p}$}} flux is derived on the contrary at solar minimum, in the same
conditions as the BESS 95 + 97 flights [26] whose combined
measurements are indicated by the vertical shaded band for a {\mbox{$\bar{\rm p}$}}
energy of 0.24 GeV. The correlation between the antiproton and
antideuteron signals is strong.
}
\label{fig:ndbar_pbar_scatter}
\end{figure*}
Supersymmetric antiprotons are four orders of magnitude more abundant
in cosmic--rays than antideuterons -- see table~\ref{table:susy}.
However, as already discussed, they may be swamped in the background
arising from the secondaries. The AMS experiment will collect a large
number of antiprotons on board ISSA. Our concern is whether an
hypothetical supersymmetric {\mbox{$\bar{\rm p}$}} signal may be disentangled from
the background. Because the latter still suffers from large theoretical
uncertainties, we are afraid that antiproton searches in cosmic--rays are
not yet the ultimate probe for the existence of supersymmetric relics in
the Milky Way. As discussed in Refs.~\cite{bottino98,bergstrom99,bieber99},
the distribution of secondary antiprotons turns out to be flatter than
previously estimated. Therefore, it is still a quite difficult task to
ascertain which fraction of the measured antiproton spectrum may
be interpreted as a supersymmetric component. Notice however
that as soon as the secondary {\mbox{$\bar{\rm p}$}} flux is reliably estimated,
low--energy antiproton searches will become a more efficient tool.
Meanwhile, we must content ourselves with using observations
as a mere indication of what a supersymmetric component cannot
exceed. The vertical shaded band of
Figs.~\ref{fig:ndbar_pbar_scatter} and \ref{fig:dbar_pbar_scatter}
corresponds actually to the 1--$\sigma$ antiproton flux which the
BESS 95 + 97 experiments \cite{bess_2} have measured at a {\mbox{$\bar{\rm p}$}}
kinetic energy of 0.24 GeV.
In Fig.~\ref{fig:ndbar_pbar_scatter}, the supersymmetric
antideuteron yield $N_{\mbox{$\bar{\rm D}$}}$ has been derived at solar
maximum. This corresponds to the conditions of the future
AMS mission on board the space station. The antideuteron
yield is plotted as a function of the associated supersymmetric
{\mbox{$\bar{\rm p}$}} flux at Earth. The latter is estimated at solar minimum
to conform to the BESS data to which the vertical band refers.
The scatter plot of Fig.~\ref{fig:ndbar_pbar_scatter} illustrates the
strong correlation between the antideuteron and the antiproton
signals, as may be directly guessed from
Eq.~\ref{dNdbar_on_dEdbar_susy}. The horizontal dashed line
indicates the level of sensitivity which AMS/ISSA may reach.
Points located above that line but on the left of the shaded vertical
band are supersymmetric configurations that are not yet excluded
by antiproton searches and for which the antideuteron yield is
potentially detectable. The existence of such configurations illustrates
the relevance of an antideuteron search at low energies.
\begin{figure*}[h!]
\centerline{
\resizebox{0.7\textwidth}{!}
{\includegraphics*[1.5cm,6.5cm][18.5cm,23.cm]{dbar_pbar_scatter.ps}}
}
\caption{
Both supersymmetric antideuteron and antiproton fluxes at
the Earth are plotted against each other. They are modulated
at solar minimum, while the energy per nucleon is
$T_{\mbox{$\bar{\rm D}$}}^{\oplus} / {\rm n} = 0.24$ GeV/n. As in
Fig.~\ref{fig:ndbar_pbar_scatter}, the configurations are clearly
aligned, hence a strong correlation between the antiproton and
antideuteron signals.
}
\label{fig:dbar_pbar_scatter}
\end{figure*}
As shown in Fig.~\ref{fig:dbar_pbar_scatter}, the number
of interesting configurations is largest at solar minimum.
Both {\mbox{$\bar{\rm D}$}} and {\mbox{$\bar{\rm p}$}} fluxes at Earth are plotted against
each other. Energies have been set equal to 0.24 GeV/n.
The correlation between the antideuteron and antiproton
cosmic--ray fluxes is once again noticeable.
Once the energy spectrum of the secondary component
is no longer spoilt by considerable theoretical uncertainties,
measurements of the antiproton cosmic--ray flux will
be a powerful way to test the existence of supersymmetric
relics in the galactic halo. In the mean time, searches for
low--energy antideuterons appear as a plausible alternative, worth
being explored. A dozen spallation antideuterons should be
detected by the future AMS experiment on board ISSA
above a few GeV/n. For energies less than $\sim$ 3 GeV/n,
the {\mbox{$\bar{\rm D}$}} spallation component becomes negligible and may
be supplanted by a potential supersymmetric signal. We
conclude that the discovery of a few low--energy antideuterons
should be taken seriously as a clue for the existence of
massive neutralinos in the Milky Way.
\vskip 1.cm
\noindent {\bf Acknowledgements}
\vskip 0.5cm
\noindent We would like to express our gratitude toward S.~Bottino for
stimulating discussions. We also wish to thank R.~Battiston and
J.P.~Vialle for supplying us with useful information pertaining to
the AMS experiment. This work was supported by DGICYT under grant
number PB95--1077 and by the TMR network grant ERBFMRXCT960090 of the
European Union.
\newpage
|
\section{Introduction}
Speckle interferometry observations, made just 30 and 38 days after the
explosion of supernova SN1987A (which was first seen on February 23, 1987),
showed evidence for a bright source, separated from the SN by only 60 mas
(Nisenson et al, 1987). This source (we will call it BS1) was seen at least
three times: by us 30 and 38 days after the explosion and by (Meikle et al,
1987) 50 days after the explosion. Two months later, (Chalabaev et al, 1989)
detected asymmetries in infrared speckle observations that were in
approximately the same north-south direction as BS1 which may have been
related to BS1. We observed SN1987A again 98 days after the explosion, but the
images failed to show conclusive evidence that BS1 was redetected. While at
the time, there was much speculation as to plausible mechanisms for the
observations, see, for example (Rees, 1987; Piran and Nakamura, 1987; Colgate
et al, 1990), no fully satisfactory model was put forward to explain the
results. In this paper, we present a new reconstructed image that shows not
only a much cleaner reconstruction of BS1, but it also shows evidence of a
second bright spot on the opposite side of the SN. We will demonstrate that
this image is consistent with all the present data if we assume relativistic
jets emanating from the the SN. In addition, two groups have recently
suggested that at least some Gamma Ray Bursts (GRBs) could be explained as
highly collimated jets from supernovae (SNe) (Wang and Wheeler, 1998; Cen,
1998). A theoretical model for these jets, developed by Cen (Cen, 1999) to
explain GRBs is not only consistent with our original observations, but also
provides an explanation why BS1 faded by day 98.
\section{Observations}
Extensive observations of SN1987A were performed over the first two years
after the explosion, using the CfA PAPA detector and speckle interferometry
optics with the CTIO 4-meter telescope. Details of observing procedures and
data reduction are in our earlier papers (Karovska et al, 1989; Papaliolios et
al, 1989; Karovska et al, 1991). These observations provided a unique high
angular resolution data base on the early history of SN1987A. The changing
size and shape of the expanding envelope was monitored, and substantial
asymmetries were detected. Observations of the SN debris with the WF/PC
camera and the FOC over the last few years have confirmed the accuracy of our
measurements of the expansion velocity, as well as the asymmetries in the
debris shape and direction of the elongation (Pun, C.S.J. 1995; Pun, C.S.J
1997). The direction of polarization ($200^{\circ}$) of the SN (Schwarz and
Mundt, 1987) was also found to be close to the direction of elongation of the
SN debris as measured by us ($200^{\circ}$), suggesting asymmetries in the
expanding debris. However, our observations of BS1, that were seen at several
wavelengths, 30 and 38 days after the explosion, were viewed with some
skepticism, particularly since new observations made two months later did
not provide a certain redetection of BS1.
\section{Data Processing}
Recently, we went back to the original data and reprocessed them using newer,
more sensitive algorithms, to see if anything new could be learned about BS1
that would allow us to understand more about its nature. The original
processing had used the Knox-Thompson (KT) algorithm (Knox and Thompson, 1974)
to recover both the modulus and phase of the image Fourier transform, in order
to obtain true image reconstructions. We used a large image magnification in
our speckle camera, so that the telescope diffraction limit would be well
oversampled. This resulted in a field of only 1.9 arseconds for the SN
observations, which was overfilled by the 2 to 3 arcseconds seeing disk. For
the KT algorithm to work, the individual speckle frames must have a field size
that is at least twice the diameter of the seeing disk, so that adjacent
frequencies in the Fourier transform of each speckle frame are correlated in
phase.
In order to meet the KT requirement of field size and seeing disk size, each
speckle frame was multiplied by a gaussian mask which effectively narrowed the
"seeing disk" to less than one arcsecond. This allowed the phase to converge
in the speckle average, but it also had the deleterious effect of reducing the
signal-to-noise ratio in the reconstructions, since all the speckles towards
the outside of the mask were discarded.
Between the time of our original processing of the data and the present, we
have implemented a new image reconstruction algorithm that does not require
masking the data with a gaussian. Instead, we generate the speckle
autocorrelation using the unmasked data and then apply a modified version of
an iterative transform algorithm (ITA) (Fienup, 1984) to find the phase in the
Fourier transform and produce an image. The major drawback in the ITA method
is that there is a $180^{\circ}$ ambiguity in the orientation of the
reconstruction. However, the masked KT reconstructions clearly show BS1,
allowing a determination of the orientation and eliminating the ambiguity,
despite their poorer SNR.
The results of this processing on the SN data are consistent with our earlier
KT results but give cleaner reconstructions of BS1. In addition, the cleaner
images reveal a new feature that we believe could play an important role in
understanding not only BS1 but also the geometry of the whole SN system.
Figure 1 shows our new reconstruction from the data set taken with a 10 nm
wide filter centered on 654 nm (10 nm bandpass) from day 38. On the left is a
false color image and on the right, a contour plot with plate scale and
orientation on it. The 654 nm data were far and away the best for SNR of all
our observations, since they were aquired with the smallest zenith angle, so
the atmospheric dispersion was small, the seeing was the best, and the signal
levels were the greatest (the peak sensitivity of the detector is in the red).
The plot shows the lowest 1\% contours with a linear display up to 10\%, and
then has contours at 20, 40, 60 and 80\%. It very clearly shows BS1 separated
by four pixels from the SN, corresponding to $60 \pm 8$ mas angular
separation, with a magnitude difference from the SN of $2.7 \pm .2$
magnitudes. It also reveals a second feature, in-line with BS1, on the
opposite side of the SN, separated by $160 \pm 8$ mas and somewhat fainter
than BS1 (a $4.2 \pm .2$ magnitude difference from the SN). We refer to this
feature as BS2. While BS2 is substantially fainter than BS1, it is well
above the noise floor in the new image (but not in the original KT
reconstructions). BS2 is $9.5 \sigma$ above the background where $\sigma$
is the standard deviation calculated from regions adjacent to the images.
There are a few additional point-like peaks in the image that are almost
certainly reconstruction noise. These peaks have a different character from
BS1 and BS2 since they do not show the roughly east-west elongation that is
seen in the SN, BS1 and BS2 features. This elongation is due to imperfectly
corrected atmospheric dispersion. The brightest of these other peaks
(north-east of the SN) is 0.7 magnitude fainter than BS2, and the other peaks
are more than 1 magnitude fainter (so they fall below the 1\% contour level in
the plot). These noise peaks are introduced when we deconvolve the SN image
with data recorded on a nearby bright, unresolved star, using a Wiener
filter. This deconvolution is critical in correcting the speckle
interferometry transfer function (Nisenson, 1989). Adjusting the Wiener filter
smooths the image and changes the noise characteristics. For the image in
Figure 1, the filter was adjusted to cleanly separate BS1 from the SN. If the
filter is set so that the high frequencies in the image are reduced, the noise
peaks are eliminated. BS1 and BS2 remain but have a broader shape and BS1
starts to merge with the SN image.
Three other data sets were processed in the same way: the images taken on day
38 with filters centered on 533 nm and 450 nm, and the image taken on day 30,
with the filter centered at 654 (all filters had a 10nm bandpass). The 533 nm
image from day 38 and the 654 nm image from day 30 image show the BS1 feature
more clearly than the KT reconstructions, but the reconstruction noise does
not allow a reliable identification of a feature corresponding to BS2. There
is a feature in the day 30 image in the location predicted if there were
linear motion from the date of the explosion. However, this image is noisier
than the 654 nm day 38 image, and the 2nd spot is at the same level
as the reconstruction noise. The day 30 image also shows that the separation
of the centroid of BS1 and the SN appears to be about 1 pixel less than the
separation in the day 38 image, consistent with linear outward motion of the
feature for the 8 day time difference. The image taken with the 450 filter is
very noisy and no reliable identification of even BS1 could be made from
it. We also note that (Miekle et al, 1987) detected a bright feature 50 days
after the explosion, separated from the SN by 74 $\pm$ 8 mas in the same
direction as BS1. Assuming their detection and ours were the same phenomenon,
this measurement is also consistent with linear motion.
\section{Analysis}
If we interpret BS1 and BS2 as evidence for a jet and counter-jet, and we
assume that the two sources were ejected at the time of the
explosion, we can then calculate the geometry of the system. The distance to
SN1987A is nominally 50 kpc. At that distance, the angular separation of BS2
from the SN results in an apparent velocity that is superluminal
(V$_{apparent} = 1.24c$) where c is the velocity of light. This requires
that BS2's velocity is relativistic and that it appears to be superluminal
because it has a component of its velocity coming towards us
(blue-shifted). Since all relativistic velocities saturate at c, it is
reasonable to assume the actual velocities of BS1 and BS2 are equal. With that
assumption, the equation for the angle to the line of sight, $\theta$, is
given by (Pearson et al, 1981)
\begin{equation}
\tan \theta = \frac{2 V_1 V_2}{(V_2 - V_1)}
\end{equation}
where V$_1$ and V$_2$ are the apparent velocities (normalized to c) of the
red-shifted and blue-shifted components, respectively. From the
apparent distances of BS1 and BS2 from the SN in Figure 1, the known time
between the explosion and the observations, and the distance to the SN, we can
immediately calculate $\theta$, the angle that the line joining BS1, the SN
and BS2 makes with the line-of-sight to the supernova. We can then calculate
the actual velocity, $v/c$, of the components with the formula
\begin{equation}
v/c = \frac{V_1} {\sin\theta_1 - V_1 \cos\theta_1}
\end{equation}
In this case, V$_1$ = $.46 \pm .06$ and V$_2$ = $1.24 \pm .06$. Then, from
equation (1), $\theta$ is calculated to be $55^{\circ} \pm 3$ from the
line-of-sight, and, from equation (2), $v/c$ is .80. The position angle for
BS1 is $194^{\circ} \pm 3$ and for BS2 is $14^{\circ} \pm 3$.
\section{Discussion}
This result is particularly interesting since HST (FOC) images of SN1987A
recorded in O {\sc III} (500.7 nm) show that the SN appears to be at the
center of a bipolar nebula in the shape of an hourglass with a bright ring at
its waist (Plait et al, 1995). This ring appears to be centered on the SN and
is thought to be the boundary between the Red Supergiant wind from the earlier
stages of the SN progenitor evolution, and the faster wind from the Blue
Supergiant which was the last evolutionary stage before the explosion. The EUV
flash from the explosion excited the gas in the ring, and it has been slowly
dropping in brightness with time. The narrow elliptical shape probably
indicates that the outflow from the star had a preferential plane (likely to
be the SN equatorial plane) which is tipped by $44^{\circ}$ from the line of
sight if one assumes that the ring is circular. This is not very different
from the angle ($55^{\circ}$) from the line-of sight that was calculated for
the axis along which BS1, the SN, and BS2 lie. In addition, the position
angle on the sky of the minor axis of the ellipse is $179^{\circ} \pm 3$,
close to the position angle of BS1 and BS2 on the sky ($194^{\circ}$). Also,
the position angle of the jet is nearly aligned with the direction of
elongation of the SN debris.
This is highly suggestive that BS1 and BS2 are within 11 degrees from
perpendicularity to the equatorial plane of the SN if that plane is defined by
the observed ring. Since the blue-shifted spot (BS2) is to the north, this
would imply that southern part of the ring is closer to us. However, if our
geometry is correct, then it conflicts with conclusions drawn from recent
measurements made with the Space Telescope Imaging Spectrograph (STIS)
(Michael et al, 1998). They see a blue-shifted hot spot on the northern part
of the ring, and a reverse shock from high-speed debris hitting the $H_2$
region inside the ring which is also blue-shifted to the north. Since the hot
spot and the higher density material are likely to be in the plane of the
ring, this implies that the northern part of the ring is nearer to us. The
geometry implied by the STIS measurements makes interpretation of the
phenomenon as a polar jet much less likely. As the debris starts to reach the
vicinity of the ring and starts to interact with it, these ambiguities should
be resolved.
Cen's model (Cen, 1999) predicts not only the color dependence and brightness
of BS1 and BS2, but also that their visual brightness would fade in the
interval between day 38 and day 98 after the explosion. His model also shows
that the brightness of the jet is proportional to its mass, so the brightness
ratio of BS1 and BS2 should be proportional to their mass ratio. If the
intrinsic flux from each component was equal, the Lorentz factor would show
increased brightness ratio of the blue-shifted component compared to the
red-shifted component. Since the measured intensity of BS2 is approximately
four times fainter than BS1, the mass of BS1 must be much greater than the
mass of BS2 to compensate for the Lorentz effect. The spectral index of the
emission from the jet will also somewhat affect the brightness ratio and thus
the mass ratio. This difference in mass of the two components fits with the
model for a supernova jet (Cen, 1998). In his analysis, the jet is caused by
large asymmetries in supernova explosions, so one might expect unequal masses
in the jet and counter-jet.
We are reprocessing additional data, taken 98-100 days after the explosion,
with our new reconstruction algorithms to see if we can find a feature at the
position predicted by the velocity of the jet. By June 1, 1987, the SN had
increased in brightness by almost one magnitude, so even if BS1 had not faded,
as predicted, any certain detection will be very difficult because of the
expected large magnitude difference from the SN. Results of this reprocessing
will be published in a subsequent paper.
We thank Renyue Cen and John Raymond for many helpful discussions.
\newpage
|
\section{INTRODUCTION}
The number of strong gravitational lenses has grown tremendously in
the last five years, with well over 40 examples of strong
gravitational lenses produced by galaxies
(e.g.~\markcite{keeton1996}Keeton \& Kochanek 1996 and the CASTLES
webpage \footnote{http://cfa-www.harvard.edu/castles/}). When the
separations are smaller than 3\farcs0, the lens galaxy has now been
detected in all but one system: the small-separation, high-luminosity
contrast lens Q~1208+1011. The lens galaxies have normal photometric
properties and their mass-to-light ratios are typical of early-type
galaxies embedded in dark matter halos
(\markcite{kochanek1999b}Kochanek et al.\ 1999b,
\markcite{keeton1998}Keeton et al.\ 1998). For the 18 systems with
separations larger than 3\farcs0 and nearly identical redshifts (see Table
1), a normal lens galaxy is detected in only four systems (RXJ~0911+0551,
Q~0957+561, HE~1104--1805 and MG~2016+112). No lens galaxy has been
detected in any of the remaining 14 systems, with typical lower bounds on
the mass-to-light ratio of the lens about 10 times higher than those
measured for the normal lenses (\markcite{jackson1998}Jackson et al.\
1998). Whether the {\it quasar pairs}\footnote{The definition of quasar
pairs also includes close physical systems which have been proven to be
non-lenses, but exclude known lenses.} with nearly identical redshifts
separated by 3\farcs0 to 10\farcs0 are examples of ``dark'' gravitational
lenses or binary quasars remains largely a mystery. Proving either scenario
has very interesting ramifications on the dark mass distribution in the
universe, or the nature of interacting quasars (Schneider
1993\markcite{schneider1993}). There are about 2 such quasar pairs for
every 1000 optically selected quasars (\markcite{hewett1998}Hewett et al.\
1998).
\markcite{kochanek1999a}Kochanek, Falco \& Mu\~noz (1999a) used a
comparison of the optical and radio properties of the quasar pairs to show
that the majority of the systems must be binary quasars. First, five pairs
(PKS~1145--071, HS~1216+5032, Q~1343+2640, MGC~2214+3550 and
FIRST~J1643+315) are ``O$^2$R'' pairs in which only one of the quasars is a
radio source and we can be certain that the system is a binary quasar.
Second, even though most of the known lenses were radio selected and the
angular completeness of the radio lens surveys is better than that of the
optical lens surveys, there is no example of a ``dark'' radio lens (an
``O$^2$R$^2$'' pair). These two facts are inconsistent with the dark lens
hypothesis. Quantitatively, \markcite{kochanek1999a}Kochanek et al.\
(1999a) set a 2$\sigma$ (1$\sigma$) upper limit of 22\% (8\%) on the
fraction of the quasar pairs that could be gravitational lenses, the
most likely scenario being that none is a gravitational lens.
The high incidence and angular separations of the binary quasars can be
quantitatively explained in terms of merger induced quasar activity
(\markcite{kochanek1999a}Kochanek et al.\ 1999a). The enhancement of the
quasar-quasar correlation function by a factor of $10^2$ on the angular
scales of the pairs is exactly as predicted by the locally observed
enhancement by a factor of $10$ in the quasar-galaxy correlation function on
the same physical scales (\markcite{fisher1996}Fisher et al.\ 1996,
\markcite{yee1987}Yee \& Green 1987, \markcite{french1983}French \& Gunn
1983). The concentration of the pairs on scales smaller than 10\farcs0
($<50$~kpc) is a natural consequence of the need for a close passage and
tidal interactions to trigger renewed quasar activity. The absence of
binaries on scales smaller than 3\farcs0 is a natural consequence of the
rapid increase in the orbital decay rate due to dynamical friction as the
binaries shrink.
These limits on the existence of dark lenses are, however, statistical.
While the maximum likelihood solution is that there are no dark lenses, the
statistical limit corresponds to allowing the existence of 3 dark lenses at
the 2$\sigma$ limit. The existence of even 3 dark lenses would mean that
the densities of dark lens halos and normal clusters are the same on these
mass scales (\markcite{maoz1997}Maoz et al.\ 1997,
\markcite{kochanek1995}Kochanek et al.\ 1995,
\markcite{wambsganss1995}Wambsganss et al.\ 1995). Moreover, the binary
hypothesis must somehow explain the remarkable similarity of the quasar pair
spectra in the optical (e.g. \markcite{hawkins1997}Hawkins et al.\ 1997,
\markcite{hewett1989}Hewett et al.\ 1989,
\markcite{steidel1991}Steidel \& Sargent 1991,
\markcite{michalitsianos1997}Michalitsianos et al.\ 1997).
Q~1634+267A,B remains one of the best candidates for exploring the issue of
dark lenses and binary quasars. After the discovery by
\markcite{sramek1978}Sramek \& Weedman (1978) in a slitless spectroscopy
survey, \markcite{djorgovski1984}Djorgovski \& Spinrad (1984) and
\markcite{turner1988}Turner et al.\ (1988) found the optical spectra of the
two quasars to be very similar, with an upper limit on the velocity
difference of $150~\hbox{km~s}^{-1}$, at the redshift of $z=1.961$. The
flux ratio of the pair was 4.4 at R-band, although in images with seeing
comparable to the image separation (\markcite{djorgovski1984}Djorgovski \&
Spinrad 1984). No lens galaxy has been detected down to limits of
$K\approx$ 21.5 (\markcite{steidel1991}Steidel \& Sargent 1991) and
$R=23.5$ (\markcite{djorgovski}Djorgovski \& Spinrad 1984).
Steidel \& Sargent (1991; hereafter SS91) used the Palomar 5-meter telescope
to obtain very high signal-to-noise spectra of Q~1634+267 that illustrate the
striking similarity of the spectra of components A and B, modulo a constant
scaling factor of 3.28. They also found a number of absorption lines in the
brighter component A, most notably the Mg~{\sc ii} doublet
$\lambda\lambda$2796,2803 at $z=1.1262$, which perhaps hinted at the
presence of a lens galaxy. When the emission line spectra of A and B were
compared in detail, there were slight differences in the Ly~$\alpha$ + {\sc
N~v}, Si~{\sc iv} + {\sc O~iv]}, and {\sc C~iv} line profiles, and the line
velocities were shifted by as much as 300--500~km~s$^{-1}$
(\markcite{small1997}Small et al.\ 1997). The time delay created by the
different travel times for the two rays can lead to spectral differences
under the lens hypothesis, if the source quasar is variable. The temporal
variations of quasar spectra have been examined by
\markcite{filippenko1989}Filippenko (1989), \markcite{wisotzki1995}Wisotzki
et al.\ (1995),
\markcite{impey1996}Impey et al.\ (1996), and
\markcite{small1997}Small et al.\ (1997). In particular,
\markcite{small1997}Small et al.\ (1997) demonstrated that the
spectral differences observed in Q~1634+267A,B were typical of the
differences seen in spectra of the same quasar taken after an interval
of several years.
With the advent of a second generation of instruments on the Hubble
Space Telescope (HST), NICMOS has offered unique capabilities to
discover and identify missing lens candidates by observing them at
rest optical wavelengths where they are bright, while providing
unprecedented high angular resolution in the infrared necessary for
studying these systems in detail. The
CfA/Arizona~Space~Telescope~Lens~Survey (CASTLES) project is currently
conducting a study of roughly 40 gravitational lens systems and
candidates using optical/near infrared imaging with the HST. The
selection sample consists of ``simple'' systems where the strong
lensing is believed to be caused predominantly by a single galaxy.
Some of the goals are: to place tighter constraints on the lens
geometry; to study the mass and light distribution of the lens galaxy;
to study the evolutionary history and environment of the lens galaxy;
to better constrain cosmological parameters through lens modeling and
statistics, and finally, in some cases to discover the missing lens
galaxy.
In Section 2 we present our observations of Q~1634+267A,B, the photometry,
and the determination of the magnitude limit at the anticipated lens
position. We also present new limits on the radio flux of Q~1634+267A,B and
two other quasar pairs. Section 3 presents parameters derived for a
hypothesized lens galaxy from a simple isothermal sphere model. We discuss
the properties of the metal line absorber at $z=1.1262$ in Section 4, and
the statistical similarity of quasar spectra in Section 5. Conclusions
follow in Section 6. We adopt $H_0=65~\hbox{km~s}^{-1}~\hbox{Mpc}^{-1}$
throughout the paper and display key results for both $\Omega_0=0.1$ and
$\Omega_0=1$.
\section{OBSERVATIONS AND ANALYSIS}
\subsection{Observations}
We observed Q~1634+267 on 15 August 1997 with HST using NICMOS Camera 2 (NIC2)
and the $H$ (F160W) filter. We obtained four 690-second integrations in a
10.5-pixel dither pattern, for a total integration time of 2560 seconds. We
reduced the data with our own software package ``nicred''
(\markcite{mcleod1997}McLeod 1997; \markcite{lehar1999} Leh\'ar et al.\
1999). Figure 1 shows the reduced, combined image of the field of
Q~1634+267. Q~1634+267 A and B are approximately centered in the NIC2
field, and a star is visible at the edge of the $\sim 19\arcsec\times
19\arcsec$ field of view.
\subsection{Astrometry and Photometry}
The plate scales of the NICMOS cameras changed as the IR array
underwent thermal expansion. The variation was monitored
\markcite{cox1998}(Cox et al.\ 1998), and the NIC2 plate scales were
0\farcs0760926 and 0\farcs0754090 per pixel in the X and Y directions,
respectively for the measurement nearest the date of observation. The
difference in the X and Y plate scales is due to a small tilt of the
IR array relative to the focal plane. We adopt a zero point of
$M_{H}=21.79\pm0.02$ for F160W for infinite apertures, based on a
comparison of HST archival and ground-based observations of the
standard star P330E (Persson et~al.\ 1998). It agrees closely with
the value of 21.83 mag obtained by the Space Telescope Science Institute
\footnote{http://www.stsci.edu/ftp/instrument\_news/NICMOS/NICMOS\_phot/keywords.html}.
The foreground Galactic extinction in the direction of Q~1634+267 is only
$E(B-V)=0.072$ for $R_V=3.1$ (Schlegel, Finkbeiner \& Davis 1998); hence, we
applied no extinction corrections to our photometric estimates.
We fitted a point-spread function (PSF) to our combined image, to determine
the relative coordinates and the magnitudes of Q~1634+267 A and B. Because
A and B are well separated, we derived a PSF estimate from both components
with the IRAF\footnote{IRAF(Image Reduction and Analysis Facility) is
distributed by the National Optical Astronomy Observatories, which are
operated by the Association of Universities for Research in Astronomy, Inc.,
under cooperative agreement with the National Science Foundation.} DAOPHOT
package. Table 2 shows the resulting relative coordinates and H-band
magnitudes of A, B and the star within our field of view. The quasar images
are separated by 3\farcs852$\pm$0.003; the brightness ratio A/B is $3.34 \pm
0.06$.
We obtained 10-minute snapshots of three quasar pairs, Q~1634+267,
Q~1120+0195, and LBQS~1429--008, on 9 Dec 1998 with the VLA in the C array at
$\lambda = 3.6$ cm. We detected no radio flux to $1\sigma$ limits of
$0.028$, $0.030$, and $0.045$~mJy/Beam, respectively. With these low flux
limits, there is little hope that further radio observations can shed much
light on these systems.
\subsection{Magnitude Limit at the Expected Lens Position}
Visual inspection of the images revealed no lens candidate near the expected
position. To search for a faint lens that might be hidden by A or B, we
fitted the PSF derived in the previous section to the quasar components. We
then subtracted the fitted PSF from these components. The fainter one, B,
was marginally broader than A. The difference might be caused by intrinsic
PSF differences at different positions on the detector, and similar small PSF
mismatches were seen in other NICMOS archival images containing two or
more stars. Thus, we cannot conclude that the slight broadening is due to a
faint galaxy underneath quasar B. Such a galaxy would have $m_H\gtrsim21.8$
mag, which is too faint and at the wrong location to be able to produce
lensed images with the observed splitting of $\sim 3\farcs8$.
To determine our detection limit for a lens galaxy, we generated circular
galaxies with a de Vaucouleurs profile and an effective radius corresponding
to an $L^*$ galaxy ($\sim 3h_{65}^{-1}$~kpc) over the redshift range
$0.4\lesssim z\lesssim1.8$ and for $\Omega_0=0.1$ and $1.0$ cosmologies. We
convolved the redshifted galaxies with the PSF derived from A and B, and
added them to our NIC2 frame on a grid of positions within the frame. The
resulting images were smoothed with a Gaussian kernel with $\sigma = 3$
pixels, and inspected for signs of the added galaxies. We found a magnitude
limit of $m_H = 22.5$ ($1\sigma$) for an early-type lens galaxy. For the
sky background in our frame, the standard deviation of the noise level per
pixel corresponds to a surface brightness of roughly 23.3 mag arcsec$^{-2}$.
\section{LOWER LIMIT ON A LENS M/L RATIO}
We compute the expected mass to light ratio for a putative lens galaxy for
Q~1634+267 from our detection threshold. The anticipated mass of the lens
can be derived from simple lens theory assuming it has a singular isothermal
sphere (SIS) mass profile. The lens equation relates the angular
coordinates $\beta$ ($\theta$) of the source (images),
$\vec{\beta}=\vec{\theta} - \vec{\alpha}(\theta)$, where $\alpha(\theta)$ is
the deflection angle for an image at $\theta$, with the lens galaxy at the
center of coordinates. The SIS assumption implies: $\alpha(\theta) =
(1/2)\Delta\theta = 4 \pi (D_{ds}/ D_s) ({\sigma_v}^2/c^2)$, where $D_s$
($D_{ds}$) is the distance from the observer (lens) to the source;
$\Delta\theta$ is the observed angular separation of the lensed images, and
$\sigma_v$ is the velocity dispersion of the lens galaxy. Therefore, using
the observed image separation one can determine the velocity dispersion
$\sigma_v$ for a lens galaxy as a function of redshift. We show in Figure
2a the velocity dispersion ${\sigma_v}(z)$ using cosmologies $\Omega_0=0.1$
and 1, and $H_0=65~\hbox{km~s}^{-1}~\hbox{Mpc}^{-1}$. For an A/B magnification
ratio of $\mu$, the lens galaxy should lie $\Delta\theta\{\mu/(\mu+1)\}$ from
image A along the A-B separation vector.
We translate our magnitude limit of $H=22.5$ into an upper bound on the
luminosity of the missing lens galaxy, placed at various redshifts. We
follow the method implemented in Keeton et al.\ (1998) to derive $H$--band
luminosity of elliptical galaxies at various redshifts using
\markcite{charlot1993}Charlot \& Bruzual (1993) models. The models account
for k-correction and passive evolution from formation redshift $z=5$.
Coupled with the masses required to produce lensing, determined via the lens
equations, the luminosity limits yield the mass-to-light ratio (M/L) as a
function of redshift, shown in Figure 2b for cosmologies of $\Omega_0=0.1$
and $1$. The most likely (comoving) distance to the lens is half the
distance from the observer to the source, where the galaxy has the largest
cross section for multiply imaging a source. For Q~1634+267 at redshift
1.961 the most probable lens location is therefore at redshift $0.45$. An
elliptical galaxy at $z=0.5$ would have $M/L\gtrsim1000$ in a low density
universe, which is $\gtrsim50$ times higher than a normal lens galaxy
(\markcite{keeton1998}Keeton et al.\ 1998,
\markcite{jackson1998}Jackson et al.\ 1998), and 50 times higher than a
limit placed by \markcite{turner1988}Turner et al.\ (1988) in the optical.
Therefore, if obscuration is responsible for hiding a lens galaxy with a
normal mass to light ratio from view, then the dust must cause at least 4
magnitudes of extinction in the $H$-band. If, however, we accept the
Mg~{\sc ii} absorption lines as evidence for a lens galaxy at $z\approx1.1$
(\markcite{steidel1991}SS91), then the detection limit of $H>22.5$
corresponds to about $0.15L^*_H$, or $M/L_H\gtrsim 690$ for an E/S0 type
galaxy. But if the lens galaxy type is an Sb, at a redshift of $z=1.1$ it
would have a $M/L_H\gtrsim 430$. The minimum $M/L_H$ is $400$ for an Sb
galaxy at a redshift of $z=1.3$.
If the lens is an early-type galaxy, we can predict its luminosity from the
Faber-Jackson relation \markcite{faber1976}(1976) between velocity dispersion
$\sigma_v$ and luminosity, ${\sigma_v /\sigma^*}=(L/L^*)^{1/4}$, where
$\sigma^*$ and $L^*$ are the velocity dispersion and luminosity of an $L^*$
galaxy. Studies of lens statistics (e.g. \markcite{kochanek1996}Kochanek
1996) show that the velocity dispersion estimated from the image separation
should closely match the central velocity dispersion of the stars. Figure
2a shows the velocity dispersion estimated from the lens geometry as a
function of the lens redshift. At $z\approx0.5$ the predicted velocity
dispersion is $\sigma\approx350~\hbox{km~s}^{-1}$, considerably higher than
the characteristic velocity dispersion of an $L^*$ galaxy of $\sigma^*
\simeq 220\hbox{km~s}^{-1}$. A galaxy at $z\approx0.5$ would be a
$L\approx6L^*_H$ galaxy, far brighter than normal early-type galaxies,
where $1L^*_H$ would correspond to $H\approx17.4$ $(16.9)$ mag for a
passively evolving stellar population formed at $z_f\approx5$ using the
\markcite{charlot1993}Charlot \& Bruzual (1993) models for $\Omega=0.1$
$(1.0)$. Such a bright galaxy would be at least $650$ $(1000)$ times above
our detection limit. At the redshift of the Mg~{\sc ii} absorber of
$z\approx1.1$, the required velocity dispersion rises to
$\approx500~\hbox{km~s}^{-1}$, at the highest end of the range observed for
bright galaxies (see for example, \markcite{bender1996}Bender et al.\
1996). It would have an extreme luminosity of $15L^*_H$ if the scaling
relation continued to hold.
We conclude that the absolute lower limit on the mass-to-light ratio of the
putative lens is $M/L_H\approx690$ for an early type galaxy at a redshift
$z\approx1.1$, and $M/L_H\gtrsim 400$ for an Sb type at a redshift of
$z\approx1.3$. The limit is higher for any other redshift and for
$\Omega_0>0.1$. For the most probable lens redshift of $z\approx0.5$ and
the cosmology of a high-density universe, the limit is $M/L_H>1000$.
Depending on the redshift of the presumed galaxy, the observed $H$ band
corresponds roughly to rest $R$ band ($z\approx1.5$) or $I$ band
($z\approx0.4$). Since $(B-H)\approx5.7$ mag for ellipticals at redshift
0.5, and $(B-H)_\odot=1.91$ mag, the limit would be about 30 times higher
for $M/L_B$ than for $M/L_H$. If Q~1634+267A,B is the split image of a
single quasar, the lens must be a galaxy with an unprecedentedly high
mass-to-light ratio.
In fact, as we discuss in \S4, there are no galaxies above a detection limit
of $\approx 0.1L^*_H$ within $40h_{65}^{-1}$~kpc and no galaxies above a
detection limit of $\approx 1L^*_I$ within $100h_{65}^{-1}$~kpc of the
quasars. Thus the depth of the imaging also rules out any normal cluster of
galaxies centered near the expected lens position. Moreover, the limit on
the optical $M/L$ of $M/L_B \approx 10^4h_{65}$ greatly exceeds even the
usual estimates for groups and clusters of galaxies, where virial $M/L
\approx 200$--$400h$ (e.g., see \markcite{carlberg1996}Carlberg et al.\
1996, \markcite{bahcall1995}Bahcall et al.\ 1995, \markcite{david1995}David
et al.\ 1995) is typical.
\section{THE NATURE OF THE Mg~{\sc ii} ABSORBER}
In Q~1634+267A,B, metal absorption lines are observed at $z=1.839$ and
$z=1.126$ (\markcite{steidel1991}Sargent \& Steidel (1991),
\markcite{djorgovski1984}Djorgovski \& Spinrad (1984), and
\markcite{turner1988}Turner et al.\ (1988)). Mg~{\sc ii}
$\lambda\lambda2796, 2803$ absorption doublets with equivalent widths like
the EW$_0$(Mg~{\sc ii} $\lambda2796)=1.51$\AA\ doublet at $z=1.126$
(\markcite{steidel1991}SS91) are associated with chemically enriched
material in the halos of visible, luminous galaxies along the line of sight
(e.g.\ \markcite{steidel1993}Steidel 1993, \markcite{steidel1995}1995;
\markcite{steidel1997}Steidel et al.\ 1997). Because quasars show roughly
one metal line absorption system per unit redshift down to a typical rest
equivalent width limit of $\sim0.3$\AA\ \markcite{steidel1993}(e.g. Steidel
1993), the mere presence of a metal absorption system is not evidence for the
presence of a lens galaxy. In a survey of galaxies selected by the presence
of Mg~{\sc ii} absorption, \markcite{steidel1993}Steidel (1993) showed that
the absorbers are associated with galaxies of Hubble types from E to Sb.
The strength of the absorption is anti-correlated with the distance of the
quasar from the nearest luminous galaxy, and it is correlated with the
intrinsic brightness of the galaxy (\markcite{steidel1995}Steidel
1995,\markcite{steidel1993} 1993). The luminosities of the $\sim 70$
systems studied by \markcite{steidel1995}Steidel (1995) and
\markcite{steidel1997}Steidel et al.\ (1997) range from $0.02L^*$ to
$4L^*$ with 90\% (61/70) having $L > 0.1L^*$. All but one (i.e. 99\%)
of the galaxies were found within a projected radius of $40h_{65}^{-1}$~kpc.
We searched both the NICMOS H image and existing HST WFPC1 optical images
(obtained under GTO-1116 by Westphal) for signs of a galaxy associated with
the $z=1.126$ absorption feature. Since the absorbing galaxy could have an
exponential disk profile rather than the de Vaucouleurs profile considered
in \S2, we re-derived the detection limit for an exponential disk galaxy
with a scale factor of $4\ h^{-1}_{65}$~kpc. The limit of $22.5$ mag
($1\sigma$ for $\Omega_0=0.1$) was similar to that for a de Vaucouleurs
model with a $3\ h^{-1}_{65}$~kpc effective radius. In the evolutionary
models of \markcite{charlot1993}Charlot \& Bruzual (1993), a passively
evolving $L^*$, Sb-type, galaxy which formed at $z_f=5$ would have
$H_*\approx19.7$ for $\Omega_0=1.0$, and $H_*\approx20.0$ for
$\Omega_0=0.1$, at $z=1.126$. Therefore, our detection limit implies that
we could have detected a galaxy with $L > 0.1L^*_H$ anywhere within an
impact parameter at $40h_{65}^{-1}$~kpc from Q~1634+267A (to the edge of the
NICMOS array along the short direction). We derive a weaker detection limit
of $L>L^*_I$ from the F785LP (I-band) out to $100h_{65}^{-1}$~kpc. Thus,
any galaxy associated with the $z=1.126$ Mg~{\sc ii} absorption feature is
either unusually faint or distant compared to those found in the
\markcite{steidel1995}\markcite{steidel1993} Steidel (1993, 1995) surveys.
Since the PSF of quasar B is broader than that of quasar A, it is unlikely
that the null detection could be explained by a galaxy directly under
Q~1634+267A, although we cannot exclude a very sub-luminous, compact dwarf
with perfect alignment. Alternatively, a galaxy with a luminosity $\gtrsim$
$1L^*_B$ lying outside our field of view might be responsible for the
absorption. Distant ($> 40\ h^{-1}_{65}$~kpc), luminous ($L>L^*$) galaxies
are associated with absorption, but the Mg~{\sc ii} equivalent widths are
less than 0.4\AA\ (\markcite{steidel1995}Steidel 1995,
\markcite{steidel1997}Steidel et al.\ 1997), compared to the observed
1.5\AA. For distant absorbers it is also odd that no equivalent Mg~{\sc ii}
absorption feature is seen in Q~1634+267B. In summary, the absorbing galaxy
falls outside of the parameter space populated by galaxies found in Mg~{\sc
ii} absorber survey. The absorber must therefore be very under-luminous,
$\lesssim 0.1L^*_H$.
\section {SIMILARITY OF THE QUASAR SPECTRA}
In the absence of a lens galaxy, it is the remarkable spectral similarity of
many of the quasar pairs which provides the impetus for the gravitational
lens interpretation. The spectra of the Q~1634+267A,B pair show some of the
greatest similarities of the pair population, and \markcite{small1997}Small
et al.~(1997) have demonstrated that the observed differences are consistent
with the temporal variations of quasar spectra over epochs separated by
months to years. No study, however, has examined whether the spectra of the
close pairs show a degree of spectral similarity that is statistically
greater than that of randomly selected quasars. An affirmative answer
increases the credibility of the lens hypothesis, because the lens hypothesis
provides a natural explanation for the spectral similarities. While it
would not rule out the binary hypothesis, it leaves a serious puzzle as to
why the nuclear regions of two separate galaxies have such highly correlated
emission properties. We start by defining how we will measure spectral
similarities, and then estimate the likelihood of finding a single quasar
pair with spectra as similar as Q~1634+267A,B. In isolation, the significance
is hard to interpret because we picked Q~1634+267A,B for study due to its
spectral similarities. Thus, in our final step we estimate the likelihood
of finding a population of quasar pairs with the observed spectral
similarities.
\subsection {Sample Definition}
We will compare spectra using the observed differences between the
spectral slope $\alpha$ and the rest equivalent widths (EW$_0$) of the
Ly~$\alpha$, {\sc C~iv} ($\lambda\lambda$1548, 1550) and {\sc C~iii]}
($\lambda$1909) emission lines. We selected these features because
they are commonly measured in statistical surveys of quasar spectra.
For Q~1634+267 we computed the indices from the
\markcite{steidel1991}SS91 spectra. These calibrated spectra were
taken at the Palomar 5-meter, using a double spectrograph that covered
the blue (3150--4750\AA) and the red (4600--7000\AA) wavelengths.
We used either the Large Bright Quasar Survey (LBQS) or the pair
population as a whole for a comparison sample.
The LBQS consists of 1055 optically selected quasars
(\markcite{foltz1987}Foltz et al.\ 1987, \markcite{foltz1989} Foltz et al.\
1989, \markcite{hewett1991}Hewett et al.\ 1991). We use the measurements of
the spectral indices for the LBQS quasars from
\markcite{francis1996}Francis (1996), \markcite{hewett1995}Hewett et al.\
(1995), \markcite{francis1993a}Francis (1993a), and
\markcite{francis1991}Francis et al. (1991). The binned distributions
for the LBQS quasars were fitted with analytic functions (composites of
Gaussians and Lorentzians) to provide smooth probability distributions.
Some of the spectral features are correlated
(\markcite{francis1992}Francis et al.\ 1992;
\markcite{francis1993a}Francis 1993a), and we will conduct tests to
examine the effects of these correlations on the probability
calculations.
Unfortunately we lack a complete spectral database for the pairs, although
the uncanny similarity of Q~1634+267A,B is not an isolated case. There are
14 quasar pairs with angular separations
$3\arcsec\lesssim\theta\lesssim10\arcsec$ and similar redshifts (see Table
1). We will frequently refer to this as our {\it 14 pair sample}. Of these
14, we can reject 7 as lenses. In the cases of PKS~1145--071, HS~1216+5032,
Q~1343+2640, J~1643+3156 and MGC~2214+3550, only one of the quasars is a
radio source; in the case of Q~0151+048 the small emission line redshift
difference is confirmed by an absorption feature in the background quasar;
in LBQS~2153--2056, the spectrum of quasar B shows a strong absorption
blueward of the {\sc C~iv} emission line. Four of these binaries
(QJ~0240--343, PKS~1141--071, HS~1216+5032, Q~0151+048) show fairly similar
spectra, although they all have at least one significant spectral
difference. Of the 7 ambiguous pairs, five (Q~2138--431 in
\markcite{hawkins1997}Hawkins et al.\ 1997, Q~1429--008 in
\markcite{hewett1989}Hewett et al.\ 1989, Q~1634+267 and Q~2345+007 in
\markcite{steidel1991}SS91, Q~1120+0195 in
\markcite{michalitsianos1997}Michalitsianos et al.\ 1997) show
remarkably similar optical spectra. For the pair sample we restrict
our comparison to the {\sc C~iv} equivalent width and the power law
slope $\alpha$ because the other two indices are not consistently
observed or are impossible to measure without the original data.
\subsubsection {Line and Continuum Slope Measurements}
To determine the spectral slope, we fit a power law (defined as
F$_\nu\propto\nu^\alpha$) excluding regions where there are emission or
absorption lines. The Fe~{\sc ii} in the UV spectra of quasars introduce
substantial uncertainty in the measurement of continuum slopes. The power
law $\alpha$ is also uncertain in the blue spectrum from contamination due
to the extended wings of the {\sc C~iv}, {\sc N~v}, and Ly~$\alpha$ emission
lines, as well as the Ly~$\alpha$ forest absorption. Because of these
complications, the width of the spectral slope distribution remains
controversial. Some (e.g. \markcite{elvis1994}Elvis et al.\ 1994;
\markcite{webster1995}Webster et al.\ 1995) find large dispersions,
while others find little or no dispersion in $\alpha$
(\markcite{francis1996}Francis 1996; \markcite{sanders1989} Sanders et al.\
1989). In our analysis we will adopt the narrow $\alpha$ distribution from
\markcite{francis1996}Francis (1996), as derived from 30 LBQS quasars. The
Francis (1996) distribution has a root-mean-squared (RMS) width of
$\sigma_\alpha=0.3$, compared to $\sigma_\alpha=0.5$ for the broader
distributions of \markcite{francis1992}Francis et al. (1992). In Q~1634+267
the mean slopes are $\alpha=-0.17$ in the blue (observed 3150--4734\AA) and
$\alpha=-0.12$ in the red (observed 4580--6985\AA) spectra respectively. In
both the blue and the red, the power law indices between A and B differ by
$\Delta\alpha=0.03\pm0.03$.
When comparing equivalent widths we must be careful to use identical
measurement methods because of the difficulties introduced by line blending
and continuum definition. \markcite{francis1993a}Francis (1993a) showed that
{\sc C~iv} equivalent widths can vary by a factor of two depending on the
measurement method, and thus broaden the distribution of equivalent widths
by the same factor. For Q~1634+267A,B we avoid these problems by using the
same definitions followed by the LBQS (\markcite{francis1993a}Francis 1993a).
For this method, continuum windows are defined on both sides of a given
line. Then the flux is summed above a line adjoining the continuum windows,
over a region known as the line window. The continuum windows are chosen to
exclude the flux contributed by nearby broad lines, so the line fluxes
should be considered a lower limit. The difference in rest equivalent widths
(EW$_0$) between A and B are 3\AA\ for {\sc C~iv}, 8\AA\ for Ly~$\alpha$
and 3\AA\ for {\sc C~iii]}. Measuring EW$_0$({\sc C~iv}) requires
connecting the red and blue halves of the spectra because the blue side
spectrum terminates on the red wing of the {\sc C~iv} emission line, causing
the continuum windows to fall on different spectrographs. We calculate a
scaling factor of 1.22 using the regions of overlap. But there is some
evidence that the same scaling factor might not apply to quasar B because of
large fluctuations in the data over that region. Therefore the EW$_0$({\sc
C~iv}) differences of 3\AA\ should be considered an upper limit. Indeed,
restricting ourselves to the blue regions alone and fitting Gaussians to the
emission lines we find quasars A and B to differ at most by
$\Delta$EW$_0$({\sc C~iv})$~=1$\AA. The two methods yield the same flux.
For completeness, and because the numbers have not been published elsewhere,
we also measure other line fluxes and summarize them in Table 2, although
some we do not use in our analysis. All of the line fluxes are measured with
the technique described in \markcite{francis1993a}Francis (1993a) except for
{\sc N~v}. Because the blending with Ly~$\alpha$, the measurement for {\sc
N~v} lines is done by fitting Gaussians to Ly~$\alpha$ simultaneously with
{\sc N~v} where we assumed that both lines are similar in shape.
For the other quasar pairs we estimated bounds on the {\sc C~iv} equivalent
width and the spectral index $\alpha$ based on the published spectra. For
the 5 most similar pairs, we estimate that they have spectral differences
$\Delta\alpha\lesssim0.1$ and $\Delta$EW$_0$({\sc C~iv})~$\lesssim6$\AA\
(see Table 3). For the two pairs whose spectra have the worst noise
(Q~1120+0195 (\markcite{michalitsianos1997}Michalitsianos et al.\ 1997) and
Q~1429--008 (\markcite{hewett1989}Hewett et al.\ 1989)) the estimates are
rough, while they are more accurate for the other three pairs. Since the
estimates of the equivalent widths may be imprecise, we will also quote
results by doubling our bound on the differences to $\Delta$EW$_0$({\sc
C~iv})~$\lesssim12$\AA. We summarize our parameter estimates for the pairs
in Table 3.
\subsubsection {Joint Probabilities and Correlations among Spectral Features}
To compute the probability that two random quasars have given differences in
{\sc C~iii]}, Ly~$\alpha$, {\sc C~iv} equivalent widths and spectral slope
$\alpha$, we randomly draw the properties of two quasars at a time from the
compiled number distributions. If the spectral features are completely
uncorrelated, the likelihood that two quasars are similar to each other, i.e.
have differences of less than $\Delta$EW$_0$({\sc C~iv}), $\Delta$EW$_0$({\sc
C~iii]}), etc., is simply the joint product of each individual probability.
However, if there are correlations, they could make the alarming
similarities of the quasar pairs a trivial selection effect. Therefore we
first explore the extent to which spectral features are correlated.
Through a principal component analysis, \markcite{francis1992}Francis et
al.\ (1992) showed that the equivalent widths of quasar emission lines are
correlated with the power law slope $\alpha$. \markcite{francis1993b}Francis
(1993b) further showed that $\alpha$ is correlated with redshift. Their
study was based on correlations with a spectral slope distribution in
\markcite{francis1992}Francis et al.\ (1992) which has a large RMS
dispersion of about $\sigma_\alpha=0.5$. The strongest correlation, Al~{\sc
iii}$+${\sc C~iii]} with $\alpha$, has a large scatter; the correlations of
this and other lines are very weak. Subsequently,
\markcite{francis1996}Francis (1996), using more accurate photometry, found
that the spectral slope distribution actually has a much narrower dispersion
of $\sigma_\alpha=0.3$. But the newer study does not affect the dispersions
of the line strengths, which are more accurate as long as they are measured
consistently. Because the scatter of the line strengths remains large for a
much reduced range in spectral slopes, the correlations are weaker in light
of the narrow distribution compared to \markcite{francis1992}Francis et al.\
(1992). Furthermore, plotting spectral slope {\it vs.} redshift from
\markcite{francis1996}Francis (1996) shows that there is no correlation
between the two parameters. In the analysis below, we will use the narrow
power law distribution found in \markcite{francis1996}Francis (1996).
Therefore, the correlations between line emission, continuum, slopes and
redshifts can be safely ignored for our purposes.
Nonetheless, to confirm that correlations are unimportant even under
extreme scenarios we conduct the following Monte Carlo experiment to choose
quasars from the number distributions found in \markcite{francis1992}Francis
et al.\ (1992) and \markcite{francis1993a}Francis (1993a). We select from the
broader spectral slope distribution where correlations were originally
found. The correlation between emission line strengths and $\alpha$,
and $\alpha$ with redshift $z$, means that the process of
drawing quasars must proceed in a sequence to account for
correlations at each step: first we draw quasars out of the LBQS redshift
distribution, then the spectral slope distribution, followed at last by the
emission line distribution. To account for the correlation of $\alpha$ with
redshift we do the following. Having picked a redshift, and given the power
law distribution $\alpha$, we offset the entire distribution by an amount
$d\alpha=0.79\times d~log(1+z)$ (See Figure 2 of
\markcite{francis1993b}Francis 1993b). We can then randomly select a value
of $\alpha$ from this new distribution. There is, however, one subtlety.
The distribution for $\alpha$ contains no redshift information because it is
summed over all redshifts. But because of the correlation, if quasars at
{\it any given} redshift have a small dispersion in their spectral slopes,
as one accumulates the distribution over broader redshift ranges, the {\it
net} distribution would also gradually become wider. We account for this
effect by arbitrarily shrinking the dispersion and then drawing from this
narrower spectral slope distribution instead of the original, broader one.
We verified that decreasing the dispersion by 25\%, more than one would
expect, would not significantly alter our discussions below. Once a value
for $\alpha$ has been selected, one can then proceed in the same fashion to
select from the number distribution of {\sc C~iii]}, {\sc C~iv}, or
Ly~$\alpha$, accounting for their roughly linear correlations with $\alpha$
by: $EW_{line}=m\alpha+b$ where the correlation coefficients $m$ and $b$
are given in \markcite{francis1992}Francis et al. (1992). Again, we
verified that any correlations would only begin to affect our conclusions if
the number distributions were all narrowed by more than 25\%.
There may also be weak correlations of quasar spectral properties with
luminosity. The pair members share a common redshift and have similar
luminosities, so we would under-estimate the likelihood of the members
possessing similar spectra if the distributions of spectral indices are
correlated with either variable. We saw that the spectral index correlation
with redshift has little or no role in our analysis. The Baldwin
(\markcite{baldwin1977}Baldwin 1977) effect, a correlation between
luminosity and {\sc C~iv} equivalent width, has been studied extensively
(e.g. more recently see \markcite{kinney1990}Kinney et al.\ 1990,
\markcite{sargent1989}Sargent et al.\ 1989,
\markcite{baldwin1989}Baldwin et al.\ 1989). Their results show that at a
given luminosity the equivalent widths of the {\sc C~iv} emission line have
a defined range in strengths, with a scatter of about 20\AA. The strength
of the {\sc C~iv} lines decreases with luminosity and the relation holds
over 7 orders of magnitude in quasar luminosity. The parameterization of
the correlation: $EW({\sc C~iv})\propto L^{\beta}_{1450}$ has $\beta$
ranging from $\beta=-0.07$ (\markcite{francis1992}Francis et al.\ 1992) to
$\beta=-0.45$ (\markcite{smetanka1991}Smetanka et al.\ 1991). LBQS quasars,
being a magnitude limited sample, have only a small spread in absolute
luminosity, $-28\lesssim M_B\lesssim-26$ \markcite{francis1992}(Francis et
al.\ 1992). Since at any luminosity there is a considerable scatter in {\sc
C~iv} and vice versa (see \markcite{osmer1998}Osmer \& Shields 1988 for
summary), we do not expect the luminosity correlations to affect our
analysis. Nonetheless, we tested for any effects by deriving the distribution
of $\Delta\hbox{EW}_0({\sc C~iv})$ both with and without limits on the
luminosity difference $\Delta L$ of the quasars making up the simulated
pairs using the $B_J$ magnitudes and EW$_0$({\sc C~iv}) values for the
439 LBQS survey quasars from Francis (1993). We found no significant
differences between the distributions without a luminosity restriction
and those where the quasars were required to have
absolute luminosities differing by less
than a factor of 5 ($1.7$ mag) or a factor of 2.5 ($1$ mag).
Because the parameter correlations are very weak, in the discussions
that follow, we make the assumption that no two spectral features are
correlated.
\subsection {The Spectral Similarity of Q~1634+267A,B}
As discussed previously, the spectra of Q~1634+267 A and B have equivalent
width differences $\Delta\leq3$\AA\ for ${\sc C~iv}$; 8\AA\ for Ly~$\alpha$;
and 3\AA\ for {\sc C~iii]}, as well as 0.03 for $\alpha$. If we select
randomly from the distributions of these features in the LBQS survey, the
probability of finding two quasars with smaller differences are 11\%, 16\%,
30\%, and 12\% respectively. Assuming no correlations between the features,
the joint probability of the differences being as small as observed is
0.06\%.
With such a low joint probability, it would be reasonable to find several
quasars with such similar spectra in the total quasar population ($\sim 10^4$
quasars), but highly unlikely to find any in the tiny sub-population of
quasar pairs ($14$ objects). The strength of this conclusion depends only
weakly on the details of the comparison, and the joint probability only rises
to 1\% if we confine the comparison to the spectral index and {\sc C~iv}
equivalent width differences. Adding additional spectral features to the
estimate would further reduce the likelihood, but it becomes increasingly
important to fully understand the spectral correlations of quasars.
Moreover, calculations of {\it a posteriori} probabilities can generally
yield arbitrarily low likelihoods by over-specifying the problem.
Nonetheless, the similarity of the Q~1634+267 A and B spectra extends across
the entire rest-UV spectra of the pair.
\subsection {The Spectral Similarity of the Pair Population}
Because we pre-selected Q~1634+267A,B for study because of its spectral
similarities, we cannot interpret it in isolation from the rest of the pair
population. Moreover, the uncanny similarities of Q~1634+267A,B are not an
isolated example among the pairs. To fully understand the significance of
the spectral similarities of the pairs, we must consider the entire pair
population.
Under the assumption that the quasars making up the pair sample are drawn
from the same parent population as the quasars in the LBQS, we can produce
mock samples of quasar pairs by randomly selecting 14 pairs of quasars from
the LBQS. We can then compare the distribution of the spectral
slope and {\sc C~iv} equivalent width differences, $\Delta \alpha$ and
$\Delta$EW$_0$({\sc C~iv}), to those observed for the pairs. We will
estimate the probability of finding at least as many similar pairs as
observed in the real sample.
We generated $10^4$ mock pair catalogs to determine the fraction of catalogs
in which N$_{sim}$ pairs have spectral differences smaller than the
specified limits on $\Delta\alpha$ and $\Delta$EW$_0$({\sc C~iv}), with the
results illustrated in Figure 3. The top panel shows the limits on spectral
similarity ($\Delta\alpha\lesssim0.1$ and $\Delta$EW$_0$({\sc
C~iv})$\lesssim6$\AA) which characterize the differences observed for the 5
pairs with similar spectra (Q~2138--431, LBQS~1429--008, Q~1634+267,
Q~1120+0195 and Q~2345+007). The chance of drawing 5 or more pairs with
such similar spectra in the sample of 14 pairs is about 0.2\%. If we were
to relax the constraints to $\Delta\alpha\lesssim0.1$ and
$\Delta$EW$_0$({\sc C~iv})$\lesssim12$\AA\ to account for possible
systematic errors in measuring $\Delta$EW$_0$({\sc C~iv}), the probability
for finding 5 or more such pairs grows only to 3\%. We conclude that the
likelihood for finding 5 pairs with such similar spectra in a sample of 14
pairs is low. If we significantly broaden the criterion for similarity
($\Delta\alpha\lesssim0.2$ and $\Delta$EW$_0$({\sc C~iv})$\lesssim20$\AA),
the likelihood of finding five or more similar pairs rises to 60\%.
However, with these weaker criteria there are now seven observed pairs
meeting the criterion for similarity (we should add QJ~0240--343 and
LBQS~2153--2056), and the likelihood of finding seven or more similar pairs
is then smaller, at 20\%. We regard this third case as a significant
under-estimate of the similarities. The low probability for finding similar
quasars randomly, using two parameters, may only be an upper bound. If our
pairs sample were partly made up of gravitational lenses, microlensing and
time delay might cause some spectral differences in those systems. It is
impossible to quantify such a conjecture with our pairs sample.
The probability for finding similar quasars is significantly lower if we use
a broad continuum slope distribution (e.g. \markcite{francis1992}Francis
1992) rather than the narrow distribution of \markcite{francis1996}Francis
(1996) used for the estimates in Figure 3. For the broad estimate of the
$\alpha$ distribution, the likelihoods for drawing at least 5 similar pairs
under the same three sets of constraints on $\Delta\alpha$ and
$\Delta$EW$_0$({\sc C~iv}) are 0.01, 0.2\% and 1.2\% instead of 0.2\%, 3.0\%
and 20\% (see Figure 4). Thus, the debate about the width of the
continuum slope distribution does not alter the conclusion that the spectral
similarities of the pairs are unlikely to occur by chance unless the
distribution is even narrower than that found by
\markcite{francis1996}Francis (1996). The distribution of {\sc C~iv}
equivalent widths, {\it as long as measured consistently}, does not suffer
from the systematic problems of the spectral slope distribution.
\section{CONCLUSIONS}
We again confirm that there is no sign of a lensing galaxy or a surrounding
group or cluster of galaxies near the Q~1634+267A,B quasar pair. Our limit
on the mass-to-light ratio is $\gtrsim690h_{65}$ in the $H$ band, tighter than
previous estimates. We also lack a candidate galaxy within
$100h_{65}^{-1}$~kpc for producing the $z=1.126$ Mg~{\sc ii} absorption
feature in the spectrum of A. Since the existence of even one large
separation lens produced by a ``dark'' mass distribution has profound
implications, confirming or disproving the lens hypothesis is highly
desirable. It appears from the data in the literature that the flux ratio
of the quasars is significantly time variable, changing from 4.4
(\markcite{djorgovski1984}Djorgovski \& Spinrad 1984) to 2.83
(\markcite{turner1988}Turner et al.\ 1988), to 3.28
\markcite{steidel1991}(SS91) and 3.34 (this work). With its large angular
separation, regular photometric monitoring of the system to search for a
time delay would be simple on modest size, ground-based telescopes.
Although we found that the pairs Q~1634+267, Q~1120+0195, and LBQS~1429--008,
are at best weak radio sources, with upper limits on their fluxes of $0.28$,
$0.030$, and $0.045$~mJy/Beam at $\lambda=3.6$ cm, deeper observations may
still detect the sources or demonstrate that both members are at the
extremes of the radio to optical flux ratio distribution. Deep X-ray
observations would also be useful, both to measure the flux ratio of the
quasars and to search for hot X-ray emitting gas associated with any
lensing potential. Deep IR imaging should detect the quasar host galaxies,
whose shapes must show tangential, arc distortions under the lens hypothesis.
The spectral similarities of some of the quasar pairs has always been the
major impetus for the ``dark'' lens interpretation of the systems. To date,
these arguments have been based on the similarities of individual pairs
rather than the overall pair population. Using the distribution of
continuum slopes and {\sc C~iv} equivalent widths, we demonstrated that it
was improbable to find as many similar pairs as observed in the existing
sample of 14 pairs. The probability estimates range from 0.01\% to 3\%,
depending on how strictly similarity is defined and depending on whether a
broad or narrow distribution was used for the continuum slope distribution.
A better quantitative analysis needs a uniform spectral survey of the pairs.
The effects of binary quasar interaction on spectral similarity have been
surmised, for example, by \markcite{schneider1993}Schneider (1993).
\markcite{kochanek1999}Kochanek et al. (1999) demonstrated that most of the
quasar pairs are binary quasars rather than ``dark lenses.'' The binary
quasar hypothesis is a successful quantitative interpretation of both the
number of quasar pairs and the distribution of their separations. If they
are all binary quasars, then there must be a physical explanation for the
spectral similarities. Binary quasars are examples of quasar activity
triggered by mergers, which represent a small fraction of all quasar
activity (about 5\% in simple models, see Kochanek et al. 1999a). We
present two hypotheses for how binary quasars can have such similar
spectra. First, if the physical properties of quasars triggered by major
mergers (e.g. mass, accretion rate, surrounding gas density) were confined
to a limited range of values compared to the overall quasar population, then
the observable properties of the binaries would be more similar than those
of randomly selected quasars. This hypothesis has a definite, testable
prediction -- the spectra of the different pairs should be more similar to
each other than for randomly selected quasars. It would also be expected
that the quasar host galaxies have properties commensurate with being the
products of major mergers -- which can be tested by deep imaging. Second,
if quasar spectra have a significant dependence on the age of the activity,
then the binary spectra are more similar than for random quasars because the
merger provides a common triggering event for the activity in the two
systems. Here we would not expect the spectra of the different pairs to
show any more similarity than randomly selected quasars. In either case,
the binary quasar population becomes a very important probe of the physics
of quasars because it provides evidence for correlations between events on
large scales and the detailed operation of the central engine.
\section{ACKNOWLEDGMENTS}
We are very grateful to C.C. Steidel for providing the Q~1634+267 spectrum
used in our analysis and to P.J. Francis for supplying the luminosity and
{\sc C~iv} data for the LBQS quasars. We also thank C.B. Foltz, L.C. Ho,
A.V. Filippenko, W.D. Li, A. Quillen, and G. Rudnick for enlightening
discussions, and the referee P. Schneider for his comments. Support for the
CASTLES project was provided by NASA through grant numbers GO-7495 and
GO-7887 from the Space Telescope Science Institute, which is operated by the
Association of Universities for Research in Astronomy, Inc. CSK and CRK
were also supported by the NASA Astrophysics Theory Program grant
NAG5-4062. HWR is also supported by Fellowship from the Alfred P. Sloan
Foundation. Our research was supported by the Smithsonian Institution.
|
\section{Introduction\protect\\}
\label{sec:intro}
Polchinski's seminal work \cite{pol} on D-brane has dramatically changed
our view on
perturbative
superstrings. Yet, we can use many tools developed in the perturbative
framework of superstrings to do D-brane calculations. These help us, at
least in certain cases, to
attack some hard problems in physics such as the information loss puzzle
and the entropy problem in black hole physics. The D-brane picture is also
the basis
for the recent $AdS/CFT$ conjectures of Maldacena\cite{mal}.
By definition, a D-brane is a
hypersurface carrying a RR charge in type II string theory on which
an open string can end.
From the D-brane worldvolume point of view, such an ending of a
fundamental string (for
short, F-string) is characterized
by the non-vanishing U(1) gauge field strength on the brane at least
in the low energy limit. A configuration of an F-string ending
on a Dp-brane
for every allowable $p$ can actually be BPS saturated,
preserving a quarter of
the spacetime supersymmetries. At the linearized approximation,
this has been
demonstrated by Callan and Maldacena \cite{calm} for $p \ge 2$ cases and by
Dasgupta and Mukhi\cite{dasm} for $p = 1$ case. The
interpretations for $p \ge 2$ and
$p = 1$ cases are, however, quite different. In the former case,
the excitation of a
worldvolume scalar field along a transverse direction is
interpreted as the
F-string attached to the Dp-brane. Whereas, for the latter one,
the excitation of this scalar field due to the introduction of an
F-string ending indicates that one half of the original D-string
must bend rigidly to form
a 3-string junction.
In spite of our reasonably well understanding of an
F-string ending on a Dp-brane from
the worldvolume point of view, our understanding of this same ending
from the spacetime point of view is still
unsatisfactory\footnote{ For some very recent efforts in this
direction see \cite{you}.}.
The well-known $p$-brane solitonic solutions of type II supergravity
theories \cite{hows,dufl}, nowadays called Dp-branes,
are merely hypersurfaces carrying RR charges
each of which preserves one half of the spacetime supersymmetries of
type II string theories.
The mass per unit $p$-brane volume for such a configuration carrying
unit RR charge is just the
Dp-brane tension. This BPS configuration can also be described
by its worldvolume Born-Infeld
action in its simplest form with flat background and vanishing
worldvolume gauge
field strength which clearly indicates that
each of the spacetime Dp-brane solitonic solutions does not have
an F-string
ending on it (this also explains why they preserve 1/2 rather
than 1/4 of the spacetime
supersymmetries).
As just pointed out, a non-vanishing worldvolume gauge field
strength is an indication of a string
ending on the corresponding Dp-brane. In general we expect
that such a configuration preserves
a quarter of the spacetime supersymmetries. The question that we
intend to address here is:
Does there exist a BPS state for each Dp-brane that has a
non-vanishing worldvolume gauge
field strength and yet preserves one half of the spacetime
supersymmetries? We will argue in this
paper that the answer is yes based on known Dp-brane results
and the 3-string junction.
Each of these BPS states is actually a non-threshold bound
state of a Dp brane carrying certain units
of quantized constant electric field strength or a non-threshold (F, Dp)
bound state with F
representing the F-strings. There actually exist more general non-threshold
bound states. For example, by the type IIB S-duality, we should have
D3 branes carrying both quantized constant electric and magnetic fields.
We will
discuss the $p = 3$ case in this paper and others in the subsequent
publications. In the following section,
we will review relevant
Dp-brane results for the purpose of this paper. In section 3,
we will present our arguments
for the existence of such BPS states and conclude this paper.
\section{Review of Some D-Brane Results\protect\\}
\label{sec:RDP}
This section is largely based on the discussion of BPS states of
a fundamental
string (F-string) ending on a Dp-brane by Callan and
Maldacena \cite{calm}
for $p \ge 2$ and by Dasgupta and Mukhi \cite{dasm} for $p = 1$
in the linearized approximation. The linear arguments should be
trusted since we are here
interested only in BPS states. As in \cite{calm}, we assume that
the massless excitations of
a Dp-brane are described by the dimensional reduction of the
10-dimensional supersymmetric
Maxwell theory.
The supersymmetry variation of the gaugino is
\begin{equation}
\delta \chi = \Gamma^{MN} F_{MN} \epsilon,
\label{eq:susy}
\end{equation}
where $M,N$ are the 10-dimensional indices. A BPS configuration is
the one in which
$\delta \chi = 0$ for some non-vanishing killing spinor.
The ending of an F-string
on a Dp-brane is equivalent to placing a point charge on the brane.
The Coulomb potential
due to such a point charge will give rise to a non-vanishing $F_{0r}$
with $r$ the radial coordinate of the $p$ spatial dimensions of
the worldvolume. With a
non-vanishing $F_{0r}$, it is obvious from Eq.\ (\ref{eq:susy})
and $\delta \chi = 0$
that the existence of
non-vanishing Killing spinors (i.e., the preservation of some unbroken
supersymmetries), requires the excitation of
one of the scalar fields, say $X^9$, such that
$F_{9r} = - \partial_r X^9 = F_{0r}$. Then
$\delta \chi = 0$ can be expressed in a familiar form as
\begin{equation}
(1 - \Gamma^0 \Gamma^9) \epsilon = 0,
\label{eq:susyc}
\end{equation}
which says that one half of the worldvolume supersymmetries are
broken by this
configuration. In other words, this configuration of an F-string
ending on a Dp-brane is
still a BPS state which preserves one half of the worldvolume
supersymmetries or
a quarter of the spacetime supersymmetries. It is easy to check that
$F_{0r} = F_{9r} = c_p (p -2)/r^{p -1}$ for $p > 2$,
$F_{0r} = F_{9r} = c_2 / r$ for $p = 2$,
and $F_{01} = F_{91} = c_1$ for $x^1 > 0$ and
$F_{01} = F_{91} = 0$ for $x^1 < 0$\footnote{For
concreteness, we assume that the original D-string is placed along
the $x^1$-axis.}
for $p =1$ satisfy the corresponding linearized equations of
motion, respectively. This has to be true to guarantee
the existence of the corresponding BPS
states. In the above, the constant $c_p$ is related to the
point charge and can be fixed by
some charge quantization which will be discussed later.
To have a clear picture about an F-string ending on a Dp-brane,
we need to examine
the above BPS configuration closely. The cases for $p \ge 2$ and $p = 1$
are quite different.
So we discuss them separately. Let us discuss $p > 2$ first.
Here we can solve $X^9$ from
$F_{9r} = c_p (p - 2) / r^{p - 1} = - \partial_r X^9$
as $X^9 = c_p /r^{p - 2}$. As explained
in \cite{calm}, the excitation of $X^9$ amounts to giving
the brane a transverse `spike'
protruding in the 9 direction and running off to infinity.
This spike must be interpreted
as an F-string attached to the Dp-brane. Callan and Maldacena have
shown that the energy
change due to the introduction of a point charge to the
Dp-brane worldvolume equals precisely
to the F-string tension times $X^9$ which is the energy of an
F-string if the spike is
interpreted as the F-string. This also says that attaching an
F-string to a Dp-brane
does not cost any energy which is not true in the case of $p = 1$.
The $p = 2$ case is not much different from $p > 2$ cases apart
from the fact that $X^9$
now behaves according to $X^9 = c_2~ {\rm ln}~ r/\delta$
with $\delta$ a small-distance cutoff
rather than like a `spike'. This $X^9$, a sort of inverse `spike',
should also be
interpreted as an F-string because of the underlying
D-brane picture and the similar energy
relation.\footnote{The energy change of D2 due to the
introduction of a point charge to the worldvolume
can also be expressed as the F-string tension times $X^9$.
Here a large-distance cutoff needs to
be introduced to make the calculation meaningful.}
So far, we have considered only the single-center Coulomb
solution in the
above BPS states describing an F-string ending on a Dp-brane
for $p \ge
2$. The BPS nature of these configurations allows multi-center
solutions. For example, for $p > 2$, $X^9$ is now
\begin{equation}
X^9 = \sum_i \frac{c_p^i}{|\vec{r} - \vec{r}_i|^{p - 2}},
\label{eq:mcs}
\end{equation}
where $c_p^i$ can be positive or negative, depending on to which
side of
the Dp-brane an F-string is attached.
This solution represents multiple strings along $X^9$ direction
ending at
arbitrary locations on the brane. This solution is still BPS, preserving
also a quarter of the spacetime supersymmetries. The energy change of
Dp-brane due
to the endings of multiple strings is again equal to the summation
of the F-string tension times individual F-string length and is
independent of the locations of the end points. Therefore, no
attachment energy is spent for such endings. These multi-center
solutions are one of the important properties which we need in section 3.
The story for $p = 1$ case is quite different. As discussed in
\cite{dasm}, the excitation of $X^9$ is no longer interpreted as an
F-string ending on a D-string but as an indication that one
side of the original
infinitely long and straight D-string or (0,1)-string must be bent rigidly
(with vanishing axion) with respect to the point on the
D-string where a point charge is inserted. Let us look at it in some detail
since the physics picture for this case consists of the starting point
of our arguments for the existence of the Dp-brane bound states in the
next section.
In the presence of this point charge, Gauss' law in one spatial dimension
states, in the case of vanishing axion, that $F_{01} = c_1$ for $x^1 > 0$
and $F_{01} = 0$ for $x^1 < 0$\footnote{One can also have an
alternative solution of
$F_{01} = 0$ for $x^1 > 0$ and $F_{01} = - c_1$ for $x^1 < 0$.}
when the original
D-string or (0,1)-string is along $x^1$ axis. Unbroken susy condition
$F_{91} = - \partial_1 X^9 = F_{01}$ says
\begin{eqnarray}
X^9 &=& - c_1 x^1, \qquad x^1 > 0, \nonumber\\
&=& 0,\qquad x^1 < 0.
\label{eq:pqs}
\end{eqnarray}
Because of the special properties of 1 + 1 dimensional electrodynamics,
the above solution is linearly increasing away from the inserted charge,
in contrast
to $p \ge 2$ cases. Before the work of Dasgupta and Mukhi, Aharony et al.
\cite{ahasy} concluded that three strings are allowed to meet at one point
provided there exist the corresponding couplings and the charges at the
junction
point are conserved. Schwarz \cite{sch} then went one step further to
conjecture, based on his ($m$,$n$)-string \cite{schone} in type IIB theory,
that
there exists a BPS state for such 3-string junction provided the three
strings are semi-infinite and the angles are chosen such that tensions,
treated as vectors, add up to zero. With this, one should not be surprised
about the above solution and the natural interpretation, as indicated
already in \cite{sch} for an F-string ending on a D-string, that
the insertion of the point charge at the origin of the D-string
causes one half
of the string to bend rigidly. The solution itself does not spell out the
ending of F-string or (1,0)-string. But a consistent picture requires that
the point charge represents the ending of a semi-infinitely long F-string
or
(1,0)-string (chosen here along positive $x^9$ direction) coming in
perpendicular to the original D-string or
(0,1)-string along $x^1$ direction. The bent segment described by
Eq.\ (\ref{eq:pqs}) that goes out from the junction is a D-string
carrying one
unit of quantized electric flux or Schwarz's (1,1)-string
(or ($-1$,$-1$)-string
depending on the orientation) in type IIB theory which follows
from the charge conservation.
The 3-string junction has also been studied by Sen\cite{sen} from
spacetime point of view based on Schwarz's ($m$,$n$)-strings in
type IIB theory. He
showed that a 3-string junction indeed preserves 1/4 of the spacetime
supersymmetries and a string network which also preserves 1/4 of the
spacetime supersymmetries can actually be constructed using 3-string
junctions as building blocks. Such a string network may, to our
understanding, correspond to the multi-center solutions in $p \ge
2$ cases. The energy change of the D-string due to the ending of an
F-string, unlike the $p \ge 2$ cases, is no longer equal to the
F-string tension times the attached F-string length, primarily due to
the formation of the (1,1)-string bound state.
In summary, the 3-string junction, as the BPS state of an F-string
ending on a D-string, is just the consequence of
D-brane picture, 1 + 1 dimensional electrodynamics and the
non-perturbative SL(2,Z) strong-weak duality symmetry in
type IIB string theory.
One important point to notice, which is well-known nowadays and will be
useful in our later discussion, is that $m$ F-strings in the
($m$,$n$)-string
bound state in type IIB theory are just $m$ units of the quantized
electric flux.
Therefore an ($m$,1)-string can be viewed as a D-string
carrying $m$
units of quantized electric flux. The ($m$,1)-string tension is
$\sqrt{1/g^2 + m^2} T_f$ with $T_f = 1/{2\pi\alpha'}$ the F-string
tension. For
small string coupling $g$ and small $m$, the ($m$,1)-string tension can be
approximated as $(1/g + g m^2/2) T_f$. Therefore, $(g m^2/2) T_f$ should
correspond to the linearized energy per unit length
of the worldsheet constant gauge field
strength $F_{01}$, i.e., $ ((2\pi \alpha'F_{01})^2 /2g)T_f$. So
we have $F_{01} = g m T_f $ which fixes $c_1 = g T_f$ for a single F-string.
By
T-dualities, the electric field $F$ due to the ending of F-strings
on a Dp-brane is quantized according to
\begin{equation}
\frac{1}{(2\pi)^{p - 2} \alpha'^{(p - 3)/2}} \int_{S_{p - 1}} \ast F = g m,
\label{eq:fq}
\end{equation}
where $\ast$ denotes the Hodge dual in the worldvolume.
This is precisely the condition used in \cite{calm} to fix the constant
$c_p$.
\section{${\rm D}_p$ Brane Bound States\protect\\}
\label{sec:dpb}
Until now, we understand that the obvious reason for the existence
of ($m$,$n$)-strings in non-perturbative type IIB string theory
is the SL(2,Z)
strong-weak duality symmetry under which the NSNS and RR 2-form potentials
transform as a doublet. However, as we will explain below, an
($m$,$n$)-string
bound state is not special at all if it is interpreted as $n$ D-strings
carrying $m$ units of quantized electric flux or field strength
as discussed in the previous
section. We will argue in this section that such a kind of bound states,
i.e., a Dp-brane carrying certain units of quantized electric flux or field
lines,
actually exist for all Dp branes for $1 \le p \le 8$. All these bound states
are BPS saturated and
preserve one half of the spacetime supersymmetries just like an
($m$,$n$)-string. The fact that the ($m$,$n$)-strings were
discovered earlier is
because they can be easily recognized in the non-perturbative type IIB string
theory and it happens that $m$ units of quantized electric flux or field
strength can
be identified as $m$ F-strings.
Now to present our argument let us begin with a 3-string junction.
Without loss of generality and
for simplicity, let us focus on an F-string ending on a D-string
with zero axion. As discussed in the previous section, the third
string is a D-string carrying one unit of quantized electric flux.
This is a stable BPS configuration which preserves a quarter of
the spacetime supersymmetries.
Suppose that we do not have an {\it a priori} knowledge of
Schwarz's ($m$,$n$)-strings and we do the linear study as Dasgupta and Mukhi
\cite{dasm} described in the previous section. The D-brane picture makes it
certain that there must exist a stable BPS configuration of
an F-string ending
on a D-string which preserves a quarter of the
spacetime supersymmetries. So we
must conclude from our linear analysis that this BPS state is a 3-string
junction. The F-string remains as an F-string in the junction
but the electric
charge at the end of the F-string will create a constant electric
field or flux
flowing along either side of the D-string with respect to the end point.
At the final stable state, one side of the D-string remains as the original
D-string but the other side becomes a D-string carrying one unit of quantized
electric flux. This appears as 3 different kinds of strings
meeting at one point.
We know that the D-string carrying one unit of quantized
electric flux or field strength in the 3-string junction is semi-infinite.
Now let us push the
junction point to spatial infinity in such a way that the
D-string carrying one
unit of quantized electric flux is along one of the axes while
the F-string and
the D-string are all at spatial infinity. To a local observer,
this D-string carrying one unit of quantized electric flux must appear to
be a stable BPS configuration\footnote{If one thinks carefully,
each of the two ends of an ($m$,$n$)-string of Schwarz at
spatial infinity must be either
associated with a 3-string junction or attached to any other
allowable object.}.
Further, we must conclude that the D-string carrying certain units
of quantized
electric flux must be a BPS one preserving one half of the spacetime
supersymmetries based on the facts that the 3-string junction preserves a
quarter of the spacetime supersymmetries and there exist BPS saturated
configurations for both the F-string and D-string each
preserving one half of
the spacetime supersymmetries.
In the 3-string junction, we must also conclude that supersymmetry conditions
from any two constituent strings can be independent and the supersymmetry
conditions from the remaining string must be related to those from the other
two strings\footnote{ We know that these are all true from Schwarz's
($m$,$n$)-strings
\cite{schone} and Sen's
analysis of spacetime supersymmetry for 3-string junctions\cite{sen}.}.
So we conclude that there exist a bound state of $n$ D-strings
carrying $m$ units of quantized electric flux based on
D-brane picture, charge
conservation and the linear study discussed in the previous section. We now
know that this bound state is just Schwarz's ($m$,$n$)-string
which provides one way to identify one unit of quantized electric flux or
field line as
an F-string\footnote{Another way of such an interpretation is given in
\cite{polb}. }.
The only thing special for the bound state of
$n$ D-strings carrying
$m$ units of quantized electric flux is the 1 + 1
dimensional electrodynamics which states that the gauge field strength is
constant on one side of a point charge. If we can
consistently have a constant electric field
in a Dp-brane worldvolume, we find no reason that a Dp-brane
carrying certain
units of quantized electric flux should not exist from the above discussion.
To make our arguments for the existence of such Dp-brane bound
states clear, let us first consider a specific $p = 3$ case.
We take $p = 3$ partially
because of the current fashion of $AdS_5/CFT_4$ correspondence and partially
because of the familiarity of the 1 + 3 dimensional electrodynamics. We will
discuss the general cases for $ 1 \le p \le 8$ afterwards.
In the case of D3-brane, we do not have the property of 1 + 1 dimensional
electrodynamics. In general, when an F-string ends on a D3-brane,
the F-string
will be spike-like, not rigid, near the end point. But this will not
prevent us from doing the same as we did above for $p = 1$ case.
As we will see,
insisting a constant electric field in any finite spatial region of
worldvolume in a consistent fashion will automatically push
the endings of F-strings to spatial infinity.
Therefore, the `spike' will appear to be a rigid F-string to any finite
region of space.
The first
question is
what kind of electric charge distribution in 1 + 3 dimensions
gives rise to a
constant electric field\footnote{It happens in this case that we can also
have a bound state of a D3 brane carrying certain units of
quantized constant magnetic field by the Type IIB S-duality. There actually
exist such bound states for $2 \le p \le 8$ \cite{bremm}. For p = 3, we can have
a bound state of a D3 brane carrying both quantized constant electric and
magnetic
fields. We will discuss the p = 3 bound states later in this section. There
actually exist similar and more general bound states which will be
discussed in forthcoming papers\cite{lurthree,lurfour}.}.
We know that a uniform 2-d surface charge distribution will do the job.
The next
question is where this surface should be placed.
When we say a constant electric
field, we mean that the field is
constant not only in magnitude but also in direction in any finite region of
space. So we have to place this charge surface at spatial infinity.
Otherwise,
the direction of the electric field will be opposite on the two sides
of the
surface. For concreteness, let us say that we
label $x^1, x^2$ and $x^3$ as the 3-space
of D3 brane and take the charge surface as $x^2 x^3$-plane and place it at
$x^1 = - \infty$. Now where does the surface charge come from?
It all comes from the
endings of parallel NSNS-strings, say along $x^9$ direction,
on the $x^2x^3$-plane
such that the resulting surface charge density is a constant.
This is possible
because of the existence of the multi-center solution discussed
in the previous
section. Since these NSNS-strings are parallel to each other,
the whole system
is still a BPS one, preserving a quarter of the spacetime supersymmetries.
Note that these endings of F-strings are now at spatial infinity.
Therefore
the `spikes' describing the endings of these F-strings have no influence on
the electric field
in any finite region of space. So everything fits together nicely.
In any finite
region, we can detect only the D3-brane carrying a constant
electric field in
it. By the same token as in $p = 1$ case, we must conclude that
this bound state
preserves also one half of the spacetime supersymmetries.
Because the charge at the end of each of these NSNS-strings is quantized, we
expect that the electric field should also be quantized.
If each of these NSNS-strings is $m$ F-strings, we should have
here $F_{01} = g m T_f$
with $g$ the corresponding string coupling constant.
This can be obtained by
T-dualities from the $F_{01} = g m T_f$ in $p = 1$ case\footnote{To be more
precise, we T-dualize the D3 brane Born-Infeld action with flat background
and non-vanishing constant worldvolume field $F_{01}$ along $x^2$ and $x^3$
directions. We then end up with a D-string Born-Infeld action.
Therefore we can
read $F_{01} = g m T_f$. Noticing the relationship between the exact tension
and linearized tension for $p = 1$ case, we must have the tension for the D3
brane bound state as given in Eq.\ (\ref{eq:btension}) since
the two cases are
related to each other by T-dualities.}.
The discussion for a general $p$ for $ 2 \le p \le 8$ is not
much different
from the $p = 3$ case. To be concrete, let us take the spatial
dimensions of a Dp-brane
along $x^1, \cdots, x^p$. The $(p - 1)$ dimensional surface
with uniform charge
distribution resulting from the endings of parallel
NSNS-strings, say, along
$x^9$-direction is taken as a $(p - 1)$-plane along
$x^2, \cdots, x^p$ directions and
is placed at $x^1 = - \infty$. Then the electric field resulting
from this charge surface
will be constant and along $x^1$-direction in any finite region of space.
It is also
quantized as $F_{01} = g m T_f$ for an NSNS-string (to be thought of as
$m$ F-strings). The rest will
be the same as in the case of $p = 3$. Since $F_{01} = g m T_f$,
we can use the corresponding
Dp-brane action to determine the corresponding tension
$T_p (m,n)$ describing $n$ Dp-branes
carrying $m$ units of quantized constant electric field which is
\begin{equation}
T_p (m,n) = \frac{T^p_0}{g} \sqrt{n^2 + g^2 m^2},
\label{eq:btension}
\end{equation}
where $T^p_0 = 1/(2\pi)^p \alpha'^{(p + 1)/2}$.
This expression clearly indicates that the configuration of
$n$ Dp-branes carrying $m$ units of quantized constant electric field
with $m$ and $n$
relatively prime integers is a non-threshold bound state.
So we conclude that $n$ Dp-branes carrying $m$ units of quantized constant
electric field consist of a BPS non-threshold
bound state which preserves one half of the spacetime supersymmetries.
Since the quantized electric flux or field lines can be interpreted
as F-strings,
these bound states should be identified with the (F, Dp) bound states which
are also related to the ($m$,$n$)-string or (F, D1) by T-dualities along the
transverse directions. But here we must be
careful about the notation `F' in (F, Dp). This F actually represents an
infinite number of parallel NS-strings along, say, $x^1$ direction, which
are distributed evenly over a $(p-1)$-dimensional plane
perpendicular to the $x^1$-axis
(or the strings).
As indicated above, each of these NS-strings is $m$ F-strings
if $F_{01} = g m T_f$. The tension
formula Eq.\ (\ref{eq:btension}) implies that we should have
one NS-string (or $m$ F-strings)
per $(2\pi)^{p-1} \alpha'^{(p - 1)/2}$ area over
the above $(p - 1)$-plane.
Since T-dualities preserve supersymmetries, we can see in a
different way that
these bound states preserve one half of the spacetime supersymmetries
since the
original (F,D1) preserves one half of the spacetime supersymmetries.
We will use this identification and perform T-dualities to construct
explicitly the
spacetime configurations for these bound states in a forthcoming
paper\cite{lurtwo}. We will show there that the tension formula
Eq.\ (\ref{eq:btension}) holds and there are indeed $m$ F-strings per
$(2\pi)^{p -1} \alpha'^{(p -1)/2}$ area of ($p -1$)-dimensions.
The spacetime configurations for (F, Dp) for $p = 3, 4, 6$ have been given in
\cite{rust,grelpt,cosp}, respectively.
Once we have the above, it should not be difficult to have a non-threshold
bound state of $n$ D3 branes carrying $q$ units of quantized
constant magnetic field with $n, q$
relatively prime.
All we need is to replace the F-strings in the above for $p = 3$
case by D-strings.
If we also choose the quantized constant magnetic field along
the $x^1$-axis, we must have
$F_{23} = q T_f$ from the discussion in \cite{calm} about
a D-string ending on a D3
brane. The corresponding tension is
\begin{equation}
T_3 (q, n) = \frac{T_0^3}{g} \sqrt{n^2 + q^2}.
\label{eq:d1d3t}
\end{equation}
This tension formula implies that the linearized approximation
on the D3 brane
worldvolume is good only if $n \gg q$. This bound state
should correspond to the
so-called (D1, D3) bound state. Again, we should have an
infinite number of D-strings
in this bound state and there should be $q$ D-strings per
$(2\pi)^2 \alpha'$ area over
the $x^2x^3$-plane.
Similarly, if we replace the F-strings or D-strings by ($m$,$q$)-strings
in the
above, we should end up with a non-threshold bound state of $n$
D3 branes carrying $m$ units of
quantized electric flux lines and $q$ units of quantized magnetic
flux lines with any two of
the three integers relatively prime. The tension for this bound state is
\begin{equation}
T_3 (m,q, n) = \frac{T^3_0}{g} \sqrt{n^2 + q^2 + g^2 m^2}.
\label{eq:d3dft}
\end{equation}
The linearized approximation on the worldvolume is good if either
$n \gg q, n \gg m$ for
fixed and finite $g$ or $n \gg q$ for small $g$ and finite $m$.
We denote this bound state as
((F, D1),D3). We should also have an infinite number of
($m$,$q$)-strings in this bound state.
We also have one ($m$,$q$)-string per $(2\pi)^2 \alpha'$ area
over the $x^2x^3$-plane.
In \cite{lurthree}, we will construct explicit configuration
for ((F,D1),D3) bound state which gives the (D1, D3) bound state as a special
case. We will confirm all the above mentioned properties for them. The spacetime
configurations for (Dp, D(p + 2)) for $0 \le p \le 4$ have been given in
\cite{bremm,grelpt,cosp}.
Note added: After the submission of this paper to hep-th, we were informed that the
existence of the bound states of Dp branes carrying constant electric fields was also
discussed in \cite{arfs} but in a completely different approach of the mixed boundary
conditions.
\acknowledgments
JXL acknowledges the support of NSF Grant PHY-9722090.
|
\section{. INTRODUCTION}
\end{center}
Quantum theories, particularly quantum gravity in $(2+1)$- dimensions provide
us with a useful field of investigation not only for theoretical and
mathematical issues, but also in some cases, for actual physical problems \cite{Des,Dese}.
In the past decade many interesting physical problems in $(2+1)$-gravity
have been solved, such as classical scattering, quantum scattering, bound states of a
scalar and spinor point particle both in the presence and absence of a magnetic
monopole and also magnetic vortex \cite{De,Ger,Gerb,Jaf}.
In reference \cite{Jaf} some interesting results have been obtained in studying quantum
scattering and bound states of a scalar charged particle in the background
metric corresponding to $(2+1)$- manifold both with local and global constant
curvature in the presence of a magnetic monopole, which satisfy the coupled
Einstein-Maxwell equations.
Here in this article we investigate the quantum mechanics of
a charged spin $\frac{1}{2}$ point particle on a $(2+1)$- spacetime with
spatial part of local constant curvature in the presence of
a constant magnetic field of a magnetic monopole. We show that these
2-dimensional Hamiltonians have the degeneracy group of $SL(2,c)$ type and
para-supersymmetry of arbitrary order or shape
invariance. Using these symmetries we have obtained their spectra algebraically.
Also, the Dirac quantization follows naturally from the representation theory.
In the case of local constancy of the curvature, the presence of angular
deficit suppresses both the degeneracy and shape invariance.
The paper is organised as follows: In section II we briefly describe the $(2+1)$-
spacetime metric of reference \cite{Jaf} and assume that angular deficit is absent.
Section III is devoted to the algorithm of the manipulation
of the Dirac operator in these spacetimes. The Dirac operator has been given
in terms of the generators of the $sl(2,c)$ Lie algebra, which reduces the
familiar Dirac operator on the $S^2$ in special case \cite{Gro}. In section IV
we obtain the left and right invariant generators of the $SL(2,c)$ Lie group
in terms of Euler's angles \cite{Eguchi}, where after eliminating the $\psi$
coordinate we get the eigenspectrum of massless Dirac operator together with its
degeneracy which is in agreement with those of reference \cite{Gro} in special case of
$S^2$. Also, as a special case, we obtain the massless Dirac operator in the presence of a magnetic vortex \cite{Gerb,Thaller}.
In section V we add the constant magnetic field of the magnetic monopole.
Using again the representation of the $SL(2,c)$ Lie group we obtain the
eigenspectrum of a charged spin $\frac{1}{2}$ particle algebraically together with its
degeneracy group. As a special case we obtain the monopole harmonics \cite{Wu,Gros,Gross}
. Also we obtain the familiar Dirac quantization from the
representation theory. In section VI, using the right invariant
generators and eliminating the coordinate $\psi$, we obtain the raising and
lowering operators of the magnetic charge. Using them we show the presence of the
para-supersymmetry of arbitrary order $p$, or equivalently, the shape invariance
symmetry associated with the Dirac operator in the presence of a magnetic
monopole. Finally in section VII we add the angular deficit to the background
spacetime metric which leads to the suppression of both degeneracy and the
shape invariance symmtery. Thus we obtain the eigenspectrum by solving the Dirac
operator by usual method which is in agreement with the result of the
reference \cite{Ger} for special case of the cone.
\begin{center}
\section{. (2+1)- Spacetime with Local Spatial Constant Curvature}
\end{center}
In $(2+1)$ dimensions the Einstein-Hilbert action of gravity coupled to matter
and electromagnetic field, together with the cosmological term can be written as
\cite{Jaf}
\begin{equation}
S=\int d^3x \sqrt{|\det g_{_{\mu\nu}}|} \{ \frac{1}{4 \pi G}(R+2\Lambda) +
\frac{1}{4} F_{\mu\nu}F^{\mu\nu}-{\cal L}_M \},
\end{equation}
where $\cal L_{M}$ is the matter Lagrangian corresponding to a very
massive point particle with mass $M$ located in the origin. We have rescaled G by a
factor of $4$.
As it is shown in reference \cite{Jaf} the following $(2+1)$- spacetime metric
corresponds to a massive point mass $M$ together with nonnegative cosmological
constant $\Lambda$ and magnetic monopole field
\begin{equation}
ds^{2}=dt^{2}-\frac{1}{2 \lambda}(d \theta^{2}+(1-GM)^2
\frac{\sin^{2}\alpha\theta}{\alpha^{2}}d\phi^{2}).
\end{equation}
where $\alpha$ and $\lambda$ satisfy the following relation
\begin{equation}
\Lambda=\alpha^2 \lambda,
\end{equation}
and $GM$ satisfies the condition $GM<1$.
The parameter $\alpha$ in equation (2.3) chooses one of the values $0,1,i$.
In the case of $\alpha=1$, $\lambda$ is a positive real number and we have
$$
0 \leq \theta < \pi,
$$
$$
0 \leq \phi < 2\pi.
$$
For $\alpha=i$, $\lambda$ is a negative real number and $(2+1)$- spacetime metric
(2.2) is Euclidean. In this case we have $0\leq \theta < \infty$ and the range
of $\phi$ is the same as $\alpha=1$ case. Finally for $\alpha=0$, $\lambda$ is
a positive real number again and $\theta$ plays the role of the radial variable
\cite{Jaf} and the range of the coordinate $\phi$ is the same as $\alpha=1,i$
case.
The magnetic monopole field corresponding to the system (2.1) is
\begin{equation}
{\cal B}=g(1-GM)\alpha\sin\alpha\theta,
\end{equation}
with
\begin{eqnarray}
g=\left \{\begin{array}{ll}
\frac{1}{2\sqrt{\pi G\lambda}} & \mbox{if $\alpha=0$ and $1$} \\
\frac{-1}{2\sqrt{\pi G|\lambda|}} & \mbox{if $\alpha=i$}\end{array} \right.
\end{eqnarray}
which extremizes the Einstein-Hilbert action given in (2.1). In other words they are the solution of
the Einstein-Maxwell equation which extremizes this action.
The magnetic field given in (2.4) is the magnetic field of a magnetic monople
located in the origin of the $R^3$ Euclidean space where the constant
curvature spatial part of the spacetime is embedded in it. The corresponding
magnetic potential one-form $A$ in the coordinates $\theta$ and $\phi$ is
\begin{equation}
A=-g(1-GM)\cos \alpha\theta d\phi.
\end{equation}
In the next section after introducing the abstract Dirac operator we write
it on the spatial part of metric (2.2) in the case $M=0$.
\vspace{1cm}
\begin{center}
\section{. Dirac Operator on two dimensional spaces with global Constant
Curvature}
\end{center}
\setcounter{equation}{0}
The massless Dirac operator on a given d-dimensional Riemannian manifold with metric
$g_{_{\mu\nu}}$ can be written as \cite{Gro,Eguchi}:
\begin{equation}
D=-i\gamma^{a}E_{_{a}}^{\; \mu}(\partial_{_{\mu}}+\frac{1}{8}\omega_{_{\mu ab}}
[\gamma^{a},\gamma^{b}]),
\end{equation}
where $\gamma^a,s$ are the generators of the flat Clifford algebra which satisfy
the following anticommutation relation:
\begin{equation}
\{\gamma^{a},\gamma^{b}\}=2\delta^{ab} \hspace{3cm} a,b=1, \cdots ,d.
\end{equation}
Also $E_{_{a}}^{\; \mu}$ , $\omega_{_{\mu ab}}$ are d-beins and spin connections,
respectively, which satisfy the following relations:
\begin{eqnarray}
\nonumber
&&E_{_{a}}^{\; \mu}g_{_{\mu\nu}}E_{_{b}}^{\; \nu}=\delta_{_{ab}} \\
\nonumber
&&E_{_{a}}^{\; \mu}e_{_{\mu}}^{\; b}=\delta_{_{a}}^{\; b} \\
&&\partial_{_{\mu}}e_{_{\nu}}^{\; a}-\Gamma_{_{\mu\nu}}^{\lambda}
e_{_{\lambda}}^{\; a}+\omega_{_{\mu ab}}e_{_{\nu}}^{\; b}=0.
\end{eqnarray}
Here in this article we are concerned with the manifolds described by metric
(2.2) and gauge potential (2.6). We also assume that $M$ vanishes in the rest of the article
except in the last section.
Then the spatial part of the metric (2.2) reads:
\begin{equation}
ds_{_{(2)}}^{2}=\frac{1}{2 \lambda}(d \theta^{2}+\frac{\sin^{2}\alpha\theta}
{\alpha^{2}}d\phi^{2}),
\end{equation}
it is clear form the above metric that the spatial part consits of a
2-dimensional manifold of global constant curvature. Using the equations (3.3) we
obtain the following expression for the nonvanishing components of zwei-beins
and spin connections associated with metric (3.4):
\begin{eqnarray}
\nonumber
&&E_{_{1}}^{\; \theta}=\sqrt{2\lambda},
E_{_{2}}^{\; \phi}=\sqrt{2\lambda}\frac{\alpha}{\sin\alpha\theta} \\
&&\omega_{_{\phi 12}}=-\cos\alpha\theta.
\end{eqnarray}
By using the equation (3.1), we obtain the Dirac operator on a manifold described
by metric (3.4) as follows:
\begin{equation}
D_2=-i\sqrt{2\lambda}\gamma^{1}(\partial_{_{\theta}}+\frac{1}{2}\frac{\alpha}
{\tan\alpha\theta})-i\sqrt{2\lambda}\gamma^{2}\frac{\alpha}{\sin\alpha\theta}
\partial_{_{\phi}}.
\end{equation}
For $\alpha=1$, $D_2$ becomes the Dirac operator on the 2-dimensional sphere $S^2$
\cite{Gro}.
It is more convenient to consider the 2-dimensional manifold (3.4) as a
submanifold of the 3-dimensional manifold $M_3$ with the line element
\begin{equation}
ds_{_{(3)}}^{2}=dr^{2}+\alpha^2 r^2(d \theta^{2}+\frac{\sin^{2}\alpha\theta}
{\alpha^{2}}d\phi^{2}),
\end{equation}
which is parametrized with the following local coordinates:
\begin{eqnarray}
\nonumber
&&x_{_{1}}=r\sin\alpha\theta \cos\phi \\
\nonumber
&&x_{_{2}}=r\sin\alpha\theta \sin\phi \\
&&x_{_{3}}=r\cos\alpha\theta .
\end{eqnarray}
Now we consider $r=$ constant submanifold of $M_3$ with the following metric:
\begin{equation}
ds_{_{(2)}}^{2}=\alpha^2 r^2(d \theta^{2}+\frac{\sin^{2}\alpha\theta}
{\alpha^{2}}d\phi^{2}).
\end{equation}
For $\alpha=1$ and $r=\frac{1}{\sqrt{2\lambda}}>0$ this submanifold coincides
with the special case of $\alpha=1$ of two dimensional manifold with metric
(3.4), while for $\alpha=0$, with the assumption of $\lim_{_{\alpha\rightarrow
o,r\rightarrow \infty}}\alpha r=finite=\frac{1}{\sqrt{2\lambda}}$, it is
the same as $\alpha=0$ case of metric (3.4). Finally for $\alpha=i$, with the
assumption $r=\frac{1}{\sqrt{2|\lambda|}}>0$ it changes to $\alpha=i$
case of metric (3.4).
Now we try to define the Dirac operator on the manifold $M_3$. Using the
equations (3.3), for nonvanishing components of 3-beins and spin connections
we get
\begin{eqnarray}
\nonumber
&&E_{_{1}}^{\; \theta}=\frac{1}{\alpha r},
E_{_{2}}^{\; \phi}=\frac{1}{r\sin\alpha\theta},
E_{_{3}}^{\; r}=1 \\
&&\omega_{_{\theta 13}}=\alpha, \omega_{_{\phi 21}}=\cos\alpha
\theta, \omega_{_{\phi 23}}=\sin\alpha\theta.
\end{eqnarray}
Finally, using the relations (3.1) and (3.10) the Dirac operator associated with
metric (3.7) reads
\begin{equation}
D_3=-i\gamma^{1}\frac{1}{\alpha r}(\partial_{_{\theta}}+\frac{1}{2}\frac{\alpha}
{\tan\alpha\theta})-i\gamma^{2}\frac{1}{\alpha r}\frac{\alpha}{\sin\alpha\theta}
\partial_{_{\phi}}-i\gamma^{3}(\partial_{_{r}}+\frac{1}{r}).
\end{equation}
It is straightforward to see that the Dirac operator $D_3$ yields
\begin{eqnarray}
\nonumber
&&(-i\gamma^{3}D_3 + \frac{1}{r})\mid_{r=constant}= \\
&&-i\Gamma^{1}\frac{1}{\alpha r}(\partial_{_{\theta}}+\frac{1}{2}\frac{\alpha}
{\tan\alpha\theta})-i\Gamma^{2}\frac{1}{\alpha r}\frac{\alpha}{\sin\alpha\theta}
\partial_{_{\phi}},
\end{eqnarray}
with $\Gamma^a$ defined as
$$
\Gamma^{a}=-i\gamma^{3}\gamma^{a}, \hspace{3cm} a=1,2
$$
which satisfy the following Clifford algebra
\begin{equation}
\{\Gamma^{a},\Gamma^{b}\}=2\delta^{ab}.
\end{equation}
Assuming the equivalence of the metric of submanifold given in (3.9) with the
two dimensional metric (3.4) and also replacing $\alpha r$ with $\frac{1}{
\sqrt{2\lambda}}$, we can deduce that the operator (3.12) is the same as Dirac
operator $D_2$ given in (3.6).
In terms of local coordinates (3.8) the operator $D_3$ can be written as
\begin{equation}
D_3=-i\sigma_{_{i}}\partial_{_{i}},
\end{equation}
where $\sigma_{_{i}},s$ are Pauli matrices.
Using the identity $(\frac{\sigma_{_{i}}x_{_{i}}}{r})^2=I$, we have
\begin{equation}
D_3=(\frac{\sigma_{_{i}}x_{_{i}}}{r})^2 D_3=-i\frac{\sigma_{_{i}}x_{_{i}}}{r}
(\partial_{_{r}}+\frac{i}{r}\epsilon_{_{ijk}}\sigma_{_{i}}x_{_{j}}
\partial_{_{k}}).
\end{equation}
Now, comparing the operator (3.15) with (3.11), it follows that the operator $\gamma^3$
has the following form
\begin{equation}
\gamma^{3}=\frac{\sigma_{_{i}}x_{_{i}}}{r}.
\end{equation}
Using the relations (3.12) and (3.15) together with the relation (3.11) the Dirac
operator $D_2$ over two dimensional manifold with metric (3.4) can be
represented as
\begin{equation}
D_2=\frac{1}{r}-\frac{i}{r}\epsilon_{_{ijk}}\sigma_{_{i}}x_{_{j}}
\partial_{_{k}}=\sqrt{2\lambda}(\sigma_{_{1}}I_{_{1}}+\sigma_{_{2}}I_{_{2}}
+\alpha \sigma_{_{3}}I_{_{3}}+\alpha I),
\end{equation}
where I is a $2\times 2$ identity matrix and the differential operators $I_{_{i}}$ with $i=1,2,3$ in (3.17) have
the following form:
\begin{eqnarray}
\nonumber
&& I_{_{1}}=i(\sin\phi \partial_{_{ \theta}}+
\frac{\alpha}{\tan \alpha \theta}\cos\phi\partial_{_{\phi}}) \\
\nonumber
&& I_{_{2}}=i(-\cos\phi \partial_{_{\theta}}+
\frac{\alpha}{\tan \alpha \theta}\sin\phi\partial_{_{\phi}}) \\
&&I_{_{3}}=-i\partial_{_{\phi}},
\end{eqnarray}
and satisfy the following $sl(2,c)$ Lie algebra
\begin{eqnarray}
\nonumber
&&[I_{_{1}},I_{_{2}}]=i\alpha^2 I_{_{3}} \\
\nonumber
&&[I_{_{2}},I_{_{3}}]=iI_{_{1}} \\
&&[I_{_{3}},I_{_{1}}]=iI_{_{2}}.
\end{eqnarray}
It is clear that for $\alpha=1$ this algebra becomes an $su(2)$ Lie algebra,
for $\alpha=i$ we get $su(1,1)$ Lie algebra, and finally for $\alpha=0$
we get $iso(2)$ Lie algebra \cite{Jaf,Jafa}.
\begin{center}
\section{. Degeneracy group of the Dirac operator on two dimensional manifolds
with global constant curvature}
\end{center}
\setcounter{equation}{0}
In order to obtain the degeneracy group of the Dirac operator $D_2$ over two
dimensional manifold with metric (3.4) we need to know the left and right
invariant generators of $SL(2,c)$ group which have the following form in the
Eulerean coordinates \cite{Jaf}
\begin{eqnarray}
\nonumber
&&L_{_{1}}^{(L)}=i(\sin\phi\partial_{_{\theta}}+
\frac{\alpha}{\tan\alpha\theta}\cos\phi\partial_{_{\phi}}-
\frac{\alpha}{\sin\alpha\theta}\cos\phi\partial_{_{\psi}}) \\
\nonumber
&&L_{_{2}}^{(L)}=i(-\cos\phi\partial_{_{\theta}}+
\frac{\alpha}{\tan\alpha\theta}\sin\phi\partial_{_{\phi}}-
\frac{\alpha}{\sin\alpha\theta}\sin\phi\partial_{_{\psi}}) \\
\nonumber
&&L_{_{3}}^{(L)}=-i\partial_{_{\phi}}, \\
\nonumber
&&L_{_{1}}^{(R)}=i(\sin\psi\partial_{_{\theta}}+
\frac{\alpha}{\tan\alpha\theta}\cos\psi\partial_{_{\psi}}-
\frac{\alpha}{\sin\alpha\theta}\cos\psi\partial_{_{\phi}}) \\
\nonumber
&&L_{_{2}}^{(R)}=i(-\cos\psi\partial_{_{\theta}}+
\frac{\alpha}{\tan\alpha\theta}\sin\psi\partial_{_{\psi}}-
\frac{\alpha}{\sin\alpha\theta}\sin\psi\partial_{_{\phi}}) \\
&&L_{_{3}}^{(R)}=-i\partial_{_{\psi}},
\end{eqnarray}
where $0\leq\phi <2\pi$, $0\leq\psi <4\pi$ and $0\leq\theta < \pi$ for $\alpha=
1$, while $0\leq\theta<\infty$ when $\alpha=0,i$.
It is rather well known that both left and right invariant generators satisfy
$sl(2,c)$ Lie algebra denoted by $sl(2,c)_{_{L}}$ and $sl(2,c)_{_{R}}$,
respectively and also they commute with each other. That is we have
\begin{eqnarray}
\nonumber
&&[L_{_{1}}^{(L)},L_{_{2}}^{(L)}]=i\alpha^2 L_{_{3}}^{(L)} \hspace{15mm}
[L_{_{2}}^{(L)},L_{_{3}}^{(L)}]=iL_{_{1}}^{(L)} \hspace{15mm}
[L_{_{3}}^{(L)},L_{_{1}}^{(L)}]=iL_{_{2}}^{(L)}, \\
\nonumber
&&[L_{_{1}}^{(R)},L_{_{2}}^{(R)}]=i\alpha^2 L_{_{3}}^{(R)} \hspace{15mm}
[L_{_{2}}^{(R)},L_{_{3}}^{(R)}]=iL_{_{1}}^{(R)} \hspace{15mm}
[L_{_{3}}^{(R)},L_{_{1}}^{(R)}]=iL_{_{2}}^{(R)}, \\
&&[\vec{L^{(L)}},\vec{L^{(R)}}]=0.
\end{eqnarray}
Now, using the generators (4.1) we define the following new bases:
\begin{eqnarray}
\nonumber
&&K_{_{1}}^{(L)}:=L_{_{1}}^{(L)}\otimes I+\frac{1}{2}\alpha\sigma_{_{1}}
\hspace{12mm}
K_{_{2}}^{(L)}:=L_{_{2}}^{(L)}\otimes I+\frac{1}{2}\alpha\sigma_{_{2}}
\hspace{12mm}
K_{_{3}}^{(L)}:=L_{_{3}}^{(L)}\otimes I+\frac{1}{2}\sigma_{_{3}}, \\
&&K_{_{1}}^{(R)}:=L_{_{1}}^{(R)}\otimes I
\hspace{25mm}
K_{_{2}}^{(R)}:=L_{_{2}}^{(R)}\otimes I
\hspace{25mm}
K_{_{3}}^{(R)}:=L_{_{3}}^{(R)}\otimes I.
\end{eqnarray}
Using the commutation relations (4.2) and the properties of Pauli matrices it is
rather straightforward to show that the newly defined left and right invariant
operators (4.3) also satisfy $sl(2,c)$ Lie algebra separately and commute
with each other.
Now, the operator $F$ defined as
\begin{equation}
F:=\sqrt{2\lambda}(\sigma_{_{1}}L_{_{1}}^{(L)}+\sigma_{_{2}}L_{_{2}}^{(L)}
+\alpha\sigma_{_{3}}L_{_{3}}^{(L)}+\alpha I),
\end{equation}
commute with all the generators given in (4.3), that is
\begin{eqnarray}
\nonumber
&&[F,\vec{K^{(L)}}]=0, \\
&&[F,\vec{K^{(R)}}]=0.
\end{eqnarray}
Therefore, in order to obtain the eigenspectrum of operator $F$ , we need the
set of commutative operators expressed in terms of operators (4.3),which are
$$
\{K_{_{3}}^{(R)},K_{_{3}}^{(L)},K_{_{1}}^{(L)\;2}+K_{_{2}}^{(L)\;2}+
\alpha^2 K_{_{3}}^{(L)\;2},K_{_{1}}^{(R)\;2}+K_{_{2}}^{(R)\;2}+
\alpha^2 K_{_{3}}^{(R)\;2}\}.
$$
Then, we have the following simultaneous eigenvalue equations
\begin{eqnarray}
\nonumber
&&K_{_{3}}^{(R)}\Psi=q\Psi \\
\nonumber
&&K_{_{3}}^{(L)}\Psi=m\Psi \\
\nonumber
&&(K_{_{1}}^{(R) \;2}+K_{_{2}}^{(R) \;2}+\alpha^2K_{_{3}}^{(R) \;2})\Psi=
\alpha^2 l(l+1)\Psi \\
&&(K_{_{1}}^{(L) \;2}+K_{_{2}}^{(L) \;2}+\alpha^2K_{_{3}}^{(L) \;2})\Psi=
\alpha^2 j(j+1)\Psi.
\end{eqnarray}
Obviously the operators $K_{_{3}}^{(R)}$ and $K_{_{3}}^{(L)}$ have the following
differential form
\begin{eqnarray}
\nonumber
&&K_{_{3}}^{(R)}=\left(\begin{array}{cc}
-i\partial_{_{\psi}} & 0 \\
o & -i\partial_{_{\psi}}
\end{array}\right), \\
&&K_{_{3}}^{(L)}=\left(\begin{array}{cc}
-i\partial_{_{\phi}}+\frac{1}{2} & 0 \\
o & -i\partial_{_{\phi}}-\frac{1}{2}
\end{array}\right).
\end{eqnarray}
Therefore, the two component spinor $\Psi$ reads
\begin{equation}
\Psi=\left(\begin{array}{c}
ae^{i(m-\frac{1}{2})\phi+iq\psi} f_{_{1}}(\theta) \\
be^{i(m+\frac{1}{2})\phi+iq\psi} f_{_{2}}(\theta)
\end{array}\right).
\end{equation}
In equation (4.8) $a$ and $b$ are constants and $f_{_{1}}$ and $f_{_{2}}$ are
functions of variable $\theta$.
In order the spinor $\Psi$ to become a periodic function of $\phi$ with period
$2\pi$ , the quantum number $m$ must be a half integer number.
Now, using the left and right invariant generators (4.1) we have:
\begin{eqnarray}
\nonumber
&&L_{_{1}}^{(L) \;2}+L_{_{2}}^{(L) \;2}+\alpha^2 L_{_{3}}^{(L) \;2}=
L_{_{1}}^{(R) \;2}+L_{_{2}}^{(R) \;2}+\alpha^2 L_{_{3}}^{(R) \;2}= \\
&&-\{\partial_{_{\theta}}^{2}+
\frac{\alpha}{\tan\alpha\theta}\partial_{_{\theta}}+
\frac{\alpha^{2}}{\sin^{2} \alpha\theta}(\partial_{_{\phi}}^{2}+
\partial_{_{\psi}}^{2}-2\cos\alpha\theta\partial_{_{\phi}}\partial_{_{\psi}})\}.
\end{eqnarray}
The operator (4.9) yields the following eigenvalue equation \cite{Vilenkin}:
\begin{eqnarray}
\nonumber
&&-\{\partial_{_{\theta}}^{2}+
\frac{\alpha}{\tan\alpha\theta}\partial_{_{\theta}}+
\frac{\alpha^{2}}{\sin^{2} \alpha\theta}(\partial_{_{\phi}}^{2}+
\partial_{_{\psi}}^{2}-2\cos\alpha\theta\partial_{_{\phi}}\partial_{_{\psi}})\}
Y_{_{nq}}^{l}(\theta ,\phi ,\psi)= \\
&& \alpha^2 l(l+1)Y_{_{nq}}^{l}(\theta ,\phi ,\psi),
\end{eqnarray}
where the eigenfunction (4.10), that is $Y_{_{nq}}^{l}(\theta,\phi,\psi)$
reads
\begin{equation}
Y_{_{nq}}^{l}(\theta ,\phi ,\psi)=e^{in\phi +iq\psi}P_{_{nq}}^{l}
(\cos\alpha\theta).
\end{equation}
On the other hand, the operators
$K_{_{1}}^{(L)\;2}+K_{_{2}}^{(L)\;2}+\alpha^2 K_{_{3}}^{(L)\;2}$ and
$K_{_{1}}^{(R)\;2}+K_{_{2}}^{(R)\;2}+\alpha^2 K_{_{3}}^{(R)\;2}$ may be
represented in the form
$$
K_{_{1}}^{(R) \;2}+K_{_{2}}^{(R) \;2}+\alpha^2K_{_{3}}^{(R) \;2}=
\left(\begin{array}{cc}
L_{_{1}}^{(R) \;2}+L_{_{2}}^{(R) \;2}+\alpha^2L_{_{3}}^{(R) \;2} & 0 \\
0 & L_{_{1}}^{(R) \;2}+L_{_{2}}^{(R) \;2}+\alpha^2L_{_{3}}^{(R) \;2}
\end{array}\right),
$$
$$
\hspace{-100mm}K_{_{1}}^{(L) \;2}+K_{_{2}}^{(L) \;2}+\alpha^2K_{_{3}}^{(L) \;2}=
$$
\begin{equation}
\left(\begin{array}{cc}
L_{_{1}}^{(L) \;2}+L_{_{2}}^{(L) \;2}+\alpha^2L_{_{3}}^{(L) \;2}+\frac{3}{4}
\alpha^2 +\alpha^2 L_{_{3}}^{(L)} & \alpha(L_{_{1}}^{(L)}-iL_{_{2}}^{(L)}) \\
\alpha(L_{_{1}}^{(L)}+iL_{_{2}}^{(L)}) &
L_{_{1}}^{(L) \;2}+L_{_{2}}^{(L) \;2}+\alpha^2L_{_{3}}^{(L) \;2}+\frac{3}{4}
\alpha^2 -\alpha^2 L_{_{3}}^{(L)}
\end{array}\right),
\end{equation}
where the operators $L_{_{1}}^{(L)}+iL_{_{2}}^{(L)}$ and
$L_{_{1}}^{(L)}-iL_{_{2}}^{(L)}$ are the raising and lowering operators of index
$n$ of eigenfunction (4.11), that is we have
$$
(L_{_{1}}^{(L)}+iL_{_{2}}^{(L)})Y_{_{nq}}^{l}(\theta ,\phi ,\psi)=
\sqrt{\alpha^2 (l+n+1)(l-n)}Y_{_{n+1q}}^{l}(\theta ,\phi ,\psi),
$$
$$
(L_{_{1}}^{(L)}-iL_{_{2}}^{(L)})Y_{_{nq}}^{l}(\theta ,\phi ,\psi)=
\sqrt{\alpha^2 (l-n+1)(l+n)}Y_{_{n-1q}}^{l}(\theta ,\phi ,\psi).
$$
Summarizing the above explanation the eigenfunction $\Psi$ of eigenvalue
equations (4.6) reads
\begin{equation}
\Psi=\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)=\Omega\left(\begin
{array}{c}
\pm \sqrt{l\pm m+\frac{1}{2}}Y_{_{m-\frac{1}{2}q}}^{l}(\theta ,\phi ,\psi) \\
\sqrt{l\mp m+\frac{1}{2}}Y_{_{m+\frac{1}{2}q}}^{l}(\theta ,\phi ,\psi)
\end{array}\right),
\end{equation}
where $\Omega$ is constant of normalization.
Note that in equation (4.13) $q$ takes integer values because $m$ can take
on only half integer values as said before \cite{Vilenkin}.
Now, by taking the operator $F$ of eq (4.4) to the power two we arrive at:
$$
F^2=2\lambda(K_{_{1}}^{(L) \; 2}+K_{_{2}}^{(L) \; 2}+\alpha^2K_{_{3}}^{(L)
\; 2}+\frac{1}{4}\alpha^2 I)=2\lambda(j+\frac{1}{2})^2 \alpha^2 I.
$$
Therefore, we have the following eigenvalue equation
\begin{equation}
F\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)=\pm \sqrt{
2\lambda(j+\frac{1}{2})^2 \alpha^2}\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi).
\end{equation}
Now, transfering the factor $e^{iq\psi}$ which appears in the wavefunction (4.13)
to the left of the operator $F$ and eliminating it from both sides of
equation (4.14), we get:
\begin{equation}
F(q)\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi)=\pm \sqrt{
2\lambda(j+\frac{1}{2})^2 \alpha^2}\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi),
\end{equation}
where $F(q)$ and $\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta,\phi)$ are:
\begin{eqnarray}
\nonumber
&&F(q)=\sqrt{2\lambda}\left(\begin{array}{cc}
\alpha(1-i\partial_{_{\phi}}) & e^{-i\phi}(-\partial_{_{\theta}}+i
\frac{\alpha}{\tan\alpha\theta}\partial_{_{\phi}}+q\frac{\alpha}{\sin\alpha\theta}) \\
e^{i\phi}(\partial_{_{\theta}}+i
\frac{\alpha}{\tan\alpha\theta}\partial_{_{\phi}}+q\frac{\alpha}{\sin\alpha\theta}) &
\alpha(1+i\partial_{_{\phi}})
\end{array}\right), \\
&&\Psi_{_{l,j=l\pm\frac{1}{2},m,q}}(\theta,\phi)=\Omega \left(\begin{array}{c}
\pm\sqrt{l\pm m+\frac{1}{2}}e^{i(m-\frac{1}{2})\phi}P_{_{m-\frac{1}{2} q}}
^{l}(\cos\alpha\theta) \\
\sqrt{l\mp m+\frac{1}{2}}e^{i(m+\frac{1}{2})\phi}P_{_{m+\frac{1}{2} q}}
^{l}(\cos\alpha\theta)
\end{array}\right).
\end{eqnarray}
Now we consider the limiting case of $q\rightarrow 0$. In this limit the operator
$F(q)$ becomes the same as the operator $D_2$ in (3.17), that is we have
$$
\lim_{_{q\rightarrow o}}F(q)=D_2.
$$
Specially for $\alpha=1$ we get the Dirac operator on $S^2$ \cite{Gro}, and the
eigenfunction introduced in (4.16) becomes the wellknown spinor
harmonics \cite{Wu}.
There is another interesting limiting case: to let $\alpha
\rightarrow 0$ and $l\rightarrow\infty$ but $\alpha l$ to remain finite,
that is
$$
\lim_{_{\alpha\rightarrow 0,l\rightarrow\infty}} \alpha l=k.
$$
In this limit $\theta$ plays the role of radial coordinate and we have
\cite{Vilenkin}
$$
\lim_{_{\alpha\rightarrow 0,l\rightarrow\infty}}P_{_{nq}}^{l}(\cos\alpha
\theta)=J_{_{|n-q|}}(kr),
$$
where $J_{_{|n-q|}}(kr)$ is the Bessel function with index $|n-q|$.
In brief, we have
\begin{equation}
Z_{_{k,m,q}}(r,\phi)=\lim_{_{\alpha\rightarrow 0,l\rightarrow\infty}}
\Psi_{_{l,j=l\pm\frac{1}{2},m,q}}(\theta,\phi)=\Omega^{\prime}
\left(\begin{array}{c}
\pm e^{i(m-\frac{1}{2})\phi}J_{_{|m-\frac{1}{2}-q|}}(kr) \\
e^{i(m+\frac{1}{2})\phi}J_{_{|m+\frac{1}{2}-q|}}(kr)
\end{array}\right),
\end{equation}
where $\Omega^{\prime}$ is the new constant of normalization.
The operator $F(q)$, in the limit of $\alpha\rightarrow 0$ reads
\begin{equation}
\lim_{_{\alpha\rightarrow 0}}F(q)=\sqrt{2\lambda}\left(\begin{array}{cc}
0 & e^{-i\phi}(-\partial_{_{r}}+\frac{i}{r}\frac{\partial}{
\partial \phi}+\frac{q}{r}) \\
e^{i\phi}(\frac{\partial}{\partial r}+\frac{i}{r}\frac{\partial}{
\partial \phi}+\frac{q}{r}) & 0
\end{array}\right).
\end{equation}
In this limit the operator $F(q)$ has the following eigenvalue
$$
E=\pm \sqrt{2\lambda}k.
$$
The operator (4.18) is exactly the Dirac operator of a very light spin
$\frac{1}{2}$ particle in the presence of magnetic vortex with gauge potential
$\vec{A}=e_{_{\phi}}\frac{q}{r}$ \cite{Thaller}. The wavefunction (4.17) is
the eigenstate associated with the scattering of a massless fermion from a vortex.
it is obvious that in the case $q=0$, (4.17) and (4.18) represent the wavefunction
and the Dirac operator of a free massless fermion on two dimensional flat space respectively.
In the next section we obtain the eigenspectrum of the Dirac operator in the
presence of the magnetic monopole (2.4) for the special case of $M=0$.
\begin{center}
\section{. The Dirac operator on a 2-dimensional manifold with global constant
curvature in the presence of magnetic field of a magnetic monopole}
\end{center}
\setcounter{equation}{0}
The massless Dirac operator on a Riemannian manifold with metric $g_{_{\mu\nu}}$ in the presence
of gauge field $A_{_{\mu}}$ is \cite{Nakahara}
\begin{equation}
D(A)=-i\gamma^{a}E_{_{a}}^{\; \mu}(\partial_{_{\mu}}+\frac{1}{8}\omega_{_{\mu ab}}
[\gamma^{a},\gamma^{b}]+ieA_{_{\mu}}).
\end{equation}
Therefore, using the beins and spin connections given in (3.5) and considering
the gauge potential $A=-g\cos\alpha\theta d\phi$ we obtain the following
expression for the Dirac operator
\begin{equation}
D_2(A)=-i\sqrt{2\lambda}\gamma^{1}(\partial_{_{\theta}}+\frac{1}{2}\frac{\alpha}
{\tan\alpha\theta})-i\sqrt{2\lambda}\gamma^{2}\frac{\alpha}{\sin\alpha\theta}
(\partial_{_{\phi}}-ieg\cos\alpha\theta).
\end{equation}
Now we try to obtain the Dirac operator $D_2(A)$ given in (5.2) from the Dirac
operator on the manifold described by the metric (3.7) and by the gauge field with
connection
\begin{eqnarray}
\nonumber
&&A_{_{r}}=0 \\
\nonumber
&&A_{_{\theta}}=0 \\
&&A_{_{\phi}}=-g\cos\alpha\theta.
\end{eqnarray}
Using the beins and spin connections given in (3.10) and gauge field connection
(5.3), the Dirac operator on the manifold (3.7) in the presence of gauge field
(5.3) reads:
\begin{equation}
D_3(A)=-i\gamma^{1}\frac{1}{\alpha r}(\partial_{_{\theta}}+\frac{1}{2}\frac{\alpha}
{\tan\alpha\theta})-i\gamma^{2}\frac{1}{\alpha r}\frac{\alpha}{\sin\alpha\theta}
(\partial_{_{\phi}}-ieg\cos\alpha\theta)-i\gamma^{3}(\partial_{_{r}}+\frac{1}{r}).
\end{equation}
It is straightforward to show that the following relation between the operators
$D_2(A)$ and $D_3(A)$ holds:
\begin{equation}
D_2(A)=(-i\gamma^{3}D_3(A)+ \frac{1}{r})\mid_{r=constant}.
\end{equation}
The gauge field (5.3) has the following form in the cartesian-like coordinates
(3.8)
\begin{equation}
A_{_{i}}=g\epsilon_{_{ij3}} \frac{x_{_{j}}x_{_{3}}}{r(x_{_{1}}^2+x_{_{2}}^2)},
\hspace{2cm} i,j=1,2,3,
\end{equation}
where it satisfies the following gauge condition
$$
\vec{r}.\vec{A}=0.
$$
In the local coordinates (3.8) together with the gauge connection (5.6), the Dirac operator
can be written as
\begin{equation}
D_3(A)=\sigma_{_{i}}(\frac{1}{i}\partial_{_{i}}+eA_{_{i}}).
\end{equation}
With further little algebra one can show that the Dirac operator given in
(5.7) takes the following form:
\begin{equation}
D_3(A)=(\frac{\sigma_{_{i}}x_{_{i}}}{r})^2D_3(A)=-i\gamma^{3}(\partial_{_{r}}-
\frac{1}{r}\vec{\sigma}.\vec{r}\times(\frac{1}{i}\vec{\nabla}+e\vec{A})).
\end{equation}
Finally using the relation (5.5) between the operators $D_2(A)$ and $D_3(A)$,
the Dirac operator on the 2-dimensional manifold (3.4) and in the presence of
magnetic field of a magnetic monopole is
\begin{equation}
D_2(A)=F(eg)-\alpha\sqrt{2\lambda}eg\gamma^{3}.
\end{equation}
Therefore, according to section 4, we have the following eigenvalue equation
\begin{equation}
D_2(A)\Psi_{_{l,j=l\pm\frac{1}{2},m,q}}(\theta ,\phi)=\pm\sqrt{2\lambda
\alpha^2 [(j+\frac{1}{2})^2 -q^2]}\Psi_{_{l,j=l\pm\frac{1}{2},m,q}}(\theta ,\phi),
\end{equation}
where $q$ is equal to the product of electric and magnetic charge, that is
$$
q=eg.
$$
Therefore, the Dirac quantization condition follows naturally from the finite
representation of $SL(2,c)$ Lie group.
Also $j+\frac{1}{2}\geq q$ and for $j+\frac{1}{2}=q$ the operator (5.9)
becomes noninvertible.
It is clear that for $\alpha=1$, the operator $D_2(A)$ becomes the Dirac operator
on $S^2$ in the presence of magnetic field of magnetic monopole \cite{Gros,Gross}
with monopole harmonics as its eigenfunctions.
\begin{center}
\section{. Para-supersymmetry and shape invariance of Dirac equation}
\end{center}
\setcounter{equation}{0}
In this section using the left and right invariant generators introduced in
section 4, we try to investigate the shape invariance symmetry and
para-supersymmetry of 2-dimensional Dirac operator.
Here it is more convenient to work with bases $\{J_{_{+}}^{(R)},J_{_{-}}^{(R)},J_{_{3}}^{(R)}\}$ rather than with $\{L_{_{1}}^{(R)},L_{_{2}}^{(R)},
L_{_{3}}^{(R)}\}$ which are defined as:
\begin{eqnarray}
\nonumber
&&J_{_{\pm}}^{(R)}=L_{_{1}}^{(R)}\pm iL_{_{2}}^{(R)}=
e^{\pm i\psi}(\pm \partial_{_{\theta}}+i\frac{\alpha}{\tan\alpha\theta}
\partial_{_{\psi}}-i\frac{\alpha}{\sin\alpha\theta}
\partial_{_{\phi}}), \\
&&J_{_{3}}^{(R)}=L_{_{3}}^{(R)}=-i\partial_{_{\psi}}.
\end{eqnarray}
Clearly these new bases have the following commutation relations
\begin{eqnarray}
\nonumber
&&[J_{_{+}}^{(R)},J_{_{-}}^{(R)}]=2\alpha^2 J_{_{3}}^{(R)}, \\
&&[J_{_{3}}^{(R)},J_{_{\pm}}^{(R)}]=\pm J_{_{\pm}}^{(R)}.
\end{eqnarray}
Using the relations (4.9) and (4.10) we arrive at:
\begin{eqnarray}
\nonumber
&&(J_{_{+}}^{(R)}\otimes I)(J_{_{-}}^{(R)}\otimes I)
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)=\alpha^2(l-q+1)(l+q)
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi), \\
&&(J_{_{-}}^{(R)}\otimes I)(J_{_{+}}^{(R)}\otimes I)
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)=\alpha^2(l+q+1)(l-q)
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi).
\end{eqnarray}
The above relations indicate that $J_{_{+}}^{(R)}\otimes I$ and $J_{_{-}}^{(R)}\otimes I$
are raising and lowering operators of index $q$ respectively, that is:
\begin{eqnarray}
\nonumber
&&J_{_{+}}^{(R)}\otimes I
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)=\sqrt{\alpha^2
(l+q+1)(l-q)}\Psi_{_{l,j=l\pm \frac{1}{2},m,q+1}}(\theta ,\phi ,\psi), \\
&&J_{_{-}}^{(R)}\otimes I
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)=\sqrt{\alpha^2
(l-q+1)(l+q)}\Psi_{_{l,j=l\pm \frac{1}{2},m,q-1}}(\theta ,\phi ,\psi).
\end{eqnarray}
Now by transfering $e^{iq\psi}$, which is only $\psi$ dependent factor in
the eigenspinor $\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi ,\psi)$, to
the left-hand sides of the lowering and raising operator in (6.4) we arrive at
\begin{eqnarray}
\nonumber
&&J_{_{+}}^{(R)}(q)\otimes I
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi)=\sqrt{\alpha^2
(l+q+1)(l-q)}\Psi_{_{l,j=l\pm \frac{1}{2},m,q+1}}(\theta ,\phi), \\
&&J_{_{-}}^{(R)}(q)\otimes I
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi)=\sqrt{\alpha^2
(l-q+1)(l+q)}\Psi_{_{l,j=l\pm \frac{1}{2},m,q-1}}(\theta ,\phi),
\end{eqnarray}
where $J_{_{\pm}}^{(R)}(q)$ read:
\begin{equation}
J_{_{\pm}}^{(R)}(q)=\pm \frac{\partial}{\partial\theta}-i\frac{\alpha}{\sin
\alpha\theta}\frac{\partial}{\partial\phi}-q\frac{\alpha}{\tan\alpha\theta}.
\end{equation}
But $J_{_{+}}^{(R)}(q)\otimes I$ and $J_{_{-}}^{(R)}(q)\otimes I$ are still
raising and lowering operators of index $q$ of the eigenspinors
$\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi)$ and the relations (6.5)
indicate that $\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi)$ can form
the basis for a representation of para-supersymmetry of order $p$, where $p$ is an arbitrary
integer.
According to \cite{Jafa,Jafar}, the non-unitary para-supersymmetric algebra of order $p$
can be generated by parafermionic generators of order $p$, denoted by
$Q_1$ and $Q_2$ and a bosonic generator $H$, which satisfy the following
relations
\begin{eqnarray}
\nonumber
&&Q_1^pQ_2+Q_1^{p-1}Q_2Q_1+ \cdots +Q_1Q_2Q_1^{p-1}+Q_2Q_1^p=2pQ_1^{p-1}H \hspace{40mm} (6.7a) \\
\nonumber
&&Q_2^pQ_1+Q_2^{p-1}Q_1Q_2+ \cdots +Q_2Q_1Q_2^{p-1}+Q_1Q_2^p=2pQ_2^{p-1}H \hspace{40mm} (6.7b) \\
\nonumber
&&Q_1^{p+1}=Q_2^{p+1}=0 \hspace{114mm} (6.7c) \\
\nonumber
&&[H,Q_1]=[H,Q_2]=0. \hspace{105mm} (6.7d)
\end{eqnarray}
\setcounter{equation}{7}
By introducing the operators
\begin{eqnarray}
\nonumber
&&X_{_{+}}(q):=J_{_{+}}^{(R)}(q)\otimes I \\
\nonumber
&&X_{_{-}}(q):=J_{_{-}}^{(R)}(q)\otimes I \\
&&{\cal H}_{q}:=H_{q}\otimes I,
\end{eqnarray}
we can represent the generators $Q_1$, $Q_2$ and $H$ by the following
$(p+1)\times(p+1)$ matrices of the form
\begin{eqnarray}
\nonumber
&&(Q_1)_{_{qq^{\prime}}}:=X_{_{-}}(q)\delta_{_{q+1,q^{\prime}}} \\
\nonumber
&&(Q_2)_{_{qq^{\prime}}}:=X_{_{+}}(q^{\prime}-1)\delta_{_{q,q^{\prime}+
1}} \\
&&(H)_{_{qq^{\prime}}}:={\cal H}_{_{q}}\delta_{_{q,q^{\prime}}}
\hspace{3cm} q,q^{\prime}=1, \cdots ,p+1,
\end{eqnarray}
where each element of these matrices is a $2\times 2$ matrix.
In (6.9) we need to choose the Hamiltonians ${\cal H}, with q=1, \cdots ,p+1$ so that
the generators (6.9) satisfy the para-supersymmetric algebraic relations
(6.7).
The generators $Q_1$, $Q_2$ and $H$, as defined in (6.9), satisfy the equation
(6.7c), but the equations (6.7a,b) lead to the following equations:
$$
X_{_{+}}(p-2) \cdots X_{_{+}}(1)X_{_{+}}(0)X_{_{-}}(1)+ \cdots +X_{_{+}}(p-2)X_{_{-}}(p-1)X_{_{+}}(p-2)X_{_{-}}(p-3) \cdots X_{_{+}}(0)+
$$
\vspace{-7mm}
\renewcommand{\theequation}{\arabic{section}.\arabic{equation}{a}}
\begin{equation}
X_{_{-}}(p)X_{_{+}}(p-1)X_{_{+}}(p-2) \cdots X_{_{+}}(0)=2pX_{_{+}}(p-2)X_{_{+}}(p-3) \cdots X_{_{+}}(0){\cal H}_{_{1}}
\end{equation}
$$
X_{_{+}}(p-1) \cdots X_{_{+}}(1)X_{_{+}}(0)X_{_{-}}(1)+X_{_{+}}(p-1) \cdots X_{_{+}}(2)X_{_{+}}(1)X_{_{-}}(2)X_{_{+}}(1)+ \cdots +
$$
\vspace{-7mm}
\renewcommand{\theequation}{\arabic{section}.\arabic{equation}{b}}
\setcounter{equation}{9}
\begin{equation}
X_{_{+}}(p-1)X_{_{-}}(p)X_{_{+}}(p-1)X_{_{+}}(p-2) \cdots X_{_{+}}(1)=2pX_{_{+}}(p-1)X_{_{+}}(p-2) \cdots X_{_{+}}(1){\cal H}_{_{2}}
\end{equation}
$$
X_{_{-}}(1) \cdots X_{_{-}}(p-1)X_{_{-}}(p)X_{_{+}}(p-1)+X_{_{-}}(1) \cdots X_{_{-}}(p-2)X_{_{-}}(p-1)X_{_{+}}(p-2)X_{_{-}}(p-1)+ \cdots +
$$
\vspace{-7mm}
\renewcommand{\theequation}{\arabic{section}.\arabic{equation}{c}}
\setcounter{equation}{9}
\begin{equation}
X_{_{-}}(1)X_{_{+}}(0)X_{_{-}}(1)X_{_{-}}(2) \cdots X_{_{-}}(p-1)=2pX_{_{-}}(1)X_{_{-}}(2) \cdots X_{_{-}}(p-1){\cal H}_{_{p}}
\end{equation}
$$
X_{_{-}}(2) \cdots X_{_{-}}(p-1)X_{_{-}}(p)X_{_{+}}(p-1)X_{_{-}}(p)+ \cdots +X_{_{-}}(2)X_{_{+}}(1)X_{_{-}}(2)X_{_{-}}(3) \cdots X_{_{-}}(p)+
$$
\vspace{-7mm}
\renewcommand{\theequation}{\arabic{section}.\arabic{equation}{d}}
\setcounter{equation}{9}
\begin{equation}
X_{_{+}}(0)X_{_{-}}(1)X_{_{-}}(2) \cdots X_{_{-}}(p)=2pX_{_{-}}(2)X_{_{-}}(3) \cdots X_{_{-}}(p){\cal H}_{_{p+1}}.
\end{equation}
Finally equations (6.7d) imply the following equations
\renewcommand{\theequation}{\arabic{section}.\arabic{equation}}
\begin{eqnarray}
&&{\cal H}_{_{q}}X_{_{-}}(q)=X_{_{-}}(q){\cal H}_{_{q+1}} \nonumber \\
&&{\cal H}_{_{q+1}}X_{_{+}}(q-1)=X_{_{+}}(q-1){\cal H}_{_{q}}.
\end{eqnarray}
Now, defining the Hamiltonians ${\cal H}_{_{q}},q=1, \cdots ,p$ as:
\begin{eqnarray}
\nonumber
&&{\cal H}_{_{q}}=\frac{1}{2}X_{_{-}}(q)X_{_{+}}(q-1)+\frac{1}{2}C_{_{q}} I \hspace{1cm} q=1,2,\cdots,p \\
&&{\cal H}_{_{p+1}}=\frac{1}{2}X_{_{+}}(p-1)X_{_{-}}(p)+\frac{1}{2}C_{_{p}} I.
\end{eqnarray}
By the definitions (6.12) relations (6.11) are satisfied for the special case $q=p$. In order
the relations (6.11) to be satisfied for $q=1, \cdots ,p-1$, too we need
to choose the constants $C_{_{q}}$ as:
\begin{equation}
E_{_{q+1}}-E_{_{q}}=C_{_{q}}-C_{_{q+1}},
\end{equation}
where
\begin{equation}
E_{_{q}}:=\alpha^2[l(l+1)-q(q-1)].
\end{equation}
To obtain relation (6.13) we have used the following shape
invariance property between the operators $X_{_{\pm}}(q)$
\begin{equation}
X_{_{+}}(q-1)X_{_{-}}(q)-X_{_{-}}(q+1)X_{_{+}}(q)=E_{_q}-E_{_{q+1}}.
\end{equation}
Substituting (6.12) in formula (6.10a), and also using the shape invariance
property (6.15) we obtain
\begin{equation}
C_{_1}=\frac{1}{p}[(1-p)E_{_{1}}+E_{_{2}}+E_{_{3}}+ \cdots +E_{_p}].
\end{equation}
Finally combining (6.13) with (6.16) we obtain
\begin{equation}
C_{_{q}}=\frac{1}{p}\sum_{q^{\prime}=1}^{p}E_{_{q^{\prime}}}-E_{_{q}}.
\end{equation}
From (6.17) we can see that the following relation
among the constants $C_{_{q}}$ holds
$$
C_{_{1}}+C_{_{2}}+ \cdots +C_{_{p}}=0.
$$
Using the substitution (6.12) and the shape invariance property (6.15) together
with the constants $C_{_{q}}$ given in (6.17) one can straightforwardly show
that all the equations (6.10b,c,d) are satisfied, too. Also using the result
given in (6.17) and the shape invariance relation (6.15) it follows that
the Hamiltonians ${\cal H}_{_{q}}$ are isospectral and we have
\begin{equation}
{\cal H}_{_{q}}\Psi_{_{l,j=l\pm \frac{1}{2},m,q-1}}(\theta ,\phi)=E
\Psi_{_{l,j=l\pm \frac{1}{2},m,q-1}}(\theta ,\phi), \hspace{3cm} q=1, \cdots ,p+1.
\end{equation}
with
\begin{equation}
E=\frac{1}{2p}\sum_{q=1}^{p}E_{_{q}}.
\end{equation}
Substituting $E_{_{q}}$ in (6.19) and by using the relation (6.14) we get
\begin{equation}
E=\frac{1}{6}\alpha^{2}[3l(l+1)+1-p^{2}].
\end{equation}
In a similar manner by substituting (6.14) in (6.17) we have
\begin{equation}
C_{_{q}}=\frac{1}{3}\alpha^{2}[3q(q-1)+1-p^{2}].
\end{equation}
Substituting the constants $C_{_{q}}$ in (6.21), and also using the relations
(6.6) and (6.8) after the substituting (6.12) we obtain the explicit differential
form of the Hamiltonian ${\cal H}_{_{q}}$ as
\begin{equation}
{\cal H}_{_{q}}=-\frac{1}{2}[\frac{\partial^{2}}
{\partial\theta^{2}}+\frac{\alpha}{\tan\alpha\theta}\frac{\partial}{\partial\theta}
+\frac{\alpha^{2}}{\sin^{2}\alpha\theta}\frac{\partial^{2}}{\partial\phi^{2}}
-\frac{2i(q-1)\alpha^{2}}{\sin\alpha\theta \tan\alpha\theta}\frac{\partial}{\partial\phi}
-\frac{(q-1)^2\alpha^{2}}{\sin^{2}\alpha\theta}+\frac{1}{3}\alpha^2(p^2-1)]\otimes I.
\end{equation}
The bases of the representation of para-supersymmetric algebra of order $p$
can be represented by column matrix with $(p+1)$ row, that is
\begin{equation}
(\Psi_{_{l,j=l\pm \frac{1}{2},m}}(\theta ,\phi))_{_{q}}:=
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi), \hspace{3cm}
q=0,1, \cdots ,p,
\end{equation}
where $\Psi_{_{l,j=l\pm \frac{1}{2},m}}(\theta ,\phi)$ is the eigenvector of paraboson operator $H$ with eigenvalue $E$, that is
\begin{equation}
H\Psi_{_{l,j=l\pm \frac{1}{2},m}}(\theta ,\phi)=E
\Psi_{_{l,j=l\pm \frac{1}{2},m}}(\theta ,\phi).
\end{equation}
It follows rather trivially from the commutation relations (6.7d) and also from the
relations $(6.7c)$ that $Q_1^{q}\Psi_{_{l,j=l\pm\frac{1}{2},m}}(\theta,\phi)$
and $Q_2^{q}\Psi_{_{l,j=l\pm\frac{1}{2},m}}(\theta,\phi)$ for $q=1, \cdots ,p$
are eigenstates of the bosonic generator $H$ with the corresponding eigenvalue
$E$. Hence, it follows that
\begin{equation}
\Psi_{_{l,j=l\pm \frac{1}{2},m,q}}(\theta ,\phi)=
\frac{X_{_-}(q+1)}{\sqrt{E_{_{q+1}}}}\frac{X_{_-}(q+2)}{\sqrt{E_{_{q+2}}}}
\cdots \frac{X_{_-}(q^{\prime}+q)}{\sqrt{E_{_{q^{\prime}+q}}}}
\Psi_{_{l,j=l\pm \frac{1}{2},m,q^{\prime}+q}}(\theta ,\phi),
\hspace{10mm} q=0,1, \cdots ,p-q^{\prime}.
\end{equation}
The representation of para-supersymmetry algebra of order $p$ is valid even
for the special case of $\alpha=0$. Using the relation (4.17) in the limiting cases
of $\alpha\rightarrow 0$ and $l\rightarrow\infty$ we arrive at the following
shape invariance relations
\begin{eqnarray}
\nonumber
&&(\frac{\partial}{\partial r}-\frac{i}{r}\frac{\partial}
{\partial \phi}-\frac{q}{r})\otimes I Z_{_{k,m,q}}(r,\phi)=k
Z_{_{k,m,q+1}}(r,\phi) \\
&&(-\frac{\partial}{\partial r}-\frac{i}{r}\frac{\partial}
{\partial \phi}-\frac{q}{r})\otimes I Z_{_{k,m,q}}(r,\phi)=k
Z_{_{k,m,q-1}}(r,\phi).
\end{eqnarray}
Note that in the limiting case of $\alpha=0$ there is neither any
highest nor lowest state in the realization of para-supersymmetry, since there
is no first order differential operator which can kill any of the Bessel
functions. In this case we can have a para-supersymmetry of infinite order where
the Bessel functions form its bases.
\begin{center}
\section{. Solution of Dirac equation on 3-dimensional manifolds with a local
constant curvature}
\end{center}
\setcounter{equation}{0}
For nonvanishing $M$, the spatial part of the metric given in (2.2) describes
a 2-dimensional manifold with local constant curvature. In this section we try
to solve the Dirac equation of a point mass ${\cal M}$ and charge $e$ on the
manifold with metric (2.2), in the presence of magnetic field with connection
(2.6). The angular deficit due to the presence of a very heavy point mass $M$
destroys the global constancy of the curvature, hence we lose both the
degeneracy and para-supersymmetry. Therefore, we can't solve the Dirac equation
by the algebraic methods anylonger. Thus we have to solve it by an ordinary method
of solution of coupled first order differential equations.
In a flat $(2+1)$- spacetime one can represent the Dirac $\gamma$ matrices as
\cite{Ger,Gerb}
$$
\gamma^0=\sigma_{_{3}}
$$
$$
\gamma^1=i\sigma_{_{2}}
$$
\begin{equation}
\gamma^2=-i\sigma_{_{1}},
\end{equation}
where $\gamma^{a}, \; a=0,1,2$ close the Clifford algebra as
\begin{equation}
\{\gamma^a ,\gamma^b\}=2\eta^{ab} \hspace{3cm} a,b=0,1,2.
\end{equation}
The Minkowski metric $\eta_{_{ab}}$ has the following signature
$$
\eta_{_{ab}}=diag(+,-,-).
$$
One can write the metric (2.2) in the following form
\begin{equation}
ds^{2}=dt^{2}-\rho^2(\beta^{-2}d \theta^{2}+
\frac{\sin^{2}\alpha\theta}{\alpha^{2}}d\phi^{2}),
\end{equation}
where $\rho^2$ and $\beta$ are defined as:
$$
\beta=1-GM
$$
$$
\rho^2=\frac{(1-GM)^2}{2\lambda}.
$$
Since $\lambda$ is negative when $\alpha=i$ we have $\rho^2 <0$, hence the
metric (7.3) is Euclidean while for other values of $\alpha$ it is Minkowskian.
We can choose the three-beins associated with the metric (7.3) as:
\begin{equation}
e_{_{\mu}}^{\; a}=\left(\begin{array}{ccc}
1 & 0 & 0 \\
0 & \rho\frac{1}{\beta}\cos\phi & \rho\frac{1}{\beta}\sin\phi \\
0 & -\rho\frac{\sin\alpha\theta}{\alpha}\sin\phi & \rho\frac{\sin\alpha\theta}{\alpha}\cos\phi
\end{array}\right).
\end{equation}
With the above choice of 3-beins
we can compare our results in a special case with those of reference \cite{Ger}. The inverse of matrix (7.4) is:
\begin{equation}
E_{_{a}}^{\; \mu}=\left(\begin{array}{ccc}
1 & 0 & 0 \\
0 & \frac{\beta}{\rho}\cos\phi & -\frac{1}{\rho}\frac{\alpha}{\sin\alpha\theta}\sin\phi \\
0 & \frac{\beta}{\rho}\sin\phi & \frac{1}{\rho}\frac{\alpha}{\sin\alpha\theta}\cos\phi
\end{array}\right).
\end{equation}
According to reference \cite{Ger}, the Dirac equation in $(2+1)$- spacetime for a fermion with mass
${\cal M}$ and electric charge $e$, in the presence of a gauge field with gauge
connection $A_{_{\mu}}$ is:
\begin{equation}
[i\gamma^aE_{_{a}}^{\; \mu}(\partial_{_{\mu}}-\frac{i}{2}\omega_{_{\mu}}^{\; b}\gamma_{_{b}}+
ieA_{_{\mu}})-{
\cal M}]\Psi(t,\theta ,\phi)=0,
\end{equation}
where $\omega_{_{\mu}}^{\; a}$ is given by
\begin{equation}
\epsilon^{\kappa\mu\nu}\partial_{_{\mu}}e_{_{\nu}}^{\; a}=
\epsilon^{\kappa\mu\nu}\epsilon^{a}_{_{bc}}\omega_{_{\mu}}^{\; b}e_{_{\nu}}^{\; c}.
\end{equation}
Hence
\begin{equation}
\omega_{_{\mu}}^{\; a}=\left(\begin{array}{ccc}
0 & 0 & 0 \\
0 & 0 & 0 \\
\beta\cos\alpha\theta-1 & 0 & 0
\end{array}\right).
\end{equation}
Therefore, the Dirac equation (7.6) can be written as
\begin{equation}
\{i[\gamma^0 \partial_{_{t}}+\frac{1}{\rho}\gamma^{\theta}(\beta
\partial_{_{\theta}}-\frac{\alpha}{2\sin\alpha\theta}(1-\beta\cos\alpha\theta))
+\frac{1}{\rho}\frac{\alpha}{\sin\alpha\theta}\gamma^{\phi}(\partial_{_{\phi}}+i
eA_{_{\phi}})]-{\cal M}\}\Psi(t,\theta ,\phi)=0,
\end{equation}
where $\gamma^{\theta}$ and $\gamma^{\phi}$ are defined as:
$$
\gamma^{\theta}=\cos\phi\gamma^{1}+\sin\phi\gamma^{2}
$$
$$
\gamma^{\phi}=-\sin\phi\gamma^{1}+\cos\phi\gamma^{2}.
$$
In the limiting case of $\alpha\rightarrow 0$, the coordinate $\theta$ becomes
similar to a radial coordinate $r$ and the Dirac equation (7.9) in the absence
of gauge field $A_{_{\phi}}$ becomes exactly the Dirac equation associated
with a massive fermion on a $(2+1)$- spacetime dimension with conical spatial
part \cite{Ger}.
Now let the Dirac spinor has the following time dependence
$$
\Psi(t,\theta ,\phi)=e^{-iEt}\Psi(\theta ,\phi),
$$
together with the following $\phi$ dependence
\begin{equation}
\Psi(\theta,\phi)=\left(\begin{array}{c}
e^{i(m-\frac{1}{2})\phi}f_{_{1}}(\theta) \\
e^{i(m+\frac{1}{2})\phi}f_{_{2}}(\theta)
\end{array}\right),
\end{equation}
where $m$ is a half integer number. One can show that the functions $f_{_{1}}(\theta)$
and $f_{_{2}}(\theta)$ satisfy the following differential equations
$$
\{(1-z^2)\frac{d^2}{dz^2}-2z\frac{d}{dz}-\frac{1}{1-z^2}((\frac{m}{\beta})^2+
(eg\pm\frac{1}{2})^2-2\frac{m}{\beta}(eg\pm\frac{1}{2})z)\}f_{_{1,2}}(z)=
$$
\begin{equation}
-(\frac{R^2}{\alpha^2 \beta^2}(E^2-{\cal M}^2)+(e^2g^2-\frac{1}{4}))f_{_{1,2}}(z),
\end{equation}
with $z$ defined as
$$
z=\cos\alpha\theta.
$$
Now defining
$$
\frac{\rho^2}{\alpha^2 \beta^2}(E^2-{\cal M}^2)+(e^2g^2-\frac{1}{4})=c(c+1),
$$
with $c$ as a real number, one can give the solutions of equations (7.11) in
terms of hypergeometric functions \cite{Grad}.
In the limiting cases of $\alpha,e\rightarrow o$ and $c\rightarrow\infty$
such that the product $\alpha c$ remains constant we obtain \cite{Grad}
\begin{equation}
\lim_{_{\alpha\rightarrow 0,c\rightarrow \infty,e\rightarrow 0}}f_{_{1,2}}
(\theta)=J_{_{|\frac{m}{\beta}\mp\frac{1}{2}|}}(\alpha cr).
\end{equation}
Writing the half integer number $m$ as
$$
m=n+\frac{1}{2},
$$
with $n$ as an arbitrary integer the Dirac equation (7.9) takes the following solution
in the above mentioned limit
\begin{equation}
\Psi(t,\theta,\phi)=e^{-iEt}\exp{i(n+\frac{1}{2}-\frac{1}{2}\sigma_{_{3}})\phi}
\left ( \begin{array}{c}
AJ_{_{|\frac{1}{\beta}(n+\frac{1-\beta}{2})|}}(\alpha cr) \\
BJ_{_{|\frac{1}{\beta}(n+\frac{1-\beta}{2})+1|}}(\alpha cr)
\end{array}\right ),
\end{equation}
where $A$ and $B$ are arbitrary constants.
The solution (7.13) is eigenstate of Dirac Hamiltonian associated with a fermion
on $(2+1)$ spacetime with conical spatial part \cite{Ger} with corresponding
eigenvalues:
\begin{equation}
E=\pm\sqrt{{\cal M}^2+\frac{1}{\rho^2}\beta^2\alpha^2 c^2}.
\end{equation}
\vspace{40cm}
|
\section{Introduction}
The dust and high stellar density of the Milky Way obscures
up to 25\% of the optical extragalactic sky,
creating a Zone of Avoidance (ZOA).
The resulting incomplete coverage of surveys of external
galaxies leaves open the possibility that dynamically important
structures, or even nearby massive galaxies, remain undiscovered.
Careful searches in the optical and infrared wave bands can narrow the ZOA,
(see Kraan-Korteweg, this volume)
but in the regions of highest obscuration and infrared confusion, only
radio surveys can find galaxies.
The 21 cm line of neutral hydrogen (HI) passes readily through the obscuration,
so galaxies with sufficient HI can be found through detection of their 21 cm
emission.
Of course, this method will miss HI-poor, early-type galaxies, and cannot
discriminate
HI galaxies with redshifts near zero velocity from Galactic HI.
Here we describe an HI blind survey for galaxies in the southern ZOA
conducted with the new multibeam receiver on the 64-m Parkes telescope.
A survey of HI galaxies in the northern ZOA is underway with the Dwingeloo
radiotelescope (Henning et al. 1998; Rivers et al. this volume).
\section{The Shallow Survey}
\subsection{Observing Strategy}
The HI Parkes ZOA survey covers the southern ZOA
($212\deg \le l \le 36\deg; \vert b \vert \le 5\deg$) over the
velocity range (cz) = $-1200$ to $12700\,\rm\,km\,s^{-1}$.
The multibeam receiver is a focal plane array with 13 beams arranged in an
hexagonal grid.
The spacing between adjacent beams is about two beamwidths, each beamwidth
being 14 arcmin.
The survey is comprised of 23 contiguous rectangular fields
which are scanned parallel to the galactic equator.
Eventually, each patch will be observed 25 times, with scans offset by
about 1.5 arcmin in latitude.
The shallow survey discussed here consists of two scans in
longitude separated by
$\Delta$b = 17 arcmin, resulting in an rms noise of about 15 mJy, equivalent
to a 5$\sigma$ HI mass detection limit of $4 \times 10^6$ d$^2\!\!_{\rm Mpc}$
M$_{\odot}$ (for a galaxy with the typical linewidth of
$200{\,\rm\,km\,s^{-1}}$).
\subsection{Data Visualization}
After calibration, baseline-subtraction, and creation of data cubes,
all done with specially developed routines based on aips++
(Barnes et al. 1998, Barnes 1998)
the data are examined by eye using the visualization
package Karma
(http://www.atnf.csiro.au/karma/).
The data are first displayed as right ascension -- velocity
planes, in strips of constant declination.
Data cubes are
then rotated, and right ascension -- declination planes are
checked for any suspected galaxies (eg. Figure 1).
\begin{figure}
\begin{center}
\hbox{
\psfig{file=fig1a,height=8.7cm}
\psfig{file=fig1b,height=8.7cm}
}
\caption{Left panel shows a right ascension -- velocity slice to
5000${\,\rm\,km\,s^{-1}}$ which includes a galaxy
discovered by the survey.
Galactic HI appears as the strong horizontal feature at zero velocity.
Note the extragalactic HI signal at $7^{\rm h}18^{\rm m}$,
$900{\,\rm\,km\,s^{-1}}$.
Right panel shows a right ascension -- declination plane at the
velocity of the suspected signal.
The galaxy is evident at $7^{\rm h}18^{\rm m}$, $-9\deg$.}
\label{fig-1}
\end{center}
\end{figure}
\section{Galaxies Found by the Shallow Survey}
The shallow 21-cm survey of the southern ZOA
has been completed, and 107 galaxies with peak HI flux densities $\geq$
about 80 mJy
have been cataloged.
Refinement of the measurement of their HI characteristics is ongoing,
but the objects
seem to be normal galaxies.
However, of three large multibeam ZOA galaxies imaged in HI with the ATCA,
two were seen to break up into complexes of HI suggestive of
tidally-interacting systems (Staveley-Smith et al. 1998)
Continued follow-up synthesis observations are planned to investigate
the frequency of these interacting systems in this purely HI-selected
sample.
Most of the galaxies are within $4000{\,\rm\,km\,s^{-1}}$,
which is about the redshift limit
for detection of normal spirals
of this shallow phase of the survey.
As the deep survey continues, spirals at higher velocities will be
recovered.
The effective depth of the shallow survey
is not quite sufficient to recover large numbers
of galaxies which might be associated with the Great Attractor
(but see Juraszek et al. this volume.)
However, a striking feature becomes apparent with the addition of
the ZOA galaxies.
An enormous filament, which crosses the ZOA twice, is clearly evident
when these ZOA data are displayed along with optically-known galaxies
above and below the plane within $3500{\,\rm\,km\,s^{-1}}$ (Fig. 2.)
This structure snakes over $\sim180\deg$ through the southern sky.
Taking a mean distance of $30{h^{-1}}$ Mpc, this implies a linear
size of $\sim100{h^{-1}}$ Mpc, with thickness of $\sim5{h^{-1}}$ Mpc
or less.
Also, note the relative emptiness of the Local Void.
Three hidden galaxies found
on a boundary of the Void (l $\sim30\deg$) lie at
$\sim 1500 {\,\rm\,km\,s^{-1}}$. Two of these objects were also recovered
by the Dwingeloo Obscured Galaxies Survey (Rivers et al. this volume.)
The positions and redshifts of these objects are consistent with their
being members of the cluster at this location proposed by Roman et al. (1998).
\begin{figure}
\begin{center}
\psfig{file=fig2,width=17.5cm}
\caption{Galaxies within v $\leq3500{\,\rm\,km\,s^{-1}}$.
The open circles mark galaxies between $500 - 1500{\,\rm\,km\,s^{-1}}$,
triangles for $1500 - 2500{\,\rm\,km\,s^{-1}}$, and dots for
those with $2500 - 3500{\,\rm\,km\,s^{-1}}$.
High latitude data are taken from the literature (LEDA).
Also plotted are galaxies from deep optical ZOA surveys
(Kraan-Korteweg et al. this volume.) The
galaxies discovered by the multibeam shallow ZOA survey
fill in the lowest galactic latitudes, where optical surveys
fail.
}
\label{fig-2}
\end{center}
\end{figure}
Of the 107 objects found,
28 have counterparts in the NASA/IPAC Extragalactic Database (NED)
with matching positions and redshifts.
Optical absorption,
estimated from the Galactic dust data of
Schlegel et al. (1998), ranges from A$_{\rm B}$ = 1 to more than 60 mag
at the positions of the 107 galaxies,
and is patchy over the survey area.
No objects lying behind more than about 6 mag of obscuration have
confirmed counterparts in NED, as expected.
The shallow multibeam HI survey connects structures all the way across
the ZOA within $3500{\,\rm\,km\,s^{-1}}$ for the first time.
The ongoing, deep ZOA survey will have sufficient sensitivity to
connect structures at higher redshifts.
While 14 of the 107 galaxies lie within
$1000{\,\rm\,km\,s^{-1}}$ and are therefore fairly nearby,
all of the newly-discovered objects have peak HI flux densities an order
of magnitude or more lower than the Circinus galaxy.
Thus, it seems our census of the most dynamically important, HI-rich nearby
galaxies is now complete, at least for those objects with velocities
offset from Galactic HI.
Simulations are currently being devised to investigate our sensitivity
to HI galaxies whose signals lie within the frequency range of the Milky
Way's HI.
This will be done by embedding artificial HI signals of varying strength,
width, position, and frequency, into real data cubes.
Then, an experienced HI galaxy finder (PH) will examine the cubes without
previous knowledge of the locations of the fake galaxies.
In this way, we hope to quantify better this remaining blind spot of the
HI search method.
\section*{Acknowledgements}
We thank HIPASS ZOA collaborators
R. D. Ekers, A. J. Green, R. F. Haynes, S. Juraszek, M. J. Kesteven,
B. S. Koribalski, R. M. Price, and
A. Schr\"oder.
This research has made use of the NASA/IPAC Extragalactic Database (NED)
which is operated by the Jet Propulsion Laboratory, Caltech, under
contract with the National Aeronautics and Space Administration.
We have also made use of the Lyon-Meudon Extragalactic Database (LEDA),
supplied by the LEDA team at the Centre de Recherche Astronomique de
Lyon, Observatoire de Lyon.
The research of P.H. is supported by NSF Faculty Early Career Development
(CAREER) Program award AST 95-02268.
\section*{References}
\parskip 0pt\par\noindent\hangindent 0.5 truecm Barnes, D.G. 1998, in ADASS VII, eds. Albrecht, R., Hook, R.N.,
\& Bushouse, H.A., San Francisco: ASP
\parskip 0pt\par\noindent\hangindent 0.5 truecm Barnes, D.G., Staveley-Smith, L, Ye, T., \& Osterloo, T.
1998, in ADASS VII, eds. Albrecht, R., Hook, R.N.,
\& Bushouse, H.A., San Francisco: ASP
\parskip 0pt\par\noindent\hangindent 0.5 truecm Henning, P.A., Kraan-Korteweg, R.C., Rivers, A.J., Loan, A.J.,
Lahav, O., \& Burton, W.B. 1998, AJ, 115, 584
\parskip 0pt\par\noindent\hangindent 0.5 truecm Roman, A.T., Takeuchi, T.T., Nakanishi, K., \& Saito, M.
1998, PASJ, 50, 47
\parskip 0pt\par\noindent\hangindent 0.5 truecm Schlegel, D.J., Finkbeiner, D.P., \& Davis, M. 1998, ApJ, 500,
525
\parskip 0pt\par\noindent\hangindent 0.5 truecm Staveley-Smith, L., Juraszek, S., Koribalski, B.S., Ekers,
R.D., Green, A.J., Haynes, R.F., Henning, P.A., Kesteven, M.J.,
Kraan-Korteweg, R.C., Price, R.M., \& Sadler, E.M. 1998, AJ, in press.
\end{document}
|
\section{Introduction}
The galaxy luminosity distribution, or luminosity function\footnote{ The
{\it luminosity function} is expressed in units of number density ({\it e.g.}\
number per Mpc$^3$), whereas the {\it luminosity distribution} simply
gives the shape of the luminosity function without density normalization.},
$\phi(L)$ plays
an important role in our understanding of the properties of galaxies,
galaxy evolution, and galaxy formation. The connection between $\phi(L)$
and galaxy formation is through the primordial
density fluctuation spectrum, $\delta\rho/\rho \propto k^n$, where
$k$ is wavenumber. If the Universe consisted only of weakly interacting
particles ({\it e.g.}\ ``cold dark matter''), then the mass function of ``halos''
would be determined solely by $n$, and could be computed using a simple
physical recipe for the gravitational clustering and merging of
halos ({\it e.g.}\ Press \& Schechter 1974). However, it is the baryons,
rather than the dark matter, that we observe directly; hence
the luminosity function $\phi(L)$ depends on gas physics and
radiation processes ({\it e.g.}\ cooling, radiative transfer, star formation,
energy input from supernovae, to name just a few). It follows that
the luminosity function
is sensitive not only to the fluctuation spectrum $\delta\rho/\rho$, but also
to the detailed history of galaxy formation and evolution in different
environments.
Generally the luminosity function of galaxies is parametrized by a Schechter
(1976) function,
\begin{equation}
\phi(L) = \phi^* e^{-(L/L^*)} (L/L^*)^{\alpha},
\end{equation}
\noindent
where $L^*$ is a characteristic luminosity defining the transition
between a power-law at faint magnitudes and an exponential cutoff at
bright magnitudes. [Further information concerning the Schechter function
can be found in Felten (1985).] $L^*$ corresponds roughly to the brightness
of the Milky Way. For a CDM fluctuation spectrum with
$n=1$, the power-law exponent $\alpha$ is theoretically predicted to be
$\approx -2$ on the scale of galaxies (Bardeen {\it et al.}\ 1986), though this
result is very sensitive to the detailed physical processes involved in
the calculation ({\it cf.}\ Babul and Ferguson 1996, Frenk {\it et al.}\ 1996, Kauffmann
{\it et al.}\ 1998).
What is known empirically about the shape of the luminosity distribution in
the nearby Universe? Values of $\alpha$ in the range $-$0.7 to $-$1.0 have
been obtained for bright galaxies within 2--3 magnitudes of $L^*$ ({\it e.g.}\
Loveday {\it et al.}\ 1992; Marzke {\it et al.}\ 1994 $a,b$; Lin {\it et al.}\ 1996), with a hint
of a turnup in the LF at magnitudes fainter than about $M_R = -17$ in the
work of Lin {\it et al.}. The field luminosity distribution derived from the SSRS2
redshift survey (Marzke {\it et al.}\ 1998) indicates a relatively flat value of
$\alpha \simeq -1$ for E/S0's (including dwarf spheroidals) and spirals,
and a steep $\alpha = -1.8$ value for dwarf irregulars and peculiars; this
steepening of the LF for late-type star forming systems also appears in
the work of Bromley {\it et al.}\ (1998), who subdivided the Las Campanas Redshift
Survey according to emission-line strength. C\^ot\'e {\it et al.}\ (1998) have
found a very steep ($\alpha=-2.1$) luminosity distribution for
nearby H~I-rich low surface brightness galaxies, and Schneider, Spitzak,
\& Rosenberg (1998) find a steep upturn in the H~I mass function for
low mass objects (M$_{H~I} \mathrel{\spose{\lower 3pt\hbox{$\mathchar"218$} 10^8$ M$_\odot$).
Turning to other environments, Loveday (1997) finds that the luminosity
distribution of dwarf galaxies surrounding luminous ($\sim L^*$)
galaxies is steep: this can be interpreted either as a turnup in a
supposedly universal field luminosity distribution, or alternately as
an enhanced probability that dwarfs form in the vicinity of luminous,
massive galaxies. Galaxies in groups exhibit a slope $\alpha \simeq
-1$ (Muriel, Valotto \& Lambas 1998). There is evidence that compact
group luminosity distributions cannot be fitted by a single Schechter
function (Hunsberger, Charlton \& Zaritsky 1998), but instead show
$\alpha =-0.5$ at the bright end and $\alpha=-1.2$ below $M_R \simeq
-16$. There is also some indication that cluster LF's must be fitted
with multiple Schechter functions ({\it e.g.}\ Trentham 1998$a$, Lopez-Cruz
{\it et al.}\ 1997), with a steep upturn at faint magnitudes ({\it e.g.}\ Trentham
1998$b$; Phillipps {\it et al.}\ 1998$a$). Phillipps {\it et al.}\ (1998$b$) suggest
the existence of an environmental dependence of dwarf-to-giant ratio
({\it i.e.}\ $\alpha$) in clusters. The faintest galaxies in the steep
luminosity distribution population in clusters are, based on their colors
(Trentham 1998$a,b$), dwarf spheroidals, whereas in the field they appear
to be gas-rich dwarf irregulars (Marzke {\it et al.}\ 1998). (It should, however, be noted
that the Marzke {\it et al.}\ dIr/peculiar sample is drawn from a small local
volume and comprises only 4\% of the total SSRS2 sample.)
On the basis of the discussion given above, it appears that
the simple paradigm of a universal Schechter function (that fits
the LF of galaxies in all environments) is now untenable. There is
growing evidence that the LF is not a simple Schechter function, that it
depends on environment, and that it also depends on galaxy morphological
class and/or gas content within a given environment. No clear physical
picture has emerged that would allow one to understand current observational
evidence on the shape and environmental dependence of the LF.
The Local Group represents a unique opportunity for the study
of the luminosity distribution in relatively low density
environments. The manner in which faint
Local Group galaxies are detected is completely different from that for
other groups (because most Local Group galaxies are easily resolved into
stars); hence surface brightness selection effects operate differently
in the Local Group than they do in more distant groups. Subtraction of
foreground and background contaminating objects is irrelevant for Local
Group galaxies, something that is of course not the case for more distant
clusters. The numbers of galaxies in poor groups are so low that it has
usually been possible to study only the {\it composite} LF of poor groups
(rather than the LF of any group individually) -- here the Local Group
again is an exception. Furthermore, the census of Local Group members
extends to considerably fainter absolute magnitudes ($M_V \simeq -8.5$) than
do any other samples for which the luminosity distribution has been
measured (though incompleteness must be severe at the faint end). Thus a study of the
Local Group luminosity distribution is of considerable importance.
What is known about the Local Group luminosity distribution? Tully (1988)
derived a composite luminosity function for six nearby groups (including
the Local Group), and found $\alpha=-1\pm0.2$. Van den Bergh
(1992) demonstrated that the integral luminosity function of the Local
Group was consistent with $\alpha = -1.1$, but did not rule out the
possibility that other values of $\alpha$ fitted the data equally
well. More recently, Mateo (1998) has shown that the LF of galaxies in
the vicinity of the Local Group (but extending out beyond the usually
accepted LG boundary of R = 1 Mpc) is consistent with that derived for
poor groups (Ferguson and Sandage 1991). Again, this statement does not
preclude the possibility that other luminosity distributions fit equally
well.
Over the past few years substantial additional data have been accumulated
on Local Group membership and absolute magnitudes, and so the time seems
ripe for a fresh, and more detailed, attack on the problem of the Local
Group luminosity distribution.
\section {The Local Group Catalog}
The Local Group of galaxies was first decribed by Hubble (1936) in
his book {\it The Realm of the Nebulae}. He listed M~31, M~32, M~33, the
Magellanic Clouds, NGC~205, NGC~6822, and IC~1613 as probable members
of the small group of galaxies associated with our Milky Way system.
Inspection of the prints of the Palomar Sky Survey (van den Bergh 1962)
shows that a large fraction of all of galaxies occur in small groups
and clusters that resemble the Local Group. This shows that our Galaxy
is located in a rather typical region of space. Since Hubble's
pioneering work the number of galaxies that are known to belong to the
Local Group has increased by 4 or 5 per decade to over thirty. A
listing of data on presently known Local Group members (van den Bergh
2000) is given in Table 1.
Selection of Local Group members proceeded in three steps. First
galaxies with distances from the Local Group centroid (Courteau and
van den Bergh 1999) less than or about 1.5~Mpc were regarded as suspected Group
members. Secondly it was required that Local Group members should lie
close to the relation between between radial velocity V$_r$ and cos
$\theta$ for well-established Local Group members, where $\theta$
is the distance from the solar apex (Courteau and van den Bergh 1999).
Finally, Local Group members should not appear to be associated with
groups of galaxies that are centered well beyond the limits of the
Local Group.
On the basis of these criteria van den Bergh (1994, 2000)
concluded that the following objects should be excluded from Local
Group membership: (1) UKS~2323-326, (2) Maffei~1 and its companions,
(3) UGC~A86, (4) NGC~1560, (5) NGC~1569, (6) NGC~5237, (7) DDO~187, (8)
Cassiopeia~1, and (9) NGC~55. A particularly strong concentration of
these Local Group suspects, which includes (2), (3), (4), (5) and (8)
listed above, occurs in the direction of the IC~342/Maffei group
(Krismer, Tully \& Gioia 1995). (1) and (9) appear in the
direction of the Sculptor (=South Polar) group; in the case of (9),
Jergen, Freeman \& Binggeli (1998) find D=$1.66 \pm 0.2$ kpc, which gives
a distance 1.65 Mpc from the Local Group centroid. Finally, the discovery
of a Cepheid (Tolstoy et al. 1995) in DDO~155 (=GR~8) suggests that
this object is located at a distance of 2.2 Mpc, which places it well
beyond the usually accepted limits of the Local Group. Dohm-Palmer {\it et al.}\
(1998) obtain a similar distance to DDO~155 from the tip of the red giant
branch.
Also excluded
from Local Group membership are the galaxies NGC~3109, Antlia, Sextans~A
and Sextans~B. These objects, which are located relatively close
together on the sky, all have distances of 1.3 -- 1.5 Mpc from the
Milky Way, and, of more relevance, distances of $\sim 1.7$~Mpc from
the Local Group centroid. Furthermore, these objects possess a mean
radial velocity of $+114 \pm 12$ km s$^{-1}$
relative to the relation between $V_r$ and cos $\theta$ for
well-established Local Group members (van den Bergh 1999).
This suggests that these galaxies form a small group just beyond the
zero velocity surface of the Local Group. (This surface is at a distance
$R(LG) = 1.18 \pm 0.15$ Mpc from the Local Group centroid
[Courteau and van den Bergh 1999].)
How does the Local Group membership defined above compare with that of
Mateo (1998)? The principal differences are that the Mateo catalog does
not contain several recently-discovered satellites of M~31, but does
include nine objects beyond 1 Mpc (NGC~55, EGB~0427+63, Sextans A,
Sextans B, NGC~3109, Antlia, GR~8, IC~5152, UKS~2323-326). Most of these
were discussed above. EGB~0427+63 has a distance of 2.2 Mpc
(Karachentsev, Tikhonov \& Sazonova 1994) and thus lies well outside
the Local Group. From a color-magnitude diagram, Zijlstra \& Minniti (1999) find that IC~5152 has a distance from the Milky Way of $1.70 \pm 0.16$ Mpc,
a result that agrees with the Cepheid distance of 1.6 Mpc (Caldwell \&
Schommer 1988); the distance of this galaxy from the Local Group centroid
is therefore 1.8 Mpc, again beyond the Local Group.
A more detailed discussion of
Local Group membership and of the outer boundary of the Local Group can be
found in van den Bergh (2000).
\section{Luminosity Distribution of the Local Group}
Because Local Group galaxies are situated so nearby it is possible to
study their luminosity distribution down to very faint absolute
magnitudes. Nevertheless these data are, no doubt, still quite
incomplete for M$_V > -10$. This is shown most clearly by the fact that only one
galaxy fainter than this limit has so far been discovered in the
Andromeda subgroup of the Local Group, whereas five such faint objects
are presently known in the Milky Way subgroup of the Local Group.
On the other hand, a
survey of a twenty thousand square degree area at high Galactic
latitudes by Irwin (1994) resulted in the discovery of only a single
new Local Group member. Furthermore, no new optically visible Local Group
galaxies have turned up in the survey of compact high latitude
high velocity clouds (Braun and Burton 1999).
Taken at face value these results might be taken to suggest
that the luminosity distribution of the Local Group no longer increases
below $M_V \simeq -8$. It is noted in passing that very large low surface
brightness galaxies in the Local Group, like those that have been discovered in the Virgo
cluster (Impey, Bothun \& Malin 1988), in the Fornax cluster (Bothun,
Impey \& Malin 1991) and in the M~81 group (Caldwell et al. 1998), may
have also eluded us.
The data in Table 1 can be used to study the luminosity
distribution of Local Group galaxies. Histograms plotting this distribution
are shown in Figure 1. The upper histogram clearly shows an increased
number of objects at faint absolute magnitudes. The separation by
morphological type (lower two panels) shows that most of this increase
is due to galaxies that are dwarf spheroidals.
A somewhat smoother visual impression of the Local Group luminosity
distribution may be obtained by plotting the {\it cumulative} luminosity
distribution, which is compared in Figure 2 to several different
cumulative Schechter functions. In Fig. 2, the cumulative numbers are
normalized at M$_V$=$-$10 because it is unlikely that the data are complete
at fainter magnitudes. This figure shows that a Schechter function with
$\alpha\simeq -1.1$ and M$_V^*=-20$ is an acceptable fit to the data ({\it cf.}\ van den Bergh
1992, Mateo 1998), and that there is some evidence for a steepening to
$\alpha < -1.3$ at faint magnitudes. (Because of incompleteness effects,
this is an upper limit on the faint-end slope).
To parametrize the luminosity distribution of the Local Group, we fit
the data from Table 1 to a Schechter (1976) function [Eqn. (1)]. As
discussed in \S 1, this function possesses a power-law luminosity
dependence with exponent $\alpha$, and an exponential cut-off at
L$>$L$^*$. In a plot of log $\phi$(M) vs. absolute magnitude M, the faint end of
the Schechter function is linear, with slope $a = -0.4 (\alpha + 1)$.
In detail, we fit the unbinned Local Group absolute magnitude data to a
Schechter function using maximum likelihood techniques. The small
number of objects involved dictates that we use a Poisson,
rather than Gaussian, estimator of likelihood. The very small population of luminous galaxies ($L \mathrel{\spose{\lower 3pt\hbox{$\mathchar"218$} L^*$) also means that it is not possible to obtain
a robust estimation of M$_V^*$; hence we
fit only $\alpha$ (and of course a normalization constant proportional to
$\phi^*$), with a few different trial values of M$_V^*$ (which make virtually
{\it no} difference to the fitted value of $\alpha$). The sparsenesss
of the dataset furthermore prevents us from considering luminosity
distributions that are a combination of two or more fitting functions (as
found for Coma and other rich clusters by Trentham 1998$b$; see also
Ferguson and Sandage 1991; Binggeli, Sandage and Tammann 1987). However,
as will be seen, it is nevertheless possible to constrain such functions
by isolating different magnitude ranges.
The maximum likelihood program was tested with artificial data sets drawn
from a distribution of absolute magnitudes that followed a Schechter
function. From thousands of simulations, the maximum likelihood program
was found to return almost precisely the input value of $\alpha$ in the
mean, even for very small numbers of objects (N $<$ 10). Furthermore,
the error estimates (see below) were also found to be accurate.
Table 2 gives the maximum likelihood value of $\alpha$ for various fits
to the Local Group data (different magnitude ranges, and selections of
morphological types), together with several different error estimators
for this quantity. The first error estimate is simply a $1\sigma$ error:
this is derived by finding that region of the maximum likelihood
probability distribution
that is centred on the fitted value of $\alpha$, and that includes 68\%
of the probability. Also given in Table 2 are 95\% and 99\% confidence
limits for the upper bound on $\alpha$ ({\it i.e.}\ the values of $\alpha$ for
which the probability is 95\% and 99\% that $\alpha$ is steeper than
this). Finally, Table 2 also shows the percentage probability that
$\alpha$ is steeper than (less than) --1.3.
Considering the entire data set (faint limit $M_V = -8$), it is
apparent that a value $\alpha=-1.07 \pm 0.05$ is a best fit, with a 95\%
probability that $\alpha < -0.98$. Since the Local Group luminosity
distribution is known to be incomplete at such faint magnitudes,
we instead consider limiting the choice of objects at the faint end.
However, regardless of the choice of parameters, the value $\alpha \simeq
-1.1 \pm 0.1$ emerges. The probability that $\alpha$ is steeper
than --1.3 is $<1$\%.
The only exception to this is for galaxies fainter than $M_V = -15$; for
such objects slopes as steep as $\alpha=-1.5$ are derived. However: (1)
these slopes have large errors ($\pm 0.2$ or even greater) because they
are based both on small numbers of objects, and also on a limited range
of $M_V$; (2) the 95\% probability upper limit for $\alpha$ (lower
limit on steepness) continues to hover around $\alpha = -1$, and the
probability that $\alpha < -1.3$ is only 85\%; and (3) the effect goes
away if one instead considers objects brighter than $M_V = -16$. Thus
we consider this apparent steepening of the luminosity distribution at
faint absolute magnitudes to be tantalizing, but {\it not} proven.
Note that the derived slope is not very sensitive to the precise value of
the faint cutoff for the fit. This is probably because of incompleteness
at the faint end, but also because, even with a faint end upturn in the
luminosity distribution, the majority of objects contributing to the
fit are at brighter magnitudes.
The derived value of the slope $\alpha$ is not sensitive to the assumed
outer boundary of the Local Group. Relative to the Local Group centroid,
the shell between 1.18 Mpc (the zero velocity surface according to
Courteau and van den Bergh 1999) and 1.6~Mpc contains only a single
galaxy, SagDIG. Removing this galaxy from our sample does not alter
any of the results above. Mateo (1998) includes nine galaxies in
his Local Group catalog that, because of their distance, do not appear
in our catalog (see Section 2). Including these nine objects in our fits makes
$\alpha$ less steep by $< 0.1$. It should be noted that even this small
effect can be entirely explained by incompleteness at the low luminosity
end of the sample of galaxies beyond 1.18 Mpc. We also stress that
the available evidence does {\it not} support the inclusion of these nine
galaxies in our Local Group catalog (see discussion in Section 2).
Most of the apparent steepening in $\alpha$ for faint objects is due to
the dwarf spheroidals in the Local Group, as can be seen from Fig. 1.
Fitting a power-law slope to these objects alone shows a steeper
value of $\alpha$ than for the entire dataset, but again the effect is only
marginally significant. From a Kolmogorov-Smirnov two-sample test,
the difference between the luminosity distributions of dIr's and
dSph's is significant only at the 90\% level. Unfortunately, the observations
of M81 group dwarfs ({\it e.g.}\ van Driel {\it et al.}\ 1998) do not enable us to throw
additional light on this problem. This is because these authors were
not able to determine morphological classes for the two faintest magnitude
bins in their survey.
Trentham (1998$c$) has derived a composite luminosity function for
galaxies in clusters, and has shown that it can be applied to galaxies
in the field as well. The cumulative form of this empirical luminosity
function (for which $\alpha$ steepens towards fainter $M_V$) is plotted as
the dashed line in Fig. 2. A Kolmogorov-Smirnov test excludes the
possibility that Local Group galaxies are drawn from this parent
population at $>99$\% probability.
Finally, we have compared the distribution of $B$ magnitudes of galaxies
in the M~81 group (van Driel {\it et al.}\ 1998) with the luminosity distribution
of M$_V$ of Local Group galaxies, under the assumptions that (m-M)$_B =
28.8$ and $<B-V> = 0.5$ for the M~81 galaxies. From a comparison between
the (presumably more-or-less complete) data on galaxies with M$_V$
brighter than --10 and $B$ brighter than 17.5, a Kolmogorov-Smirnov
test shows no significant difference between the M~81 (N=38) and Local Group
luminosity distributions (N=27). This suggests that the Local Group and M~81
LFs are broadly similar, and are drawn from similar parent populations.
\section {Discussion and Conclusions}
A Schechter function with $\alpha \simeq -1.1 \pm 0.1$ provides a good
fit to new data for the luminosity distribution of the Local Group.
This result is in agreement with the luminosity distribution found
for poor groups ({\it e.g.}\ Ferguson and Sandage 1991, Muriel {\it et al.}\ 1998),
and is probably consistent with the work of Hunsberger {\it et al.}\ (1998), who
found $\alpha \simeq -0.5$ for $M_R < -18$ and $-$1.2 for $M_R > -18$.
Our result is comparable to various determinations of $\alpha$ in the
field ({\it e.g.}\ Loveday {\it et al.}\ 1992; Marzke {\it et al.}\ 1994 $a,b$; Lin {\it et al.}\ 1996;
Marzke {\it et al.}\ 1998), and is insensitive to the manner in which the
Local Group is defined.
There is evidence for a steepening in $\alpha$ below M$_V$=$-$15; as
discussed in \S 1, this effect has been observed in other environments.
However, the steepening of the field luminosity distribution observed by
Marzke {\it et al.}\ (1998) is in the dIr population, in contrast to the
situation in the Local Group,
for which the dIr population possesses a flat $\alpha \approx -1$, and
for which the dSph population appears to be possess steeper $\alpha$
(though this difference is significant only at the $\sim90$\% level
in our work).
The steepening in $\alpha$ that we observe at faint magnitudes is limited
in significance by small number statistics, and almost certainly $\alpha$ is steeper than our fits would indicate, because of magnitude dependent
incompleteness. Clearly much further observational work is needed
to improve the completeness of the census of Local Group members at faint absolute magnitudes.
\acknowledgments
C.J.P. acknowledges the hospitality of the South African Astronomical
Observatory, where part of this work was completed, and the financial
support of the Natural Sciences and Engineering Research Council of
Canada. The authors are grateful to George Jacoby and Taft Armandroff
for pointing out an error in an earlier version of Fig. 2.
\clearpage
\begin{deluxetable}{llllrrrrr}
\scriptsize
\tablenum{1}
\tablewidth{0pt}
\tablecaption{Derived Properties of Probable Local Group Galaxies}
\tablehead{\colhead{Name} & \colhead{Alias} & \colhead{DDO Type} &
\colhead{(m-M)$_0$} & \colhead{M$_V$} & \colhead{$\ell$} &
\colhead{$b$} & \colhead{D[kpc]} & \colhead{cos$\theta$}}
\startdata
M 31 &N 224 & Sb I-II &24.4 &-21.2 &121.17 &-21.57 & 760 & 0.88 \\
Milky Way &Galaxy &S(B)bc I-II: &14.5 &-20.9: &000.00 & 00.00 & 8 &-0.15 \\
M 33 &N 598 & Sc II-III &24.5 &-18.9 &133.61 &-31.33 & 795 & 0.73 \\
LMC &... & Ir III-IV &18.5 &-18.5 &280.19 &-33.29 & 50 &-0.80 \\
SMC &... &Ir IV/IV-V &18.85 &-17.1 &302.81 &-44.33 & 59 &-0.61 \\
M 32 &NGC 221 & E2 &24.4 &-16.5 &121.15 &-21.98 & 760 & 0.88 \\
NGC 205 &... & Sph &24.4 &-16.4 &120.72 &-21.14 & 760 & 0.88 \\
IC 10 &... & Ir IV: &24.1 &-16.3 &118.97 &-03.34 & 660 & 0.94 \\
NGC 6822 &... & Ir IV-V &23.5 &-16.0 &025.34 &-18.39 & 500 & 0.29 \\
NGC 185 &... & Sph &24.1 &-15.6 &120.79 &-14.48 & 660 & 0.91 \\
IC 1613 &... & Ir V &23.3 &-15.3 &129.73 &-60.56 & 725 & 0.47 \\
NGC 147 &... & Sph &24.1 &-15.1 &119.82 &-14.25 & 660 & 0.92 \\
WLM &DDO 221 & Ir IV-V &24.85 &-14.4 &075.85 &-73.63 & 925 & 0.32 \\
Sagittarius& ... & dSph(t) &17.0 &-13.8::&005.61 &-14.09 & 24 &-0.04 \\
Fornax &... & dSph &20.7 &-13.1 &237.24 &-65.66 & 138 &-0.25 \\
Pegasus &DDO 216 & Ir V &24.4 &-12.3 &094.77 &-43.55 & 760 & 0.76 \\
Leo I &Regulus & dSph &22.0 &-11.9 &225.98 &+49.11 & 250 &-0.44 \\
And I &... & dSph &24.55 &-11.8 &121.69 &-24.85 & 810 & 0.86 \\
And II &... & dSph &24.2 &-11.8 &128.91 &-29.15 & 700 & 0.78 \\
Leo A &DDO 69 & Ir V &24.2 &-11.5 &196.90 &+52.41 & 690 &-0.14 \\
Aquarius* &DDO 210 & Ir V &25.05 &-11.3 &034.04 &-31.35 &1025 & 0.40 \\
SagDIG* &... & Ir V &25.7: &-10.7: &021.13 &-16.23 &1300: & 0.22 \\
Pegasus II&And VI & dSph &24.45 &-10.6 &106.01 &-36.30 & 830 & 0.83 \\
Pisces &LGS 3 & dIr/dSph &24.55 &-10.4 &126.77 &-40.88 & 810 & 0.71 \\
And III &... & dSph &24.4 &-10.2 &119.31 &-26.25 & 760 & 0.86 \\
And V &... & dSph &24.55 &-10.2 &126.22 &-15.12 & 810 & 0.87 \\
Leo II &... & dSph &21.6 &-10.1 &220.14 &+67.23 & 210 &-0.26 \\
Phoenix &... & dIr/dSph &23.0 & -9.8 &272.19 &-68.95 & 395 &-0.30 \\
Sculptor &... & dSph &19.7 & -9.8 &287.69 &-83.16 & 87 &-0.06 \\
Tucana &... & dSph &24.7 & -9.6 &322.91 &-47.37 & 870 &-0.44 \\
Cassiopeia&And VII & dSph &24.2 & -9.5 &109.46 &-09.95 & 690 & 0.98 \\
Sextans &... & dSph &19.7 & -9.5 &243.50 &+42.27 & 86 &-0.65 \\
Carina &... & dSph &20.0 & -9.4 &260.11 &-22.22 & 100 &-0.85 \\
Draco &... & dSph &19.5 & -8.6 &086.37 &+34.71 & 79 & 0.77 \\
Ursa Minor&... & dSph &19.0 & -8.5 &104.88 &+44.90 & 63 & 0.66 \\
\enddata
\tablecomments{Colons denote uncertain values.}
\tablenotetext{*} {Membership in Local Group not yet firmly established.}
\end{deluxetable}
\begin{deluxetable}{ccrrrr}
\scriptsize
\tablenum{2}
\tablewidth{0pt}
\tablecaption{Maximum Likelihood Fits to Local Group Data}
\tablehead{\colhead{M$_V$(br)} & \colhead{M$_V$(ft)} & \colhead{$\alpha$} &
\colhead{$\alpha$(95\%)} & \colhead{$\alpha$(99\%)} &
\colhead{\hfil P($\alpha<-1.3$)} }
\startdata
\multicolumn{6}{l}{(a) All morphological types} \\
-22 & -8 & -1.07 $\pm$ 0.05 & -0.98 & -0.94 & 0.00\% \\
-22 & -10 & -1.09 $\pm$ 0.07 & -0.96 & -0.91 & 0.50\% \\
-22 & -11 & -1.02 $\pm$ 0.08 & -0.88 & -0.83 & 0.18\% \\
-22 & -12 & -0.91 $\pm$ 0.10 & -0.74 & -0.66 & 0.02\% \\
-18 & -8 & -1.10 $\pm$ 0.07 & -0.99 & -0.93 & 0.40\% \\
-18 & -10 & -1.17 $\pm$ 0.10 & -0.99 & -0.94 & 11.31\% \\
-18 & -11 & -1.09 $\pm$ 0.13 & -0.87 & -0.77 & 6.52\% \\
-18 & -12 & -0.92 $\pm$ 0.18 & -0.61 & -0.45 & 2.13\% \\
-16 & -8 & -1.09 $\pm$ 0.09 & -0.94 & -0.88 & 1.82\% \\
-16 & -10 & -1.21 $\pm$ 0.14 & -0.95 & -0.87 & 30.11\% \\
-16 & -11 & -1.06 $\pm$ 0.20 & -0.72 & -0.57 & 15.15\% \\
-16 & -12 & -0.62 $\pm$ 0.35 & 0.05 & 0.32 & 2.38\% \\
-15 & -8 & -1.18 $\pm$ 0.11 & -0.98 & -0.92 & 17.93\% \\
-15 & -10 & -1.50 $\pm$ 0.21 & -1.14 & -1.01 & 85.87\% \\
-15 & -11 & -1.48 $\pm$ 0.33 & -0.94 & -0.71 & 73.27\% \\
\multicolumn{6}{l}{(b) Sph/dSph data only} \\
-17 & -9 & -1.33 $\pm$ 0.12 & -1.13 & -1.05 & 62.53\% \\
-17 & -10 & -1.31 $\pm$ 0.16 & -1.04 & -0.94 & 58.55\% \\
-17 & -11 & -1.11 $\pm$ 0.22 & -0.73 & -0.57 & 23.66\% \\
-15 & -9 & -1.52 $\pm$ 0.18 & -1.24 & -1.12 & 90.72\% \\
-15 & -10 & -1.66 $\pm$ 0.29 & -1.23 & -1.05 & 92.01\% \\
-15 & -11 & -1.44 $\pm$ 0.46 & -0.72 & -0.42 & 66.92\% \\
\multicolumn{6}{l}{(c) Ir/dIr data only} \\
-16 & -9 & -1.02 $\pm$ 0.17 & -0.70 & -0.58 & 7.90\% \\
-16 & -10 & -1.09 $\pm$ 0.22 & -0.72 & -0.56 & 20.69\% \\
-16 & -11 & -1.01 $\pm$ 0.31 & -0.48 & -0.26 & 20.49\% \\
-15 & -10 & -1.40 $\pm$ 0.33 & -0.90 & -0.66 & 66.20\% \\
\enddata
\tablecomments{M$_V$(br) and M$_V$(ft) give the range of absolute magnitude
over which the fit was done. The error in $\alpha$ corresponds to a 1$\sigma$
error. $\alpha$(95\%) and $\alpha$(99\%) are the values for which
the probability is 95\% (99\%) that the true value of $\alpha$ lies below
the tabulated values. P($\alpha<-1.3$) is the probability that that
$\alpha < -1.3$.}
\end{deluxetable}
|
\section{Introduction}
\label{IntroSection}
Although both gravity and gauge theories contain a local symmetry,
they have rather disparate physical properties. QCD, for example, is
a confining theory while gravity is not. Similarly, in four dimensions
QCD is renormalizable while field theories of gravity are
non-renormalizable. Nonetheless, within the context of the
perturbative expansions some remarkable `squaring' relations exist
between the tree-level $S$-matrix elements of gravity and gauge
theories in both string and field
theories~\cite{KLT,BGK,BDDPR,AllPlusGravity,MHVGravity}. These
squaring relations imply that gauge symmetry is more closely related
to the diffeomorphism invariance of gravity than one might suspect
based on the respective Lagrangians. Recently, there has also been a
resurgence of interest in perturbative gravity both in anti-de Sitter
spaces~\cite{AdS} and in phenomenological applications of large
compact dimensions~\cite{Dimopoulos}, further motivating
an investigation of these relations.
In this letter we will show that one can construct a low energy
Lagrangian for gravity directly from QCD $S$-matrix elements and then
discuss how this Lagrangian relates to the usual Einstein-Hilbert
Lagrangian. The starting point of our investigation is the Kawai,
Lewellen and Tye (KLT) relations expressing closed string tree
amplitudes in terms of open string amplitudes~\cite{KLT}. These
relations arise from the property that the integrand of a closed
string amplitude is composed of left- and right-mover open string
components (see e.g.~ref.~\cite{GSW}). In the low energy limit where
string theory reduces to field theory, the KLT relations express
gravity tree amplitudes in terms of a sum of products of `left' gauge
theory amplitudes and `right' gauge theory amplitudes.
The KLT relations were first exploited by Berends, Giele and Kuijf to
obtain an infinite sequence of maximally helicity violating pure
gravity tree amplitudes~\cite{BGK} using known gauge theory
results. By unitarity, tree-level relations necessarily imply that
loop amplitudes in the two theories are also related. Indeed, the
method of constructing $S$-matrix elements via their analytic
properties (see refs.~\cite{Review}) provides a means for
exploiting this to obtain loop-level (super) gravity amplitudes.
Using these ideas, the divergence structure of $N=8$ supergravity has
been investigated with the result that it appears to be less divergent
than previously thought~\cite{BDDPR}. Two infinite sequences of
maximally helicity violating one-loop amplitudes in gravity theories
have also been constructed~\cite{MHVGravity}. These
may be viewed as results in an effective field theory of gravity
which necessarily must match the low energy limit of a more
fundamental theory, such as string or $M$ theory. The ability to use
the squaring relation between gravity and gauge theory to perform
non-trivial gravity computations suggests that one can develop a
deeper understanding of perturbative gravity by exploring this
relationship.
{}From the point of view of the field theory Lagrangians, the KLT
relations are rather mysterious: the Einstein-Hilbert Lagrangian does
not factorize in any obvious way in terms of the Yang-Mills
Lagrangian. It is not even completely clear what the notion of `left' and
`right' parts of the theory mean given that the graviton is a
symmetric tensor. Another obvious difficulty is that the Yang-Mills
Lagrangian has only three- and four-point interactions while the
Einstein-Hilbert Lagrangian has an infinite set of interactions.
Nevertheless, the gravity $S$-matrix does have the property that it is
composed of products of two gauge theory amplitudes and that a
graviton, very roughly speaking, is composed of `left' and `right'
gauge fields, i.e., $h_{\mu\nu} \sim {\bar A}_\mu A_\nu$. When one
has a property of the $S$-matrix, a natural idea is to organize the
underlying formalism so that it reflects that property. As a well
known example, the general superiority of Feynman diagrams
as compared to time ordered perturbation theory follows from their
preservation of manifest Lorentz symmetry. Here we wish to rearrange
the gravity Lagrangian so it captures properties associated with the
factorization of the $S$-matrix into sums of products of gauge theory
amplitudes.
The analysis of this letter will consist of two parts. In the first
part we will systematically construct a gravity Lagrangian from QCD
gluon amplitudes using the KLT relations. We explicitly carry out the
procedure through five graviton interactions; in principle, the
procedure can be carried out to arbitrarily high orders, although as a
practical matter the complexity of the computations increases rapidly
with the number of gravitons. Nevertheless, the procedure shows that
a Lagrangian for gravity can be obtained using QCD amplitudes. The
gravity Lagrangian obtained in this way has the properties that (a) by
construction it produces the correct tree-level amplitudes and (b)
graviton Lorentz indices associated with the `left' gauge theory do
not contract with ones associated with the `right' gauge theory. A
Lagrangian with the property that the associated Feynman diagrams
factorize into left and right parts has been presented previously by
Siegel~\cite{Siegel}. A spin-off from our analysis is that it can be
used to provide a simpler set of gravity Feynman rules than the
conventional~\cite{DeWitt} de Donder gauge rules. Moreover, in our
construction, the gravity three vertex is given directly as sums of
squares of gauge theory vertices.
In the second part, we will find a set of field variables which allows
us to interpret the gravity Lagrangian obtained from QCD in terms of
the conventional Einstein-Hilbert Lagrangian. Non-linear field
redefinitions and gauge fixings have been used previously by van de
Ven~\cite{vandeVen} to simplify the computation of two-loop
divergences in gravity. Here we wish to consider such field redefinitions
and gauge fixings in order to find a particular set of
field variables which helps clarify the connection to QCD. This
construction indicates that there is a closer correspondence between
ordinary gauge invariance and the diffeomorphism invariance of
Einstein gravity than one might have suspected.
\section{Review of the KLT relations and applications}
A basic property of integrands for closed string amplitudes is that
they factorize into open string integrands (except
for the momentum zero mode). At tree-level, KLT~\cite{KLT} used this
property to find simple relationships expressing closed string
amplitudes in terms of products of open string amplitudes. In the
infinite string tension limit where string theory reduces to an
effective field theory, for the four- and five-point amplitudes these
relations are,
$$
\eqalign{
{\cal M}_4^{\rm tree} (1,2,3,4) & =
- i \Bigl({\kappa \over 2} \Bigr)^2
s_{12} A_4^{\rm tree} (1,2,3,4) \, A_4^{\rm tree}(1,2,4,3)\,, \cr
{\cal M}_5^{\rm tree}
(1,2,3,4,5) & = i \Bigl({\kappa\over 2} \Bigr)^3 \Bigl[
s_{12} s_{34} A_5^{\rm tree}(1,2,3,4,5)
A_5^{\rm tree}(2,1,4,3,5) \cr
& \hskip 1.5 cm
+ s_{13}s_{24} A_5^{\rm tree}(1,3,2,4,5) \,
A_5^{\rm tree}(3,1,4,2,5) \Bigr]\,, \cr}
\equn\label{KLTExamples}
$$
where, $\kappa^2 = 32 \pi G_N$, $s_{ij} = (k_i + k_j)^2$, the $A_n$
are color-ordered gauge theory partial amplitudes~\cite{ManganoReview}
and the ${\cal M}_n$ are gravity amplitudes. Color does not appear in
these relations because the gauge theory partial amplitudes are
independent of color. The arguments of the amplitudes label the
external legs. These relations hold for any particle states appearing
in any closed string theory since they follow from the basic
factorization of the string integrands into `left' and `right'
sectors. In particular, they hold in pure Einstein gravity, gravity
coupled to a dilaton, anti-symmetric tensor or gauge field, and in
supergravity theories. Moreover, these KLT relations generalize to an
arbitrary number of external legs. (Explicit expressions may be found
in Appendix~A of ref.~\cite{MHVGravity}.) A basic property of the KLT
formulas is that any Lorentz indices associated with the `left' gauge
theory amplitude do not contract with the Lorentz indices of the
`right' gauge theory amplitude.
The KLT relations give tree-level gravity amplitudes directly from
known gauge theory amplitudes, which are generally far easier to
compute. This was first applied by Berends, Giele and
Kuijf~\cite{BGK} to obtain an infinite sequence of maximally helicity
violating $n$-point Einstein gravity tree amplitudes using known
QCD gluon amplitudes. Beyond tree-level, one may exploit the KLT
relations using the cutting method developed for performing QCD amplitude
calculations of phenomenological interest~\cite{Review}. With this
method one can reconstruct complete loop amplitudes (in a given
dimensional regularization scheme) from their $D$-dimensional
unitarity cuts. The one-loop amplitudes are specified in terms of
tree amplitudes which satisfy the KLT relations; one can then obtain
higher loop amplitudes by iterating the procedure. In this way, one
can obtain loop-level (super) gravity $S$-matrix elements without
reference to a Lagrangian or Feynman
rules~\cite{BDDPR,AllPlusGravity,MHVGravity}. The efficiency of the
computational method follows from the fact that one is {\it recycling}
previously performed gauge theory calculations to obtain new
results in gravity theories.
\begin{figure}[ht]
\centerline{\epsfxsize 1.9 truein \epsfbox{TwoParticle.eps}}
\vskip -.3 cm
\caption[]{
\label{TwoParticleFigure}
\small One can recycle tree amplitudes into loop amplitudes via
$D$-dimensional unitarity cuts.}
\end{figure}
For example, consider the unitarity cuts of a one-loop four-point amplitude in
any gravity theory coming from the low energy limit of a string
theory. As depicted in \fig{TwoParticleFigure}, a gravity tree
amplitude appears on each side of the cut. Thus, on the cut one must
evaluate the product,
$$
\sum_{\rm gravity \atop states}
{\cal M}_4^{\rm tree} (1, 2, \ell_2, -\ell_1) \times
{\cal M}_4^{\rm tree} (3,4, \ell_1, -\ell_2) \,,
\equn
$$
summed over all states in the gravity theory that can cross the cut.
From the KLT relations (\ref{KLTExamples}) we may re-express this
product in terms of gauge theory amplitudes,
$$
\eqalign{
&\sum_{\rm gauge\atop states}
s_{12} \, A_4^{\rm tree} (1, 2, \ell_2, -\ell_1) \times
A_4^{\rm tree} (3,4, \ell_1, -\ell_2) \cr
\times \null &
\sum_{\rm gauge\atop states}
s_{12} \, A_4^{\rm tree} (1,2, \ell_1, -\ell_2) \times
A_4^{\rm tree} (3,4, -\ell_2,\ell_1) \,.\cr}
\equn
$$
The sum over the gravity theory states necessarily factorize into a
sum over the states of two gauge theories because the underlying
string theory has this property. This is a generic property
holding for all cuts.
In this way, one can re-express cut gravity
amplitudes in terms of gauge theory ones. This has led to the
construction of a number of non-trivial gravity loop
amplitudes~\cite{BDDPR,AllPlusGravity,MHVGravity}.
\section{Construction of a gravity Lagrangian from QCD}
We now discuss a procedure for constructing an off-shell low energy
Lagrangian for pure gravity starting from QCD gluon amplitudes.
We explicitly carry this out for up to five graviton interactions. By
construction, the Lagrangian that we obtain is equivalent to the
Einstein-Hilbert Lagrangian in that it produces identical
tree-level $S$-matrix elements.
In field theory one usually takes a given Lagrangian and constructs
Feynman rules that can then be used to obtain the $S$-matrix elements.
Here we wish to reverse this process, since from the KLT relations we
have the gravity $S$-matrix in terms of QCD amplitudes, but wish
instead to obtain a Lagrangian that preserves the `left'-`right'
factorization of Lorentz indices.
To obtain the $n$-graviton term in the Lagrangian we start with the
tree-level $n$-graviton amplitudes, given in terms of the QCD
amplitudes, and subtract all diagrams containing up to
$(n-1)$-graviton interaction vertices. By iterating this procedure
we can systematically build a gravity Lagrangian. However, to
jump-start this process we first need a propagator and three
vertex.
Consider the graviton propagator. The standard de Donder gauge
propagator,
$$
P_{\mu\nu; \alpha\beta} (k) = \langle h_{\mu \alpha} h_{\nu\beta} \rangle_0
= {1 \over 2}
\Bigl[\eta_{\mu\nu} \eta_{\alpha\beta} + \eta_{\mu\beta} \eta_{\nu\alpha}
- {2\over D-2} \eta_{\mu\alpha} \eta_{\nu\beta} \Bigr] {i\over k^2 + i\epsilon}
\,,
\equn\label{deDonderPropagator}
$$
is unacceptable since it contracts `left' and `right' indices; for example,
the last term contracts $\mu$ with $\alpha$. Moreover, it contains
explicit dependence on the dimension $D$, which must somehow cancel from
the tree-level $S$-matrix elements since there is no such dependence in
the KLT relations or in the gauge theory amplitudes.
(We have chosen $\eta_{\mu\nu}$ to have signature $(+,-,-,-)$.)
A better propagator is~\cite{BDS}
$$
P_{\mu\nu; \alpha\beta} (k) =
{i\over 2}
{\eta_{\mu\nu} \eta_{\alpha\beta} + \eta_{\mu\beta} \eta_{\nu\alpha}
\over k^2 + i \epsilon} \,.
\equn\label{SymmetricPropagator}
$$
However, even this propagator is unacceptable because of the second
term, which contracts a left index with a right one. To prevent
such contractions, we wish to use instead the propagator,
$$
P_{\mu\nu; \alpha\beta} (k) =
\eta_{\mu\nu} \eta_{\alpha\beta}\, {i \over k^2 + i \epsilon} \, .
\equn\label{NewPropagator}
$$
In a sense, for each graviton $h_{\mu\alpha}$ we assign the index
$\mu$ to be a `left' index and $\alpha$ to be a `right' index. Of
course, since $h_{\mu\alpha}$ is symmetric it does not matter which
index is assigned to the left and which to the right, but once the
choice is made we demand that left indices never contract with right
ones. At first sight the gravity propagator (\ref{NewPropagator})
might seem rather peculiar since it appears to violate the fundamental
property of the graviton being a symmetric tensor field.
Nevertheless, it is not difficult to show that one may use the
propagator~(\ref{NewPropagator}) instead
of~(\ref{SymmetricPropagator}) with no effect on the $n$-graviton
$S$-matrix elements, provided that all vertices satisfy a {\it rigid}
left--right interchange symmetry. By a rigid symmetry we mean that
each $n$-point vertex be symmetric under a simultaneous interchange of
all left and right Lorentz indices,
$$
V_n^{\mu_1 \mu_2 \cdots \mu_n; \alpha_1 \alpha_2 \cdots \alpha_n}
(k_1, k_2, \ldots , k_n)
= V_n^{\alpha_1 \alpha_2 \cdots \alpha_n; \mu_1 \mu_2 \cdots \mu_n}
(k_1, k_2, \ldots , k_n) \,,
\equn
$$
where the $\mu_i$ are the left indices associated with each
graviton and the $\alpha_i$ are the right indices.
With this requirement, one can show that there is no need to
symmetrize the graviton propagator in the left and right indices.
Consider, for example, the diagram in \fig{PropagFigure}. The
external legs of the diagrams do not need to be explicitly symmetrized
since the external graviton polarization tensors automatically symmetrize the
legs. The interchange of the $\mu$ and $\alpha$ indices can be undone
by performing a rigid interchange of left and right indices on the
left-most vertex. Although the rigid interchange will also flip the indices on
the external legs, this can be undone since the indices of each
external leg are contracted with a symmetric tensor polarization. The
same type of argument works for general diagrams. Thus, the propagator
(\ref{NewPropagator}) can be used instead of the propagator
(\ref{SymmetricPropagator}) since the symmetrization of indices
is automatically taken care of by the rigidly symmetrized vertices
and by the symmetric polarization tensors on the external legs.
\begin{figure}[ht]
\centerline{\epsfxsize 1.2 truein \epsfbox{Propag.eps}}
\vskip -.2 cm
\caption[]{
\label{PropagFigure}
\small A sample diagram for showing that we may use propagator
(\ref{NewPropagator}) instead of (\ref{SymmetricPropagator}).}
\end{figure}
Given the propagator (\ref{NewPropagator}) we must also choose a three
vertex. Any three vertex which agrees on shell with the
three-graviton vertex is suitable for our purposes. (Kinematic
restrictions prevent a massless particle from decaying into two
others, but in terms of formal polarization tensors the vertex does
not vanish; below we shall also obtain the same vertex from the
Einstein-Hilbert action.) From string theory (see
e.g. ref.~\cite{GSW}), the on-shell three-graviton vertex can be
expressed in terms of products of gauge theory vertices. This
motivates the choice of the three graviton vertex,
$$
\eqalign{
i G_3^{\mu\alpha, \nu\beta, \rho\gamma}(k, p , q) & = - {i \over 2}
\Bigl( {\kappa \over 2} \Bigr)
\Bigl[
V_{\rm GN}^{\mu\nu\rho}(k, p, q)
\times V_{\rm GN}^{\alpha \beta \gamma}(k, p, q) +
V_{\rm GN}^{\nu\mu\rho}(p, k, q) \times
V_{\rm GN}^{\beta \alpha \gamma}(p, k, q) \Bigr]\,, \cr }
\equn\label{GravityYMThreeVertex}
$$
where
$$
V^{\mu\nu\rho}_{\rm GN}(1,2,3) =
i\sqrt{2} \bigl( k_1^\rho \, \eta^{\mu\nu}
+ k_2^\mu \eta^{\nu\rho} + k_3^\nu \eta^{\rho\mu} \bigr)\,,
\equn
$$
is the color ordered Gervais-Neveu~\cite{GN} Yang-Mills three vertex,
from which the color factor has been stripped. Our main reasons for
using the vertex (\ref{GravityYMThreeVertex}) is its simplicity and
the fact that it makes the relationship to gauge theory manifest. By
construction, this vertex is symmetric under the rigid interchange of
left and right indices.
Armed with the propagator~(\ref{NewPropagator}) and the three vertex
(\ref{GravityYMThreeVertex}) we may then use the KLT relations to find
a gravity Lagrangian using QCD gluon amplitudes as input. At the first step
of the process one defines a four-vertex by subtracting from the
four-point $S$-matrix obtained via the KLT relations all diagrams
containing a kinematic pole, as shown in \fig{BootStrapFigure}. This
four vertex automatically has the property that left Lorentz indices
do not contract with right ones, since the gravity $S$-matrix and
diagrams containing the three-point vertex
(\ref{GravityYMThreeVertex}) have this property. Moreover, the vertex
defined in this way also has the rigid symmetry under an interchange
of left and right indices.
\begin{figure}[ht]
\centerline{\epsfxsize 3.8 truein \epsfbox{BootStrap.eps}}
\vskip -.2 cm
\caption[]{
\label{BootStrapFigure}
\small One can obtain a four vertex with the left-right
factorization property by subtracting the diagrams containing the
kinematic poles from the $S$-matrix.}
\end{figure}
We may then convert this four-vertex to an $h^4$ term in the
Lagrangian by inverting the usual procedure of obtaining Feynman
vertices from Lagrangians. Of course, there is some ambiguity in this
process since one can always add terms which vanish on shell. As long
as these terms do not mix left and right Lorentz indices and satisfy a
rigid left-right symmetry, which can always be imposed by hand, they
are acceptable terms; differences in the four-vertex then induce
differences in the deduced higher-point vertices.
Once the four-graviton terms in the Lagrangian have been chosen we can
then continue the process to obtain a five vertex by subtracting from
the five-graviton amplitude all diagrams containing three- and
four-point vertices. In principle, one can continue in this way to an
arbitrary number of external legs, allowing one to build a Lagrangian
for gravity order by order in perturbation theory using QCD gluon
amplitudes. By construction this Lagrangian has the property that it
generates all tree-level $S$-matrix elements and that it never
contracts left indices with right ones.
Starting from QCD gluon tree amplitudes, we have carried out the
construction for up to five points yielding the local
gravity Lagrangian $L = \sum_i L_i$,
where
$$
\eqalign{
L_2 & = - {1\over 2} \, h_{\mu\nu} \partial^2 h_{\mu\nu}\,, \cr
L_3 & = \kappa \Bigl[{1\over 2}h_{\mu\nu} h_{\rho\sigma,\mu\nu} h_{\rho\sigma}
+ h_{\nu\mu} h_{\rho\mu, \sigma} h_{\rho\sigma, \nu} \Bigr]\,, \cr
L_4 & = -\kappa^2 \Bigl[
{1\over 32} h_{\mu\nu,\lambda} h_{\mu\nu,\lambda}
h_{\rho\sigma} h_{\rho\sigma}
+{1\over 2} h_{\mu \nu,\lambda}h_{\mu\rho}h_{\sigma\nu,\rho}
h_{\sigma\lambda}
+{1\over 8} h_{\mu \nu}h_{\mu \lambda,\rho}h_{\sigma \nu,\rho}
h_{\sigma \lambda}
-{1\over 4} h_{\mu\nu}h_{\mu\lambda}h_{\rho\sigma,\nu}
h_{\rho\sigma,\lambda}\cr
& \hskip 1.3 cm
-{1\over 4} h_{\mu\nu} h_{\lambda\rho}h_{\lambda\sigma,\rho}
h_{\mu\sigma,\nu}
+{1\over 16} h_{\mu\nu} h_{\mu\nu}h_{\lambda\rho}
h_{\lambda\rho,\sigma\sigma}
+{1\over 24} h_{\mu\nu} h_{\lambda\nu} h_{\lambda\rho}
h_{\mu\rho,\sigma\sigma}\Bigr] \,, \cr
L_5 & = \kappa^3 \Bigl[
{3\over 64} h_{\mu\nu} h_{\mu\nu} h_{\rho\sigma}
h_{\rho\sigma,\kappa\lambda} h_{\kappa\lambda}
-{1\over 12} h_{\mu\nu} h_{\mu\nu,\kappa} h_{\rho\sigma,\lambda}
h_{\rho\kappa} h_{\lambda\sigma}
-{1\over 48} h_{\mu\nu,\kappa} h_{\mu\nu} h_{\rho\sigma}
h_{\rho\kappa,\lambda} h_{\lambda\sigma} \cr
& \hskip 1.3 cm
-{1\over 12} h_{\mu\nu,\kappa} h_{\mu\nu,\lambda} h_{\rho\sigma}
h_{\rho\kappa} h_{\lambda\sigma}
-{3\over 32} h_{\mu\nu} h_{\mu\nu} h_{\rho\sigma,\kappa\lambda}
h_{\rho\kappa} h_{\lambda\sigma}
-{1\over 12} h_{\mu\nu} h_{\rho\nu,\mu} h_{\rho\sigma}
h_{\lambda\sigma,\kappa} h_{\lambda\kappa} \cr
& \hskip 1.3 cm
-{1\over 6} h_{\mu\nu} h_{\rho\nu} h_{\rho\sigma}
h_{\lambda\sigma,\mu\kappa} h_{\lambda\kappa}
+{1\over 6} h_{\mu\nu} h_{\rho\nu,\kappa} h_{\rho\sigma,\mu}
h_{\lambda\sigma} h_{\lambda\kappa}
-{1\over 6} h_{\mu\nu} h_{\rho\nu,\kappa} h_{\rho\sigma}
h_{\lambda\sigma,\mu} h_{\lambda\kappa} \cr
& \hskip 1.3 cm
+{1\over 12} h_{\mu\nu} h_{\rho\nu} h_{\rho\sigma,\mu\kappa}
h_{\lambda\sigma} h_{\lambda\kappa}
+{1\over 8} h_{\mu\nu} h_{\rho\nu,\lambda} h_{\rho\sigma}
h_{\mu\sigma,\kappa} h_{\lambda\kappa}
+{1\over 24} h_{\mu\nu} h_{\rho\nu,\lambda\kappa} h_{\rho\sigma}
h_{\mu\sigma} h_{\lambda\kappa} \Bigr]\,, \cr}
\equn
\label{GaugeLagrangian}
$$
where $h_{\mu\nu, \lambda} = \partial_\lambda h_{\mu\nu}$.
In Minkowski space, one of any two contracted indices should be taken
to be a raised index using $\eta^{\mu\nu}$, but we have suppressed
this. In principle it might have been necessary to introduce
auxiliary fields for locality to hold; indeed, as discussed below, when
comparing this Lagrangian to the Einstein-Hilbert gravity Lagrangian
it is useful to introduce an auxiliary dilaton. Although the
Lagrangian (\ref{GaugeLagrangian}) is not unique since the terms can
be modified by adding or subtracting contributions that vanish on
shell and appropriately modifying the higher-point contributions, it
is a relatively simple one. More importantly, as we shall see below,
it allows for a relatively straightforward match to the conventional
Einstein-Hilbert Lagrangian. By adjusting the four-point terms in the
Lagrangian we have found a solution for $L_5$ that contains only six
terms, but then the connection to the Einstein-Hilbert Lagrangian is a
bit more complicated.
For compactness we have removed the rigid symmetrization between left
and right indices; however, if one uses the propagator
(\ref{NewPropagator}), each vertex must be rigidly symmetrized between
left and right indices, e.g.,
$$
L_3 \rightarrow {\kappa\over 2}
\Bigl[h_{\mu\nu} h_{\rho\sigma, \mu\nu} h_{\rho\sigma}
+ h_{\nu\mu} h_{\rho\mu, \sigma} h_{\rho\sigma, \nu}
+ h_{\mu\nu} h_{\mu\rho, \sigma} h_{\sigma\rho, \nu} \Bigr]\,.
\equn
$$
Although we do not explicitly give the generated Feynman rules here,
it is quite straightforward to obtain these. The advantages of using
Feynman rules generated by this Lagrangian instead of the conventional
de Donder gauge rule are clear: besides the fact that one can use the
propagator (\ref{NewPropagator}) instead of the more complicated de
Donder gauge propagator (\ref{deDonderPropagator}) the three-, four-
and five-point vertices are quite a bit simpler than the corresponding
de Donder gauge vertices. Once the interaction terms have been
rigidly symmetrized, when deriving the Feynman rules we can in a sense
treat $h_{\mu\nu}$ as a tensor field with no special index symmetry.
Note that the Feynman diagrams generated by this Lagrangian do not
contain explicit factors of $D$. In the conventional de Donder gauge
such factors do appear, but somehow cancel from the $S$-matrix
elements.
The relative simplicity of these Feynman rules is related to the
preservation of the $S$ matrix property that left and right Lorentz
indices should not contract with each other. Although the KLT
relations might appear to suggest that field variables exist where gravity
can be reformulated as a polynomial theory with no more than
four-point interactions as in the gauge theory case, we
have been unsuccessful in finding such a Lagrangian.
The Lagrangian (\ref{GaugeLagrangian}) does contain general coordinate
invariance although it is not manifest since the symmetry has been
gauge fixed. To make this explicit, we now relate the Lagrangian
(\ref{GaugeLagrangian}) to the usual Einstein-Hilbert one. In order
to do so we must first eliminate all terms in the Einstein-Hilbert
Lagrangian which contract left with right graviton indices, i.e,
$$
h_{\mu\mu}\, , \hskip 1 cm
h_{\mu\nu} h_{\nu\lambda} h_{\lambda\mu} \,, \hskip 1 cm
\cdots, \hskip 1 cm \, {\rm Tr}[h^{2m+1}] \,,
\equn\label{BadTerms}
$$
where $\, {\rm Tr}[h^n] \equiv h_{\mu_1 \mu_2} h_{\mu_2 \mu_3} \cdots h_{\mu_n \mu_1}$.
Because of the way that these types of terms are tangled in the
Einstein-Hilbert Lagrangian, it is not obvious how one can accomplish
this. Nevertheless, the existence of the Lagrangian
(\ref{GaugeLagrangian}) implies that there must be some rearrangement
with the desired property.
In ref.~\cite{BDS} the dilaton was introduced to
allow for a field redefinition which removes the graviton trace from
the quadratic terms in the Lagrangian. The appearance of the dilaton
as an auxiliary field to help rearrange the Lagrangian is motivated by
string theory which requires the presence of such a field. Following
the discussion of ref.~\cite{BDS}, we then consider the Lagrangian for
gravity coupled to a dilaton,
$$
L^{\rm EH} = {2\over \kappa^2} \sqrt{-g} \,
R + \sqrt{-g} \, \partial^\mu\phi\partial_\mu\phi \,.
\equn
$$
(Our conventions are those of Weinberg~\cite{Weinberg}, except that
we have flipped the signature of the metric, $g_{\mu\nu} \rightarrow
- g_{\mu\nu}$; this then induces an overall sign flip in the Lagrangian.)
In de Donder gauge, for example, taking $g_{\mu\nu} = \eta_{\mu\nu} +
\kappa h_{\mu\nu}$, the quadratic part of the Lagrangian is
$$
L_2 =
- \frac{1}{2} h_{\mu\nu} \partial^2 h_{\mu\nu}
+ \frac{1}{4} h_{\mu\mu} \partial^2 h_{\nu\nu}
- \phi \partial^2 \phi \,.
\equn
$$
The term involving $h_{\mu\mu}$ can be eliminated with the
simultaneous field redefinitions~\cite{BDS},
$$
h_{\mu\nu} \rightarrow h_{\mu\nu}
+ \eta_{\mu\nu}{\sqrt{\frac{2}{D-2}}}\, \phi\,,
\hskip 2 cm
\phi \rightarrow \frac{1}{2} h_{\mu\mu} + \sqrt{\frac{D-2}{2}}\, \phi\,,
\equn\label{FieldRedef}
$$
so that the Lagrangian reduces to
$$
L_2 \rightarrow
- \frac{1}{2} h_{\mu\nu} \partial^2 h_{\mu\nu}
+ \phi \partial^2 \phi\,.
\equn\label{TwoLagrange}
$$
(This field redefinition is a bit different than the one in ref.~\cite{BDS}
because we have chosen to normalize the dilaton kinetic term differently
so as to slightly simplify the induced interaction terms discussed below.)
For the case of purely graviton external states the dilaton will not
contribute to the tree-level $S$-matrix because of the selection rule
that $\phi$ must be created or annihilated in pairs. Of course, if
the dilaton, or any matter fields
appearing in the theory, are taken as external states they can
propagate in intermediate states. In this case, one would need to
specify the precise matter content of the theory before proceeding
with the analysis. Nevertheless, the fundamental factorization of the
$S$-matrix implied by the KLT relations would still hold, since it
follows from the underlying string theory.
Here we wish to generalize the field redefinitions (\ref{FieldRedef})
to all orders in $\kappa$ so as to remove all terms of the form
(\ref{BadTerms}) from the Lagrangian. Our generalization for the case of
no gauge fixing is
$$
g_{\mu\nu} = e^{\sqrt{\frac{2}{D-2}} \kappa \phi}
e^{\kappa\, h_{\mu\nu}} \equiv
e^{\sqrt{\frac{2}{D-2}} \,\kappa \phi}
\Bigl(\eta_{\mu\nu} + \kappa h_{\mu\nu} + {\textstyle \kappa^2\over 2}
h_{\mu\rho} h_{\rho\nu} + \cdots \Bigr) \,,
\equn\label{StringVars}
$$
where $h_{\mu\nu}$ is the graviton field,
followed by the change of variables
$$
\phi \rightarrow -\sqrt\frac{2}{D-2} \,
\Bigl( \phi + \frac{1}{2} h_{\rho\rho}\Bigr) \,.
\equn\label{PhiRedef}
$$
We have verified through ${\cal O}(h^6)$ that this choice eliminates all
terms (\ref{BadTerms}) which mix left and right Lorentz indices,
even before fixing the gauge.
As yet we have not performed
any gauge fixing so the action is generally coordinate invariant, even
if the choice of field variables obscures this.
One might be concerned that the field redefinition (\ref{StringVars})
would alter the gravity $S$-matrix. However, the $S$-matrix is
guaranteed to be invariant under non-linear field redefinitions or
under linear ones that do not alter the coupling to the external
traceless polarization tensors. Indeed, our explicit calculations
respect this property, as required.
An important remaining question is how one can choose a gauge fixing
so that the terms in the Einstein-Hilbert Lagrangian resemble the
terms of the Lagrangian (\ref{GaugeLagrangian}) deduced from the QCD
amplitudes. We have found a solution, which is to replace the field
redefinition (\ref{PhiRedef}) with a non-linear generalization,
$$
\phi \rightarrow - \sqrt\frac{2}{D-2}\biggl[
\biggl( \phi + \frac{1}{2} {\rm Tr}\, h \biggr) +
\kappa \biggl( \frac{1}{4} \phi^2 - \frac{1}{8} {\rm Tr}\,(h^2) \biggr)+
\kappa^2 \biggl( \frac{1}{12} \phi^3 - \frac{1}{8} \phi\, {\rm Tr}\,(h^2)
\biggr) +\cdots\biggr] \,.
\equn\label{PhiRedefB}
$$
Then we add a gauge fixing term to the Lagrangian, $(F_\mu)^2$ following
the standard procedure, where
$$
\eqalign{
F_\mu & = \biggl( h_{\mu\nu,\nu} + \phi,_\mu \biggr)
+ \kappa \biggl( -\frac{1}{4} {\rm Tr}\,(h^2)_{,\mu}
-\frac{1}{2} \phi h_{\mu\nu,\nu} - h_{\mu\nu} \phi,_\nu \biggr)
+ \kappa^2 \biggl( \frac{1}{16} {\rm Tr}\,(h^2) h_{\mu\nu,\nu}
+\frac{1}{8} {\rm Tr}\,(h^2)_{,\nu} h_{\mu\nu}\cr
& \hskip 3 cm
-\frac{1}{12} h_{\mu\nu} h_{\lambda\nu,\rho} h_{\lambda\rho}
+\frac{1}{6} h_{\mu\nu,\rho} h_{\lambda\nu} h_{\lambda\rho}
+\frac{1}{24} h_{\mu\nu} h_{\lambda\nu} h_{\lambda\rho,\rho}
- \frac{1}{8} ( \phi\,{\rm Tr}\,(h^2) )_{,\mu} \biggr) +\cdots\,. \cr}
\equn
$$
With these choices we then obtain the desired form of the Lagrangian
through ${\cal O}(h^4)$, which we express in terms of the Lagrangian
(\ref{GaugeLagrangian}) derived from QCD amplitudes,
$$
\eqalign{
L_2^{\rm EH} & = L_2 + \phi \partial^2 \phi \,, \cr
L_3^{\rm EH} & = L_3 -
{\kappa\over 2} h_{\mu\nu,\kappa} h_{\mu\nu,\kappa} \phi \,, \cr
L_4^{\rm EH} & = L_4
+ \kappa^2 \Bigl[ {1\over 32} h_{\mu\nu,\lambda} h_{\mu\nu,\lambda}
h_{\rho\sigma} h_{\rho\sigma}
+{1\over 2} \phi h_{\mu\nu} h_{\lambda\sigma,\mu} h_{\lambda\sigma,\nu}
-{1\over 4}\phi_{,\mu} h_{\mu\nu} h_{\lambda\nu} h_{\lambda\sigma,\sigma}
+{1\over 2}\phi_{,\mu} h_{\mu\nu} h_{\lambda\sigma,\nu} h_{\lambda\sigma} \cr
& \hskip 1.5 cm
-{1\over 2}\phi_{,\mu} h_{\mu\nu} h_{\lambda\nu,\sigma} h_{\lambda\sigma}
-\phi h_{\mu\nu} h_{\mu\lambda,\sigma} h_{\sigma\lambda,\nu}
-{1\over 8} \phi_{,\mu} h_{\mu\rho,\rho} h_{\lambda\sigma} h_{\lambda\sigma}
+{1\over 8} \phi^2 h_{\mu\nu,\lambda} h_{\mu\nu,\lambda} \Bigr] \,.\cr}
\equn
$$
For the case of pure graviton external states we may integrate out
$\phi$ from the path integral. Equivalently, we may substitute the
equation of motion for $\phi$,
$$
\phi = {\kappa\over 4} \, {1\over \partial^2}
(h_{\mu\nu,\kappa} h_{\mu\nu,\kappa}) + \cdots
\equn
$$
into the Lagrangian. For the two- and three-graviton terms in the
Lagrangian, this gives exactly the terms in the Lagrangian
(\ref{GaugeLagrangian}). For the four-point we obtain exactly $L_4$
plus a non-local piece that vanishes for on-shell gravitons,
$$
\Delta L_4 = {\kappa^2 \over 16} h_{\mu\nu,\kappa} h_{\mu\nu,\kappa}
\, {1\over \partial^2} \, (h_{\rho\sigma, \lambda\lambda} h_{\rho\sigma})
\equn
$$
Since the four-point Lagrangians differ a bit off-shell,
at the five-point level, terms that are non-zero on-shell need to
be added to $L_5$,
$$
\Delta L_5\Bigr|_{\rm on\ shell} = \kappa^3 \Bigl[
{1\over 64} h_{\mu\nu} h_{\mu\nu} h_{\rho\sigma}
h_{\rho\sigma,\kappa\lambda} h_{\kappa\lambda}
-{1\over 16} h_{\mu\nu,\kappa} h_{\mu\nu} h_{\rho\sigma}
h_{\rho\kappa,\lambda} h_{\lambda\sigma}
-{1\over 32} h_{\mu\nu} h_{\mu\nu} h_{\rho\sigma,\kappa\lambda}
h_{\rho\kappa} h_{\lambda\sigma} \Bigr] \,.
\equn
$$
In a sense we may take $L^{\rm EH}$ as the more fundamental one since
it comes directly from the Einstein-Hilbert Lagrangian, while the
Lagrangian in \eqn{GaugeLagrangian} provides the guidance needed to
obtain it. This shows the connection of the Lagrangian obtained from
QCD (\ref{GaugeLagrangian}) and the Einstein-Hilbert Lagrangian for a
very particular set of field variables and and gauge fixing.
Although the two Lagrangians differ somewhat even after
integrating out the auxiliary dilaton, we have explicitly shown that
for up to five gravitons they generate the same on-shell scattering
amplitudes.
Presumably, similar results can be obtained without introducing the
dilaton; nevertheless, we found it useful for clarifying the required
reorganization of the Einstein-Hilbert Lagrangian.
Other reorganizations based on introducing
other fields such as an anti-symmetric tensor, which
is also motivated by string theory, are also possible~\cite{Siegel}.
At loop level there are also ghost contributions that can be
obtained via the usual methods. (The Jacobian generated by the field
redefinition is trivial, at least in perturbation theory, since it is
a point transformation.) Additionally the auxiliary dilaton would
propagate in the loops, which would then need to be subtracted (see
e.g., ref.~\cite{BDS}). At loop level the unitarity method advocated
in refs.~\cite{Review,BDDPR,AllPlusGravity,MHVGravity} is, however, an
efficient way to obtain the $S$-matrix without the need for Feynman
rules.
\section{Sample applications}
As one simple application, one may obtain the soft factor for gravity
amplitudes directly from the soft behavior of QCD amplitudes. Gravity
tree amplitudes have the well known behavior~\cite{WeinbergSoftG},
$$
{\cal M}_n^{\rm tree}(1,2,\ldots,n^+)\ \mathop{\longrightarrow}^{k_n\to0}\
{\kappa\over2} \mathop {\cal S} \nolimits^{\rm gravity}(n^+) \times\
{\cal M}_{n-1}^{\rm tree}(1,2,\ldots,n-1) \,,
\equn\label{GravTreeSoft}
$$
as the momentum of graviton $n$, which we have taken to carry positive
helicity, becomes soft. Using the three-graviton vertex
(\ref{GravityYMThreeVertex}) which is expressed in terms of the QCD
three-gluon vertex,
the gravity soft factor can then be expressed in terms of the QCD soft factor:
$$
\mathop {\cal S} \nolimits^{\rm gravity}(n^+) = \sum_{i=1}^{n-1} s_{ni}
\mathop {\cal S} \nolimits^{\rm QCD}(q_l, n^+, i) \times
\mathop {\cal S} \nolimits^{\rm QCD}(q_r, n^+, i) \,,
\equn\label{FinalSoftGrav}
$$
where $\mathop {\cal S} \nolimits^{\rm QCD}(q, n^+, i) = \spa{q}.i/\spa{q}.{n}\spa{n}.i$
is the eikonal factor for a positive helicity soft gluon in QCD. The
$\spa{i}.j= \langle i^- | j^+\rangle$ are spinor inner products and
$|i^{\pm}\rangle$ are massless Weyl spinors of momentum $k_i$, labeled
with the sign of the helicity (see e.g.,~\cite{ManganoReview}).
Although not manifest, the soft factor (\ref{FinalSoftGrav}) is
independent of the choices of null helicity `reference momenta' $q_l$ and
$q_r$. By choosing $q_l = k_1$ and $q_r = k_{n-1}$ we recover the
form of the soft graviton factor for $k_n \rightarrow 0$ used
in, for example, refs.~\cite{BGK,AllPlusGravity,MHVGravity}.
As a less trivial example, the explicit factorization of the left and
right indices of the interaction vertices would imply
that the all-plus helicity graviton current satisfying
eqs.~(B.10) and (B.11) of ref.~\cite{MHVGravity} do actually
follow from Einstein gravity. In ref.~\cite{MHVGravity} the hitch in
obtaining these recursion relations from Einstein gravity was the
different choices of helicity reference momenta on the left and on the
right. (Although, not derived directly from Einstein gravity these
recursion relations were useful for defining `half-soft' functions
which serve as building blocks for one-loop gravity amplitudes.) In
this case, $\varepsilon_+^{\mu\alpha} \varepsilon_+^{\beta\mu} \not = 0$, which
prevented a derivation of the recursion relations from Einstein
gravity. With the factorization of left and right Lorentz indices, such
contractions simply do not occur.
\section{Discussion}
Since the tree-level $S$-matrix is generated by solving perturbatively
the classical equations of motion, one might suspect that it is
possible to relate more general solutions of the classical equations
of motion for gravity to gauge theory solutions. The property that
the $S$-matrix can be expressed in terms of `left' and `right' sectors
is generic in string theory. For this reason, it should be possible
to extend the discussion in this letter to theories containing, for
example, the anti-symmetric tensor or to supergravity theories. It
also seems reasonable that gravity theories in curved spaces
can be reformulated to express tree-level
amplitudes in terms of gauge theory ones. Moreover,
the intimate connection of perturbative gravity amplitudes with gauge
theory ones suggests a closer connection of diffeomorphism invariance
with non-abelian gauge symmetry than one might have suspected. We
feel that these issues deserve further attention.
\vskip .2 cm
We thank S. Cherkis, L. Dixon and J. Schwarz for helpful discussions.
|
\section{Introduction}
During the recent years branes were recognized
as an important ingredient
in description of the nonperturbative
sector of string and field theories.
Essentially speaking the branes are
multi-dimensional objects having tension
and bearing some quantum numbers
which can be considered as the
higher-dimensional generalizations of the
electric and magnetic charges (see for instance \cite{polchinski}
as a review).
In general the branes are very
natural objects to study, since they are natural generalizations of
the
familiar pointlike particles. Including them into the string
theory considerations has enriched the string
physics as much as Maxwell
theory physics is enriched by adding charges and monopoles.
In the context of string theory the branes are mainly considered as
background for string dynamics.
Some dynamical questions of brane
physics have also been studied in the
classical approximation. Consideration
of the Schwinger-type processes in
external fields involves the simplest
quasiclassical behavior and the paper
is devoted to their analysis.
The spontaneous Schwinger-type processes have previously been
considered in \cite{Teitelboim},
\cite{Porrati}, \cite{dggh}, \cite{emparan},
\cite{MS}
solutions to the equations of motion of Born-Infeld (BI)
brane action which support the string like configuration, representing
electrically or magnetically charged
pointlike object on the brane worldvolume,
have been discussed in \cite{cm}
\cite{gibons}, \cite{Savvidy}.
Recently similar solutions involving string junctions were
found in \cite{ga}.
The basic example of the brane tunneling phenomena
was analyzed in \cite{Teitelboim} where the brane
creation in the external RR field has been described.
The RR fields couple to the higher dimensional
generalization of electric charges and like the point particle
Lagrangian includes a term involving the potential one form
integrated over the
worldline of the particle, the $p$-brane action includes the RR
(or NS) $p+1$-form integrated over $p+1$-dimensional worldvolume of
the
brane; a coefficient in front of this term is the higher dimensional
generalization of the electric charge.
We remind that the Schwinger's pair production in the
external electric field is intuitively described as virtual
charge anticharge pair being accelerated by the
external field in the classically
forbidden region until their energy reaches their mass threshold.
The acceleration in the classically
forbidden region is neatly formulated in
terms of imaginary time trajectories, that is, in terms of
trajectories
in Euclidean space which solve the Euler-Lagrange equations obtained
from
the original Euler-Lagrange equations by the analytical continuation
of time. In the case of uniform external electric field the
corresponding
trajectories are circles of a fixed radius. Analogously, brane
production
in the external
RR fields is described in terms of the brane world surfaces
which minimize (extremize) the Euclidean brane action.
In the case of
uniform RR field the corresponding surfaces
describing $p$-brane production
are $p+1$-dimensional spheres (bubbles) of a
given radius. The action
computed on such spheres defines the exponential
factor in the production rate.
The process described can be characterized as a
spontaneous brane production
in the external RR field.
Apart from the spontaneous brane production one can consider {\it
induced}
brane production, that is production of branes in the presence of
some original brane configuration, which is our
main concern here. Since in the
nonperturbative string theory branes can couple to branes
the external branes can couple to the bubbles and hence can deform
them
changing the brane production rate. The
physics arising is a higher dimensional
generalization of the one discussed in \cite{SV}
(see also consideration
of the case of nonzero temperature \cite{temp}
or density \cite{gk}), where it was shown that in the
particle induced vacuum decay the bubble is
deformed in the presence of
external particle(s). The key point concerning the induced decay is
the existence of the localized mode of the particle on the kink or
antikink. Due to zero mode the initial
particle can transfer its energy
into the energy of tunneling. The same phenomenon takes place in the
brane bubble creation thanks to the existence of junction
\cite{junction} which substitutes
the particle-kink vertex in the higher dimensional case.
Indeed the initial
brane looses almost all its initial energy after the junction
point and this energy leads to the increase of the nucleation rate.
Hence the crucial point for our further analysis is the existence
of the
string junctions when three (p,q) strings join in the vertex which
is subject to the charge conservation and zero total tension condition
\cite{junction}. It was shown \cite {dm,ahashimoto}
that the string junction
keeps 1/4 of the original SUSY and can
therefore be treated classically
so that the generic string networks can be
developed \cite{sen}. More
recently junctions have been promoted to the
M-theory configurations
\cite{M-theory} where the vertex is resolved
smoothly. Note also that
the analogous webs have been elaborated for the 5-branes within
string or M-theory approach \cite{kol}. In what follows we shall
use junctions to find out the explicit
form of the Euclidean solution
corresponding to the induced tunneling.
When considering strings stretched between external D3 branes
interesting interpretation of the induced tunneling emerges. The key
point is that the end of the fundamental (F) string on the D3 brane
can be considered as the point charge \cite{strom}, the
end of the D string as the monopole \cite{diac} and the
end of (p,q) string as the dyon.
Hence any process involving the strings between D3 branes amounts
to the specific process in the SUSY gauge theory
on the D3 worldvolume \cite{witten}.
For instance the processes with the exchange of the quantum
numbers in the bulk have the
interpretation of the nonperturbatively
mediated phenomena involving the generic dyonic states, such as a
well known process of the scattering
of the monopoles to dyons \cite{ah}.
In what follows we shall use
junction configuration with the background D3 branes
\cite{hashimoto} to describe such processes in the
external field.
It is perhaps important that the induced brane
production can be given
a slightly different interpretation - the one of
quantization of the external
branes in an external RR field. The bubbles then describe
contribute to imaginary
part in self-energy (``self-tension'') of the
external branes. Contribution
of such "brane loops" has been discussed
recently for topological strings
in the M theory framework \cite{vafa}.
Furthermore, in terms of the field theory on the inducing brane,
the external field is a perturbation
of the theory. Therefore the induced
bubble has something to do with the renormalization group flow.
The rest of the paper is organized as follows.
In section 2 we introduce
some necessary background. In section 3 we
discuss our main example -
the induced brane production with one external brane.
The production rate in the leading exponential approximation
is calculated . Section 4 contains
the consideration of the $(p,q)$
strings between D3 branes in
the external field and the
interpretation of the induced tunneling
in terms of the gauge theory on the D3 brane.
Discussion on the related issues and
open questions can be found in Section 5.
\section{Spontaneous brane production}
Consider a $p$-brane omitting the fermionic degrees
of freedom. Its action, essentially, consists of two pieces
\begin{equation}
\label{action}
S=S_{tension}+S_{charge},
\end{equation}}
where $S_{tension}$ is
\begin{equation}
\label{tension}
S_{tension}=T \int_{V} \sqrt{det({\hat G}_{\mu \nu}+
\cal{F}_{\mu \nu})},
\end{equation}
the integration is over the $p+1$-dimensional world-volume $V$
of the brane,
$G_{\mu \nu}$ is the metric induced on the
worldvolume via its embedding into
the target space and the coefficient $T$ is the tension of the brane.
Additional contribution $\cal{F}_{\mu \nu}=F_{\mu \nu}-B_{\mu \nu}$
is due the U(1) YM field and the pullback of the
background NS two-form B field .
The target space is taken to be the flat
Euclidean one in the this section,
its metric being
\begin{equation}
\label{metric}
ds^{2}= \Sigma_{i}(dx^{i})^{2}.
\end{equation}
The second term $S_{charge}$ is different
for the fundamental and D branes.
For fundamental brane it has a simple form
\begin{equation}
\label{charge}
S_{charge}=Q \int_{V} {\hat C},
\end{equation}
where ${\hat C}$ is NS $p+1$ form potential
integrated over the worldvolume of the brane.
For D brane the additional CS term comes
from the RR external fields and
has the structure
\begin{equation}
S_{charge}=\int {\exp {\cal{F}}} \wedge \tilde{C},
\end{equation}
where $\tilde{C}$ collects all relevant RR forms.
In the most interesting
case of D string it includes the excited two-form and axion fields
\begin{equation}
S_{D-str}=\int \tilde{C_2}+\tilde{C_0}\wedge \cal{F}.
\end{equation}
In external fields the branes are produced in the form of bubbles.
Hence if the brane worldvolume forms a closed manifold (the bubble),
the CS term can
be rewritten as
\begin{equation}
\label{volume}
S_{charge}= Q \int_{U} {\hat H},
\end{equation}
where $H=dC$ and $U$ is the manifold with the boundary $V$.
In the simplifying assumption that $H$ is a uniform $D=p+2$-form
in the flat Euclidean space of dimension $D$
\footnote{The assumption about
dimension is not restrictive because
the $D$-dimensional Euclidean space can be a subspace of the
target space supporting the flux
of the field $H$; it is energetically
profitable that the bubbles are produced in this subspace and in the
quasiclassical consideration the
rest of the target space is irrelevant.}
the last term reads
\begin{equation}
\label{volume}
S_{charge}={\pm} Q{\Phi} \int_{U} \sqrt{det({\hat G}_{\mu \nu})},
\end{equation}
where ${\Phi}$ is a flux density of the field $H$. The
two possible signs
in Eq.(\ref{volume}) refer to possible orientations of $V$.
Thus, in view of Eqs.(\ref{tension},\ref{volume}),
the brane bubble action
Eq.(\ref{action}) becomes a sum of
surface and volume terms (the latter having
negative coefficient for the relevant bubbles)
\begin{equation}
\label{actionprim}
S=T \int_{V} \sqrt{det({\hat G}_{\mu \nu})}-
Q{\Phi} \int_{U} \sqrt{det({\hat G}_{\mu \nu})},
\end{equation}
where $Q{\Phi}$ is positive for the relevant bubbles.
Hence there is a competition
of the two terms: the surface term
suppresses small bubbles while the
volume term blows up sufficiently large bubbles.
The one which extremize
the action Eq.(\ref{action}) is the
critical bubble. In the present case
the critical bubble is a $p+1$-dimensional sphere of radius
\begin{equation}
\label{radius}
R=\frac{(p+1)T}{Q{\Phi}}.
\end{equation}
The critical bubble can be obtained in many ways.
We shall now briefly
describe two of them because we believe
each of them is instructive.
Probably the simplest one
is to make the spherically symmetric anzatz and
to extremize the action
Eq.(\ref{actionprim}) with respect to the radius.
This gives an algebraic
equation giving the critical radius $R$ from
Eq.(\ref{radius}).
The extremal value of the action
Eq.(\ref{actionprim})
\begin{equation}
\label{exponential}
S_{c}=\frac{1}{p+1} {\Omega}_{p+1}TR_{c}^{p+1}
\end{equation}
defines the exponential of the production rate
\begin{equation}
{\Gamma} {\sim} e^{-S_{c}}
\end{equation}
where ${\Omega}_{p}=\frac{2{\pi}^{\frac{p+1}{2}}}{{\Gamma}(\frac{p+1}{2})}$ is
volume of unit $p$-sphere.
This type of argument was given in \cite{Langer}.
Note that under variation of the form of the bubble
the surface term in the
action Eq.(\ref{actionprim}) gives the so-called ``Laplace pressure''
(or, in other words, trace of the external curvature)
multiplied by tension $T$, while the volume term
gives $-Q{\Phi}$, so in this case one gets the equation
\begin{equation}
\label{pressure}
T(\Sigma_{i} \frac{1}{R_{i}})=Q{\Phi},
\end{equation}
where $\frac{1}{R_{i}}, i=1, \ldots ,p+1$
are principal curvatures of
the world-volume of the brane. In the case of spherical symmetry
all $R_{i}$'s are equal and one comes back to Eq.(\ref{radius}).
From this picture it is immediately seen that
for $p=0$ brane the bubble
in any case can only be glued from
arcs of the circle of radius $R$
Eq.(\ref{radius}) \cite{SV}.
The other way, followed in \cite{VKO}, consists
in choosing among the
Euclidean
coordinates a ``time'' coordinate and
assuming the spherical symmetry only
for the remaining ones. Then Eq.(\ref{actionprim}) reduces to
the effective action
\begin{equation}
\label{effective}
S_{eff}=T{\Omega}_{p} {\int}dt {\rho}^{p} \sqrt {1+{\dot {\rho}}^{2}}-
\frac{Q{\Phi}}{p+1}{\Omega}_{p} {\int}dt {\rho}^{p+1}
\end{equation}
where ${\rho}$ is
the radial coordinate in the spherical coordinate system:
\begin{equation}
\label{spherical}
ds^{2}={\Sigma}_{i}(dx^{i})^{2}=dt^{2}+dr^{2}+r^{2}(d{\Omega}_{p})^{2}
\end{equation}
and $(d{\Omega}_{p})^{2}$ is the metric on the unit $p$ sphere.
Extremization of $S_{eff}$ from Eq.(\ref{effective}) gives an
equations
for ${\rho}(t)$ which defines the critical bubble. One can
use time translation
symmetry of the effective Lagrangian
Eq.(\ref{effective}) to reduce the order
of the equation for ${\rho}(t)$ from second one to the first one.
Namely, the first integral reads
\begin{equation}
\label{energy}
{\cal E}=-\frac{R{\rho}^{p}}{\sqrt{1+{\dot {\rho}}^{2}}}+{\rho}^{p+1},
\end{equation}
where $R$ is as in Eq.(\ref{radius}).
Since the tunneling proceeds at
$E=0$ Eq.(\ref{energy}) becomes an equation of the sphere of
the radius $R$. The section $t=0$ of the critical bubble is the
configuration which is born in Minkowski space and then blown
up by the external field.
\section{One brane induced brane production}
Essentially, the picture of the induced
brane production in the external
field looks as follows. There is an infinite
brane which do not interact
with the external field so asymptotically
its worldvolume is flat.
If the worldvolume of this brane
(we shall call it external) can glue to
(or, end on)
worldvolume of branes which interact
with the external field then the
bubble made of the worldvolume(s) of the
new brane(s) can arise somewhere
in the middle of the world volume of the
external brane. The brane action
computed on such configuration of branes
(minus action computed on the
configuration which consists of the external brane alone)
defines exponential factor in the brane production rate.
A configuration which emerges (and then blows up)
in Minkowski space consists
of a spacelike slice
of the external brane worldvolume with
the equatorial slice of the bubble
somewhere in the middle of it.
In string theory, the brane worldvolume can
glue to worldvolume of a
brane of the same dimension or to a worldvolume
of a brane of higher
dimension.
In the former case, gluing is possible for the
$(P,Q)$-branes, in the latter case gluing
is possible when the
higher dimensional brane bears an excitation
of internal degrees of freedom,
basically, an excitation of the gauge field living on the brane .
In the present paper we
concentrate on the first case.
In IIB string there are $(P,Q)$ strings (1-branes) and
$(P,Q)$ 5-branes \footnote{Actually there are also $(P,Q)$ 7 branes
but we shall not discuss them later.}. The $(P,Q)$-branes are
branes bearing two types of charges -
RR and NS ones , correspondingly
there are two types of CS terms in $(P,Q)$-brane action.
As to the tension of the $(P,Q)$-brane, it depends on the charges
in the following way:
\begin{equation}
\label{tensionpq}
T_{P,Q}=T|P+Q{\tau}|,
\end{equation}
where the complex parameter ${\tau}$ encodes axion and dilaton v.e.v.
The week string coupling limit
corresponds to ${\tau} \rightarrow i\infty$
case.
Note that in what follows we shall assume that the RR axion
field is not excited.
From the analysis of the string junctions
and (P,Q) webs \cite{sen, kol},
it is known that (P,Q) branes worldvolumes can glue under two
conditions:
1)conservation of the charges;
2)mechanical equilibrium of tension forces.
Thus, e.g., (1,0)-brane couples to (1,1) and (0,-1) branes, and
at ${\tau}=i$ the angle between (1,0)
and (1,1) ones will be equal to
$3{\pi}/4$,
and the angle between the (1,0) and (0,-1)
branes will be equal to ${\pi}/2$.
Obviously, the conditions of gluing are not locally affected by the
external field.
What is affected by the external filed
is the shape of the worldvolume
(as seen from Eq.(\ref{pressure})). However, for any brane of a
given configuration of charges one
can choose such a linear combination of
RR and NS external fields that the brane does not feel it in the
leading approximation. Below we
assume that the external field is of NS type, while the external brane
is a D-brane, that is with charges of $(L,0)$ type, and hence it does
not
interact with the external field. $(L,0)$-brane can couple to
$(P,Q)$ and $(L-P,-Q)$ branes and
worldvolumes of the latter are curved
by the external NS field. Locally
they are glued according to the same
condition 1) and 2) above, that is
the angle ${\alpha}_{(L,0)(P,Q)}$
between $(L,0)$ and $(P,Q)$ is defined according to
\begin{equation}
\label{alpha}
cos {\alpha}_{(L,0)(P,Q)}= -\frac{P+Q Re{\tau}}{|P+Q{\tau}|}
\end{equation}
and analogously the angle ${\alpha}_{(L,0)(R-P,-Q)}$ is given by
\begin{equation}
\label{alpha1}
cos {\alpha}_{(L,0)(R-P,-Q)}= -\frac{L-P-Q Re{\tau}}{|P-L+Q{\tau}|}.
\end{equation}
Note that there is an essential difference here from the (P,Q)-webs
case - the internal branes are not flat, their worldvolumes form sort
of ``caps'' (see Fig.1) which are glued to the
external brane worldvolume at the
angles defined by Eqs.(\ref{alpha}),(\ref{alpha1}), with one of the
caps
is glued at the
angle defined by Eq.(\ref{alpha}) from above and the other one -
at the angle defined by Eq.(\ref{alpha1}) from below.
\begin{figure}[t]
\hspace*{5cm}
\epsfxsize=7cm
\epsfbox{bub1.eps}
\caption[x]{Stringy induced Euclidean solution.}
\label{IA}
\end{figure}
Let us turn to the discussion of the gauge field contribution.
Thanks to a point junction Gauss law
on the worldsheet D string theory
has to be modified
since the F string inserts the point source.
Due to the discontinuity of
the electric field $\Delta E=g$,where g is the IIB string coupling,
The $A_0$ component has to be piecewise
linear and according to BPS condition
has to be
correlated with one of the scalars representing
the transverse fluctuation of the string \cite{dm,ahashimoto}.
This condition is fulfilled
at the junction and is equivalent to the total zero tension condition.
In the case considered instead of the
point like junction point we have
junction circle around the bubble. The Gauss law now reads as
\begin{equation}
divE=g(\delta (x - x_0(t))- \delta (x + x_0(t))
\end{equation}
where $\rho$ denotes the radius of the circle in the (x,t) plane
(the plane of the external string),
$x_{0}^2+t^2=\rho^2$. Solution to the Gauss law constraint
provides the discontinuity of the
electric field. Once again BPS condition
amounts to the total zero tension along
the circle which is consistent
with the equations of motion to the gauge field and scalar. Let us
emphasize that we use the BI action
to describe the D string. Introducing
the canonical momentum for the gauge field $ \Pi=\frac{\delta L_{BI}}
{\delta E}$ and performing the
Legendre transform to the Hamiltonian
description one immediately recognizes
the action of the string with
tension $T_{P,Q}$ since the canonical momentum
along the "caps" is constant due to the
equations of motion.
To find the shape of the caps it is most
convenient to turn the second of the
two possibilities described at the
end of section 2. A coordinate $z$,
orthogonal to the
external brane worldvolume (see footnote in section 2),
will be considered as ``time'' in the effective problem.
\footnote{Perhaps, it is worth to
stress that this is unphysical time,
which is
convenient to describe the shape of
the caps, because the caps are obviously
spherically symmetric in the rest of
coordinates. The physical time - the
one in which the brane production should be interpreted - is, in
fact, one of the coordinates along the external brane.}
With this choice, the effective action reads
\begin{equation}
\label{action1}
S=S_{external}+S_{(P,Q)}+S_{(L-P,-Q)},
\end{equation}
where $S_{external}$ includes only tension
term for the external brane and
$S_{(P,Q)}$ and $S_{(L-P,-Q)}$ include both the tension and CS terms
(cf. Eq.(\ref{effective}):
\begin{equation}
\label{action1}
S_{(P,Q)}=T_{(P,Q)}{\Omega}_{p} {\int}_{0}^{t_{+}}dt {\rho}^{p}
\sqrt {1+{\dot {\rho}}^{2}} -
\frac{Q{\Phi}}{p+1} {\Omega}_{p} {\int}_{0}^{t_{+}}dt {\rho}^{p+1}
\end{equation}
and
\begin{equation}
\label{action2}
S_{(L-P,-Q)}=T_{(L-P,-Q)}{\Omega}_{p} {\int}_{t_{-}}^{0}dt {\rho}^{p}
\sqrt {1+{\dot {\rho}}^{2}} -
\frac{Q{\Phi}}{p+1} {\Omega}_{p} {\int}_{t_{-}}^{0}dt {\rho}^{p+1},
\end{equation}
where tensions $T_{(P,Q)}$ and $T_{(L-P,-Q)}$ are defined in
Eq.(\ref{tensionpq}), and $t_{+}, {\rho}=0$ ( $t_{-}, {\rho}=0$) is a
top
(bottom) point of the upper (lower) cap.
Notice that the volume terms in Eq.(\ref{action1}) and
Eq.(\ref{action2})
have the same coefficient which is traced
back to the charge conservation.
Instead of Eq.(\ref{energy}) of section 2 we now have two
first integrals - the one for the
upper cap and the other - for the lower cap.
Moreover it is clear that the first integral
constant ($E$) vanishes, so the caps are segments of spheres.
The radii of the spheres are those
of the spheres in the spontaneous brane
production:
\begin{equation}
\label{radius1}
R_{(P,Q)}=\frac{(p+1)T|P+Q{\tau}|}{Q{\Phi}}
\end{equation}
for the $(P,Q)$-cap and
\begin{equation}
\label{radius2}
R_{(L-P,-Q)}=\frac{(p+1)T|P-L+Q{\tau}|}{Q{\Phi}}
\end{equation}
for the $(L-P,-Q)$ cap. Given the radii Eqs.(\ref{radius1}),
(\ref{radius2}),
the segments are uniquely defined by the angles
Eqs.(\ref{alpha}),(\ref{alpha1}).
To verify that the caps fit into a bubble one
has to check that the $p$-spheres on the
boundaries of the segments have the
same radius which is indeed the case.
To find the exponential of the brane production rate one should take
the value of the effective action Eq.(\ref{action1})
on the configuration
described and subtract from it action
computed on the configuration which
consists of the external brane alone.
The contribution from the external
string in Eq.(\ref{action1})is, of course,
infinite. However, the decay rate is defined by the difference
between the critical value of the action from Eq.(\ref{action1})
and the value of the action when only
flat external brane (without any bubble)
is present. This difference is finite and defines the following rate
of
the induced brane production
\begin{eqnarray}
\label{rate}
{\Gamma}=exp \{- \frac{1}{(p+1)(p+2)}{\Omega}_{p+1}
\frac{ {\left ((p+1)|P+Q{\tau}|T \right )}^{p+2}}{(Q{\Phi})^{p+1}}
\frac{1}{\pi} arcsin \left | \frac{QIm{\tau}}{P+Q{\tau}}\right|
\nonumber\\
- \frac{1}{(p+1)(p+2)}{\Omega}_{p+1}
\frac{ {\left ((p+1)|P-L+Q{\tau}|T \right )}^{p+2}}{(Q{\Phi})^{p+1}}
\frac{1}{\pi} arcsin \left | \frac{QIm{\tau}}{P-L+Q{\tau}}\right|
\nonumber\\
+ \frac{1}{(p+1)(p+2)}
{\Omega}_{p}L
\frac{ {\left ((p+1)T \right )}^{p+2}}{({\Phi})^{p+1}}
|Im{\tau}|^{p+1}\}
\end{eqnarray}
${\Gamma}$ can equally be considered as
an imaginary part of the $(L,0)$
tension in the external NS field.
The effective contribution from the external string
comes from the difference between the action of the external
brane without and with the bubble
solution and in the string case
reads
\begin{equation}
S_{ext}=-T_{1,0}\pi \rho^2,
\end{equation}
where $\rho$ is the intersection radius.
Let us emphasize that the
contribution $S_{ext}$ can be treated
differently as the contribution
due to the Wilson loop along the matching
circle. This interpretation
is in perfect accord with the Gauss law
if we attribute the Wilson loop
to the effective "charges". Note once
again that the effects due
to the classical gauge fields on the caps are taken into account
by tensions of the branes.
It is in order here to compare the rates of the spontaneous and
induced brane nucleation. Of course, this comparison cannot be taken
literally since the final states are different. However it is
instructive to identify the enhancement factors for the induced processes
in different regimes.
Let us recall the situation for the false vacuum decay induced by
a particle of mass $m$ in (1+1) dimensions. The relevant parameter
is the ratio ${\epsilon}=m/2{\mu}$ where $\mu$ is the kink mass
(the tension of the bubble). At small $m$ the external particle doesn't
disturb classical bubble solution significantly. Therefore the
enhancement factor is just exponential of the particle action
inside the bubble, $exp^{2mR_{0}}$, where $R_{0}={\mu}/{\epsilon}$
is the radius of the unperturbed bubble. Opposite limit
corresponds the so-called ``sphaleron'' region
$\epsilon \sim 1$ where the exponential factor disappears since the
initial particle can decay into the kink-antikink pair without
external field. Note however that in the sphaleron region large quantum
corrections make inapplicable the saddle point approximation.
Now turn to the brane induced processes. The relevant parameter
analogous to the parameter $\epsilon$ above is the parameter $\tau$
defining the ratio of tensions of $D$ and $NS$ branes. The region of the
slightly disturbed bubble corresponds to the limit
$Im{\tau} \rightarrow \infty$ and the enhancement factor in this limit
is again given by exponential of the external brane action
inside the bubble,
\begin{equation}
\label{rate1}
{\Gamma}_{enh}=exp\{\frac{1}{(p+1)}
{\Omega}_{p}L
\frac{ {\left ((p+1)T \right )}^{p+2}}{({\Phi})^{p+1}}
|Im{\tau}|^{p+1}\}
\end{equation}
The ``sphaleron'' region is where the initial brane can pass into
a pair of branes without external field. The corresponding
``line of marginal stability''( where the tension
of the initial brane equals sum of the tensions of pair
of the produced branes)
is reached at $Im \tau =0$. Discussion of quantum
corrections to the ``line of marginal stability'' is beyond
the scope of the paper.
Eq.(\ref{rate}) predicts that there is no suppression of the induced
brane production at this line.
However, in analogy with the particle induced vacuum decay,
Eq.(\ref{rate}) cannot be trusted in the sphaleron region because
of strong string corrections.
Let us now comment on the choice of the initial D-brane state. The
subtlety concerns the possibility to have
the initial brane at rest. If
we consider the initial D-brane one can wonder about the dependence
of the general BI action on the NS B field
and therefore the corresponding
interaction of D brane. To handle this issue for a string
we can suppose that gauge fields are not
excited in the initial state
and there is no axion $\tilde{C_0}$ field at
all. However in the generic
case the noncommutative geometry induced by NS
B field is involved
hence the noncommutative BI Lagrangian actually governs the dynamics.
It is natural to discuss the more general
case when two external branes
are involved in the induced process.
Let us consider two $(1,0)$ branes as external
branes (see fig.2) inducing the tunneling.
In this case
the bubble describing branes production
consists of two caps, one above
and one below , and
a barrel in the middle. One of the caps
has charges $(1,1)$, the other one -
$(1,-1)$, and the barrel - $(0,1)$.
The external branes (``legs''),
the caps, and the
barrels are glued at the angles defined by Eq.(\ref{alpha}).
Notice, that
in vicinity of the bubble the external
branes are not parallel and not flat.
The effective action consists of various pieces - those for the legs,
for the caps and for the barrel. One can see that the caps are again
segments of spheres, while for
the barrel the first integral - analogue
of Eq.(\ref{energy}) is not zero. Its external
curvature has two different
eigenvalues. What concerns to the legs, there is no ``volume term''
in the effective action for them which allows
them to spread to infinity,
and the first integral is again nonzero which makes them curve.
There is no principal difficulties in
describing the bubble in this case,
but the formulae are not as transparent
as in the one brane induced
case, and we will not bother the reader by them.
Now, after describing all these induced bubbles,
we would like to come back
to the spontaneous brane creation. It is now
clear, that in addition to the
basic case of round bubble made out of one
brane, there are other bubbles
of branes spontaneously arising
in the external field. For example,
one can have a bubble glued from two curved branes bearing
$(P.Q)$ and $(-P-L,-Q)$ charges
(``caps'') and one flat brane
in the middle with $(L,0)$ charge .
The production rate of such bubbles is described by Eq.(\ref{rate})
with the change of $L$ to $-L$.
\begin{figure}[t]
\hspace*{5cm}
\epsfxsize=7cm
\epsfbox{bub2.eps}
\caption[x]{Time slice of the Euclidean solution with two external
branes .}
\label{IA}
\end{figure}
\section{Nucleation of brane bubbles in D3 background}
So far we considered the one-brane
induced process in the flat space.
It is interesting to consider the string induced process in the
background
of two D3 branes far apart from each other.
Namely consider the initial F or D string stretched
between two D3 branes along some direction.
In this case the ends of the
string on D3 branes represent monopoles \cite{diac}
(D string) or charges \cite{strom}
(F string) in the 4d SYM theory
on the D3 brane worldvolume.
If distance between D3 branes corresponding to the vacuum
expectation
value of the scalar in 4d theory is large
we can ignore the metric
deformation due to D3 branes and consider the
perturbative domain in 4d theory. The configuration
under consideration is presented
at Fig.3.
In the external B field the process looks as follows.
The initial D string
creates a bubble in the Euclidean
space which then evolves in Minkowski
space. Since the simplest decay mode
for D string is to F string and
(1,-1) string, from the point of
view of the 4d observer on D3 brane
the process looks like the
exponentially suppressed decay of
monopole to charge and dyon
in the external field during the finite time
determined by the bubble
action.
The time-evolution for the process looks as follows.
The critical (turning point) configuration is the one
which arises in the Minkowski region. Then the external field
accelerates the (1,1) and (-1,0) strings and at some time
size of the bubble becomes comparable with the distance
between D3 branes. At this moment in terms of the $SU(2)$
theory on D3 branes (1,1) dyon and (-1,0) are created.
Note that we can believe such consideration
if the vacuum expectation
value of the scalar is much larger then
the critical radius, otherwise
the $AdS_{5}\times S^5$ structure of the near horizon
D3 brane metric
\begin{equation}
ds^2=H^{-\frac{1}{2}}dx_{||}^2 + H^{\frac{1}{2}}
(dr^2+ r^2d\Omega_{5}^2)
\end{equation}
\begin{equation}
H=1+ \frac{4\pi g\alpha^{'2}}{r^4},
\end{equation}
where $X_{||}$ are four coordinates
along the D3 worldvolume and
$d\Omega_{5}^2$ is the five-sphere metric
has to be taken into account.
\begin{figure}[t]
\hspace*{5cm}
\epsfxsize=7cm
\epsfbox{bub3.eps}
\caption[x]{Time slice of the Euclidean solution
for the induced process in D3 background}
\label{IA}
\end{figure}
Let us note that the process looks somewhat unexpected from the
point of view of 4d action on D3 brane where two-form NS field
enter through the BI term while the RR one through CS one.
Since the BI action supports the F and D string excitations
\cite{cm,gibons} we
can claim that these excitations are
unstable in the RR or NS two form
background respectively.
Having this
in mind we can interpret the process from the D3 brane observer
viewpoint as the
"instanton" like monopole decay in the NS B field (for D string)
or charge decay in the RR B field (for F string).
Both processes are exponentially
suppressed due to the dyon production
rate by the Euclidean tunneling
exponent.
The decay rate can be presented in the form
\begin{equation}
\Gamma =f_{mink}\exp(-S_{Eucl}),
\end{equation}
where $S_{Eucl}$ has been found in the previous section while the
$f_{Mink}$ stands for the Minkowski part
of evolution till the moment when
dyonic strings touch D3 branes.
Minkowski contribution to the action is purely imaginary
\begin{equation}
S_{Mink}=i\int p_{Mink}(r)dr,
\end{equation}
where $p_{Mink}$ is the canonical momentum conjugate to the
the radial variable. The subtle point is the integration region.
In principle there are two different situations; in the first case
the effective potential for the radius has no any additional
extremum in the D3 brane background and the contribution of
the Minkowski part of the evolution doesn't play the important role.
However one can't exclude the possibility
that there is such extremum
and the Minkowski evolution allows the
finite motion. In this case
the amplitude develops the poles related to the energy levels
for the radial variable in the
Minkowski region. The possibility
of such scenario requires the
careful analysis of all gravitational
effects and deserves further investigation.
Let us turn to the case of two D strings. Without external field
we have two monopoles with the nontrivial
moduli space. The brane configuration
fits perfectly Nahm description of the moduli space \cite{diac}.
Nahm equations
themselves
\begin{equation}
D_sT_i=-\epsilon_{ijk}[T_j,T_k]
\end{equation}
describe the condition of the BPS
invariance of the configuration and
s is identified with the direction along the strings.
Matrixes $T_i$ correspond to the positions of D string ends on the
D3 branes.
Solution to the Nahm equations with
the proper boundary conditions amounts to the moduli space of the
two-monopole configuration and provides the hyperkahler metric.
External field yields the deformation of the picture. Consider the
SU(2) theory on the worldvolume of D strings. The
external field enters the Lagrangian
through the BI action and hence deforms
Nahm equations.
The new B field dependent metric can
in principle be found in a standard way
using the construction of the spectral
curve from the solution of the
Nahm equation or using the hyperkahler structure \cite{ah}.
From the generic point of view the
state of two monopoles can be represented
by the point in the moduli space while their slow relative motion
is governed by the geodesic motion in the hyperkahler metric.
In the external B field the unstable
submanifold in the moduli space
is developed. Indeed assuming that the bubble size is smaller
than the distance between D3
branes we have to conclude that any
point at this submanifold is unstable due to the decay
of two static monopoles to the
dyons due to the bubble solution.
The instability reflects the
presence of the negative mode which
results in the exponentially suppressed process.
One more interesting
comment is in order here. The Nahm equations are the monopole
generalization of the ADHM construction
for instantons. It is also known
that the monopole can be thought of as the chain of
instantons. In the brane picture
above this can be recognized viewing
the D1 string as the bound state of the infinite number of D(-1)
branes representing instantons. Since the distance from D3 brane
corresponds to the instanton size \cite{bg}
we can conjecture that the bubble
creation can be considered as the nontrivial deformation of the
instantons of the sizes related with the radius of the
bubble .
The case of the nontrivial momenta of monopoles
looks more complicated. Even in the absence of the external
fields there are some nonperturbative
phenomena in this case. We can
mention the scattering at the right
angle in the forward collision or the
transition to the dyons at the
generic kinematics \cite{ah}. In the external
field the picture is even more rich. The inspection
of the matching conditions for
the bubble solution with the forward collision
of two D string in the B field shows that there is
a possibility to have different kinematics
at the final state therefore besides
the scattering at the right angle
there is the exponentially suppressed processes
with generic angle kinematics.
Moreover there is a plenty of
possibilities in the generic
kinematics for the nonperturbative
phenomena. For instance there
is the process of transition of
the pair of initial monopoles or
charges into the generic state of
several dyons through the more complicated "bubble" type solution.
\section{Discussion}
In this paper we developed
quasiclassical approach to a higher dimensional generalization
of the induced tunneling processes.
Utilizing the existence of the brane
junction configuration the explicit
quasiclassical tunneling exponent has
been calculated. Our main example
involves the stringy induced amplitudes
but the 5 brane case is treated along the
same routing. The solutions
found above can be combined with the
string-like solutions to the BI equations
of motion to get the generic webs in the
external fields. Similar
arguments can be applied to more
generic processes analogous to
the charge and monopole decays mentioned
above. Let us also remark that
in the usual field theory framework
there are slightly different processes
similar to the induced false vacuum decay via bubble creation,
for instance the processes
yielding the baryon number non-conservation.
The corresponding
nonperturative solution involves
instanton-antiinstanton pair and the collective
coordinate relevant for tunneling is
the Chern number. It seems that
the results of the paper might be
useful for the quantitative description of
their higher dimensional analogies.
Not much can be said about the quantum corrections to the induced
processes. It is not clear if the quantum correction can be
reduced to the geometric characteristics of the solution. The only
remark available concerns the effect of the quantum fluctuations
of the gauge fields
to the stringy induced process since the relevant 2d YM action is
almost topological and depends only on the area of the
manifold which the YM theory is
defined on. In our case the relevant manifold
has the topology of the sphere therefore the expectation value of
the Wilson loop in 2d YM theory is of interest. The Wilson loop
expectation can be formulated in term of group characters \cite{gt}
and even calculated at the large N limit \cite{dk}. It appears
that the dependence on the area exhibits some phase
transition behavior but for large
areas it manifests the standard area law and therefore actually
yields the tension renormalization.
Let us mention that the proper playground for the theory with
the external NS B field is the
noncommutative geometry. Having in mind that
the rate of the process of the creation of branes depends
nonanalitically on the B field
one can expect that the partition function of the theory
can manifest some singular structure similar to the singularities
at the complex coupling plane in the usual field theory due to
instantons. In particular, since the ratio of the process can be
considered as imaginary (nonanalitical in $B$ !)
part of the brane tension, one can expect a change of the notion
of BPS states in the noncommutative case, as compared to \cite{SK}.
We shall be back to this subject elsewhere \cite{GS}.
Let us turn now to the gauge theories
on the worldvolume on the emerging branes. The situation is
most transparent in the case of 5 brane webs. The particular
configuration of 5 branes with some "external 5 brane legs"
amounts to SU(N) 5d gauge theory on the worldvolume
of N 5 branes stretched between the
external ones \cite{kol}. In the external
field some brane production is possible which means the process of
nonperturbative change of the rank of the gauge group whose rate
depends nonanalitically on the external field. Moreover the
brane diagram itself can be treated as a kind of Euclidean solution
corresponding to the nonperturbative phenomena from the point of view
of the gauge theories on the {\it external } 5 branes. Finally let
us mention that the Minkowski evolution of the 5 brane bubbles
acquires the meaning of some RG flow in the theory on their
worldvolumes.
It would be interesting to discuss the case
when temperature or density
play the role of the inducing factor. In the temperature case it is
necessary to find the Euclidean solution periodic in the Euclidean
time. The temperature fixes the size of the matching p-sphere and
therefore gives rise to the action on the solution. The formulae
in this case are similar with the substitution of the temperature
instead of the tension of the external brane. It is not clear how
the density case has to be treated since the chemical potential
for the state with branes has to be
defined properly. Let us also note
that the induced process can be considered in the AdS spaces
generalizing the consideration in \cite{MS}.
We are thankful to A.Mikhailov, M.Shifman and M.Voloshin for the
useful
discussions. A.G. thanks to TPI at University of Minnesota were the
paper
was completed for the hospitality. The work was supported in part by
grants
INTAS-96-0482 and RFBR 97-02-16131
(A.G) and INTAS-97-0103 (K.S).
|
\section{General Appearance}
\vspace*{-0.5pt}
\noindent
\noindent
\noindent
\newpage
The supersymmetric procedures are an interesting
and fruitful extension of (one-dimensional) quantum mechanics. For recent
reviews see [\refcite{susy}]. These techniques are, essentially,
factorizations of one-dimensional Schr\"odinger operators, first
discussed in the supersymmetric context
by Witten in 1981 [\refcite{w81}], and well known in the mathematical
literature in the
broader sense of Darboux covariance of Schroedinger equations
[\refcite{d82}].
In 1984, Mielnik [\refcite{mn}] introduced a different factorization of the
quantum harmonic oscillator based on the general Riccati solution. As a result,
Mielnik obtained a one-parameter
family of potentials with {\em exactly} the same spectrum as that of the
harmonic oscillator. However, even though in the same year Nieto discussed the
connection
of such a factorization with the inverse scattering approach, and Fern\'andez
applied it to the hydrogen atom case, Mielnik's result remained a curiosity
for a decade during which only a few authors paid attention to it.
On the other hand, constructing families of strictly isospectral potentials is
an important possibility with many potential
applications in physics [\refcite{susy}]. This explains the recent surge
of interest in this supersymmetric issue [\refcite{rosu}].
My goal in this work is to give a multiple-parameter generalization of
Mielnik's procedure based on the ground-state function of any soluble
one-dimensional quantum mechanical problem. This is just
a form of Crum's iterations, i.e., repeated Darboux transformations.
Some work along this line has already been done by
Keung {\em et al.} [\refcite{k}], who performed an iterative construction for
the reflectionless, solitonic, ${\rm sech}$ potentials and the attractive
Coulomb potential presenting
relevant plots as well. However, they first go $n$ steps away from a given
ground state and only afterwards perform the $n$ steps backwards.
On the other hand, Pappademos {\em et al.} [\refcite{psp}],
working in the continuum part of the spectrum,
got one- and two-parameter supersymmetric families of potentials strictly
isospectral with respect to the half-line free particle and Coulomb potentials
and focused on the supersymmetric bound states in the continuum. Their procedure
is closer to the method I will present in the following.
For more recent works see
Bagrov and Samsonov [\refcite{bs}], Fern\'andez {\em et al} [\refcite{fvm}],
Junker and Roy [\refcite{jr}], and Rosas-Ortiz [\refcite{roo}].
In the following, I first briefly recall the mathematical background of
Mielnik's method and next pass to a simple multiple-parameter
generalization for the particular but physically relevant zero-mode case.
I begin with the ``fermionic" Riccati (FR) equation
$y^{'}=-y^2+V_{1}(x)$ [the ``bosonic" one being $y^{'}=y^2+V_{0}(x)$] for which
I suppose to know a particular solution $y_{0}$. Notice also that I do not
put any free constant in the Riccati equations, that is, I work at zero
factorization energy. Let us seek the general solution in the form
$y_1=w_1+y_{0}$.
By substituting $y_{1}$ in the FR equation one gets the Bernoulli equation
$-w_{1}^{'}=w_{1}^2+(2y_{0})w_{1}$. Furthermore,
using $w_2=1/w_1$, we obtain the simple first-order linear differential equation
$w_{2}^{'}-(2y_{0})w_{2}-1=0$, which can be solved
by employing the
integration factor $F_{0}(x)=e^{-\int _{c}^{x}2y_{0}}$, leading
to the solution
$w_2(x)=(\lambda +\int _{c}^{x}F_{0}(z)dz)/F_{0}(x)$,
where $\lambda$ occurs as an integration constant. In applications the lower
limit $c$ is either $-\infty$ or 0 depending on whether one deals with full-line
or half-line problems, respectively. In the latter case, $\lambda$ is
restricted to be a positive number.
Thus, the general FR solution reads
$$
y_{1}=y_{0}+
\frac{e^{-\int _{c}^{x}2y_{0}}}{\lambda +\int _{c}^{x}
e^{-\int _{c}^{z}2y_{0}}}\equiv
y_{0}+\frac{F_{0}}{\lambda +\int _{c}^{x}(F_{0})}=y_{0}+
D\ln\left(\lambda +\int _{c}^{x}(F_{0})\right)~,
\eqno(1)
$$
where $D=\frac{d}{dx}$.
It is easy to reach the conclusion that the particular FR
solution $y_{0}$ corresponds to
Witten's superpotential [\refcite{w81}],
while the general FR solution
$y_1=y_{0}+\frac{F_{0}}{\lambda +\int _{c}^{x}F_{0}}$ is of
Mielnik type [\refcite{mn}]. This is especially clear when one is able to
identify $f_{0}=F_{0}^{1/2}$ with the quantum mechanical
ground-state wavefunction $u_{0}$ of the problem at hand.
This requires suitable asymptotic behaviour of the Riccati solution $y_{0}$
and applying the normalization condition to $f_{0}$,
turning it into a true zero mode. As is well known, this case corresponds to
the so-called unbroken SUSYQM, which will be assumed to hold henceforth.
Moreover, $-2y_{0}^{'}$
($\equiv -2\frac{d^2}{dx^2}\ln F_{0}^{1/2}$)
is the Darboux transform contribution to the initial
Schr\"odinger potential, i.e., $V_1=V_0-2y_{0}^{'}$. Also, the modes
$$
u _{\lambda}(x)=
\frac{F_{0}^{1/2}}{\lambda +\int _{c}^{x} F_{0}}=\frac{u _{0}}{\lambda +
\int _{c}^{x}u_{0}^2}
\eqno(2)
$$
can be normalized and therefore considered as ground-state wavefunctions
of the bosonic family of potentials
corresponding to Mielnik's parametric superpotential.
The one-parameter true zero modes read
$$
v _{\lambda}(x)=
\frac{\sqrt{\Lambda}F_{0}^{1/2}}{\lambda +\int _{c}^{x} F_{0}}
=\frac{\sqrt{\Lambda}u _{0}}{\lambda +
\int _{c}^{x}u_{0}^2}~,
\eqno(3)
$$
where $\sqrt{\Lambda}=\sqrt{\lambda(\lambda +1)}$ is the normalization
constant.
Moreover,
$-2y_{1}^{'}$ can be thought of as the general Darboux transform contribution
to the initial potential generating the bosonic strictly isospectral
family, which reads
$$
V_{\lambda}^{M}=
V_{0}(x)-2\frac{d^2}{dx^2}\ln \left(\lambda +\int ^{x}u^{2}_{0}\right)
=V_{0}(x)-\frac{4u_{0}u^{'}_{0}}{\lambda +\int^{x}u^2_{0}}
+\frac{2u^4_{0}}{(\lambda +\int^{x}u^2_{0})^2}~.
\eqno(4)
$$
This family of potentials can be seen as a continuous deformation of the
original potential, because the latter is included in the infinite limit of
the deforming parameter $\lambda$ and $v_{\pm\infty}=u_{0}$ as well.
In more intuitive terms, Mielnik's method based on an intial Schroedinger
true zero mode may be called
a double Darboux technique of deleting followed
by reinstating a nodeless ground-state wavefunction
$u_0(x)$ of a potential $V_{0}(x)$ by
means of which one can generate a one-parameter family of isospectral
potentials
$V_{\lambda}(x)$, where $\lambda$ is a labeling,
real parameter of each member potential in the set.
One can go on with one of the strictly isospectral bosonic
zero modes $u_{\lambda _{1}}=\frac{u_0}{\lambda _{1}+\int _{c}^{x} u_{0}^{2}}$
(i.e., by choosing $\lambda =\lambda _{1}$)
and repeat the strictly isospectral construction, getting a
new two-parameter
zero mode
$$u _{\lambda _1,\lambda _{2}}=\frac{u_{\lambda _1}}{\lambda _2
+\int _{c}^{x} u_{\lambda _1}^{2}}=\frac{u_{0}}{(\lambda _1+\int _{c}^{x}
u_{0}^{2})
(\lambda _2 +\int _{c}^{x}u_{\lambda _1}^{2})}~.
\eqno(5)
$$
The two-parameter true zero modes read
$$
v_{\lambda _1,\lambda _{2}}=
\frac{\sqrt{\Lambda _1\Lambda _2}
u_{0}}{(\lambda _1+\int ^{x}u_{0}^{2})
(\lambda _2 +\int ^{x}v_{\lambda _1}^{2})}~.
\eqno(6)
$$
The resulting
two-parameter family of strictly isospectral potentials will be
$$
V_{\lambda _{1}, \lambda _{2}}=
V_{0}-2\frac{d^2}{dx^2}\ln\Big[\left(\lambda _{1}+\int ^{x}u_{0}^{2}\right)
\left(\lambda _2 +\int ^{x}v_{\lambda _1}^{2}\right)\Big]~.
\eqno(7)
$$
At the $i$th -parameter level, one will have
$$
V_{\lambda _{1}, \lambda _{2},... \lambda _{i}}=
V_{0}-2\frac{d^2}{dx^2}\ln\Big[\left(\lambda _{1}+\int ^{x}u_{0}^{2}\right)
\left(\lambda _2 +\int ^{x}v_{\lambda _1}^{2}\right)...
\left(\lambda _{i}+\int^{x}v_{\lambda _{1}...\lambda _{i-1}}^{2}\right)\Big]
\eqno(8)
$$
and
$$
v_{\lambda _{1}...\lambda _{i}}=
\frac{\sqrt{\Lambda _1 \Lambda _2...\Lambda _{i}}
u_{0}}{(\lambda _{1}+\int ^{x}u_{0}^{2})...(\lambda _{i}+
\int ^{x}v_{\lambda _{1}...\lambda _{i-1}}^{2})}~.
\eqno(9)
$$
Explicit formulas for the parametric zero modes can be obtained if
one uses a notation based on the integration factor
$\int _{c}^{x}F_{0}={\cal F}(x)-{\cal F}(c)=\Delta _{x}{\cal F}$. Then
$$
v_{\lambda _1}(x)=
\frac{\sqrt{\Lambda _1}u_0}{
\lambda _1 +\Delta _{x}{\cal F}}~.
\eqno(10)
$$
Next, one can calculate
$$
\int _{c}^{x}\frac{u_{0}^{2}}{(\lambda +\int _{c}^{x}
u_{0}^{2})^2}=\int _{c}^{x}
\frac{{\cal F}^{'}dx}{(\lambda -{\cal F}(c)+{\cal F}(x))^{2}}
=\int _{{\cal F}(c)}^{{\cal F}
(x)}\frac{dz}{(\lambda -{\cal F}(c)+z)^2}=
$$
$$
=\frac{1}{\lambda}
-\frac{1}{\lambda +\Delta _{x}{\cal F}}=\frac{\Delta _{x}{\cal F}}
{\lambda(\lambda +\Delta _{x}{\cal F})}~.
\eqno(11)
$$
Thus
$$
v_{\lambda _1,\lambda _2}(x)=\frac{\sqrt{\Lambda _1\Lambda _2}u_0}{
\lambda _1 \lambda _2+(\lambda _1+\lambda _2+1)\Delta _{x}{\cal F}}~.
\eqno(12)
$$
At the next step one gets
$$
v_{\lambda _1,\lambda _2, \lambda _3}(x)=
\frac{\sqrt{\Lambda _1\Lambda _2\Lambda _3}u_0}{
\lambda _1 \lambda _2 \lambda _3+
(\lambda _1\lambda _2+\lambda _2\lambda _3+\lambda _3\lambda _1+
\lambda _1+\lambda _2+\lambda _3 +1)\Delta _{x}{\cal F}}
\eqno(13)
$$
and the general formula at the $i$ level can be written down in the form
$$
v_{\lambda _1,\lambda _2,..., \lambda _{i}}(x)=
\frac{\sqrt{\Lambda _1....\Lambda _{i}}u_{0}}{C_{1}^{(i)}+
C_{2}^{(i)}\Delta _{x}{\cal F}}~,
\eqno(14)
$$
where the first coefficient in the denominator is the product of all
parameters, whereas the second coefficient is just the sum over all the rest
of lower order Viete-type expressions in the parameters.
By the same token, one can write a general formula for the strictly
isospectral potentials
$$
V_{\lambda _1,\lambda _2,...,\lambda _{i}}=V_{0}-
2D^2\ln[C_{1}^{(i)}+C_{2}^{(i)}\Delta _{x}{\cal F}]=
V_{0}-\frac{4C_{2}^{(i)}u_{0}u_{0}^{'}}{C_{1}^{(i)}+
C_{2}^{(i)}\Delta _{x}{\cal F}}+\frac{2(C_{2}^{(i)})^2u_{0}^{4}}{
(C_{1}^{(i)}+C_{2}^{(i)}\Delta _{x}{\cal F})^2}~,
\eqno(15)
$$
which may be considered as the generalization of the furthest right-hand side
of Eq.~(4) and
represents a simple generalization of Mielnik's one-parameter potentials.
Since $\Lambda _1...\Lambda_{i}=C_{1}^{(i)}(C_{1}^{(i)}+C_{2}^{(i)})$, one
might
think that there is nothing new in (14) and (15) with respect to a common
Mielnik solution with an effective parameter $\lambda _{eff}^{(i)}
=C_{1}^{(i)}/C_{2}^{(i)}$. However, I will argue that by performing such an
equivalence one loses a certain type of information.
This information is a consequence of the symmetry of
(14) and (15) in the space of parameters. One can see that the subindices of
any pair of parameters can be interchanged without affecting the formulas.
Thus, each $\lambda$ parameter can be varied independently of the others,
making it possible to put questions related to the following type of situation.
Suppose we construct two Mielnik potentials corresponding to $\lambda _1$
and $\lambda _2$ and ask what is the potential bearing true zero modes that
for $\lambda _1\rightarrow \pm\infty$ goes to the Mielnik case
for $\lambda _2$,
whereas for $\lambda _2\rightarrow \pm\infty$ it goes to the Mielnik case for
$\lambda _1$. The answer is provided by the construction of this work and
corresponds to the particular case $V_{\lambda _1,\lambda _2}$ bearing the
true zero modes $v_{\lambda _1,\lambda _2}$. Indeed, as one can easily check,
$v_{\lambda _1,\pm\infty}=v_{\lambda _1}$ and
$V_{\lambda _1,\pm\infty}=V_{\lambda _1}^{M}$,
whereas $v_{\pm\infty, \lambda _2}
=v_{\lambda _2}$ and $V_{\pm\infty,\lambda _2}=V_{\lambda _2}^{M}$. In the
general case, one starts with a set of $i$ Mielnik potentials corresponding
to $i$ fixed values of Mielnik's parameter and asks the same question, this
time for the set of $i$ asymptotic limits. The answer is given by
(14) and (15) and cannot be provided if one works with only one effective
parameter unless its multiple-parameter value found above is used.
Another interesting remark is the one-to-one relationship between any
polynomial equation $a_0x^{i}+a_{1}x^{i-1}+...+a_{i}=0$ and the present
iterative construction. If we consider the $\lambda$ parameters as the
zeros of such arbitrary polynomials, we can write (14) as
$$
v_{\lambda _1,\lambda _2,..., \lambda _{i}}=
\frac{\sqrt{(-1)^{i}a_{i}(\sum _{0}^{i}(-1)^{i}a_{i})}u_{0}}{(-1)^{i}a_i+
(\sum _{0}^{i-1}(-1)^{i}a_i)\Delta _{x}{\cal F}}~,
\eqno(16)
$$
and in (15) one can substitute the same type of denominator as in (16).
There is only one constraint on the employed polynomials, which one
should impose in order to avoid possible singularities.
Usually the integral $\Delta _{x}{\cal F}$
in the denominators of (15) and (16) is of the kink type, i.e., it may be
written in the form $\alpha +\beta K(x)$, where the function $K(x)$ has the
kink behavior,
taking values between -1 and +1 and $\alpha$ and $\beta$ are some constants, of
which $\alpha$ may be zero. Then the allowed intervals for the
effective parameter are $\lambda _{eff}^{(i)}>\beta -\alpha$ and
$\lambda _{eff}^{(i)}<-(\beta +\alpha)$. When $\alpha =0$ one gets
$|\lambda _{eff}^{(i)}|> \beta$.
It is worthwhile to mention that the previous iteration process can be
understood most easily from
the Riccati equation standpoint as follows. To get, for example,
the two-parameter zero
mode, one should start again with the FR equation $y^{'}=-y^{2}+V_{1}(x)$
and take as the known particular solution $y_{p}^{(1)}=y_{0}+y_{\lambda _{1}}$,
where $y_{\lambda _{1}}=\frac{F_{0}}{\lambda _{1} +\int _{c}^{x}(F_{0})}$.
The intermediate Bernoulli equation will
be $-w_{1}^{'}=w_{1}^{2}+2y_{p}^{(1)}w_{1}$. This is turned into a
first-order
differential equation by the inverse function method. The integration factor
of the latter is
$F_{\lambda _1}={\rm exp}^{(-\int _{c}^{x}2y_{p}^{(1)})}$ and the solution for
the first-order differential equation can be written $w_2=(\lambda _2+\int _{c}^{x}
F_{\lambda _1}dz)/F_{\lambda _1}$. From this presentation it is clear how one
should proceed for an arbitrary step. Also, the logarithmic derivative notation
in Eq. (1) is equally convenient to have a clear image of the iteration
process. Thus, one can generate hierarchies of parametric
Schr\"odinger zero modes of any desired order by means of the Riccati
connection.
The parametric
normalization deletes the interval $[-1,0]$ from the parameter space
of $\lambda _{eff}^{(i)}$.
At the $-1$ limit, one can make a connection with the Abraham-Moses
isospectral technique [\refcite{am}], whereas at the $0$ limit the connection
can be done
with another isospectral construction developed by Pursey [\refcite{p}].
This connection is only from the point of view of the potentials; the
zero modes as worked out here just disappear.
In conclusion, I have shown explicitly the way Crum's iteration works when
the general Riccati solutions (general superpotentials) at zero factorization
energy are based on the
corresponding Schr\"odinger ground-state wavefunctions, obtaining
general formulas for this simple `generalization' of Mielnik's
one-parameter SUSYQM isospectrality. Plots of the two-parameter
formulas for the harmonic oscillator case are presented
in Fig. 1-5.
One may consider the
results of this work as pointing to an interesting hierarchical
structure within the general Riccati solution
produced by a particular type of repeated Darboux transformations
when the normalization condition of quantum mechanics is taken care of at each
iterative step.
\newpage
\nonumsection{Acknowledgments}
\noindent
This work was partially supported by the CONACyT Project 458100-5-25844E.
I wish to thank D.J. Fern\'andez and B. Mielnik for very useful
correspondence.
\nonumsection{References}
|
\section{Introduction}
\label{sect:intro}
Although there are quantum systems which cannot be obtained by
quantizing a classical system, the route through a classical
theory, via canonical quantization or
path integrals, is by far the most important way of
constructing a quantum theory. One of the most profound problems in
physics today is to quantize the classical theory of relativity, an
issue to which enormous and diverse efforts have been devoted
in the past 50 years. It may therefore be useful to re-examine this
quantization procedure with care. Here we discuss one interesting point
discovered in such a re-examination, although its implications (if any)
specifically for the quantization of gravity are as yet unclear.
Rather, it appears to bear on the matter of quantizing classical systems
in general.
Given a classical system, we can directly observe
the particle trajectories.
Thus we can determine the
classical equations of motion, but not directly the Hamiltonian
or the Lagrangian. From the equations of motion, we (i) construct
the Hamiltonian, Lagrangian or action, which
(ii) by canonical quantization
or a path integral, gives us the
quantum description. The quantum mechanics of the system
is determined by the action, but the classical paths picked out by the
equations of motion concern only the extrema; one certainly
does not expect, in general, a functional to be uniquely
determined from its extrema. From this point of view,
it is remarkable that the
quantum mechanics can be determined to a large extent
from the classical equations of motion. The issue we address in this paper
is: {\em To what extent?}
There are well-known ambiguities in both steps (i)
and (ii) of this procedure.
In step (i), different Hamiltonians or Lagrangians
can be constructed from the
same equations of motion. A famous example is the Aharonov-Bohm
effect \cite{ab}. For an electron with charge $-e$
which is kept from entering a
long solenoid, the classical equation of motion is the same
whether the magnetic field inside the solenoid is
turned on or not --- but the Hamiltonians (or Lagrangians)
are different.
In the absence of a magnetic field,
outside the solenoid the Hamiltonian is
\begin{equation}
H = \frac{{\bf p}^ 2}{2m} \quad ,
\eeql{eq:h1}
\noindent
while with the magnetic field turned on, it is
\begin{equation}
H = \frac{ ( {\bf p} + e {\bf A} )^ 2} {2m} \quad .
\eeql{eq:h2}
\noindent
Although classically this difference in the Hamiltonians is not
observable, it leads to observable effects upon quantization.
In step (ii) of the quantization procedure, we have the well-known
ordering ambiguity, which appears in various guises.
In canonical quantization, it appears as different ways of
ordering products of the operators ${\hat x}$ and ${\hat p}$;
different orderings are possible as long as they lead
to Hermitian operators with the same classical
limit \cite{fn2}. In the configuration-space path-integral formulation,
it appears as the ambiguity in choosing the measure in the space
of paths, while in the phase-space path integral it appears as the
ambiguity in skeletonization \cite{hartle1}. Corresponding to these
different manifestations are various proposals to fix this
ambiguity \cite{hall}.
In this paper we address an ambiguity in step (i) of the
procedure, similar to but not the same as that of the
Aharonov-Bohm effect. It is similar in that different Hamiltonians
or Lagrangians can be constructed
for the same classical equations of motion. However, in the case
of the Aharonov-Bohm effect the Hamiltonians (\ref{eq:h1}) and (\ref{eq:h2}) are
related by a canonical transformation
\begin{equation}
{\bf x} \mapsto {\bf x}' = {\bf x} \quad,
\eeql{eq:canx}
\begin{equation}
{\bf p} \mapsto {\bf p}' = {\bf p} - e {\bf A} ( {\bf x} ) \quad ,
\eeql{eq:canp}
\noindent
with a generating function of the second type \cite{goldstein},
\begin{equation}
F_2 = {\bf p}' \cdot {\bf x}
+ e \int {\bf A} ( {\bf x} ) \cdot d{\bf x} \quad .
\eeql{eq:genfn}
\noindent
With (\ref{eq:canx}) -- (\ref{eq:genfn})
we can readily obtain \cite{goldstein} the
Hamiltonian (\ref{eq:h2}) from
form (\ref{eq:h1}). Since the transformation is canonical, Poisson brackets are
preserved \cite{goldstein2}. With the Poisson brackets turned into
commutators upon quantization \cite{fn2}, it is guaranteed that the two
Hamiltonians lead quantum-mechanically to the same local physics, although
globally they are different, as manifested in the Aharonov-Bohm effect.
In comparison, the ambiguity we examine here is that, in general, for
the same classical motions, there
exist different Hamiltonians which are {\em not} canonically
equivalent (even locally); these will
lead to different quantum behaviors.
It is amazing that such an ambiguity, which could have been
discovered 70 years ago, has not to our knowledge previously appeared in
the literature.
In Section \ref{sect:free} we describe the construction of different
Hamiltonians from a given equation of motion,
for systems with one degree of freedom. It is shown
that the difference cannot be eliminated, even locally,
by a canonical transformation. Section \ref{sect:ex} gives simple
examples, demonstrating that the different
Hamiltonians lead to different quantum descriptions.
Section \ref{sect:lag} gives the Lagrangian treatment,
in which the result can in many cases be relatively simple.
Section \ref{sect:dis} is a brief conclusion, including remarks
on coupled systems with more than one degree of freedom. Where needed,
units with $\hbar = c = 1$ are used.
\section{Freedom in constructing the Hamiltonian}
\label{sect:free}
Suppose the equations of motion of a system
with one degree of freedom are found experimentally to be
\begin{equation}
\dot{p} + \frac{\partial H}{\partial x} = 0 \quad ,
\eeql{eq:eqp}
\begin{equation}
\dot{x} - \frac{\partial H}{\partial p} = 0 \quad ,
\eeql{eq:eqx}
\noindent
for a certain function $H =H(p,x)$. No doubt one can identify
this function $H$ as the Hamiltonian, and base canonical
quantization on it. However, classically (2.1) and (2.2) are
indistinguishable from
\begin{equation}
g_1 (p,x) \left[ \dot{p}+ \frac{\partial H}{\partial x} \right] +
g_2 (p,x) \left[ \dot{x}- \frac{\partial H}{\partial p} \right]
= 0 \quad ,
\eeql{eq:neweoma}
\begin{equation}
g_3 (p,x) \left[ \dot{p}+ \frac{\partial H}{\partial x} \right] +
g_4 (p,x) \left[ \dot{x}- \frac{\partial H}{\partial p} \right]
= 0 \quad ,
\eeql{eq:neweomb}
\noindent
as long as
\begin{equation}
\Delta \equiv g_1 g_4 - g_2 g_3 \ne 0
\eeql{eq:det}
\noindent
obtains. In general, we can form not just linear combinations of
(\ref{eq:eqp})
and (\ref{eq:eqx}), and the $g$'s can depend on $p$,
$x$ and the time $t$. For simplicity, we shall restrict to
linear combinations in (\ref{eq:eqp})
and (\ref{eq:eqx}), and time-independent $g$'s. This will suffice
to demonstrate the
ambiguity in constructing the Hamiltonian.
We examine the possibility that (\ref{eq:neweoma})
and (\ref{eq:neweomb}) are the Hamiltonian equations
of motion for another Hamiltonian $H'(p',x)$,
where the coordinate $x$ (being directly observable)
is unchanged, i.e., $x' = x$, but the momentum $p'$ could be different
from $p$. This then requires
\begin{equation}
\dot{p}' + \left. \frac{\partial H'}{\partial x} \right|_{p'}
= g_1 \left[ \dot{p} + \frac{\partial H}{\partial x} \right]
+ g_2 \left[ \dot{x}- \frac{\partial H}{\partial p} \right] \quad ,
\eeql{eq:neweleq1}
\begin{equation}
\dot{x} - \left. \frac{\partial H'}{\partial p'} \right|_{x}
= g_3 \left[ \dot{p} + \frac{\partial H}{\partial x} \right]
+ g_4 \left[ \dot{x}- \frac{\partial H}{\partial p} \right] \quad .
\eeql{eq:neweleq2}
\noindent
The partial derivatives on the RHS of (\ref{eq:neweleq1}) and
(\ref{eq:neweleq2}) are understood to be
at fixed $p$ or $x$. By taking
\begin{equation}
p' = p' (p,x) \quad ,
\eeql{eq:newp}
\begin{equation}
H' = H' ( p'(p,x), x) \quad ,
\eeql{eq:newh}
\noindent
the LHS of (\ref{eq:neweleq1}) and (\ref{eq:neweleq2}) can be rewritten
as functions of $x, \dot{x}, p$ and $\dot{p}$, as can the RHS.
Since $x, \dot{x}, p$ and $\dot{p}$ are independent, the
coefficients of $\dot{x}$ and $\dot{p}$ in (\ref{eq:neweleq1}) must separately
match. By converting the partial derivatives on the LHS
from fixed $p'$ to fixed $p$ and $x$, we find
\begin{equation}
\frac{\partial p'}{\partial x} = g_2 (p,x) \quad ,
\eeql{eq:tr1}
\begin{equation}
\frac{\partial p'}{\partial p} = g_1 (p,x) \quad .
\eeql{eq:tr2}
\noindent
The other terms in (\ref{eq:neweleq1}) give
\begin{equation}
\frac{\partial H'}{\partial x}
- \frac{\partial p' / \partial x} {\partial p' / \partial p}
\frac{\partial H'}{\partial p}
= g_1 \frac{\partial H}{\partial x}
-g_2 \frac{\partial H}{\partial p} \quad .
\eeql{eq:tr3}
\noindent
Likewise, (\ref{eq:neweleq2}) leads to
\begin{equation}
g_4 ( p,x) = 1 \quad ,
\eeql{eq:tr4}
\begin{equation}
g_3 ( p,x) = 0 \quad ,
\eeql{eq:tr5}
\begin{equation}
\frac{1}{\partial p' / \partial p}
\frac{\partial H'} {\partial p}
= \frac{\partial H}{\partial p} \quad .
\eeql{eq:tr6}
\noindent
The conditions (\ref{eq:tr1}) -- (\ref{eq:tr6}) imply the pair of equations
\begin{equation}
\frac{\partial H'}{\partial x}
= g_1 (p,x) \frac{\partial H}{\partial x} \quad ,
\eeql{eq:newh1}
\begin{equation}
\frac{\partial H'}{\partial p}
= g_1 (p,x) \frac{\partial H}{\partial p} \quad .
\eeql{eq:newh2}
\noindent
Hence the integrability condition for $H'$ is
\begin{equation}
\frac{\partial H}{\partial x}
\frac{\partial g_1}{\partial p}
- \frac{\partial H}{\partial p}
\frac{\partial g_1}{\partial x}
\equiv \left[ H , g_1 \right]_{x,p} = 0 \quad ,
\eeql{eq:intcond}
\noindent
with $[ \;\; ]_{x,p}$ being the Poisson bracket with respect to the
independent variables $x$ and $p$. This means that $g_1(p,x)$ is
a constant of motion.
In one dimension, there is only one nontrivial constant
of motion, namely $H$. Thus, we must have
\begin{equation}
g_1 = F(H) \equiv \frac{d {\tilde F}(H)}{dH} \quad .
\eeql{eq:defnF}
\noindent
From (\ref{eq:newh1}) and (\ref{eq:newh2}),
$\partial H' / \partial x =
\partial {\tilde F} / \partial x$, and
$\partial H' / \partial p =
\partial {\tilde F} / \partial p$,
so that we get immediately
\begin{equation}
H' = {\tilde F}(H) \quad ,
\eeql{eq:newhint}
\noindent
up to an irrelevant additive constant.
This expresses $H'$ in terms of $(p,x)$. To express
$H'$ in terms of $(p', x)$, we note that (\ref{eq:tr2})
determines $p'$ as
\begin{equation}
p' = \int_0^p dp \, g_1(p,x) \,+\, S(x) \quad ,
\eeql{eq:intnewp}
\noindent
up to the arbitrary function $S(x)$. In view of
(\ref{eq:tr1}), this is the only remaining freedom
in $g_2$. In many examples, $H$
is even in $p$; if we want to maintain the same
condition for $H'$ in terms of $p'$, then we would
have to choose $S(x) =0$.
To be more explicit, the formal solution can be expressed
as follows. We write
\begin{equation}
F(H) = \sum_n a_n(x) p^{2n} \quad .
\eeql{eq:powerF}
\noindent
(For simplicity we take only even powers, though this
restriction is easily lifted.) This yields
\begin{equation}
p' = \sum_n \frac{1}{2n+1} a_n(x) p^{2n+1} \quad ,
\eeql{eq:newp2}
\noindent
where, consistent with the assumption
that $H$ is even in $p$, we have set the
integration constant to $S(x) =0$.
It is easily checked that (\ref{eq:newp2}) solves (\ref{eq:tr1})
as well, because we have guaranteed the integrability
condition. The inverse transformation is formally
\begin{equation}
p = \sum_n b_n(x) p'^{2n+1} \quad ,
\eeql{eq:newpinv}
\noindent
with
\begin{eqnarray}
b_0 &=& \frac{1}{a_0} \quad ,
\nonumber \\
b_1 &=& - \frac{1}{3} \frac{a_1}{a_0^4} \quad ,
\nonumber \\
b_2 &=& \frac{1}{3} \frac{a_1^2}{a_0^7}
- \frac{1}{5} \frac{a_2}{a_0^6} \quad ,
\eeal{eq:b}
\noindent
etc.
Thus, given any $H$, we can obtain
a host of other Hamiltonians by choosing (i)
an arbitrary fucntion $F(H)$
and (ii) an arbitrary
function $S(x)$.
The resulting
Hamiltonian equations are guaranteed to be equivalent to the
original ones, (\ref{eq:eqp}) and (\ref{eq:eqx}),
as long as $F$ is nonzero
[cf. Eqs. (\ref{eq:det}), (\ref{eq:tr4}) and
(\ref{eq:tr5})].
Finally consider
the Jacobian of the transformation
$(x, p) \mapsto (x' , p')$, where $x' = x$:
\begin{equation}
J = \det
\pmatrix {
\partial x' / \partial x &
\partial x' / \partial p \cr
\partial p' / \partial x &
\partial p' \partial p }
= [ x' , p' ]_{x,p}
= g_1(p,x) \quad .
\eeql{eq:jac}
\noindent
A canonical transformation must have $J = 1$. The so-called
extended canonical transformation \cite{goldstein} (with scale
transformation included) has $J = \mbox{constant}$,
but not a function
of $p$ or $x$. Provided $g_1 = F(H)$ is not chosen to
be a constant over all of phase space, the transformation
$(x,p) \mapsto (x', p')$ is not canonical.
\section{Examples}
\label{sect:ex}
\subsection{Free particle}
\label{subsect:fp}
A Newtonian free particle is described by the Hamiltonian
\begin{equation}
H = \frac{p^2}{2m} \quad .
\eeql{eq:hfree}
\noindent
The choice
\begin{equation}
F(H) = \left( 1 - 2H/m \right)^{-3/2}
= \left( 1 - p^2/m^2 \right)^{-3/2} \quad ,
\eeql{eq:fex1}
\noindent
leads to
\begin{equation}
H' = {\tilde F} = \left(m^2 - p^2 \right)^{-1/2} \quad ,
\eeql{eq:newhex1c}
\begin{equation}
p' = \frac{p}{ (1 - p^2 /m^2 )^{1/2} } \quad .
\eeql{eq:newpex1}
\noindent
In terms of $p'$, we have
\begin{equation}
H' = \left( p'^2 + m^2 \right)^{1/2} \quad ,
\eeql{eq:newhex1b}
\noindent
which is the Hamiltonian of a {\em relativistic}
free particle! It is physically obvious
that by observing classical free particles [provided
the particles do not travel at speeds greater than unity, i.e.,
provided $p<m$, as per (\ref{eq:newhex1c})],
the two theories cannot be distinguished.
Although
this relationship between the Newtonian and the relativistic
free particle might not have a deep physical meaning,
this example clearly demonstrates that drastically different-looking theories
can correspond to the same classical motions.
It is well known that the quantum theory of a
relativistic point particle is very different from that of a
Newtonian free particle, with the former sharing some common
features with quantum gravity \cite{hartle1}.
Although the Hamiltonian (\ref{eq:newhex1b}) is nonpolynomial
in $p$, and is therefore
a non-local operator upon quantization, there is no reason why
canonical quantization cannot be based on it. This has been
studied by Newton and Wigner \cite{newton} (see also Hartle and
Kuchar \cite{hartle1}).
The result can be stated simply: a Fourier
component with wave number~$k$ evolves in time~$t$ with a phase
$\sqrt{k^2 + m^2}\, t$. This leads to the propagator
function $G'$ for $H'$, viz.,
\begin{eqnarray}
G'(x,t;x_0 , t_0 )
&=& \int_{-\infty}^{\infty} \frac{dk}{2 \pi} \,
e^{ik(x-x_0)}
e^{-i(k^2 + m^2 )^{1/2} (t-t_0 )}
\nonumber \\
&=& \lim_{\epsilon \rightarrow 0^+ }
\left[ \frac{im(t-t_0 -i \epsilon) }{\pi \Delta \lambda}
K_1 (m \Delta \lambda )
\right] \quad ,
\eeal{eq:green1}
\noindent
with $\Delta \lambda = [(x-x_0)^2 - (t- t_0 -i\epsilon )^2 ]^{1/2}$,
and $K_1$ the usual modified Bessel function \cite{redmount}. This
$G'$ is the probability amplitude for the wavefunction originally
localized at $x_0$ at time $t_0$ to be localized at $x$ at the later
time $t$, and has the same physical meaning as the familiar Newtonian
free-particle propagator $G$:
\begin{equation}
G (x,t; x_0 , t_0 )
= \sqrt { \frac{m}{2i \pi (t- t_0 )} }
\exp \left[ \frac{i m(x-x_0)^2}{2(t-t_0 )} \right] \quad ,
\eeql{eq:green2}
\noindent
corresponding to the Hamiltonian (\ref{eq:hfree}).
So although
both the Hamiltonians (\ref{eq:hfree}) and (\ref{eq:newhex1b})
give straight lines as
classical trajectories, the quantum theories they yield via canonical
quantization are very different.
\subsection{A system with a non-trivial potential}
\label{subsect:pot}
Consider a one-dimensional system with the classical Hamiltonian
\begin{equation}
H = x^2 p^2 + x^3 p + \mbox{\small $\frac{1}{4}$} x^4 \quad ,
\eeql{eq:hex2}
\noindent
and take
\begin{equation}
F(H)= H^{-1/2} =
\left(xp + \mbox{\small $\frac{1}{2}$} x^2 \right)^{-1} \quad .
\eeql{eq:g1ex2}
\noindent
The conditions (\ref{eq:tr1}) and (\ref{eq:tr2}) then give
\begin{equation}
p' = \frac{1}{x} \ln (xp + \mbox{\small $\frac{1}{2}$} x^2 ) + S(x) \quad ,
\eeql{eq:newpex2}
\noindent
for some function $S(x)$. We choose, again for simplicity,
$S(x)=0$. The new Hamiltonian is
\begin{eqnarray}
H' &=& 2H^{1/2} = 2 \left( xp + \mbox{\small $\frac{1}{2}$} x^2 \right)
\nonumber \\
& =& 2 e^{x p'} \quad .
\eeal{eq:newhex2a}
The easiest way to analyse the properties of this Hamiltonian is
to perform an additional canonical transformation
\begin{equation}
{\tilde p}= x p' \quad , \quad {\tilde x} = \ln x \quad .
\eeql{eq:canex2}
\noindent
The generating function is of the second type \cite{goldstein}:
\begin{equation}
F_2(x, {\tilde p} ) = {\tilde p} \ln x \quad .
\eeql{eq:genfnex2}
\noindent
The Hamiltonian (\ref{eq:newhex2a}) becomes
\begin{equation}
{\tilde H} = 2 e^{{\tilde p}} \quad .
\eeql{eq:newhex2b}
\noindent
The Hamiltonian equations of motion are then trivial, yielding the classical
paths
\begin{equation}
{\tilde x} (t) = 2 e^{{\tilde p}} t + \mbox{const} \quad ,
\eeql{eq:xcex2a}
\noindent
which, via (\ref{eq:canex2}), gives
\begin{equation}
x(t) = e^{C_1 t + C_2} \quad ,
\eeql{eq:xcex2b}
\noindent
where $C_1$ and $C_2$ are the two integration constants.
It can be checked that trajectory (\ref{eq:xcex2b})
indeed solves the Hamiltonian
equations of the original Hamiltonian (\ref{eq:hex2}).
The canonical quantization of the Hamiltonian (\ref{eq:hex2})
yields
a different quantum theory than that of (\ref{eq:newhex2a})
[which is
the same as that of (\ref{eq:newhex2b})]; this
can be seen by comparing the Poisson
brackets in the two cases. For any pair of functions
\begin{eqnarray}
u &=& u( p', x) = u\Bigl( p' (p,x),x\Bigr) \quad ,
\label{eq:u}\\
v &=& v( p', x) = v\Bigl( p' (p,x),x\Bigr) \quad ,
\eeal{eq:v}
\noindent
it is straightforward to verify
\begin{eqnarray}
\left[u , v \right]_{x, p'}
&\equiv& \frac{\partial u}{\partial x}
\frac{\partial v}{\partial p'}
- \frac{\partial u}{\partial p'}
\frac{\partial v}{\partial x}
\nonumber \\
&=& e^{x p'}
\left( \frac{\partial u}{\partial x} \frac{\partial v}{\partial p}
- \frac{\partial u}{\partial p} \frac{\partial v}{\partial x}
\right) = e^{xp'} \left[ u , v \right]_{x,p} \quad ,
\eeal{eq:uv}
\noindent
where the factor $e^{xp'}$ is just the inverse of
the Jacobian $g_1$ of the transformation. As the Poisson
bracket gives the leading term in order of $\hbar$ of the quantum
commutator, the Hamiltonians (\ref{eq:hex2}) and (\ref{eq:newhex2a})
must correspond to
different quantum theories. In particular, since the commutators
are different, the eigenvalue spectra for observables will be
different in the two theories.
\subsection{The harmonic oscillator}
\label{subsect:ho}
Even thoroughly understood systems can be subjected to this type of
transformation. For the harmonic oscillator, experimental
evidence implies that the Hamiltonian must essentially be
\begin{equation}
H = \frac{p^2}{2} + \frac{x^2}{2} \quad .
\eeql{eq:hex3}
\noindent
We consider three examples of transformations.
\noindent
{\em Example 1}
\noindent
We take
\begin{equation}
F = 1 + \frac{\epsilon}{2H}
= 1 + \epsilon \left( p^2 + x^2 \right)^{-1} \quad ,
\eeql{eq:defFex3}
\noindent
with $\epsilon \ll 1$. Integrating this in $H$, we find
\begin{equation}
{\tilde F} = H + \frac{\epsilon}{2} \ln H \quad .
\eeql{eq:newhex3c}
\noindent
Using (\ref{eq:tr2}),
we find
\begin{equation}
p' = p + \frac{\epsilon}{x} \arctan \frac{p}{x} \quad .
\eeql{eq:newpex3b}
\noindent
If $H'$ is now expressed in terms of $p'$ and $x$, we
get
\begin{equation}
H' = \frac{p'^2}{2} + \frac{x^2}{2}
+ \epsilon \left[ \frac{1}{2} \ln \frac{p'^2+x^2}{2} -
\frac{p'}{x} \arctan \frac{p'}{x} \right] + O(\epsilon^2) \quad .
\eeql{eq:newhex3e}
\noindent
{\em Example 2}
\noindent
As another example, we choose
\begin{equation}
F = 1 + \epsilon (2H) = 1 + \epsilon(p^2 + x^2) \quad .
\eeql{eq:fex3b}
\noindent
Follwing the same steps, we find
\begin{equation}
H' = \frac{p'^2}{2}
+ \frac{x^2}{2}
+ \epsilon \left[ \mbox{\small $\frac{1}{4}$} ( p'^2 + x^2 )^2
- \sfrac{1}{3}p'^4
- x^2 p'^2 \right] \quad ,
\eeql{eq:newhex3d}
\begin{equation}
p' = p + \epsilon
\left( \sfrac{1}{3}p^3 + x^2 p \right) \quad .
\eeql{eq:newpex3d}
\noindent
{\em Example 3}
\noindent
In this case we take
$F = H$
and ${\tilde F} = \mbox{\small $\frac{1}{2}$} H^2$.
The relation between $p'$ and $p$ is given by (\ref{eq:tr2}),
leading to
\begin{equation}
p' = \sfrac{1}{6}p^3 + \mbox{\small $\frac{1}{2}$} x^2 p \quad ,
\eeql{eq:newpex3h}
\noindent
where we have set the integration constant to zero
under the assumption that parity is maintained
(i.e., when $p \mapsto -p$, then $p' \mapsto -p'$).
The inverse relation cannot be expressed in closed
form, but is formally given by the power series
\begin{equation}
p = \frac{2}{x^2} p' - \frac{8}{9x^8} p'^3 + \cdots \quad .
\eeql{eq:pex3inv}
When (\ref{eq:pex3inv}) is substituted into
$H' = {\tilde F} = (p^2 + x^2 )^2/8$, we obtain $H' = H'(p',x)$, which
is guaranteed to give the same classical motion.
However, this is a complicated expression.
We shall see later that the complication arises
entirely from $p'$, and the same situation has a simple
description in the Lagrangian formulation because
$p'$ will not appear.
\section{Lagrangian formulation}
\label{sect:lag}
The preceding examples show that at least part of the
complication arises from the transformation from $p$ to
$p'$, and the inverse transformation, e.g.,
(\ref{eq:pex3inv}). This would suggest that the formalism
may be simpler in a Lagrangian description, in so far
as neither $p$ nor $p'$ need appear.
\subsection{Formalism}
\label{subsect:lagform}
The Lagranian equation of motion is
\begin{equation}
\frac{d}{dt} \left( \frac{\partial L}{\partial \dot{x}} \right)
- \frac{\partial L}{\partial x} = 0 \quad .
\eeql{eq:lag1}
\noindent
with $L = L(x, \dot{x})$. In greater detail, and adopting the
notation $( x, \dot{x} , \ddot{x} ) \mapsto (x, y, z)$,
and $L_x = \partial L / \partial x$ etc., we have
\begin{equation}
z L_{yy} + y L_{xy} - L_x = 0 \quad .
\eeql{eq:lag2}
\noindent
For this to be a non-degenerate second-order
differential equation, we shall assume $L_{yy} \ne 0$.
(In the Newtonian case, $L_{yy} = m$.)
This could just as well be written with an extra multiplcative
factor of $F = F(x,y)$. If this modified equaton is to
be the Euler-Lagrange equation derived from another Lagrangian
$L'$, then we must have
\begin{equation}
z L'_{yy} + y L'_{xy} - L'_x
= F \left( z L_{yy} + y L_{xy} - L_x \right) \quad .
\eeql{eq:lag3}
\noindent
We have used the notation $F$ in anticipation
that this function will turn out to be the same
as $g_1$ in the Hamiltonian formulation.
In the above equation, $z$ appears only where shown explicitly,
and since (\ref{eq:lag3}) must hold as an identity, the terms
with and without $z$ have to be separately equal, leading
to two conditions
\begin{equation}
L'_{yy} = F L_{yy} \quad ,
\eeql{eq:lag4}
\begin{equation}
yL'_{xy} - L'_x = F \left( yL_{xy} - L_x \right) \quad .
\eeql{eq:lag5}
In these expressions, we write
the quantities on the LHS as
\begin{eqnarray}
L'_{yy} &=& \partial_y L'_y = \partial_y p' \quad ,
\nonumber \\
L'_{xy} &=& \partial_x L'_y = \partial_x p' \quad .
\eeal{eq:lag8}
\noindent
Their compatibility requires the integrability
condition that $\partial_x \partial_y p'
= \partial_y \partial_x p'$. Some arithmetic then
leads to
\begin{equation}
y F_x L_{yy} - y F_y L_{xy} + F_y L_x = 0 \quad .
\eeql{eq:lag9}
On the other hand, we can consider the quantity
\begin{eqnarray}
&&L_{yy} \dot{F}
\nonumber \\
&=&L_{yy} \left( F_x \dot{x} + F_y \dot{y} \right)
\nonumber \\
&=&L_{yy} \left( F_x y + F_y z \right) \quad ,
\eeal{eq:lag10}
\noindent
and use the equation of motion (\ref{eq:lag2})
to eliminate $z L_{yy} $. This shows that
$L_{yy} \dot{F}$ is exactly the LHS of (\ref{eq:lag9}),
implying $\dot{F}=0$, i.e., $F$ is
a constant of motion.
In fact, it is easy to identify
what $F$ is. Note that in (\ref{eq:lag5}), we can write
the bracket as
\begin{equation}
yL_{xy} - L_x
= \partial_x \left( yL_y - L \right)
= \partial_x H \quad ,
\eeql{eq:lag6}
\noindent
given $L_y = p$. The same holds for the analogous
expression involving $L'$. Thus (\ref{eq:lag5})
is equivalent to
$\partial_x H' = F \partial_x H$,
confirming that $F$ is indeed the same as $g_1$ introduced
previously [cf. (\ref{eq:newh1})]. However, in the rest of this Section, we
shall not rely on any results from the Hamiltonian treatment.
Thus, once we choose any constant of motion $F$, we can
integrate to find $L'$, up to a function $S(x)$.
\subsection{Free particle}
\label{subsect:lagfree}
We return to the free particle,
with
$L = \mbox{\small $\frac{1}{2}$} my^2$,
$L_y = my$,
$L_{yy} = m$.
We choose $F = ( 1- y^2)^{-3/2}$, which
is a constant of motion, and solve for
\begin{equation}
L'_{yy} = F L_{yy} = m ( 1- y^2)^{-3/2} \quad ,
\eeql{eq:lag12}
\noindent
giving, up to integration constants which we set to zero,
\begin{equation}
L' = - m ( 1- y^2 )^{1/2} \quad ,
\eeql{eq:lag13}
\noindent
which is the Lagrangian for the relativistic
free particle.
\subsection{Harmonic oscillator}
\label{subsect:lagsho}
For the harmonic oscillator, we have
$L = \mbox{\small $\frac{1}{2}$} y^2 - \mbox{\small $\frac{1}{2}$} x^2$,
$L_y = y$,
and $L_{yy} = 1$.
We choose
$F = \mbox{\small $\frac{1}{2}$} y^2 + \mbox{\small $\frac{1}{2}$} x^2$,
which is a constant. Thus, $L'$ is to be found
by integrating
\begin{equation}
L'_{yy} = F L_{yy} = \mbox{\small $\frac{1}{2}$} y^2 + \mbox{\small $\frac{1}{2}$} x^2 \quad ,
\eeql{eq:lag16}
\noindent
leading to
\begin{equation}
L' = \sfrac{1}{24} y^4 + \sfrac{1}{4} y^2 x^2 + f(x) \quad .
\eeql{eq:lag17}
\noindent
There should be two integration constants, but we have
set the term linear in $y$ to zero using the assumption
of parity. The function $f(x)$ is easily determined
by considering the terms without any $y$ in (\ref{eq:lag5}):
\begin{equation}
-f'(x) = \mbox{\small $\frac{1}{2}$} x^2 \left( +x \right) \quad .
\eeql{eq:lag18}
\noindent
Thus we have
\begin{equation}
L' = \sfrac{1}{24} \dot{x}^4 +
\sfrac{1}{4} \dot{x}^2 x^2 - \sfrac{1}{8} x^4 \quad .
\eeql{eq:lag19}
\noindent
It can be verified directly, without any of the
general formalism given above, that this Lagrangian
gives the same classical motion, with the Euler-Lagrange
equation being the usual one multiplied
by $F$.
In fact, this is the same as the last example presented
under the Hamiltonian treatment of the simple harmonic
oscillator, but now expressed in closed form.
Comparison shows that the Lagrangian
formulation is much simpler is this case, since
all the complications lie in $p' = L'_y$.
\subsection{Path integral}
\label{subsect:path}
In all these cases the quantum
mechanics derived from $L$ and $L'$ are different.
This is most easily seen by verifying that,
in the path-integral formulation,
the phases associated with paths are different
in nontrivial ways.
Consider two paths $x_a(t)$
and $x_b(t)$ with the same end-points at $t = t_1$
and $t=t_2$ [i.e.,
$x_a(t_1) = x_b(t_1) =x_1$,
$x_a(t_2) = x_b(t_2) =x_2$],
and the action integrals
$S_a = \int_{t_1}^{t_2} dt \, L[ x_a ]$ etc.
The differences
$\Delta S = S_a - S_b$ and
$\Delta S' = S'_a - S'_b$
are clearly different in general.
Therefore the quantum
mechanics as defined by the path integral will
be different in the two cases.
\section{Discussion}
\label{sect:dis}
We have shown that just as in the case of Aharonov-Bohm effect,
different Hamiltonians can be constructed for the same classical
equation of motion. However, unlike the case of the
Aharonov-Bohm effect, these Hamiltonians are {\em not} related by
canonical transformations. Although they must be regarded as the same
classical theory --- since it is the trajectories or
equations of motion, not the Hamiltonian, which are observed classically
--- these variant
Hamiltonians give different quantum theories upon canonical quantization.
We may place the ambiguity discussed here into context by
classifying the known quantization ambiguities into three
levels. First, there are {\em small} differences in the
sense that $H$ and $H'$ agree classically but differ
by $O(\hbar)$ quantum-mechanically; the operator-ordering
ambiguity is of this type. Second, there are cases where
$H - H'$ is not small (in the above sense), but can be made small
by performing a canonical transformation on one of them.
The canonical transformation may be global (in which case the
situation is the same as the first type), or only local,
as in the Aharonov-Bohm effect. The third level
are differences between $H$ and $H'$
that cannot be made small by a canonical transformation.
The ambiguities discussed in this paper are of this type,
and in a sense are the most ``serious."
This possibility of {\em in}equivalent
Hamiltonians raises several
interesting questions, to which we do not at present
have complete answers.
One might be tempted to impose the requirement that the Hamiltonian must be
polynomial in the momentum. However, this seems to be rather
{\em ad hoc}
as a fundamental principle. Moreover, in the case of a free particle,
nature favors the more complicated relativistic Hamiltonian
over the simpler
form (\ref{eq:hfree}). In the case of the harmonic oscillator,
Hamiltonian\ (3.24),
which is quadratic in $p$, seems to be the preferred one. It is not
clear whether either of the alternatives
(\ref{eq:hex2}) or (\ref{eq:newhex2a}) is more
appropriate than the other.
Is there any guiding principle in choosing
which one to use, in general?
Are all the other possibilities truly forbidden?
The discussion in this paper, and all the examples,
refer to systems with only one degree of freedom.
This may already be of some interest, since
some models of mini-superspace contain only
one degree of freedom \cite{super}.
However, the interesting
question is, of course, whether these ambiguities extend
to systems with more degrees of freedom.
The most general argument is that, for any number of degrees
of freedom, one has no right to expect a functional (the action)
to be determined by its extrema (the classically allowed
trajectories), and therefore the existence of alternate
Hamiltonians and Lagrangians should be the rule rather than
the exception. The challenge will be to discover interesting
examples, and if possible to characterize and classify
all the possible ambiguities.
\acknowledgments
This problem may be regarded as one of {\em inversion}:
how a system can be determined from the observables.
The work of W.~M.~S. and K.~Y. on inversion is supported in part
by a grant (CUHK 4006/98P) from the Hong Kong Research Grants Council.
|
\section{Introduction}
The standard cosmology is based on the assumption that our Universe is
homogeneous and isotropic. However, the present our Universe is not
homogeneous but has clumpy structures like stars, galaxies, clusters of
galaxies, and superclusters. The recent observation \cite{COBE} of the
highly isotropic cosmic microwave background radiation is usually
regarded as an evidence of the assumption, but this is the case only up
to the stage of decoupling. Nonetheless, one may want to think that the
homogeneous and isotropic universe model, known as the
Friedmann-Lema\^{\i}tre-Robertson-Walker (FLRW) model, reflects the
averaged nature of the true Universe over a scale larger than a
supercluster. In particular, we usually expect that the global
expansion of the Universe is well approximated by the FLRW model with
the energy distribution given by the volume average.
However, this ``averaging hypothesis'' is never justified in a trivial
way \cite{El}, since Einstein's equation is highly nonlinear. In
particular, averaging over a (spatial) volume does not commute with the
time-evolution in general, so the averaged initial data can develop in
time in a quite different way from the true data. One is therefore
required to clarify the meaning of the averaging and establish the
domain of applicability of it.
One of the earliest work connected to this problem is due to Futamase
\cite{Fu}, who built a formalism which gives the back reaction of small
scale inhomogeneities to the global expansion along the spirit of
Isaacson \cite{Isaa} using the post-Newtonian expansion. However, it is
difficult to justify the approximation used in this approach, and the
basic equations are still hard to handle. Zalaletdinov \cite{Zal}
proposed a covariant averaging formalism starting from some axioms, but
this is so complicated that it seems very difficult to draw useful
consequences. Up to now, in spite of efforts \cite{BE,RSKB,CM,IHK}
including the above, the averaging has not been well understood, because
of its complexities and conceptual difficulties.
In this paper we study the problem in a quite different way from the
conventional ones. We focus on the dynamics of the scale factor
$a(\tau)$, which is, if the space is closed \cite{note:closed},
equivalent to the dynamics of the total volume $V(\tau)$ in the sense
$V(\tau)\propto a^3(\tau)$. We examine to what extent the FLRW solution
can be a good model judging from a point of view of the time-development
of the total volume $V(\tau)$. Note that when given a space of
solutions spanned by some arbitrary functions of space, we can think of
$V(\tau)$ as a functional on it. By evaluating the variation of
$V(\tau)$ at the FLRW solution we may be able to obtain some information
about the quality of the FLRW solution as an ``averaged'' model. We
explicitly do this for the spherically symmetric dust case. As a result,
we find that the FLRW solution is a {\it critical point} for $V(\tau)$ in
an appropriate sense, which gives a good support for the validity of the
FLRW solution as an averaged model. We also evaluate the second
variation of $V(\tau)$, which gives the leading effect of the deviation
from the FLRW solution.
The exact solution for spatially spherically symmetric dust spacetimes
is known as the Lema\^{\i}tre-Tolman-Bondi (LTB) solution. We add to
this solution the assumption that the spatial manifold be $S^3$.
We give a description of this subclass of the LTB solution first.
\section{The spherically symmetric dust solution on $S^3$}
The metric is written with the synchronous and
comoving coordinates as:
\begin{equation}
\label{eq:1}
{\rm d} s^2={\rm d} \tau^2-e^{\lambda(\tau,R)}{\rm d}
R^2-r^2(\tau,R)({\rm d}\theta^2+\sin^2\theta{\rm d}\varphi^2),
\end{equation}
where $\lambda(\tau,R)$ and $r(\tau,R)$ are the functions to be
determined from the Einstein equation. The general solution to this
metric is well known as the Lema\^{\i}tre-Tolman-Bondi (LTB) solution
(see e.g. \cite{LL,Kr}), which possesses three arbitrary functions
$f(R)$, $F(R)$, and $\tau_0(R)$. (Our notation follows Ref.\cite{LL},
except for the sign of $f(R)$.) With these the function $\lambda$ is
given by $e^\lambda=r'^2/(1-f(R))$, where the dash stands for the
derivative with respect to $R$. The function $r$ is given in three
separate forms, depending upon whether the arbitrary function $f(R)$ is
negative, positive, or zero, and each solution possesses the FLRW limit
of negative ($k=-1$), positive ($k=1$), and flat ($k=0$) constant
curvature, respectively. (Such a limit is achieved when $r(\tau,R)$
separates as $r(\tau,R)=\Phi(\tau)\Psi(R)$ \cite{Kr}.) Since our model
is spatially $S^3$, the spatial manifold does admit a constant curvature
limit and it should be positive. So, it is natural to choose the
positive sign of $f(R)$ \cite{note:fsign}, for which the the function
$r$ is given by
\begin{equation}
\label{eq:2}
r={F\over 2f}(1-\cos\eta),\; \tau-\tau_0(R)={F\over
2f^{3/2}}(\eta-\sin\eta).
\end{equation}
The arbitrary function $\tau_0(R)$ is called the function of ``big bang
time'', since each Killing orbit ($R=$constant) degenerates at time
$\tau=\tau_0(R)$.
Now we have seen that the solution is parametrized by three
``arbitrary'' functions, but not all solutions are suitable for the
spatial manifold $S^3$. The aim of the remaining part of this section
is to describe the conditions imposed on the three functions to obtain
``regular'' solutions on $S^3$. (By a regular solution we mean that it
is regular on the spatial manifold $S^3$ {\it during a finite interval
of time} from the big bang.) We, by describing them, show that there
does exist sufficiently large set of the regular solutions in the space
of all formal solutions. This will be needed to make the formal
calculations in the next section realistic.
Let us think of the three-sphere as a sum of two balls, $S^3\simeq
D^3\cup D^3$. This is similar to the decomposition of two-sphere into
two disks, $S^2\simeq D^2\cup D^2$, achieved by cutting the two-sphere
along the equator. We can understand the spherically symmetric
three-sphere by thinking of each ball spherically symmetric. Since each
such ball has a symmetry center, the spherically symmetric three-sphere
has {\it two} symmetry centers, which are degenerate Killing orbits, as
well. We label these points as $R=0$ and $R=\pi$, so the relevant region
of the coordinate $R$ is $I\equiv [0,\pi]$. The first condition we
should impose is therefore
\begin{equation}
\label{eq:c0}
r=0, \; \mbox{at } R=0\mbox{ and }\pi.
\end{equation}
The function $r$ should be positive except at the boundaries.
Furthermore, we must impose some regularity conditions. An efficient way
to see this is to calculate the scalar curvature ${\cal R}$ and the
curvature scalar polynomial ${\cal R}_{abcd}{\cal R}^{abcd}$, where
${\cal R}_{abcd}$ is the curvature tensor. Using our local solution, we
find ${\cal R}=\varepsilon$ and ${\cal R}_{abcd}{\cal R}^{abcd}=
12(F/r^3)^2-8(F/r^3)\varepsilon+3\varepsilon^2$, where
$\varepsilon=F'/(r'r^2)$ is the energy density of dust. Since these
scalars must be finite, we have
\begin{equation}
\label{eq:c1}
{F\over r^3}<\infty, \; R\in [0,\pi]
\end{equation}
and
\begin{equation}
\label{eq:c2}
\varepsilon={F'\over r'r^2}<\infty, \; R\in [0,\pi].
\end{equation}
While condition \reff{eq:c1} is necessary to avoid the conical
singularity possibly generated at $R=0$ and $\pi$, condition
\reff{eq:c2} is imposed to avoid the well-known shell-crossing
singularity \cite{note:1}. Finally, we have to impose the following
coordinate condition, which is necessary for the coordinates to span the
spatial manifold $S^3$ well;
\begin{equation}
\label{eq:c3}
e^\lambda={r'{}^2\over 1-f}>0,\; R\in [0,\pi].
\end{equation}
Without this condition, we possibly have fictitious solutions.
To find the solution to the boundary conditions, note that there exists
the freedom of reparameterizations $R\rightarrow \gamma(R)$. Using this
freedom we can fix the leading power of $r$ at the boundaries to the
unity, i.e., we can make
\begin{equation}
\label{eq:r}
r\propto \left\{
\begin{array}{ll}
R \quad & (R\rightarrow 0) \\
\pi-R \quad & (R\rightarrow \pi)
\end{array}
\right. .
\end{equation}
Near the big bang singularity $(\eta\rightarrow 0)$, from Eq.\reff{eq:2} we
have $r\sim (F/(4f))\eta^2$, and $\tau-\tau_0(R)\sim
(F/(12f^{3/2}))\eta^3$, so we find $r\sim
(9/4)^{1/3}F^{1/3}(\tau-\tau_0(R))^{2/3}$. Taking the condition
\reff{eq:r} into account, we have
\begin{equation}
\label{eq:4}
F(R)\propto \left\{
\begin{array}{ll}
R^3 \quad & (R\rightarrow 0) \\
(\pi-R)^3 \quad & (R\rightarrow \pi)
\end{array}
\right. .
\end{equation}
Let $\sigma$ be a real number, and suppose $\eta\propto R^\sigma$
($R\rightarrow 0$) for an arbitrary fixed $\tau$. From Eq.\reff{eq:2}, we find
that if $\sigma < 0$, the function $r$ would oscillate heavily between
positive values and zero near the boundary. So, this case is unsuitable
for the condition. On the other hand, if $\sigma\geq 0$, the boundary
condition \reff{eq:c0} is satisfied and the behavior of $f(R)$ near the
boundaries can be determined as
\begin{equation}
\label{eq:3}
f(R)\propto \left\{
\begin{array}{ll}
R^{\alpha_1} \quad & (R\rightarrow 0) \\
(\pi-R)^{\alpha_2} \quad & (R\rightarrow \pi)
\end{array}
\right. ,
\end{equation}
where $\alpha_1,\alpha_2\geq 2$. The condition \reff{eq:c1} is now trivially
satisfied.
Next, consider the regularity conditions \reff{eq:c2} and \reff{eq:c3}.
Note that the function $r$ on $I$ for a fixed $\tau$ has at least one
extremum (FIG.\ref{fig:f1}), since $r>0$ on the interior of $I$ and
$r=0$ at the boundaries. From the condition \reff{eq:c2}, we find that
the function $F'$ should vanish where $r'$ vanishes
(FIG.\ref{fig:f2}). Since $F(R)$ is independent of $\tau$, $r'$ should
also vanish on $I$ independently from $\tau$. Otherwise, the regularity
would break instantaneously. A straightforward calculation gives
\begin{equation}
\label{eq:rd}
r'=\rcp{1-\cos\eta}\bra{\frac{F'}{2f}{\cal A}(\eta)+
\frac{Ff'}{2f^2}{\cal B}(\eta)-f^{1/2}\tau_0'\sin\eta},
\end{equation}
where ${\cal A}(\eta)\equiv (1-\cos\eta)^2-\sin\eta(\eta-\sin\eta)$, and
${\cal B}(\eta)\equiv -(1-\cos\eta)^2+\frac32\sin\eta(\eta-\sin\eta)$. Since
$r'$ is a homogeneous linear combination of $f'$, $F'$, and $\tau_0'$
with distinct coefficients as functions of $\tau$, the only possible
points on $I$ for which $r'$ vanish independently from $\tau$ are the
points where $F'$, $f'$ and $\tau_0'$ vanish simultaneously. In
particular, $r'$ and $F'$ should have the same sign everywhere on $I$
since otherwise the energy density $\varepsilon$ would become negative
in some regions. We exclude such a case for physical reason. On the
other hand, the sign of $f'$ is rather arbitrary, except that $f(R)$
should be maximal and take value $1$ where $F'$ vanishes
(FIG.\ref{fig:f3}). Taking value $1$ is a consequence of the condition
\reff{eq:c3}. The sign of $\tau_0'$ should be opposite to $F'$ or zero
(FIG.\ref{fig:f4}). This is a consequence of the condition that $r'$
have the same sign as $F'$ at least for a finite interval of time from
the big bang ($\eta=0$). In fact, the leading powers of ${\cal A}(\eta)$,
${\cal B}(\eta)$, and $\sin\eta$ in Eq.\reff{eq:rd} are, respectively,
$\rcp{12}\eta^4$, $-\rcp{80}\eta^6$, and $\eta$, so the term
proportional to $\tau_0'$ dominates near the big bang singularity if
$\tau_0'$ does not vanish. It is clear that if $\tau_0'$ had the same
sign as $F'$, $r'$ would have the opposite sign when $\eta\rightarrow 0$.
Now, we have spelled out all the conditions for $F(R)\geq0$,
$f(R)\geq0$, and $\tau_0(R)$. They are Eqs.\reff{eq:4} and \reff{eq:3},
that $f(R)$ should take maximum value $1$ where $F'$ vanishes, and that
the sign of $\tau_0'$ should be opposite to $F'$ or zero. It is nice to
keep in mind that the profile of $F(R)$ decides that of $r$ if
the solution is regular.
\section{Variation of $V$ and averaging}
Given an inhomogeneous solution, in what way can we measure the
resemblance between the solution and the FLRW solution? We note that the
only dynamical content of the FLRW model is the scale factor $a(\tau)$,
which can be regarded as the (cubic root of the) total volume $V(\tau)$
if the spatial manifold is closed \cite{note:closed}. Since $V(\tau)$ is
always well defined for any spatially closed inhomogeneous spacetime
\cite{note:4} there is a large amount of theoretical advantage to
utilize it. Note that $V$ can be considered as a functional
$V[\phi_1({\bf x}),\cdots \phi_n({\bf x})]$ on a space of solutions, where
$\phi_1({\bf x}),\cdots \phi_n({\bf x})$ are the arbitrary functions of space
which span the space of solutions. For example, in the spherically
symmetric dust case, $n=3$ and we can put
$(\phi_1({\bf x}),\phi_2({\bf x}),\phi_3({\bf x}))\equiv (F(R),f(R),\tau_0(R))$. If the
inhomogeneous solution is not too far from the FLRW solution, we can
compare the two solutions by evaluating the variation $\delta V(\tau)$
at the FLRW solution. If this function $\delta V(\tau)$ vanishes for all
$\tau$, we may say that the two solutions are dynamically close to each
other and the ``averaged'' solution of the inhomogeneous solution
corresponds to the FLRW solution.
Below, we apply the above idea to the spherically symmetric dust
solution on $S^3$. Our space $P$ of solutions are assumed to be spanned
by the regular solutions on $S^3$, and the variations are taken in
it. More specifically, $P$ is a subspace of the larger space $P^*$ which
is spanned by all possible configurations of three smooth functions
$(F(R),f(R),\tau_0(R))$, and defined by subtracting from $P^*$ the
configurations which correspond to irregular solutions. The space $P$
contains the FLRW solution, since this solution is regular.
The total volume for the metric \reff{eq:1} is
\begin{equation}
\label{eq:3-1}
V(\tau) = 4\pi\int_0^\pi e^{\lambda/2}r^2{\rm d} R,
\end{equation}
where $e^{\lambda}=r'{}^2/(1-f(R))$ and $r$ is given by
Eq.\reff{eq:2}. For later use, we also define the total energy $E$:
\begin{equation}
\label{eq:3-5}
E\equiv \int_{S^3}\varepsilon{\rm d} V=4\pi\int_0^\pi
\frac{|F'|}{\sqrt{1-f}}{\rm d} R,
\end{equation}
which is a conserved quantity. Moreover, for convenience we change the
parameterization $(F(R),f(R),\tau_0(R))$ to $(A(R),f(R),\tau_0(R))$
defined by $ A(R)\equiv F/(2f^{3/2})$. The standard FLRW limit is
achieved in this parameterization when
\begin{equation}
\label{eq:3-3}
(A(R),f(R),\tau_0(R))=(a_0,\sin^2 R,\tau_{0{\rm c}}),
\end{equation}
where $a_0$ is a positive constant parameter, and $\tau_{0{\rm c}}$ is another
constant parameter. Since we can always choose $\tau_{0{\rm c}}=0$ by the coordinate
transformation $\tau\rightarrow\tau+\tau_{0{\rm c}}$, parameter $\tau_{0{\rm c}}$ is redundant as far
as the FLRW solution is concerned, but we will find that in our wider
context it is useful not to fix this parameter. At the limit
\reff{eq:3-3}, $V$ is given by
\begin{equation}
\label{eq:EV0}
V_0(\tau)=2\pi^2a_0{}^3(1-\cos\eta)^3, \; \tau-\tau_{0{\rm c}}=a_0(\eta-\sin\eta).
\end{equation}
We vary the volume $V$ with respect to $A(R)$, $f(R)$, and $\tau_0(R)$,
and evaluate at the FLRW solution \reff{eq:3-3}. The formal result is
\begin{eqnarray}
\label{eq:3-3.5}
\delta V(\tau)= && 4\pi a_0^2 (1-\cos\eta){\cal A}(\eta)\int_0^\pi \paren{
3\delta A+\tan R\delta A'}\sin^2 R\,{\rm d} R \nonumber \\
&& +2\pi a_0^3
(1-\cos\eta)^3\int_0^\pi \paren{\tan R\, \delta f}' {\rm d} R \nonumber \\
&& -4\pi a_0^2 (1-\cos\eta)\sin\eta\int_0^\pi \paren{3\delta \tau_0+\tan
R\delta \tau_0'}\sin^2 R\,{\rm d} R.
\end{eqnarray}
We may expect that the second term vanishes if we take into account the
boundary conditions in the previous section, but we have to check the
continuity of the function $\tan R\delta f$ at $R=\pi/2$ to do this. The
zeroth variation of $f(R)$ is $\sin^2 R$ and the maximal value should
always be 1 as we explained in the previous section. This means that at
the neighborhood of $R=\pi/2$, the significant (or ``dangerous'')
variation of $f(R)$ should always be approximated by $\sin^2 (R+a)$,
where $a$ is a parameter. Differentiating this with respect to $a$ and
putting $a=0$, we find that $\delta f$ is approximated by $2\sin R\cos
R\, {\rm d} a$, so that $\tan R\,\delta f$ is approximated by $2\sin^2 R\, {\rm d}
a$, which is continuous at $R=\pi/2$. We can hence safely omit the term.
(In contrast to this, we cannot apply ``integrations by parts'' to the
first and the third terms due to the discontinuities of $\sin^2 R\tan
R\delta A$ and the similar term for $\delta\tau_0$.) Thus we have
\begin{eqnarray}
\label{eq:3-4}
\delta V(\tau)= && 4\pi a_0^2 (1-\cos\eta){\cal A}(\eta)\int_0^\pi \paren{
3\delta A+\tan R\delta A'}\sin^2 R\,{\rm d} R \nonumber \\
&& -4\pi a_0^2 (1-\cos\eta)\sin\eta\int_0^\pi \paren{3\delta \tau_0+\tan
R\delta \tau_0'}\sin^2 R\,{\rm d} R.
\end{eqnarray}
Note that the function $\eta$ depends only on $\tau$ when it concerns
the FLRW solution, so the functions of $\eta$ in the above expressions
have been factored out of the integrals. Because of this feature $\delta
V(\tau)$ becomes constantly zero, if (and only if) the variation of
$A(R)$ and $\tau_0(R)$ are taken so that the integrals in
Eq.\reff{eq:3-4} vanish. This means that the FLRW solution is a critical
point for the functional $V(\tau)$, if we restrict the directions of the
variation in that way.
The meaning of the particular directions becomes clear if we consider
the variation of the total energy $E$, which coincides with the first
term of the rhs of Eq.\reff{eq:3-4} up to prefactor:
\begin{equation}
\label{eq:3-6}
\delta E = 8\pi \int_0^\pi\paren{3\delta A+\tan R\delta A'}\sin^2
R\,{\rm d} R.
\end{equation}
(We have dropped the zero term $12\pi a_0\int_0^\pi (\tan R\delta f)'{\rm d}
R$.) One may notice that the second term of the rhs of Eq.\reff{eq:3-4}
is given similarly by the variation of
\begin{equation}
\label{eq:defC}
C\equiv -4\pi\int_0^\pi \frac{|(f^{3/2}\tau_0)'|}{\sqrt{1-f}}{\rm d} R,
\end{equation}
which has been defined by replacing $A(R)$ by $\tau_0(R)$ (and
multiplying the factor $-1/2$) in the definition of $E$. In fact, we
obtain
\begin{equation}
\label{eq:3-8}
\delta C= -4\pi\int_0^\pi\paren{3\delta \tau_0+\tan R\delta \tau_0'}\sin^2
R\,{\rm d} R,
\end{equation}
so that we can write the formula \reff{eq:3-4} as
\begin{equation}
\label{eq:3-7}
\delta V(\tau)=a_0^2(1-\cos\eta)\paren{{\cal A}(\eta)\,\frac{\delta E}{2}+
\sin\eta\,\delta C}.
\end{equation}
Here, note that the space of solutions $P$ is foliated by the surface
$P_{E,C}$ of constant $E$ and $C$, while the FLRW solution $O$ is a two
dimensional subset in $P$ spanned by the two parameters $a_0$ and $\tau_{0{\rm c}}$.
We write the FLRW solution with fixed values of $a_0$ and $\tau_{0{\rm c}}$ as
$O_{a_0,\tau_{0{\rm c}}}\in O$. We can easily find that there is a unique FLRW
solution $O_{a_0,\tau_{0{\rm c}}}$ in each $P_{E,C}$, and we can define a map $Av:
P\rightarrow O$ by this correspondence:
\begin{equation}
\label{eq:Av}
Av: P_{E,C}\rightarrow O_{a_0,\tau_{0{\rm c}}}.
\end{equation}
The significance of Eq.\reff{eq:3-7} is that the FLRW solution
$O_{a_0,\tau_{0{\rm c}}}$ that best matches a given inhomogeneous solution $p\in P$
is given by $Av(p)$, the FLRW solution with the same energy $E$ and the
same ``$C$''. That is, since in each surface $P_{E,C}$ the FLRW solution
$Av(P_{E,C})$ is the critical point for the volume $V(\tau)$, all the
inhomogeneous solutions which are sufficiently close to $Av(P_{E,C})$
virtually manifest the same dynamical evolutions of volume as that of
$Av(P_{E,C})$. This is our main result. We will call $Av$ the {\it
averaging map}.
The key relation \reff{eq:3-7} seems very reasonable (though it is not
trivial) if we note the following fact about the FLRW solution. That is,
if we vary the two FLRW parameters $a_0$ and $\tau_{0{\rm c}}$ the total volume
$V_0(\tau)$ for the FLRW solution will suffer a certain change. We can
easily estimate it by differentiating $V_0(\tau)$ with respect to the
two parameters. (See Eq.\reff{eq:EV0}.) The result is
\begin{equation}
\label{eq:dV0}
{\rm d} V_0(\tau) = 6\pi^2 a_0^2(1-\cos\eta)
\paren{2{\cal A}(\eta) {\rm d} a_0-\sin\eta{\rm d}\tau_{0{\rm c}}}.
\end{equation}
Moreover, using
\begin{equation}
\label{eq:EC0}
E_0=12\pi^2a_0,\quad C_0= - 6\pi^2\tau_{0{\rm c}}
\end{equation}
for the value of $E$ and
$C$ at the FLRW limit \reff{eq:3-3}, we obtain exactly the same formula
as Eq.\reff{eq:3-7}:
\begin{equation}
\label{eq:dV02}
{\rm d} V_0(\tau) = a_0^2(1-\cos\eta)\paren{{\cal A}(\eta)
\frac{{\rm d} E_0}{2}+\sin\eta\, {\rm d} C_0}.
\end{equation}
In this sense the functionals $E$ and $C$ are ``inhomogeneous
generalizations'' of the FLRW parameters $a_0$ and $\tau_{0{\rm c}}$. Relations
\reff{eq:EC0} also imply
\begin{equation}
\label{eq:Av2}
Av(P_{E,C})=O_{a_0=\frac{E}{12\pi^2},\tau_{0{\rm c}}=\frac{-C}{6\pi^2}}.
\end{equation}
Recall that the parameter $\tau_{0{\rm c}}$ contained in the FLRW solution is merely
a kind of gauge freedom, and therefore we could fix it like
$\tau_{0{\rm c}}=0$. This seems to mean that we can restrict the space of solution
$P$ to the slice $C=0$ by gauge fixing. This is in fact the case. Note
that the coordinate transformation $\tau\rightarrow \tau-c$, where $c$ is a
constant, induces the transformation $\tau_0(R)\rightarrow \tau_0(R)+c$. Hence
the space of solutions $P$ contains gauge freedom of this type. Since
the above transformation for $\tau_0(R)$ shifts the value of $C$ by
$4\pi c\int_0^\pi \frac{|f^{3/2}|}{\sqrt{1-f}}{\rm d} R$, every surface of
constant $C$ corresponds to the same set of spacetime solutions, and
this shows the claim. Note that the gauge freedom we consider is the
freedom of choosing the origin of the time coordinate. The
restriction $C=0$ therefore provides us a natural way of choosing the
origin of the time coordinate for every distinct inhomogeneous
solutions, especially with non-simultaneous big bangs, viewing from the
comparisons of $V(\tau)$.
We should comment on a specialty of the FLRW solution. It was natural to
naively expect that the FLRW solution describes the averaged spacetime
of an inhomogeneous spacetime that is homogeneous and isotropic over a
large scale. We have seen that this is in fact the case for the
spherically symmetric case in the sense $\delta V(\tau)=0$ at the FLRW
solution. One might similarly expect that a spacetime that is smooth
over a large scale but fluctuates over small scales could be
approximated by a smoothed solution, or one might expect that a
homogeneous but anisotropic (i.e., Bianchi or Kantowski-Sachs-Nariai
type \cite{KTH}) spacetime solution could be served as a good model for
an inhomogeneous spacetime that is homogeneous over a large
scale. However, at least for the spherically symmetric case only the
FLRW solution is the good averaged or smoothed spacetime model, since
there does not exist a point where the function $\eta(\tau,R)$ becomes a
function of only time $\tau$ in $P$, except at the FLRW solution, i.e.,
$\delta V(\tau)=0$ holds only at the FLRW solution.
The accuracy of the FLRW solution as the averaged one can be estimated
from the second variation $\delta^2V$. That is, we first parameterize
the arbitrary functions with one parameter $\epsilon$ so that
$\epsilon=0$ corresponds to the FLRW solution, and expand them in power
of $\epsilon$: $A(R;\epsilon)=a_0+\epsilon \delta A(R)+(1/2)\epsilon^2
\delta^2 A(R)+\cdots$, and similarly for $f(R;\epsilon)$ and
$\tau_0(R;\epsilon)$. Then, we expand
$V[A(R;\epsilon),f(R;\epsilon),\tau_0(R;\epsilon)]$ in accordance with
these expansions;
\begin{equation}
\label{eq:20}
V=V_0+ \epsilon \delta V+\frac12 \epsilon^2\delta^2
V+\cdots.
\end{equation}
If the first order term vanishes as in our case, the second order term
$\frac12 \epsilon^2\delta^2V$ gives the leading term for the
deviation from the FLRW solution, i.e., the accuracy is given by
$v(\tau)\equiv (V-V_0)/V_0\simeq \frac12 \epsilon^2\delta^2V/V_0$.
After a lengthy calculation and a similar consideration \cite{note:omit}
as for $\delta V$, we obtain
\begin{eqnarray}
\label{eq:secV}
\delta^2 V &=& a_0^2(1-\cos\eta)\paren{{\cal A}(\eta)\, \frac{\delta^2
E}{2}+ \sin\eta\, \delta^2 C} \nonumber \\ && +12\pi a_0\bigg(
{\cal G}(\eta)J[\delta A,\delta A] +(1+2\cos\eta)J[\delta \tau_0,\delta
\tau_0] \nonumber \\ && \hspace{7em} -
2(3\sin\eta-\eta(1+2\cos\eta))J[\delta A,\delta\tau_0]\bigg),
\end{eqnarray}
where $\delta^2 E$ and $\delta^2 C$ are the second variations of,
respectively, the total energy $E$ and the conserved quantity $C$:
\begin{eqnarray}
\label{eq:secE}
\delta^2 E &=& 24\pi\int_0^\pi\paren{\delta^2 A+\rcp3\tan
R\delta^2 A'}\sin^2R\,{\rm d} R \nonumber \\
& & +24\pi\int_0^\pi \paren{\frac{\delta A\delta f}{\cos^2R}+\tan
R\delta A\delta f'+\frac{\sin R}{\cos^3 R}\paren{1-\frac23\sin^2
R}\delta f\delta A'}{\rm d} R,
\end{eqnarray}
\begin{eqnarray}
\label{eq:secC}
\delta^2 C &=& -12\pi\int_0^\pi\paren{\delta^2 \tau_0+\rcp3\tan
R\delta^2 \tau_0'}\sin^2R\,{\rm d} R \nonumber \\
& & -12\pi\int_0^\pi \paren{\frac{\delta \tau_0\delta f}{\cos^2R}+\tan
R\delta \tau_0\delta f'+\frac{\sin R}{\cos^3 R}\paren{1-\frac23\sin^2
R}\delta f\delta \tau_0'}{\rm d} R,
\end{eqnarray}
and we have defined
\begin{equation}
\label{eq:J}
J[\cdot,*]\equiv \int_0^\pi\paren{(\cdot)(*)+\rcp3\tan
R\paren{(\cdot)'(*)+(\cdot)(*)'}}\sin^2 R\,{\rm d} R,
\end{equation}
and ${\cal G}(\eta)\equiv
7-8\cos\eta+\cos^2\eta-6\eta\sin\eta+\eta^2(1+2\cos\eta)$. (If we were
allowed to apply ``integrations by parts'', the above formulae would
become much simpler, but because of the same reason as
in the calculation of $\delta V$, we could not do so. A formal application
would cause a divergence of the result.)
Since our variations are taken in the surface $P_{E,C}$ of constant $E$
and $C$ the variations of any order of $E$ and $C$ vanish, in
particular, $\delta^2 E=0$ and $\delta^2 C=0$. Hence we obtain
\begin{eqnarray}
\label{eq:secV2}
\delta^2 V(\tau) &=&
12\pi a_0\bigg( {\cal G}(\eta)J[\delta A,\delta A] +
(1+2\cos\eta)J[\delta \tau_0,\delta \tau_0]
\nonumber \\ &&
\hspace{7em}
-2(3\sin\eta-\eta(1+2\cos\eta))J[\delta A,\delta\tau_0]\bigg).
\end{eqnarray}
This depends only upon the first variations of the functions $A(R)$
and $\tau_0(R)$.
We close this section by giving an explicit example. The three functions
are $f(R)=\sin^2 R$, $A(R)=1/3+(1/900)(\sin 3R - 12\sin 5R -9\sin 7R)$,
and $\tau_0(R)=0$. (This example has been made by taking $f(R)=\sin^2 R$
and putting
\begin{equation}
\label{eq:ex1}
F(R)=\int_0^R f'(x)f^{\frac 12}(x)\psi(x){\rm d} x.
\end{equation}
We can check that all the regularity conditions for $F(R)$ are satisfied
if $\psi(R)$ is a positive function on $[0,\pi]$ such that it makes
$F(\pi)=0$. In particular, if we choose $\psi(R)=1+(1/10)((1/3)\sin
5R-\sin 7R)$, we have $F(R)=\sin^3R(2/3+(1/450)(\sin 3R-12\sin 5R-9\sin
7R))$, and the $A(R)$ presented above.) The corresponding FLRW solution
as the averaged one is given by $a_0=1/3$ and $\tau_{0{\rm c}}=0$. FIGs.\ref{fig:1}
to \ref{fig:3} are, respectively, profiles of the metric components $r$
and $e^\lambda$, and those of the energy density
$\varepsilon$. FIG.\ref{fig:4} shows the time-development of the total
volume $V$. In each figure, the profiles of the corresponding FLRW
solution are also depicted with dashed curves. FIG.\ref{fig:5} shows
$\log_{10}\frac{\epsilon^2}2|\delta^2 V|/V_0$, which is the estimation
of the accuracy $\log_{10} |v|$ by the second variation $\delta^2 V$,
and the dashed curve shows the exact one $\log_{10}|V-V_0|/V_0$. The
second variation $\delta^2 V$ is evaluated by putting $\epsilon \delta
A=(1/900)(\sin 3R - 12\sin 5R - 9\sin 7R)$, so the integral
$\epsilon^2J[\delta A,\delta A]=-11\pi/324000\simeq - 10^{-4}$. We can
see that the accuracy $v$ is in this example better than $10^{-3}$
throughout the expansion phase, though the energy fluctuation becomes
larger than $10^{-1}$.
\section{Conclusions}
We have seen that the FLRW solution is the critical point for the volume
$V(\tau)$ in the space of spherically symmetric dust solutions on $S^3$.
In accordance with the fact that the FLRW solution contains two
parameters (,though one of which is redundant in a sense), we found that
there is a natural foliation in the space of solutions defined by
constant energy $E$ and another quantity ``$C$''. The exact statement of
our result is that in each leaf of the foliation there exist a unique
FLRW solution and this point is critical with respect to the variations
taken in the same leaf. In our view, the ``averaged'' solution for all
inhomogeneous solutions in a leaf corresponds to the FLRW solution in the
same leaf.
Although our discussions have relied on the known exact solution for the
spherically symmetric case, we have seen that the correspondence between
Eqs.\reff{eq:3-7} and \reff{eq:dV02} is very natural and seems to be
independent of the spherical symmetry. In fact, we have already obtained
a preliminary result that supports a direct generalization of the
present result \cite{MT}. This will appear elsewhere.
We could not discuss a relation to observables like the
distance-redshift relation \cite{SNH}, which will be worth investigating
further.
\section*{Acknowledgments}
The author acknowledges financial support from the Japan Society for the
Promotion of Science and the Ministry of Education, Science and Culture.
|
\section{Introduction}
In QCD, the characteristic feature of the gluon mediators of the color force
is their strong self-interaction, because the gluons themselves carry color
charges. In analogy with the electric lines of force between two electric
charges, one usually assumes that color charges (quarks) are held together by
color lines of force, but the gluon-gluon interaction pulls these together into
the form of a tube (string). Therefore, the use of the string picture for the
phenomenological description of strong interactions seems to be completely
justified. In fact, the formation of chromoelectric flux tubes between static
quarks is by now a well established feature of lattice QCD simulations
\cite{lat}.
The action of the standard relativistic Nambu-Goto string with massive ends
(quarks) in the parametrization $\tau =t=x^0$ is written as $(\gamma $ is the
string tension) \cite{BN}
\beq
S=-\gamma \int _{t_1}^{t_2}dt\int _0^\pi d\sigma \sqrt{x'^2(1-\dot{x}^2)+
(\dot{x}x')^2}-\sum_{i=1,2}m_i\int _{t_1}^{t_2}dt\;\!\sqrt{1-\dot{x}_i^2},
\eeq
$$x\equiv {\bf x}={\bf x}(t,\sigma ),\;\;\;x_i\equiv x(t,\sigma _i),
\;\;i=1,2,\;\;\sigma _1=0,\;\;\sigma _2=\pi ,$$
where from now on the dot and the prime stand for the derivative with respect
to $t$ and $\sigma ,$ respectively, unless otherwise specified. The action (1)
is invariant under arbitrary transformations of $\sigma ,$ therefore the
string Lagrangian in Eq. (1) is degenerate, The corresponding constraint among
the canonical variables $x(t,\sigma )$ and $p(t,\sigma )$ $(p(t,\sigma )\equiv
{\bf p}(t,\sigma ))$ is
\beq
px'=0,
\eeq
for
\beq
p=p(t,\sigma )=\frac{\partial L_{str}}{\partial \dot{x}}=\frac{\gamma ^2}{
L_{str}}\;\!\Big[ x'(\dot{x}x')^2-\dot{x}(x')^2\Big] ,
\eeq
$$L_{str}=-\gamma \sqrt{x'^2(1-\dot{x}^2)+(\dot{x}x')^2}.$$
With the equalities
\beq
p\dot{x}-L_{str}=-\frac{\gamma ^2}{L_{str}}\;\!x'^2,\;\;\;p^2+\gamma ^2x'^2=
\left( -\frac{\gamma ^2}{L_{str}}\;\!x'^2\right) ^2,
\eeq
one obtains the expression for the canonical Hamiltonian, as follows,
\bqry
H & = & \int _0^\pi d\sigma \sqrt{p^2+\gamma ^2x'^2}+\sum _{i=1,2}\sqrt{p_i^2+
m_i^2} \NL
& = & \int _0^\pi d\sigma |p+\gamma x'|+\sum _{i=1,2}\sqrt{p_i^2+m_i^2},
\eqry
where $p_i=m\dot{x}_i/\sqrt{1-\dot{x}_i^2}.$
In the nonrelativistic limit, $|\dot{x}(t,\sigma )|\ll 1,$ $|\dot{x}_i|\ll 1,$
Eq. (1) transforms into \cite{BN}
\beq
S=-\gamma \int _{t_1}^{t_2}dt\int _0^\pi d\sigma \sqrt{x'^2}-\sum_{i=1,2}m_i
\int _{t_1}^{t_2}dt+\sum _{i=1,2}\frac{m_i}{2}\int _{t_1}^{t_2}dt\;\!\dot{
x}_i^2.
\eeq
Integral over $\sigma $ gives the length of the string (with the assumption
that there are no singularities on the string). The variation of the first
term in Eq. (6) with respect of the string coordinates leads to the
requirement on the string to have the form of a linear rod connecting the
massive ends. The effective action that leads to the equations of motion of
the massive ends is therefore
\beq
S_{eff}=\int _{t_1}^{t_2}dt\left( -\gamma |x_1(t)-x_2(t)|+
\sum _{i=1,2}\frac{m_i\dot{x}_i^2}{2}\right) .
\eeq
Hence, in the nonrelativistic limit, the string generates a linear potential
between its massive ends: $V(|x_1-x_2|)=\gamma |x_1-x_2|.$ The same result
holds for the string with massive ends in two-dimensional space-time \cite{BN}
where $x(t,\sigma )$ has only one component; therefore $p(t,\sigma )=0,$ via
(3), and, in view of (5),
$$H=\gamma \int _0^\pi d\sigma |x'|+\sum _{i=1,2}\sqrt{p_i^2+m_i^2}=
\gamma |x(0)-x(\pi )|+\sum _{i=1,2}\sqrt{p_i^2+m_i^2}.$$
The string model with constant tension is known to predict linearly rising
Regge trajectories $J=J(E^2)$ $(J$ and $E$ are the orbital momentum and energy
of the string, respectively): $J=E^2/(2\pi \gamma )$ \cite{BN}. The string
trajectories are exactly linear in the case of the massless ends, and
asymptotically linear in the case of the massive ends having some curvature
in the region $E\stackrel{>}{\sim }m_1+m_2.$ The same picture of linear
trajectories arises from a linear confining potential \cite{KS}.
However, the realistic Regge trajectories extracted from data are {\it
nonlinear.} Indeed, the straight line which crosses the $\rho $ and $\rho _3$
squared masses corresponds to an intercept $\alpha _\rho (0)=0.48,$ whereas
the physical intercept is located at 0.55, as extracted by Donnachie and
Landshoff from the analysis of $pp$ and $\bar{p}p$ scattering data in a simple
pole exchange model \cite{DL}. The nucleon Regge trajectory as extracted from
the $\pi ^{+}p$ backward scattering data is \cite{Lyu}
$$\alpha _N(t)=-0.4+0.9\;\!t+\frac{1}{2}\;\!0.25\;\!t^2,$$
and contains positive curvature. Recent UA8 analysis of the inclusive
differential cross sections for the single-diffractive reactions $p\bar{p}
\rightarrow pX,$ $p\bar{p}\rightarrow X\bar{p}$ at $\sqrt{s}=630$ GeV reveals
a similar curvature of the Pomeron trajectory \cite{UA8}:
$$\alpha _{I\!\!P}(t)=1.10+0.25\;\!t+\frac{1}{2}\;\!(0.16\pm 0.02)\;\!t^2.$$
An essentially nonlinear $a_2$ trajectory was extracted in ref. \cite{Bol2}
for the process $\pi ^{-}p\rightarrow \eta n.$ Note that the nonlinearity
of Regge trajectories was also proven on quite general grounds \cite{nonlin}.
Once the nonlinearity of Regge trajectories is an established fact, a number
of important issues immediately suggest themselves; e.g.,
(i) the understanding of the reasons for the nonlinearity of trajectories,
(ii) the development of the model(s) leading to analytic nonlinear Regge
trajectories.
Possible reasons for the nonlinearity of trajectories may be related to the
string breaking in QCD. In the absence of dynamical fermions (e.g., in zero
temperature quenched QCD simulations) does not allow the screening of the
potential between the heavy (static) $Q\bar{Q}$ by virtual color-singlet
light $q\bar{q}$ pairs, and thus the interquark potential is expected to grow
linearly with the separation $R$ for arbitrarily large $R.$ With dynamical
fermions, the static $Q\bar{Q}$ meson can decay into two heavy-light mesons.
Ignoring meson-meson interactions, one can expect the QCD string to break as
soon as the potential exceeds twice the heavy-light mass, i.e., at about 1.5
fm. (Neglecting quark mass effects on the dynamics of the binding problem,
which is a reasonable assumption once this mass is small compared to a typical
binding energy of $\sim 500$ MeV, the string breaking distance should be
shifted by $\triangle r\approx 2\triangle m/\sigma _{eff}$ when changing the
quark mass by $\triangle m,$ where $\sigma _{eff}$ is the (effective) string
tension, to be discussed explicitly below.) The expectation of the string
breaking due to the color screening described above has indeed been confirmed
by some of the data on QCD lattice simulations \cite{lat2}.
The purpose of the present letter is to present a model for nonlinear Regge
trajectories. This model is the generalization of the standard string model.
Indeed, if the string breaking really happens in QCD, as should become clear
with the forthcoming sophisticated lattice measurements, there is a
necessity to reconsider the canonical string model and modify the notion of
a constant string tension, in order to include the color screening effects.
Before we proceed with the presentation of such a modified string model, let
us note that potential models, in general, lead to nonlinear trajectories
(for potentials that are different from a linearly rising one), but these
trajectories cannot generally be cast into analytic form \cite{traj}. The
model we shall present in what follows in many cases leads to {\it analytic}
nonlinear Regge trajectories.
\section{Generalized string model}
Here we wish to generalize the standard string formulation reviewed above on
the case of arbitrary potential between the string massive ends. Such
generalization is done by the modification of the standard string tension into
the effective string tension which is a function of $|x|,$ as follows:
\beq
S_{gen}=-\int _{t_1}^{t_2}dt\int _0^\pi d\sigma \;\!\gamma (|x|)\sqrt{
x'^2(1-\dot{x}^2)+(\dot{x}x')^2}-\sum_{i=1,2}m_i\int _{t_1}^{t_2}dt\;\!
\sqrt{1-\dot{x}_i^2}.
\eeq
As we explain below, in the nonrelativistic limit the effective string tension
is the derivative of the interaction potential between the massive ends of the
string. Therefore, different choices of the effective string tension would be
related to different potentials, which makes it possible to deal, among the
others, with color-screened potentials, i.e., potentials that approach
constant values at large separations; e.g., $V(\rho )=\gamma /\mu \;\!(1-\exp
(-\mu \rho ))$ which is used to fit the lattice QCD data (e.g., first paper
of \cite{lat2}), and two examples in Section 3. As is clear from the above
discussion, such color-screened potentials may be relevant to QCD.
Obviously, the new action (8) is again invariant under arbitrary
transformations of $\sigma ,$ and therefore the constraint (2), and Eqs.
(3)-(5) hold in this case as well, with $\gamma =\gamma (|x|).$ In the
nonrelativistic limit, however, in place of (7) one will now obtain
$$S_{gen,eff}=\int _{t_1}^{t_2}dt\left( -\int _0^\pi d\sigma \;\!
\gamma (|x|)|x'|+\sum _{i=1,2}\frac{m_i\dot{x}_i^2}{2}\right) $$
\beq
=\int _{t_1}^{t_2}dt\left( -V(|x_1-x_2|)+\sum _{i=1,2}\frac{m_i\dot{
x}_i^2}{2}\right) .
\eeq
In contrast to the previous case of the standard string, it is seen in the
above relations that now
\beq
\gamma (|x|)=\frac{1}{|x'|}\;\!\frac{dV(|x|)}{d\sigma }=\frac{dV(|x|)}{d|x|},
\eeq
i.e., the effective string tension is the derivative of the potential with
respect to the distance. Obviously, in the case of a linear potential, the
effective string tension reduces to the standard (constant) one.
It should be noted that the other generalizations of the standard string have
also been discussed in the literature \cite{gen,wiggly}. Specifically, refs.
\cite{gen} deal with the linearized string Lagrangian that generally has
different coefficients of $\dot{x}^2$ and $x'^2,$ which are to be associated
with the string mass density and tension, respectively. Both of these
coefficients are functions of $\sigma .$ Refs. \cite{wiggly} deal with the
so-called wiggly string (membrane) for which the string mass density and
tension are generally different also. With $\gamma $ being a function of
$\sigma $ not $x,$ as in \cite{gen}, (i) the invariance of the action (8)
under arbitrary transformations of $\sigma $ disappears, and (ii) the
interpretation of $\gamma $ as an effective string tension, via Eq. (10),
is lost, as seen in (9). Also, there are claims in the literature that the
choice of the string mass density and tension being different from each other
contradicts the principles of relativity which require that the both coincide
\cite{BI}. In our case (upon the linearization of the Lagrangian in Eq. (8))
both the string mass density and tension would be equal to $\gamma (|x|).$
Similarly to the standard case of a constant string tension which represents
the relativization {\it \`{a} la} Poincar\'{e} of the nonrelativistic two-body
problem with linear potential \cite{Shav}, the generalized string that we are
discussing here can be considered as the relativization of a nonrelativistic
two-body problem with arbitrary potential.
\subsection{The dynamics of the generalized string model}
By varying the action of the generalized string with massive ends
(here the dot stands for the derivative with respect to $\tau ),$
\beq
S_{gen}=\int \!\!\!\!\int d\tau d\sigma L(x,\dot{x},x')+\sum _{i=1,2}
L^{(m)}(\dot{x}_i),
\eeq
one obtains the equations of motion of the generalized string,
\beq
\frac{d}{d\tau }\frac{\partial L}{\partial \dot{x}}+\frac{d}{d\sigma }
\frac{\partial L}{\partial x'}=\frac{\partial L}{\partial x},
\eeq
and the boundary conditions which represent the equations of motion of the
massive ends:
\beq
\frac{d}{d\tau }\frac{\partial L^{(m)}}{\partial \dot{x}_i}=\frac{
\partial L}{\partial x'},\;\;\;x=x_i.
\eeq
In the gauge $\tau =t$ discussed above, the equation of motion of the
generalized string reduce to
\beq
\frac{d}{dt}\frac{\partial L}{\partial \dot{x}}+\frac{d}{d\sigma }\frac{
\partial L}{\partial x'}=\frac{\partial L}{\partial x},\;\;\;x\equiv {\bf x},
\eeq
and the boundary conditions are
\bqry
m_1\;\!\frac{d}{dt}\frac{\dot{x}_1}{\sqrt{1-\dot{x}_1^2}} & = & \frac{
\partial L}{\partial x'},\;\;\;\sigma =0, \NL
m_2\;\!\frac{d}{dt}\frac{\dot{x}_2}{\sqrt{1-\dot{x}_2^2}} & = & \frac{
\partial L}{\partial x'},\;\;\;\sigma =\pi .
\eqry
Let us show that, similarly to the standard case of the string with constant
tension, there are solutions to the equations of motion of the generalized
string in the form of a rigid rod connecting the massive ends and rotating
with frequency $\omega $ about its center of mass:
\beq
x(t,\sigma )=\rho (\sigma )\Big( \cos (\omega t),\;\sin (\omega t),\;0\Big) .
\eeq
Indeed, $\gamma =\gamma (|x|)=\gamma (\rho ),$ since $x^2=\rho ^2;$ therefore
$d\rho/dx=x/\rho =(\cos (\omega t),\;\sin (\omega t),\;0),$ and $d\gamma /dx=d
\gamma /d\rho $ $d\rho /dx=d\gamma /d\rho $ $(\cos (\omega t),\;\sin (\omega
t),\;0).$ Hence
\beq
\frac{\partial L}{\partial x}=\frac{\partial L}{\partial \gamma }\;\!
\frac{d\gamma}{dx}=-\frac{d\gamma }{d\rho }\;\!\rho' \sqrt{1-\omega ^2
\rho ^2}\Big( \cos (\omega t),\;\sin (\omega t),\;0\Big) .
\eeq
Since also
\beq
\frac{d}{dt}\;\!\frac{\partial L}{\partial \dot{x}}=-\frac{\gamma \omega ^2
\rho \rho'}{\sqrt{1-\omega ^2\rho ^2}}\;\!\Big( \cos (\omega t),\;\sin
(\omega t),\;0\Big) ,
\eeq
\beq
\frac{d}{d\sigma }\;\!\frac{\partial L}{\partial x'}=\left( -\frac{d\gamma }{
d\rho }\;\!\rho' \sqrt{1-\omega ^2\rho ^2}+\frac{\gamma \omega ^2\rho \rho'}{
\sqrt{1-\omega ^2\rho ^2}}\right) \Big( \cos (\omega t),\;\sin (\omega t),
\;0\Big) ,
\eeq
(the last relation is obtained via $d\gamma /d\sigma =d\gamma /d\rho $
$\rho'),$ one sees that the equations of motion (14) are satisfied.
One can easily check that for the rotation (16), the energy of the generalized
string is given, in view of (5), by
\beq
H=\int d\sigma \sqrt{p^2+\gamma ^2x'^2}=\int d\sigma \;\!\frac{\gamma
\rho'}{\sqrt{1-\omega ^2\rho ^2}}=\int \frac{d\rho \;\!\gamma (\rho )}{
\sqrt{1-\omega ^2\rho ^2}}.
\eeq
Similarly, the orbital momentum of the generalized string is
\beq
J=J_z=\int d\sigma \left( xp_y-yp_x\right) =\int d\sigma \;\!\frac{\gamma
\omega \rho ^2\rho'}{\sqrt{1-\omega ^2\rho ^2}}=\int \frac{d\rho \;\!
\gamma (\rho )\omega \rho ^2}{\sqrt{1-\omega ^2\rho ^2}}.
\eeq
Interestingly enough, in his book \cite{Per} Perkins also presents the above
relations for the energy and orbital momentum of the generalized string. He
does not however derive these relations from the first principles Lagrangian,
as in Eq. (8).
By adding the contribution of the massive ends, one finally has the
expressions for the total energy and orbital momentum of the generalized
string with massive ends:
\beq
E=\int _0^{r_1}\frac{d\rho \;\!\gamma (\rho )}{\sqrt{1-\omega ^2\rho ^2}}+
\int _0^{r_2}\frac{d\rho \;\!\gamma (\rho )}{\sqrt{1-\omega ^2\rho ^2}}+
\frac{m_1}{\sqrt{1-\omega ^2r_1^2}}+\frac{m_2}{\sqrt{1-\omega ^2r_2^2}},
\eeq
\beq
J=\int _0^{r_1}\frac{d\rho \;\!\gamma (\rho )\omega \rho ^2}{\sqrt{1-\omega ^2
\rho ^2}}+\int _0^{r_2}\frac{d\rho \;\!\gamma (\rho )\omega \rho ^2}{\sqrt{1-
\omega ^2\rho ^2}}+\frac{m_1\omega r_1^2}{\sqrt{1-\omega ^2r_1^2}}+\frac{
m_2\omega r_2^2}{\sqrt{1-\omega ^2r_2^2}}.
\eeq
Note that the boundary conditions (15) define the separations of the massive
ends from the center of mass through the following nonlinear equations:
\beq
\frac{m_i\omega ^2r_i}{\sqrt{1-\omega ^2r_i^2}}=\gamma (r_i)\sqrt{1-\omega ^2
r_i^2},\;\;\;i=1,2.
\eeq
\subsection{Generalized massless string}
The energy and orbital momentum of the generalized massless string,
$m_1=m_2=0,$ are given by
\beq
E=2\int _0^R\frac{d\rho \;\!\gamma (\rho )}{\sqrt{1-\omega ^2\rho ^2}},\;\;\;
J=2\int _0^R\frac{d\rho \;\!\gamma (\rho )\omega \rho^2}{\sqrt{1-\omega ^2\rho
^2}},
\eeq
where $R=1/\omega $ is half of the string length for a given $\omega .$ The
condition $\omega R=1$ follows from, e.g., Eqs. (24) with $m_i\rightarrow 0.$
By eliminating $\omega $ from Eqs. (25) one can obtain $J$ as a function of
$E^2,$ the Regge trajectory. As we demonstrate below, the trajectory is
generally nonlinear, and in many cases it is given in an analytic form. It
will be shown elsewhere \cite{prep} that it is possible to uniquely recover
the potential $(V(r)\sim \int d\rho $ $\gamma (\rho ))$ from the known
analytic form of Regge trajectory, for both generalized massless and massive
strings, and the techniques of the corresponding inverse problem will be
presented in detail. It should be noticed that potential corresponding to an
analytic Regge trajectory cannot always be recovered itself in an analytic
form, but only as some power series in $\rho .$ However, in the most important
cases of analytic nonlinear Regge trajectories the corresponding potentials
happen to be recovered analytically. Below we present such potentials for
three examples of analytic nonlinear Regge trajectories that have been
discussed in the literature. For our present purposes, here we present only
the final results.
\section{Analytic nonlinear Regge trajectories}
In what follows, we consider only the generalized massless string.
\subsection{Square-root trajectory}
The square-root Regge trajectory, $J\propto E_{th}-\sqrt{E_{th}^2-E^2},$
where $E_{th}$ is the trajectory energy threshold, has been widely discussed
by Kobylinsky {\it et al.} \cite{Kob} as the simplest choice of a trajectory
for dual amplitude with Mandelstam analyticity (DAMA) \cite{DAMA}. It also
provides for the Orear (small-$t)$ regime of the amplitude. The corresponding
potential is $(\gamma ,\mu ={\rm const,}$ $V(\rho )\rightarrow \gamma /2\mu $
as $\rho \rightarrow \infty ,$ and hence $E\rightarrow E_{th}=\gamma /\mu )$
\beq
V(\rho )=\frac{\gamma }{\pi \mu }\arctan (\pi \mu \rho ),
\eeq
for which
\beq
\gamma (\rho )=\frac{dV(\rho )}{d\rho }=\frac{\gamma }{1+(\pi \mu \rho )^2},
\eeq
and leads, via (25), to
\beq
E=\frac{\pi \gamma }{\sqrt{\omega ^2+\pi ^2\mu ^2}},\;\;\;J=\frac{\gamma }{
\pi \mu ^2}\left( 1-\frac{\omega }{\sqrt{\omega ^2+\pi ^2\mu ^2}}\right) .
\eeq
Eliminating $\omega $ from the above relations gives
\beq
J=\frac{1}{\pi \mu }\left( \gamma /\mu -\sqrt{(\gamma /\mu )^2-E^2}\right) ,
\eeq
i.e., the square-root Regge trajectory. For $E\ll \gamma /\mu ,$ it reduces
to an (approximate) linear trajectory, $J\simeq E^2/(2\pi \gamma ).$ Note
that in the corresponding quantum case where $J$ takes on integer values only
and the trajectory would develop a (nonzero) intercept (the quantum defect),
$J(0),$ there would be a finite number of states lying on the trajectory,
with orbital momenta $0\leq J\leq J_{max}\leq J(0)+\gamma /(\pi \mu ^2).$
\subsection{Logarithmic trajectory}
The logarithmic Regge trajectory, $J\propto -\log (1-E^2/E_{th}^2),$ is the
ingredient of dual amplitude with logarithmic trajectories (DALT) \cite{DALT}.
There are certain theoretical reasons \cite{log} to consider logarithmic
trajectories that, for large $-t(=-E^2),$ are compatible with fixed angle
scaling behavior of the amplitude. The corresponding potential is (again
$\gamma ,\mu ={\rm const,}$ $V(\rho )\rightarrow \gamma /2\mu $ as $\rho
\rightarrow \infty ,$ and hence $E\rightarrow E_{th}=\gamma /\mu )$
\beq
V(\rho )=\frac{\gamma }{2\pi \mu }\left( 2\arctan (2\pi \mu \rho )-\frac{
\log [1+(2\pi \mu \rho )^2]}{2\pi \mu \rho }\right) ,
\eeq
for which
\beq
\gamma (\rho )=\gamma \;\!\frac{\log [1+(2\pi \mu \rho )^2]}{
(2\pi \mu \rho )^2},
\eeq
and leads to
\beq
E=\frac{\gamma }{2\pi \mu ^2}\left( \sqrt{\omega ^2+4\pi ^2\mu ^2}-\omega
\right),\;\;\;J=\frac{\gamma }{2\pi \mu ^2}\;\!\log \frac{\omega +\sqrt{
\omega ^2+4\pi ^2\mu ^2}}{2\omega },
\eeq
from which eliminating $\omega $ (viz., $\omega =\pi (\gamma ^2-\mu ^2E^2)/(
\gamma E))$ gives
\beq
J=-\frac{\gamma }{2\pi \mu ^2}\;\!\log \left( 1-\frac{E^2}{(\gamma /\mu )^2}
\right) ,
\eeq
i.e., the logarithmic Regge trajectory. For $E\ll \gamma /\mu ,$ it again
reduces to an (approximate) linear form, $J\simeq E^2/(2\pi \gamma ).$ Note
that, in contrast to the previous example of square-root trajectory, although
the trajectory has an energy threshold, in the corresponding quantum case the
number of states lying on the trajectory would be infinite.
In each of the two examples considered above, the potential belongs to the
family of the color-screened potentials mentioned in Section 2. (The exact
form of the color-screened potential, if it is realized in QCD, will be
discussed elsewhere.) Note that with a different choice of signs in Eqs.
(29),(33) (e.g., $J=\gamma /(2\pi \mu ^2) \log [1+E^2/(\gamma /\mu )^2])$
there would be a difficulty with the analytic continuation of the trajectory
in the region $E^2(=t)<0,$ since in this region the trajectory cannot have
imaginary part. Note also that besides the trajectories (29),(33), a
hybrid square-root--logarithmic trajectory, $J\propto -\log (1+\beta \sqrt{
E_{th}^2-E^2}),$ has been motivated and discussed in \cite{log,Laszlo}. We
have not menaged to obtain the analytic form of the corresponding potential.
\subsection{Hyperbolic trajectory}
The hyperbolic trajectory, $J\propto \cosh (E)-1,$ results in the
$\kappa $-deformed Poincar\'{e} phenomenology \cite{kappa}. Here the
corresponding potential is
\beq
V(\rho )=\frac{\gamma \rho \;\!\Phi (-(\pi \mu \rho )^2,\;\!2,\;\!1/2)}{4}=
\gamma \rho \sum _{k=0}^\infty \frac{[-(\pi \mu \rho )^2]^k}{(2k+1)^2}=
\gamma \rho \left( 1-\frac{(\pi \mu \rho )^2}{9}+\frac{(\pi \mu \rho )^
4}{25}-\ldots \right) ,
\eeq
where
\beq
\Phi (z,\;\!s,\;\!a)=\sum _{k=0}^\infty \frac{z^k}{(a+k)^s}
\eeq
is the so-called Lerch's transcendent. The corresponding $\gamma (\rho ),$
\beq
\gamma (\rho )=\gamma \;\!\frac{\arctan (\pi \mu \rho )}{\pi \mu \rho },
\eeq
leads to
\beq
E=\frac{\gamma }{\mu }\;\!\log \left( \pi \mu /\omega +\sqrt{1+(\pi \mu
/\omega )^2}\right) ,
\eeq
\beq
J=\frac{\gamma }{\pi \mu ^2}\left( \sqrt{1+(\pi \mu /\omega )^2}-1\right) ,
\eeq
from which eliminating $\omega $ $(\pi \mu /\omega =\sinh (\mu E/\gamma ))$
gives
\beq
J=\frac{\gamma }{\pi \mu ^2}\Big[ \cosh \left( \frac{E}{\gamma /\mu }
\right) -1\Big] ,
\eeq
i.e., the hyperbolic trajectory. For $E\ll \gamma /\mu ,$ it reduces to the
linear form, as well as in the above two cases: $J\simeq E^2/(2\pi \gamma ).$
\section{Concluding remarks}
We have presented a new generalized string model that leads to nonlinear Regge
trajectories which in many cases can be given in analytic form. We have
considered three examples of how this model generates square-root, logarithmic
and hyperbolic trajectories that have been discussed in the literature.
The main phenomenological implication of the model presented here is the
possibility to obtain Regge trajectory for an arbitrary (nonrelativistic)
potential in general, and for a color-screened potential in particular. Since
trajectory for the latter is characterized by an energy threshold, and in some
cases by a finite number of states, as in the example of square-root
trajectory considered, it may be of relevance to QCD to predict the numbers
of states lying on different trajectories, and the corresponding energy
thresholds, if a color-screened potential of, e.g., the type (26) is indeed
realized in QCD.
Of certain interest would be further exploration of the generalized string
model discussed here, and the applications of this model to the hadron
scattering (the form of the corresponding scattering amplitude) and the
thermodynamics of hot hadronic matter (the equation of state of the ensemble
of the generalized strings), as well as the quantization of this model. These,
and related issues, e.g., the understanding of the form of the color-screened
potential, if is realized in QCD, will be the subjects of further study, to
be undertaken elsewhere.
\section*{Acknowledgements}
The author wishes to thank M.M. Brisudov\'{a}, T. Goldman, L.P. Horwitz and
P.R. Page for very valuable discussions during the preparation of this work.
\bigskip
|
\section{Introduction}
This paper presents new data on a low power radio galaxy
(B2 1144+35) belonging to a complete sample of radio galaxies which we
are studying with VLBI data (\cite{g90}).
A major result from this study is the discovery
that at least in many, and perhaps in all, low power FR I radio galaxies (see
\cite{fa74}) parsec
scale jets move at relativistic velocities. The derived velocities and
orientations with respect to the line of sight (see e.g. \cite{lar97} and
references therein) support
the low-power unified scheme (see e.g. \cite{ur95}) and suggest
that
multi-epoch observations should reveal a proper motion in many FR I
galaxies. Up to now such studies have been carried out only on a small
number of low power
radio galaxies (e.g. M87, \cite{bir95}; 3C338, \cite{g98a}; NGC\ 315
\cite{cot98}),
it is therefore important to examine a larger number of
FR I radio galaxies with multi-epoch observations to measure jet
proper motions.
The low power radio galaxy B2 1144+35 discussed here, has been identified with
a faint (m$_{pg}$ = 15.7)
Zwicky galaxy (ZW186.48) in a medium-compact galaxy cluster at a redshift of
0.0630. An isocontour image taken from the Palomar Observatory Sky Survey
(POSS) is shown in Fig. 1. The contour image shows that the optical galaxy
has a boxy shape which, according to \cite{bi85}, may
occur in systems that have cannibalized low luminosity galaxies.
A nearby faint companion is embedded in its external region but
optical spectroscopy is necessary to confirm a genuine connection.
In a recent optical study of bright flat spectrum radio sources,
\cite{ma96} classify 1144+35 as a BL Lac candidate even though its
spectrum shows
H$\alpha$ and [NII] emission lines (\cite{co75,mo92}). From a comparison
between the measured line
equivalent width and the contrast \cite{ma96} suggest that
1144+35 could be a diluted
BL Lac. Moreover it was observed in an imaging and spectroscopic survey of
low and intermediate power radio galaxies (\cite{eb89}).
A CCD residual image shows a very
definite arc of dust in the galaxy nuclear region.
The radio galaxy 1144+35 was detected in the X-Ray ROSAT All-Sky Survey
(\cite{bri95}) with a flux of 7.1 $\times$ 10$^{-13}$ erg cm$^{-2}$
sec $^{-1}$ in the 0.1--2.4 Kev band. \cite{sc99} derived from HRI
ROSAT observations an X-Ray flux density of 8.1 $\times$ 10$^{-13}$ erg
cm$^{-2}$ sec $^{-1}$ using a powerlaw model and 6.1 $\times$ 10$^{-13}$ erg
cm$^{-2}$ sec $^{-1}$ using a thermal bremsstrahlung model, in the 0.1--2.4 Kev
band.
From the radio point of view, 1144+35 has a peculiar structure: \cite{mac98}
classifies it among giant radio galaxies, yet from high
resolution VLA observations (\cite{fan87}) it has been
classified a {\it naked} jet source since it shows two faint jet-like
regions and no extended emission. The radio structure is core dominated
and the two jet-like features are short and resolved.
\cite{sc99} observed this source with the Westerbork Synthesis
Radio Telescope (WSRT) at 1.4 GHz and reported WSRT observations at 325 MHz
from the WENSS survey (\cite{re97}). They present a detailed image
of the radio core and of two diffuse extended radio structures on either side
of it. The eastern extended lobe shows a leading hotspot and is clearly
associated with the central source. The northern part of the western radio
structure is a separate low power radio galaxy; the southern part is a radio
lobe with an elongated tail most likely associated with 1144+35.
The first VLBI observations of this source were carried out by \cite{g90} with
the EVN + Haystack with 2 scans, 13 minutes each. The source was detected, but
the short on source integration time and the poor $u,v$ coverage did not
allow a discussion of its structure.
It was imaged as part of the Second Caltech-Jodrell survey (\cite{he95})
at 5 GHz, where it shows two main knots with some substructure.
A preliminary study of the data from 3 epochs suggested the presence of a
possible superluminal motion (\cite{g95}).
The new VLA, MERLIN and VLBI data presented in this paper
demonstrate a complex structure over a broad range of
physical scales (1 pc -- 1 Mpc) and confirm superluminal motion.
We also discuss some
possible explanations for the structure of
this peculiar radio galaxy. We use an Hubble constant H$_0$ =
50 km sec$^{-1}$ Mpc$^{-1}$ which corresponds to a conversion factor of
1.62 pc/mas.
\placefigure{f1.eps}
\section{Observations and Data Reduction}
\subsection{VLA Data}
On November 1997, we obtained 6 hours of observing time with the VLA of the
NRAO\footnote{The National Radio Astronomy Observatory is operated by
Associated
Universities, Inc., under cooperative agreement with the National
Science Foundation}
in the D configuration at 1.4 and 5 GHz to investigate the large scale
structure of 1144+35.
The data have been calibrated in the standard way using the NRAO AIPS package
and imaged using the task IMAGR.
Calibrated data at 1.4 GHz were combined with a short (20 minutes long) VLA
observation at higher resolution (C array) obtained in September 1997, to have
better angular resolution
while 5 GHz data were combined with a long observation also in the C
configuration obtained in November 20, 1990 during a VLBI run with the
VLA used as a phased array.
In this run
the source was observed for 12 hours, with the array phased every 15 minutes
with a nearby VLA calibrator source.
Since the arcsecond radio core is variable (see Sect. 3.4), a
fraction of its flux was subtracted from the November 1990 observations
before of combining the two datasets.
The angular resolution and sensitivity of the images presented in this paper
are given in Table 1.
The arcsecond scale radio emission of this source was observed with the VLA
in the A-array configuration at 1.4, 5 and 8.4 GHz,
on April 1994 and February 1997. Moreover, on September 1997, 1144+35 was
observed at 15, 22.5 and 43 GHz for the VLA calibrator list.
The source
was observed for about 10 minutes total at each frequency in a few different
hour angles to obtain better $u,v$ coverage.
The core flux density was derived by fitting a gaussian to the
nuclear source. In these images due to the lack of short spacings and the short
integration time, only the unresolved arcsecond core is visible and therefore
we will use these data only to discuss the arcsecond core variability in the
next sections.
\placetable{arcsecvlamaps}
\subsection{MERLIN Data}
We observed 1144+35 with the MERLIN array on 1995 March 29 at 5 GHz with
a 16 MHz bandwidth for 10 hours. We used the following telescopes: Defford,
Cambridge, Knockin, Wardle, Darnhall, MK2 and Tabley. The data were edited and
amplitude calibrated in Jodrell Bank (JB) using the standard procedure based on
the OLAF programs. 3C286 was used as amplitude calibrator. The data where then
written in FITS format and loaded into AIPS where the phase calibration
was carried out using
standard MERLIN phase calibrators. The pipeline available in
Jodrell Bank
was used as a first step for the phase corrections only. The source was then
mapped in total continuum and polarization and then self-calibrated.
\subsection{VLBI Data}
In Table 2 we summarize the VLBI observations which are discussed in the
present paper.
Data marked 1, 2 and 4 are discussed here for the first time, while the
other are {\it literature} data.
The data have been correlated with the Caltech/JPL Block 2 correlator in
Pasadena (Mk2 VLBI data) and in
Socorro with the NRAO VLBA correlator (VLBA and global VLBI data). Amplitude
calibration was then done using the standard system temperature method in AIPS.
Data were globally fringe fitted and self-calibrated in the standard way.
We made some iterations of phase self-calibration, followed by a final
phase and gain self-calibration when necessary to produce the
final image. Cross-hand correlations were produced for the September
1997 global
8.3 GHz observations only. Calibration of the instrumental leakage terms was
performed using 4 widely spaced scans on the compact, unpolarized source OQ208
(Roberts et al.\ 1994). The absolute electric vector position
angles were calibrated using two 5 minute scans on 3C279 which has a long-lived
jet component of constant polarization angle (\cite{ta98}).
\placetable{vlbidata}
Multi-frequency and snap-shot (see note to Table 2) observations were
carried out switching often between frequencies or among different
sources to obtain good and uniform $u,v$ coverage. All other observations
are long integration observations typically of 8 -- 10 hours each.
Mk2 data have been obtained with the old Mark2 system with a 2 MHz
bandwidth and in single polarization mode; VLBA observations are in
single polarization mode with 32 MHz bandwidth, while the global
observation is a dual polarization, 64 MHz bandwidth observation. In
this latest global observation, we used the following array for the
total intensity images: Effelsberg, Noto, VLBA, and a single VLA antenna.
Noto was not used for polarization images.
\section{Results}
\subsection{The Megaparsec Scale}
In agreement with \cite{sc99}, we distinguish 3 different
regions in the large scale structure of 1144+35:
the core region, the East, and the West extended emission (see
Figs. 2,3 and 4). The core region coincides with the optical galaxy and
is the dominant feature in all the images. The core properties will be
discussed in detail in Sect. 3.4.
\placefigure{f2.eps}
The East extended emission
shows an {\it S} shaped morphology and an uniform low brightness.
It is clearly connected to the arcsecond core emission even if
there is a strong discontinuity in the brightness
of the inner nuclear region with respect to the
extended lobe. Two small unresolved regions (see Fig. 4) are
visible at the edge of the extended emission but are
most likely unrelated background sources. The size of this extended
emission is $\sim$6' (580 kpc).
\placefigure{f3.eps}
The extended emission on the west side can be divided into two
regions: north and south. The northern part is clearly an unrelated
double radio galaxy (see Fig. 4) with a central core and two extended
symmetric lobes. It is identified with a galaxy belonging to
the same group as 1144+35 (\cite{sc99}).
To the south of this radio galaxy we find an extended emission which cannot be
associated with the nearby field galaxy because of the strong
asymmetry and the lack of any direct connection. It is opposite to
the eastern emission of 1144+35 and we
tentatively identify it as the west lobe of 1144+35.
If 1144+35 is a giant
double radio galaxy, it has an angular size of $\sim$ 13' corresponding
to a projected linear size $\sim$ 1.3 Mpc and a total radio power at 1.4 GHz $\sim$ 1.24
$\times$ 10$^{25}$ W Hz$^{-1}$.
\placefigure{f4.eps}
To study the spectral index distribution of the extended emission, we
made two images at 1.4 and 5 GHz with $u,v$ coverage as similar as
possible. To accomplish this we removed the short baselines in the 1.4 GHz
data and the
long baselines at 5 GHz and obtained two images with a HPBW = 30'' to
enhance the sensitivity of the 5 GHz image to low brightness features.
After correcting for the primary beam attenuation we obtained a
spectral index map between these two frequencies. In Fig. 5 we show
the spectral index map with some contour levels from the 5 GHz image to
show the source structure. The central region of 1144+35 has a flat spectrum
($\alpha \sim$ 0.1 -- 0.2 with $S_\nu$ $\propto$ $\nu^{-\alpha}$). The
east extension is in general very steep; it shows a spectral index
$\alpha \sim$ 0.8 near the core that steepens continuously to the
east, reaching $\alpha \sim$ 2--2.5 in the more external regions. Only
in a small region at RA $\sim$ 11$^h$ 47$^m$ 41$^s$ where a discrete
unrelated source is present we find $\alpha \sim$ 0.9.
A Comparison of the present result with the spectral index map between 0.3 and
1.4 GHz discussed by \cite{sc99} shows that the low
frequency spectrum is flatter than the high frequency one suggesting
the presence of a steepening in the spectrum due to synchrotron losses in an
old radio structure. Future observations, better matched in angular resolution
and $u,v$ coverage, are necessary to confirm this point.
In the SW extended emission we find
$\alpha \sim$ 0.9 where a maximum of the brightness is visible and
\raise 2pt \hbox {$>$} \kern-1.1em \lower 4pt \hbox {$\sim$} 1.5 - 2.0 in the extended region. A more detailed analysis is
not possible due to poor sensitivity of the 5 GHz image to the extended
low brightness structure and because of the large primary beam
attenuation.
\placefigure{f5.eps}
We used the 1.4 GHz data and the standard formulae for synchrotron radiation
to calculate the minimum energy density (u$_{\rm min}$) and the equipartition
magnetic field H$_{\rm eq}$. We assumed a random magnetic field, equally
stored energy in relativistic electrons and heavy particles, a filling factor
= 1 and a lower and upper frequency cutoff of 10 MHz and 100 GHz respectively.
With these assumptions we found u$_{\rm min}$ $\sim$ 8 $\times$ 10$^{-14}$ erg
cm$^{-3}$ for the Eastern lobe and u$_{\rm min}$ $\sim$ 7 $\times$ 10$^{-14}$ erg
cm$^{-3}$ for the Western lobe. The H$_{\rm eq}$ is $\sim$ 1 $\mu$G in both
lobes. From the derived spectral index of radiating electrons and the magnetic
field present in the Eastern lobe we can estimate the lifetime of radiating
electrons suffering both synchrotron and inverse Compton losses. According to
the Jaffe-Perola model (\cite{ja74}) that assumes a redistribution
of electron pitch angles we derive an age in the range 5 to 9
$\times 10^7$ yrs.
\subsection{The Kiloparsec scale}
We observed 1144+35 with the VLA in C configuration at 5 GHz as part of a VLBI
experiment, obtaining a high quality map at an angular resolution of 3.5''
(see Fig. 6). This image
shows in detail the high brightness central structure visible in lower
resolution images.
We see a central emission slightly extended in the E-W direction (arcsecond core)
coincident with the optical
galaxy and faint jets emerging on both sides. We will call the SE jet
the main jet since it is longer and has a higher surface brightness.
Two faint and extended substructures are visible: J1 and J2, while in
the shorter counterjet only one (CJ1) is visible at about the same
core separation as J1. We measured the j/cj brightness ratio
at 10'' and 15'' from the core and obtained values of 3.6 and 1.1
respectively.
\placefigure{f6.eps}
To derive a spectral index map between 1.4 and 5 GHz we used the FIRST
image (\cite{be95}) and an image with same resolution and gridding
from our 5 GHz data.
Even though the 1.4 GHz data were obtained from a snapshot, we do not expect
missing flux at this frequency because of the small angular size of the
source.
The main jet has a relatively
flat spectrum: the spectral index is 0.3 and 0.45 in J1 and J2
respectively and 0.2 in between. Also CJ1 has a spectral index $\sim$0.35.
This result confirms the value derived on larger scales (Fig. 5)
and the abrupt change in the spectral index
distribution between this arcsecond scale structure and the extended
structure.
The arcsecond core spectral index will be discussed in Sect. 3.4 because of
the nuclear flux density variability.
\subsection{Polarization Data on Kiloparsec to Megaparsec scales}
The East lobe
shows a high percentage polarized flux (at 1.4 GHz we find 15 to 30\%
in the more external lobe regions, $\sim$ 5\% in the internal lobe
region and up to 60\% near the arcsecond nuclear region). At 5 GHz the
polarization percentage has the same distribution with slightly higher
values (from 30 to 70\%).
The polarization
vectors at 1.4 and 5 GHz are oriented at about the same angle.
In the extended Western region associated with 1144+35,
polarized flux is only detected in the brighter region. Here the
polarization vectors are radially oriented and the fractional
polarization is about 15 to 30\% at 1.4 GHz.
All the W radio emission does not show any change in the
vector's angle between 1.4 and 5 GHz and its depolarization is low.
To better study the central
region we obtained polarization images at 1.4 and 5 GHz with an HPBW of 11''
(see Figs. 7,8). At this resolution the signal
to noise ratio and the angular
resolution allow us to separate the different components.
At 5 GHz, in the core region, the electric field is oriented at 90$^\circ$
but at $\sim$ 15'' from the core it changes
to 0$^\circ$, and then rotates again to gradually become aligned with
the jet direction at $\sim$ 30'' -- 40'' from the core.
The 1.4 GHz vectors are rotated by $\sim$ 90$^\circ$ in the core,
and the region 15''
from the core is unpolarized. At 30'' from the core the 1.4 GHz
electric vectors are aligned with the 5 GHz
vectors. In the counterjet we see a faint polarized emission oriented at
$\sim$ 0$^\circ$ at 5 GHz and at $\sim$ 50$^\circ$ at 1.4 GHz.
\placefigure{f7.eps}
\placefigure{f8.eps}
The polarization percentage at 5 GHz is $\sim$ 20\% at 30'' from the core and
decreases slowly to a few percent in the core. At 1.4 GHz we have a comparable
percentage except in the region at 15'' which is completely
unpolarized at 1.4 GHz.
\subsection{The arcsecond core}
The arcsecond core of 1144+35 is the dominant feature of the radio
emission from this galaxy and has long been known to be variable
(\cite{ek83}). Its J2000 position is RA: 11$^h$ 47$^m$ 22.131$^s$
DEC +35$^\circ$ 01' 07.52'' (see
\cite{jo95} for a more accurate position).
In Table 3 we report the arcsecond core
flux densities available in the literature and the new results obtained by
us, in order to study the flux variability and spectrum. Only core
flux densities from high resolution data have been used so as to
avoid contamination by the extended emission. The arcsecond
core flux density is well sampled at 1.4 GHz from 1982 to 1998 and at
5 GHz from 1974 to 1998 while only sparse data at 8.4 and 15 GHz are
available from 1990 onwards. We obtained one observation at very high
frequencies (22 and 43 GHz) to study the high frequency spectrum (see
Figures 9 and 14).
At 1.4 and 5 GHz we see an increase in the
core flux density from 1974 to 1992 (1.4 GHz), and to 1994 (5 GHz).
In this time range the core flux density
increased by about a factor of 2.
After 1980 the flux density variations are smooth on time scales less than a
few years.
Starting in $\sim$ 1990-1992 the core flux density decreases and we observe a
delay in the decrease at lower frequencies possibly due to opacity effects.
We can derive the core spectral index by
comparing observations at the same time or very near in time.
Using 1997 data we have $\alpha \sim$ 0.43 between 5 and 15 GHz and 0.30
between 15 and 43 GHz with a possible flattening between 1.4 and 5 GHz
($\alpha \sim$0.13, see Fig. 14). Data in the range 1.4 to 8.3 GHz are
at the same epoch
while data in the range 15 to 43 GHz were obtained 6 months apart.
However, the slow variability and the suggestion of a constant spectral
index in time indicates that this short delay should not influence the
radio spectrum.
The present results, obtained with almost simultaneous observations rule out
the classification of the arcsecond core as a Gigahertz Peaked Spectrum (GPS)
source suggested by \cite{sn95} and \cite{sc99}. Their conclusion
was based on observations taken at too different epochs given the core flux
density variability.
\placetable{arcsecflux}
\placefigure{f9.eps}
\subsection{The parsec scale}
\subsubsection{Total intensity and spectral indices}
In Figure 10 we present the full resolution image obtained with the MERLIN array
at 5 GHz. It shows a
central source with an extended emission ($\sim$ 100 mas) in the direction
of the main jet and a marginally visible extension also in the opposite
direction.
\placefigure{f10.eps}
With the VLBA at 1.6 GHz the central source is resolved in a double structure
(Fig.~11) with a weak emission
at about 60 mas from the stronger component, in agreement with the
MERLIN image at 5 GHz.
At 5 GHz with the VLBA, the parsec scale structure is resolved in four main
substructures:
two compact components (A and B in Fig. 12) separated by 3--4 mas, a discrete
component (C) with a short symmetric extension located at about 20 mas
from component A and a faint extended emission surrounding A and B components
and clearly elongated in the direction of (C).
The shape of this extended feature is not the same
in the different epoch and frequency images due to
different $u,v$ coverage and
sensitivity to low brightness structure, however the inner peak of this
feature on the top (North) of A and B is always visible and we have named it A1
in Fig. 12.
\placefigure{f11.eps}
\placefigure{f12.eps}
\placefigure{f13.eps}
In Figure 13 we show our most recent and highest resolution image at 8.4
GHz. The radio structure is very similar to the 5 GHz image, although
C is clearly resolved into three
components, labeled C, D and E. Component A1 is resolved
into a complex structure roughly aligned with components C, D and E
and the kpc-scale jet. A fairly compact component B1 has emerged
just west of component B. There is also a clear bridge between
components A and B.
The core variability found at arcsecond resolution implies that
a flux density variability must be associated with some or all components
detected in the parsec scale.
We have compared the flux density taken at different epochs of components
A, B, and C
at the same frequency: component C is almost constant at 5 and 1.4 GHz
while at 8.4 GHz it may have increased its flux density (58 mJy on
March 1995; 65 mJy on September 1997).
Components A and B follow the trend of the arcsecond core:
their flux density slightly increases or is constant from 1990-94 and
decreases in more recent observations. Since the flux density of component A
($\sim$ 185 mJy
in the last 8.4 GHz map) is
much higher than the B (30 mJy) and C (65 mJy) flux density it is clear that
most of the arcsecond core flux density variability can be attributed
to variability of component A.
To derive the spectral index of the different components we used the
VLBA observation obtained on Nov. 26th, 1995 where the source was
observed simultaneously at 1.4, 5, and 8.4 GHz. Moreover we used
the flux densities at 15 and 22 GHz (\cite{he96})
which were not too far in time
given that the flux density variability is a smooth function of time
(see Section 3.4). The spectrum of components A, B, and C is shown in Figure 14
where we give also the arcsecond core spectrum at epoch 1997 for a comparison.
We do not show the spectrum of A1 since the flux density of this
extended feature is strongly influenced by the different $u,v$ coverages of the
observations.
Components A and B show similar spectra: we derive for component
A $\alpha^{1.4}_{8.4}$ = 0.45 and $\alpha^{8.4}_{22.0}$ = 0.92 while component
B has $\alpha^{5.0}_{8.4}$ = 0.38 and $\alpha^{8.4}_{22.0}$ = 0.98.
Component C has an inverted spectrum peaked around 8.4 GHz;
$\alpha^{1.4}_{8.4}$ = $-$0.34 and $\alpha^{8.4}_{22.0}$ = 0.45 (see Fig. 14).
Based on its spectral properties we identify component C as the center
of activity and the other components as parts of a parsec scale jet.
The core is therefore
self-absorbed with the maximum flux density at $\sim$ 8.4 GHz which
implies a magnetic field of 3--5 Gauss. The two extended
symmetric regions (D,E) visible at high resolution on both
sides of the core component C are the base of a
two sided symmetric jet structure. No further jet emission is visible on
the NW side (counter jet side), while on the SE side (the jet side) we
have the extended structure A, B, A1 and B1. Such a jet asymmetry will
be discussed in Sect. 4.
We identify the extended complex structure A,A1,B,B1 as a clear evidence of a limb
brightened jet: the maximum surface brightness is on the sides of a
resolved jet. The projected jet opening angle is of $\sim$ 10$^\circ$.
Such a structure is expected by the {\it central spine - shear layer}
jet model (\cite{la96}).
The image in Fig. 13 shows a gap of emission beyond D
implying that the shear layer is not visible at the beginning but
appears at $\sim$ 15 mas from the core.
In this context we
interpret the A, B and B1 structures as due to interaction of the
parsec scale jets with an inhomogeneous surrounding medium (clouds?).
The alternative possibility that A1, D and E components are the parsec scale jet,
as suggested by their alignment with the core, while
A, B and B1 are shocked components ejected at
a difference position angle with respect to the jet direction, appears more
problematic. The good alignment of components A1, D and E with the core shows that
this should be
the present jet position angle and therefore A, B and B1 should be an older
emission. However we note that A1 component looks more extended than A and
B components, moreover the estimated velocity of A1 (see Sect. 3.5.3), even if
affected by
a large uncertainty, is very similar to the A and B measured velocity.
Therefore we interpret the complex parsec scale morphology as a limb
brightened jet with a relevant asymmetry on one side probably due to
a strong interaction with a non symmetric external medium.
\placefigure{f14.eps}
\subsubsection{Polarization on the parsec scale}
In Fig. 15 we present the total intensity VLBI image obtained from the Sept. 97
epoch observation
at 8.4 GHz with polarized intensity vectors. Components A and B are polarized at
the $\sim$ 0.4\% and
1.2\% level respectively, while the extended northern structure (A1)
is polarized at a level of $\sim$ 5\%. A fractional polarization of 0.3\%
(170 $\pm$ 30 $\mu$Jy) was found at the core (C) position. The
polarization angles are not aligned along any preferred
orientation. This could indicate a disorder in the magnetic field
and/or a high and changing Rotation Measure, either one of which is consistent
with the low fractional polarization observed.
This result is in agreement with the presence of a dense interstellar medium
in the parsec scale region.
\placefigure{f15.eps}
\subsubsection{Proper Motion}
We used the observations made at different epochs to measure the apparent
proper motion of components A, A1, B, D, and E with respect to the core
component C.
First of all we
compared observations obtained at the same frequency at different
epochs to avoid any possible spurious result due to different spectral
indices of components and afterwards we compared the results obtained
using all the data. As expected from the uniform spectral
index and the small size of components we found that the proper motion
is only marginally related to the observing frequency and therefore we
used all the data, except the 1.4 GHz ones where the angular resolution
is too low to properly separate different components.
In Table 4 and Fig. 16 we show the distance of components A, B, and E from C at
each epoch. From these data we derive
that no proper motion is measurable within the errors for
component E ($\beta$ ~ \raise 2pt \hbox {$<$} \kern-1.1em \lower 4pt \hbox {$\sim$} ~ 0.2). Components A and B show a clear
motion from the core C in the direction of the kiloparsec
scale structure with a constant velocity. The average velocity is 2.78
$\pm$ 0.1 c for component A and 2.62 $\pm$ 0.1 c for component B.
The motion of component A1 is not well determined
because of the extended size and low surface brightness of this component.
Comparing different
epoch data we note that the radio emission centroid of A1 is always in between
A and B components suggesting a similar velocity of A1
with respect to A and B.
For component D no reliable measures were
possible because of the lack of any visible substructure.
\placetable{masdistance}
\placefigure{f16.eps}
\section{Jet Orientation and Velocity}
From the measured apparent proper motion of the parsec scale jet,
$\beta_a$ $\sim$ 2.7 h$_{50}^{-1}$,
we can constrain the jet orientation ($\theta$) and velocity
($\beta$c):
\centerline {$\beta$ = $\beta_a$/($\beta_a cos\theta + sin\theta$)}
We find that the 1144+35 jet has to move at a minimum intrinsic velocity
of $\sim$ 0.94 c and that $\theta$ has to be smaller than 40$^\circ$.
The angle $\theta$ corresponding to the minimum velocity is
$\sim$ 20$^\circ$.
Assuming an intrinsic symmetry in the parsec scale structure of 1144+35,
we identify the component E as the counterpart of the brightest region of
the main jet, on the counter jet side.
Due to the expected small viewing angle of the source with respect
to the line
of sight (see above), the size of the counterjet emission is expected to
appear smaller than that of the jet emission.
Therefore assuming
intrinsic symmetry we can use the observational
parameters to derive constraints on the velocity and orientation of this
parsec scale jet independently of the Hubble constant as following:
1) Comparing the apparent velocity of the jet $\beta_{\rm aj}$, and of the
counterjet $\beta_{\rm acj}$, we have (see e.g. \cite{mi94}):
\centerline {$(\beta_{\rm aj} - \beta_{\rm acj}$)/$(\beta_{\rm aj} + \beta_{\rm acj})$ = $\beta cos\theta$}
\vskip 0.5truecm
Given our limit on $\beta_{\rm acj}$~~ \raise 2pt \hbox {$<$} \kern-1.1em \lower 4pt \hbox {$\sim$} ~~0.2, we find:
\centerline{$\beta cos\theta$ \raise 2pt \hbox {$>$} \kern-1.1em \lower 4pt \hbox {$\sim$} 0.86.}
2) From the jet -- counterjet arm ratio (\cite{g98a,ta97}), we derive:
\centerline{$\beta cos\theta$ $\sim$ 0.82.}
Comparing different constraints, we conclude that the 1144+35 jet has an
intrinsic high velocity
$\sim$ 0.95c or higher at a small angle with respect to the
line of sight (25$^\circ$ or smaller).
Given the large size of the extended emission and assuming no large difference
in the orientation from the parsec to the Megaparsec scale structure,
we will assume a
value of $\theta$ $\sim$ 25$^\circ$. With this orientation with respect to
the line of sight and with $\beta \sim$ 0.95, the Doppler factor $\delta$ is
$\sim$ 2.25, the intrinsic jet opening angle is $\sim$ 4.2$^\circ$ and
the total linear size of the source becomes $\sim$ 3 Mpc putting it among the
largest of giant radio galaxies.
An independent constraint on $\delta$ can be derived from the
synchrotron-self-Compton (SSC) model of X-ray emission from the nuclear
region (see \cite{mar87,ghi93}). In principle when the
core angular size and the
non-thermal nuclear X-ray emission is known, the comparison
between the predicted and observed X-ray flux density gives constraints
on the Doppler factor.
From the upper limit to the core size from our VLBI data (~\raise 2pt \hbox {$<$} \kern-1.1em \lower 4pt \hbox {$\sim$}
flux density measured by \cite{bri95} is
non-thermal nuclear emission, we
derive a lower limit for $\delta$ (see also \cite{g94}). Taking
8.4 GHz as the core self-absorption frequency we find a Doppler factor
$\delta$ larger than
1.4 in agreement with our previous estimate.
From the estimated intrinsic velocity of the parsec scale jet, we can derive
that the
A+B+A1+B1 complex was ejected from the central engine about 300 years ago in
the source reference frame.
In this context the D feature is a more recent ejection from the core whose
proper motion is not yet visible due to the lack of
visible sub-structures. We would predict that in the near future it should
be possible to see this component moving away from the core.
The velocity obtained above refers to the brightest jet region (the shear
layer). The presence of a central jet region (spine) with a lower brightness
can be due to a higher velocity of the inner jet region with respect to
the brighter external jet regions. A higher velocity implies in our case
a low Doppler factor and therefore the jet central spine
moving at high velocity will be less visible than the external shear layer.
Also
the presence of a gap of emission between D and the extended jet emission
could be due to a low Doppler factor of a fast moving jet.
For the A and B structures (but A1 and B1 are very likely moving at the same
velocity) we have derived a doppler factor $\delta$ $\sim$ 2.25.
This value implies that the intrinsic radio emission is amplified of a
factor 7.5 -- 8. To justify the dimming
in the surface brightness in the central region of the jet and the gap, we
have to assume that the fast jet spine should move with a Lorentz factor
$\gamma$ $\sim$ 15.
A value of $\gamma$ $\sim$ 15 corresponds to a velocity $\sim$ 0.998 c and a
Doppler factor $\delta$ $\sim$ 0.698 which implies that the intrinsic radio
emission is dimmed by a factor $\sim$ 0.4.
This model assumes an intrinsically symmetric jet
while our images show a clear difference between components A, B, B1 and A1. As
discussed in Sect. 3.5.1 we assume that these inhomogeneities are due to the
interaction of the external jet regions with an inhomogeneous and
clumpy surrounding medium. If this is the case, the inner jet spine should
not be affected by the interaction with the external medium.
A free expanding jet is expected to show an intrinsic opening angle
$\sim 1/\gamma$, (see e.g. \cite{sa98}) while a well confined and
collimated jet will show a smaller
opening angle. For the present source the estimated value of $\gamma$ $\sim$ 15
(1/$\gamma$ $\sim$ 4$^\circ$) is in very good agreement with the intrinsic
jet opening angle derived before (4.2$^\circ$ if $\theta \sim 25^\circ$).
In this scenario, at the jet beginning there is no or very little
difference in velocity between the spine and the shear layer therefore the
whole jet is doppler deboosted
and we have a gap in the radio emission. At about 15 mas from the
core (corresponding
to a de-projected distance of $\sim$ 55 pc) the external jet regions are moving
at a lower velocity because of their interaction with the surrounding medium
and the shear layer becomes more visible
being Doppler boosted, while the inner jet region
is still moving at high velocity and consequently is de-boosted.
The presence of component D near the core could have two possible
explanations: (1) the jet is accelerating as found in NGC 315
(\cite{cot98}); or
(2) the intrinsic brightness of the jet near the core is high
and therefore component D is visible despite the dimming due to the
de-boosting effect.
With the existing data we cannot exclude that
a transversal asymmetry causes: (1) the difference between the A, B, B1 and
A1 regions; (2) the low brightness in the jet central region; and (3) the gap
near the core. Future multi-frequency observations of the polarized emission
in the jet should help to clarify the degree of
interaction between the jet and the surrounding medium. Furthermore, long
term monitoring of the source should allow the detection of a proper motion in
the counterjet side (component E). This measurement is important since it
will constrain the intrinsic jet velocity and allow for a direct
distance determination.
We found a similar parsec scale morphology with
space VLBI observations of the BL-Lac
object Mkn 501 (\cite{g98b}): in this source we see a
centrally peaked jet moving
at a velocity in the range 0.990 -- 0.999c in the inner 40 -- 45 pc
from the core, which becomes limb brightened at a de-projected distance
of $\sim$ 45 pc.
This result reinforces the scenario of a jet which appears
limb-brightened at some distance from the core ($\sim$ 40 -- 50 pc) due to the
brightness of the
shear layer which has slowed down
with respect to the central spine
because of the interaction with the surrounding medium.
Assuming a constant value for $\theta$, from the j/cj ratio on the kiloparsec
scale we find a jet velocity of $\sim$ 0.28 c at 10'' from the
core and of 0.02 c at 15'' from the core in agreement with the expected jet
velocity decrease in low power radio galaxies (see e.g. 3C449, \cite{fer99}).
Such a velocity cannot explain
the armlength asymmetry visible at arcsecond resolution (see Fig. 6);
we suggest in agreement with \cite{sc99} that the arcsecond
arm ratio asymmetry is due to the interaction with an
inhomogeneous external medium which is responsible also for the asymmetric
Megaparsec structure of 1144+35.
\section{Arcsecond Core Flux Density Variability}
The arcsecond core flux density of 1144+35 increases from
1974 to
$\sim$ 1994 and decreases from 1994 to 1998. As discussed before, the flux density
variability is not due to
the activity of the mas core but to changes in the A, B, A1, B1
complex, specially in the A component being the region with highest
flux density on the parsec scale.
A model to explain such a flux density turn-over in
GPS sources was presented by \cite{sn98} and in agreement with
{\cite{sc99} it could explain the core flux density variability in the
present source.
In fact, according to the scenario discussed in Sect. 3.5.3
the flux density variation of a factor
2 between 1974 and 1994 can be related to a slowing down of the shear layer
of the parsec
scale jet which increases the Doppler factor and therefore the observed flux.
To produce a factor 2 of increasing flux the Doppler factor had to
be $\sim$ 1.7 taking into account that now it is estimated to be $\sim$ 2.25.
Such a variation can be obtained if the velocity decreases from $\sim$ 0.98c
to 0.95 c in the last $\sim$ 15 mas behind A.
The faster flux density increase from 1974 to 1980 followed by a slower
increase from
1980 to 1994 (Fig. 9) implies a variable rate of the Doppler factor
probably due to a velocity decrease not constant in time but with a stronger
deceleration at the beginning.
The decreasing flux from 1995 to present days could be due to adiabatic
losses if A
(and B) are expanding. In an expanding component we expect a delay in the
flux density
decrease at lower frequencies due to opacity effects, as found in
the 1144+35 core (see Fig. 9).
\section{The Connection between the Large and Small Scale Structure}
The large scale structure shows a clear discontinuity between the extended
relaxed lobes and the high brightness arcsecond core and jets (Sect. 3.1).
This discontinuity is confirmed by the spectral index map which shows a
large change in the spectral index moving from the arcsecond jet
to the extended East lobe.
These observational data suggest two different phases in the life of
this radio source: the extended structure is a relic emission with an age
in the range 5 to 9 $\times$ 10$^7$ yrs as estimated in Sect. 3.1, while
the emission on the arcsecond scale has a shorter dynamical age:
assuming an average velocity of \raise 2pt \hbox {$>$} \kern-1.1em \lower 4pt \hbox {$\sim$} 0.02 c for the arcsecond
structure
we derive an age of \raise 2pt \hbox {$<$} \kern-1.1em \lower 4pt \hbox {$\sim$} 1.0 $\times 10^7$ yr, taking into account projection
effects.
We note that the Megaparsec scale East lobe of this source shows similar
physical properties and is very similar in shape to the extended
relic source 1253+275, found at the periphery of the Coma cluster (Giovannini et
al.\ 1991). We can speculate that if the core of 1144+35 had ceased its
activity, the extended emission could no longer be recognized as due to the
activity of the galaxy ZW186.48 but it would be considered a relic emission in
a group of galaxies.
A merger event with a gas rich object as suggested also by the {\it boxy}
shape of the optical image could be the origin of the restarted activity.
The large number of {\it young} CSS or CSO sources showing evidence of a recent
merger event (\cite{o'98}) confirms the correlation between these processes
and the radio activity.
Moreover we note that the parsec scale jet emission does not decrease smoothly
in our VLBI maps. The parsec scale jet appears to stop at $\sim$ 24 mas from
the core even if we have some evidence from the MERLIN and the 20 cm VLBA map
that a low level flux density emission
is present at larger distance from the core, in the same position angle.
This can be due to a dramatic
expansion of the radio jet after the position of the A component with a
correspondingly large decrease in the
surface flux density, similar to the jet expansion visible in Mkn 421 and
Mkn 501 (Giovannini et al.\ 1998b). Alternatively the A, A1, B, B1 complex
could be a
strong enhancement in the jet brightness due to a
strong increase of the nuclear activity.
This burst
should have taken place about 300 yrs ago in
the source reference frame, corresponding to $\sim$ 45 yrs in our reference
frame. This hypothesis is in agreement with the core flux density variability
discussed in the previous sections.
\section{Conclusions}
We present new VLA and VLBI observations of the low power radio galaxy
1144+35. These observations allow us to study and discuss the properties of
this source from the Megaparsec to the
parsec scale. 1144+35 is one of the few sources with observed two-sided
parsec scale
jets. These sources are of particular interest since they can be used to
place constraints on the
cosmological parameters H$_0$ and q$_0$. Moreover this source is oriented at
$\sim$ 25$^\circ$ with respect to the line of sight therefore, in agreement
with unified scheme models, presents properties in between FR I radio galaxies
and BL-Lac type objects.
On the Megaparsec scale 1144+35 appears to be a giant radio galaxy:
1.3 Mpc on the
plane of the sky corresponding to a deprojected linear size $\sim$ 3 Mpc.
It shows an East lobe with an
elongated shape, connected
to the core emission and a Western extended emission interpreted as the West
lobe of a double
radio galaxy. The age of this emission is estimated to be 5 to 9 $\times$
10$^7$ yrs.
At kiloparsec resolution, a core and a two-sided jet are present.
The surface
brightness and the spectral index distribution suggest a strong discontinuity
between this structure
and the Megaparsec scale emission. This observational result could be due
to a change in the jet direction or to renewed activity in the galaxy core
probably triggered by a galaxy merger.
The arcsecond core is the dominant feature of this source at arcsecond
resolution. Its flux density showed a large increase from 1974 to 1980
(from 280 to 500 mJy at 5 GHz),
followed by a smooth increase until 1992. From 1992 to date the core flux
density has decreased; the flux density decrement appeared first at higher
frequencies. The core spectrum obtained from same epoch observations is flat
and we rule out the
identification of the 1144+35 core as a GPS source.
At parsec resolution we identified the core source (C) and a two-sided jet
emission very
asymmetric in shape and properties. This asymmetry, when interpreted in the
light of unified scheme models and of relativistic jets, constrains the jet
velocity to be
\raise 2pt \hbox {$>$} \kern-1.1em \lower 4pt \hbox {$\sim$} 0.95c with an orientation with respect to the line of sight
\raise 2pt \hbox {$<$} \kern-1.1em \lower 4pt \hbox {$\sim$} 25$^\circ$.
These results are in agreement with the detected proper motion with $\beta_a$
$\sim$ 2.7.
The source orientation explains its properties intermediate between a FR I
radio galaxy
and a BL-Lac type object.
The parsec scale jet is limb brightened and its morphology is in agreement
with the presence of a fast inner jet spine ($\beta \sim$ 0.998) surrounded
by a shear layer in which the velocity
decreases because of the interaction with the surrounding medium.
The estimated age of this structure is of $\sim$ 300 yrs in the source
reference frame.
\acknowledgements}
We thank R. Fanti, D. Dallacasa and G. Brunetti for critical reading of the
manuscript and useful discussions.
The Digitized Sky Survey was produced at the Space Telescope Science Institute
under US Government grant NAG W-2166.
This work was partly supported by the Italian Ministry for University
and Research (MURST) under grant Cofin98-02-32
\clearpage
\begin{table}
\begin{center}
\caption{VLA image parameters \label{arcsecvlamaps}}
\medskip
\begin{tabular}{ccclcc}
\tableline \tableline
Array & Frequency & HPBW & noise & Figure & notes \\
& GHz~~~~~& $\prime\prime$ ~~~~& mJy/b & N. & \\
\hline
C+D & 1.4 & 20 & 0.06 & 2 & I map \\
C+D & 4.9 & 20 & 0.015 & 3 & I map \\
C+D & 4.9 & 20 & 0.012 & 3 & P map \\
C+D & 4.9 & 11 & 0.014 & 4 & I map \\
C & 4.9 & 3.5 & 0.06 & 6 & I map \\
\tableline
\end{tabular}
\medskip \\
\end{center}
\tablecomments{
See Figure captions for more details; I map is the total intensity image; P map
is the Polarized image used to draw E-vectors superimposed to the total
intensity images.}
\end{table}
\clearpage
\begin{table}
\begin{center}
\caption{History of VLBI Observations \label{vlbidata}}
\medskip
\begin{tabular}{cccccc}
\tableline \tableline
Epoch & Array & Array & Array & Array & Array \\
mm-yy & 1.4GHz & 5.0GHz & 8.3GHz & 15.3GHz& 22.2GHz \\
\hline
11-90 & & Mk2$^1$& & & \\
11-92 & Mk2$^1$& & & & \\
06-93 & & Mk2$^5$& & & \\
01-95 & & & & VLBA$^3$& \\
03-95 & VLBA$^2$& & VLBA$^2$ & & \\
08-95 & & VLBA$^4$ & & & \\
11-95 & VLBA$^1$ & VLBA$^1$ & VLBA$^1$ & & \\
02-96 & & & & & VLBA$^3$ \\
08-96 & & VLBA$^4$ & & & \\
09-97 & & & GLOBAL$^1$ & & \\
\tableline
\end{tabular}
\medskip \\
\end{center}
\tablecomments{
1 -- Full synthesis (8-10 hours long) observations; 2 -- 5$\times$6 min.
snap-shots; 3 -- see D. Henstock, 1996; 4 -- 8$\times$6 min.
snap-shots with 8 MHz bandwidth; 5 -- see Henstock et al. 1995.}
\end{table}
\clearpage
\begin{table}
\begin{center}
\caption{Arcsecond core flux densities \label{arcsecflux}}
\medskip
\begin{tabular}{cccccccc}
\tableline \tableline
Epoch & Flux(mJy) & Flux(mJy) & Flux(mJy) & Flux(mJy) & Flux(mJy) & Flux(mJy) &Ref. \\
mm-yy & 1.4 GHz & 5.0 GHz & 8.3 GHz & 15 GHz & 22 GHz & 43 GHz & \\
\hline
01-73 & 340 & - & - & - & - & - & 1 \\
01-74 & - & 290 & - & - & - & - & 1 \\
12-74 & - & 310 & - & - & - & - & 2 \\
02-75 & - & 310 & - & - & - & - & 2 \\
04-75 & - & 330 & - & - & - & - & 2 \\
12-75 & - & 340 & - & - & - & - & 2 \\
08-76 & - & 395 & - & - & - & - & 2 \\
01-77 & - & 425 & - & - & - & - & 2 \\
02-80 & - & 500 & - & - & - & - & 2 \\
08-82 & 529 & - & - & - & - & - & 3 \\
12-84 & 568 & - & - & - & - & - & 4 \\
05-85 & 568 & - & - & - & - & - & 5 \\
02-90 & - & - & 501 & - & - & - & 6 \\
11-90 & - & 551 & - & - & - & - & pp \\
10-91 & 600 & - & - & 359 & - & - & 7 \\
04-94 & 575 & 553 & 450 & - & - & - & pp \\
08-94 & - & 537 & - & - & - & - & 8 \\
07-95 & 570 & - & - & - & - & - & 8 \\
02-97 & 569 & 483 & 393 & - & - & - & pp \\
08-97 & - & - & - & 293 & 262 & 215 & pp \\
08-97 & 541 & - & - & - & - & - & 9 \\
\tableline
\end{tabular}
\medskip \\
\end{center}
\tablecomments{
References: 1) Colla et al.\ 1975; 2) Ekers et al.\ 1983; 3) Parma et al.\
1986; 4)
Fanti et al.\ 1986; 5) Fanti et al.\ 1987; 6) Patnaik et al.\ 1992; 7)
Snellen et al.\ 1995; 8) Taylor et al.\ 1996; 9) Schoenmakers et al.\ 1999;
pp) present paper.}
\end{table}
\clearpage
\begin{table}
\begin{center}
\caption{Distances from the core at various epochs \label{masdistance}}
\medskip
\begin{tabular}{cccccc}
\tableline \tableline
\\
Epoch & E & A & B & Frequency \\
dd/mm/yy & mas & mas & mas & GHz \\
\\
20/11/90 & - & 20.2 & 16.4 & 5.0 \\
10/06/93 & 2.3 & 21.4 & 18.0 & 5.0 \\
27/01/95 & - & 22.5 & 18.5 & 15 \\
22/03/95 & 2.1 & 22.5 & 18.6 & 8.4 \\
25/08/95 & 2.2 & 22.5 & 18.6 & 5.0 \\
26/11/95 & 2.4 & 22.7 & 18.8 & 5.0 \\
26/11/95 & 2.2 & 22.8 & 18.9 & 8.4 \\
18/02/96 & - & 23.2 & 19.3 & 22 \\
22/08/96 & 2.0 & 23.0 & 19.1 & 5.0 \\
27/09/97 & 2.4 & 23.8 & 19.8 & 8.4 \\
\tableline
\end{tabular}
\medskip \\
\end{center}
\tablecomments{
Positional uncertainties are $\pm$ 0.2 mas for the faint component
E and $\pm$ 0.1 mas for A and B components ($\pm$ 0.07 mas
at 15 and 22 GHz).}
\end{table}
\clearpage
|
\section{Introduction}
The study of molecular gas and dust is a significant observational
tool for probing the physical conditions and star formation
activity in local galaxies. Recent advances in observational
techniques at submillimeter and millimeter wavelengths now permit
such studies to be made at cosmological distances.
As a result of the IRAS survey many local objects are recognized as
containing a large mass of dust and gas, such that the objects may be
more luminous in the far-infrared (FIR) than in the optical. The
question as to how many such objects there may be at cosmological
distances, and whether they can account for the recently discovered
FIR cosmic background (\cite{Puget96}; \cite{Fixsen98};
\cite{Hauser98}), is attracting much interest and spawning new
instrument construction and new surveys (\cite{Hughes98};
\cite{Ivison98}; \cite{Kawara98}; \cite{Puget98}; \cite{Barger98};
\cite{Lilly99}). Existing
submillimeter cameras on ground-based telescopes are not yet sensitive
and large enough to detect distant objects at 350$\,\mu$m in arbitrary
blank fields, e.g., the initial Caltech Submillimeter Observatory
(CSO) SHARC survey which achieved 100$\,$mJy ($1\,\sigma$) over about
10 square arcminutes (\cite{Phillips}). However, such cameras can measure
the 350$\,\mu$m flux of objects of known position. A step forward in
the field was the recognition of IRAS~F10214+4724 as a high redshift
object ($z=2.286$) by Rowan-Robinson et al. (1991). However, it has
proved difficult to find many such objects to study. On the other
hand, quasars have sometimes proved to exist in the environs of dusty
galaxies (\cite{Haas98}; \cite{Lewis98}; \cite{Downes98}). Omont et
al. (1996b) have shown by means of a $1300\,\mu$m survey of radio-quiet
quasars that the dust emission at high redshifts can be detected in a
substantial fraction of their project sources.
In this letter we present measurements at 350$\,\mu$m towards a sample
of 20 sources with redshifts $1.8\lesssim z\lesssim 4.7$ which were
selected from different surveys and studies (references are given in
Tables \ref{detections} \& 2). The sources had
previously been detected at longer wavelengths, and are predominantly
from the work of Omont et al. (1996b) and Hughes, Dunlop \& Rawlings
(1997). A wavelength of 350$\,\mu$m roughly corresponds to the peak
flux density of highly redshifted ($z\simeq3$) dust emission of
objects with temperatures of 40 to 60$\,$K. Together with
measurements at longer wavelengths, it strongly constrains the dust
temperature and, hence, the dust mass and, especially, the luminosity
of the object. Some of these results were first presented by Benford
et al. (1998). In this paper, we assume ${\rm H}_0 = h_{100} \times
100 \, \rm km \, s^{-1} \, Mpc^{-1}$ and $\Omega_0 = 1$.
\begin{table*}[bhpt]
\caption{Sources detected at 350$\,\mu$m and derived properties.
\label{detections}}
\begin{center}
\footnotesize{
\begin{tabular}{llccccccc}
\hline
\hline \\[-0.2cm]
\multicolumn{1}{c}{Source} & \multicolumn{1}{c}{$z$}
& \multicolumn{1}{c}{R.A.} & \multicolumn{1}{c}{Dec.}
& \multicolumn{1}{c}{Flux Density} & \multicolumn{1}{c}{T$\rm _{dust}$}
& \multicolumn{1}{c}{Dust Mass\tablenotemark{\ddagger}} &
\multicolumn{1}{c}{Luminosity\tablenotemark{\ddagger}} & Ref. \\
& & \multicolumn{2}{c}{(B1950.0)} & (mJy, $\pm1\,\sigma$)& (K) &
(10$^{8} h_{100}^{-2}$ M$_{\odot}$)
& (10$^{12} h^{-2}_{100}$ L$_{\odot}$) & \\[0.2cm]
\hline \\[-0.2cm]
BR1202$-$0725 & 4.69 & 12 02 49.3 & $-$07 25 50 & 106$\pm$7\phn & 50$\pm$7
& 4.0$ ^{+0.9}_{-0.8}$ & 14.9$ ^{+0.8}_{-0.7}$ & 1,2,3,4 \\[0.1cm]
BRI1335$-$0417 & 4.41 & 13 35 27.6 & $-$04 17 21 & 52$\pm$8 & 43$\pm$6
& 3.6$ ^{+0.5}_{-0.8}$ & \phn6.0$ ^{+0.9}_{-0.5}$ & 1,2,5 \\[0.1cm]
HM0000$-$263 & 4.10 & 00 00 49.5 & $-$26 20 01 & 134$\pm$29 & 60\tablenotemark{\dagger}
& 2.0$ ^{+0.2}_{-0.2}$ & 20.7$ ^{+1.5}_{-1.5}$ & 2,6 \\[0.1cm]
4C41.17 & 3.80 & 06 47 20.8 & $+$41 34 04 & 37$\pm$9 & 52$\pm$6
& 1.0$ ^{+0.1}_{-0.1}$ & \phn4.3$ ^{+1.4}_{-1.3}$ & 7,8 \\[0.1cm]
PC2047$+$0123 & 3.80 & 20 47 50.7 & $+$01 23 56 & 80$\pm$20 &
50\tablenotemark{\dagger} & 2.3$ ^{+0.6}_{-0.6}$ & \phn8.5$^{+2.2}_{-1.6}$ & 9,10 \\[0.1cm]
Q1230+1627 & 2.74 & 12 30 39.4 & $+$16 27 26 & 104$\pm$21 &
49$\pm$12 & 2.5$ ^{+0.4}_{-0.4}$ & 8.2$^{+2.0}_{-2.0}$ & 2,11 \\[0.1cm]
Q0100+1300 & 2.68 & 01 00 33.4 & $+$13 00 11 & 131$\pm$28 & 68$\pm$5
& 1.2$ ^{+0.5}_{-0.3}$ & 23.8$ ^{+0.2}_{-5.8}$ & 12,13\\[0.1cm]
H1413+117\tablenotemark{\ddagger} & 2.54 & 14 13 20.1 & $+$11 43 38 & 293$\pm$14 & 45$\pm$7
& 8.9$ ^{+1.9}_{-1.9}$ & 17.7$ ^{+1.3}_{-0.6}$ & 14,15 \\[0.1cm]
F10214+4724\tablenotemark{\ddagger} & 2.28 & 10 21 31.1 & $+$47 24 23 & 383$\pm$51 & 55$\pm$3
& 5.5$ ^{+1.4}_{-1.4}$ & 32.8$ ^{+6.5}_{-5.3}$ & 16 \\[0.1cm]
\hline \\[-0.2cm]
\end{tabular}
}
\end{center}\vspace{-0.5cm}
\tablenotetext{\dagger}{
Because of a lack of sufficient data at other wavelengths, the
temperatures of these sources are poorly constrained. The available
upper limits constrain HM0000$-$263 to $50\lesssim \rm T_{dust}
\lesssim 75$K, so we adopt 60K; PC2047+0.123 is essentially
unconstrained, so we adopt 50K as that is roughly the median value for the dust temperature.}
\tablenotetext{\ddagger}{The dust masses and luminosities
listed above have not been corrected for any lensing amplification.
The probable intrinsic values may be found by
dividing by $A=7.6$ for H$\,$1413+117 (\cite{Alloin97}) and $A=13$ for
F10214+4724 (\cite{Downes95}).}
\tablerefs{
1. Storrie-Lombardi et al. (1996); [2.] Omont et al. (1996b);
[3.] McMahon et al. (1994); [4.] Isaak et al. (1994);
[5.] Guilloteau et al. (1997); 6. Schneider, Schmidt \& Gunn (1989);
7. Chambers, Miley \& van Bruegel (1990); [8.] Hughes, Dunlop \& Rawlings(1997);
9. Schneider, Schmidt \& Gunn (1994); [10.] Ivison (1995);
11. Foltz et al. (1987); 12. Steidel \& Sargent (1991);
[13.] Guilloteau et al. (1999); 14. Hazard et al. (1984);
[15.] Barvainis, Coleman \& Antonucci (1992);
[16.] Rowan-Robinson et al. (1993) \\
Note: the references in brackets refer to the observations shown in Fig.~2.}
\normalsize
\end{table*}
\section{Observations and Results}
The measurements were made during a series of observing runs in 1997
February and October and 1998 January and April with the 10.4$\,$m
Leighton telescope of the CSO on the summit of Mauna Kea, Hawaii,
during excellent weather conditions, with 225$\,$GHz atmospheric
opacities of $\lesssim0.05$ (corresponding to an opacity of
$\lesssim1.5$ at 350$\,\mu$m). We used the CSO bolometer camera, SHARC,
described by Wang et al. (1996) and Hunter, Benford, \& Serabyn
(1996). It consists of a linear 24 element close-packed monolithic
silicon bolometer array operating at 300$\,$mK. During the
observations, only 20 channels were operational. The pixel size is
$5''$ in the direction of the array and $10''$ in the cross direction.
The weak continuum sources were observed using the pointed observing
mode with the telescope secondary chopping in azimuth by 30$\rm
^{\prime\prime}$ at a rate of 4$\,$Hz. The telescope was also nodded
between the on and the off beams at a rate of $\sim$0.1Hz. The
point-source sensitivity of SHARC at 350$\,\mu$m is $\rm \sim 1 \,
Jy/\sqrt{Hz}$ and the beam size is $\sim9''$ FWHM. All measurements
were made at $350\mu$m with the exception of H$\,$1413+117 which was
also observed at $450\mu$m.
Pointing was checked regularly on nearby strong galactic sources which
also served as secondary calibrators, and was found to be stable with
a typical accuracy of $ \lesssim 3^{\prime\prime}$. The planets Mars,
Saturn and Uranus served as primary flux calibrators. The absolute
calibration was found to be accurate to within 20\%. Repeated
observations of H$\,1413+117$ and F$\,10214+4724$ confirmed a relative
flux accuracy of $\sim20\%$. The data were reduced using the CSO
BADRS software package. Typical sensitivities ($\rm 1 \sigma$) of
$\sim20$ mJy were achieved, after $\sim 2500\,$s of on--source
integration time.
Nine sources were detected at levels of $4\,\sigma$ and above, as
outlined in Table~\ref{detections}. Included are the $z\,>\,4$
quasars BR$\,$1202$-$0725, BRI$\,$1335$-$0417 and HM$\,$0000$-$263.
Except for the Cloverleaf (H$\,$1413+117; \cite{Barvainis92}), the
present measurements are the first reported detections for high
redshift quasars at 350$\,\mu$m. Many of the sources were measured
three or more times providing both consistency checks and improvements
in the accuracy of the flux densities. The two strongest sources,
H$\,$1413+117 and IRAS~F10214+4724, were often measured before
starting the long ($\sim$2-3 hours) integrations on the weaker
sources.
As an illustration of the data quality, Figure~1 shows the
350$\,\mu$m CSO--SHARC measurement towards BR$\,$1202$-$0725 at
$z\,=\,4.69$. This measurement corresponds to a total of 4 hours
integration on source. The source is centered at offset zero. The
other channels provide a measure of the neighboring blank sky emission
and a reference for the quality of the detection.
11 sources with redshifts between 1.8 and 4.5 were not detected at
350$\,\mu$m, with flux density upper limits at the 3$\,\sigma$ level of
30$\,$-$\,125\,$mJy. Table~2 lists their names,
redshifts, 350$\,\mu$m flux density measurements with $\pm1\,\sigma$
errors, and a 3$\,\sigma$ upper limit to their luminosities (see below).
\bigskip
\centerline{\epsfig{file=sharc_qso_fig1.eps, width=2.1in,angle=-90}}
\medskip
{\noindent\small Fig. 1 --
The 350$\,\mu$m flux density measured in the
bolometers of the SHARC array towards BR1202$-$0725 at $z$=4.69.
Offsets are given in arcsec with respect to the reference channel
number 10. The source is centered at offset zero. Emission is also
seen in the two neighboring pixels because the bolometers
sample the diffraction pattern of the telescope with a Nyquist sampling.}
\bigskip
\bigskip
\begin{center}
\small
{\sc TABLE 2} \\
{\sc Sources with upper limits at 350$\,\mu$m}
\vskip 1ex
\begin{tabular}{llccc}
\hline
\hline \\[-0.2cm]
\multicolumn{1}{c}{Source} & \multicolumn{1}{c}{$z$}
& \multicolumn{1}{c}{S$_{350}$} & L$_{\rm FIR}$ ($3\,\sigma$) & Ref. \\
\multicolumn{2}{c}{ } & \multicolumn{1}{c}{(mJy, $\pm1\,\sigma$)} &
(10$^{12} h^{-2}_{100}$ L$_{\odot}$) & \\[0.2cm]
\hline \\[-0.2cm]
BR2237$-$0607 & 4.56 & \phs$\phn5\pm15$ & $<6$ & 1 \\
BRI0952$-$0115 & 4.43 & \phs$\phn8\pm22$ & $<8$ & 1,2 \\
PSS0248+1802 & 4.43 & $-75\pm22$ & $<9$ & 3 \\
BR1117$-$1329 & 3.96 & \phs$27\pm13$ & $<4$ & 1 \\
Q0302$-$0019 & 3.28 & \phs$34\pm21$ & $<5$ & 4 \\
Q0636$+$680 & 3.18 & $-123\pm38\phn$ & $<9$ & 5 \\
Q2231$-$0015 & 3.01 & $-65\pm24$ & $<6$ & 4 \\
MG0414$+$0534 & 2.64 & $-24\pm35$ & $<8$ & 6 \\
Q0050$-$2523 & 2.16 & \phs$69\pm42$ & $<9$ & 4 \\
Q0842$+$3431 & 2.13 & \phs$16\pm10$ & $<2$ & 1 \\
Q0838$+$3555 & 1.78 & \phs$39\pm19$ & $<4$ & 1 \\[0.2cm]
\hline \\[-0.2cm]
\end{tabular}
\end{center}
\vspace*{3ex}{\parbox{3.3in}{\hskip 1em \rm\footnotesize References. ---
1. Omont et al. (1996b); 2. Guilloteau et al. (1999);
3. Kennefick et al. (1995); 4. Hewett, Foltz \& Chaffee (1995);
5. Sargent, Steidel \& Boksenberg (1989); 6. Barvainis et al. (1998) }\par}
\smallskip
\normalsize
\section{Discussion}
Figure~\ref{seds} displays the spectral energy distributions of the
six sources detected at 350$\,\mu$m for which fluxes at two or more
other wavelengths are available from the literature. In the following,
we will first comment on individual sources and then discuss the physical
properties of the objects.
The two radio-quiet, $z>4$ quasars, BR$\,$1202$-$0725 and
BRI$\,$1335$-$0417, are exceptional objects with large masses of gas
($\rm \sim 10^{11} \, M_{\odot}$) which have been detected in CO by
Omont et al. (1996a) and Guilloteau et al. (1997). They both are
clearly detected at 350$\,\mu$m with flux densities of $106\pm7$ and
$52\pm8\,$mJy, respectively. However, BRI$\,$0952$-$0115, which is
the third $z>4$ quasar in which CO has been measured
(\cite{Guilloteau99}), is not detected at 350$\,\mu$m at a 3$\,\sigma$
level of 65$\,$mJy. This upper limit is consistent with the weak flux
density of BRI$\,$0952$-$0115 at 1.3~mm, $\rm 2.8 \pm 0.6 \, mJy$
(\cite{Omont96b}; \cite{Guilloteau99}), and a temperature of $\rm \sim
50 \, K$ (see below). The radio-quiet quasar HM$\,$0000$-$263 at
$z=4.11$, which was not detected at $\rm 1.25 \, mm$ using the 30m
telescope (\cite{Omont96b}), due to its low declination, shows a
large flux density at 350$\,\mu$m (134$\pm$29$\,$mJy).
Measurements at other wavelengths would be useful to further constrain
the properties of this object.
The detection of the $z=3.8$ radiogalaxy 4C41.17 with a flux density
of $37\pm9\,$mJy at 350$\,\mu$m is one of the most sensitive
measurements of this study. This sensitivity was reached after only
3/4 of an hour of on-source time and defines the limits which can be
achieved with SHARC in the pointed observing mode under excellent
weather conditions. A marginal detection ($4\,\sigma$) was achieved
at 350$\,\mu$m of PC$\,$2047+0123, a $z=3.80$ quasar studied by
\cite{Ivison95}. Finally, the 350$\,\mu$m flux density of the
Cloverleaf (H$\,$1413+117) is significantly higher than the value
published by Barvainis, Antonucci, \& Coleman (1992), i.e.
293$\pm14\,$mJy as compared to 189$\pm56\,$mJy. We have also obtained
for the Cloverleaf a 450$\,\mu$m flux density of $\rm 226\pm34 \, mJy$
in excellent agreement with the measurement of $\rm 224\pm38\,mJy$ at
$438\,\mu$m of Barvainis, Antonucci, \& Coleman (1992), as shown in
Figure~\ref{seds}.
\setcounter{figure}{1}
\begin{figure*}[tbh]
\centerline{\epsfig{file=sharc_qso_fig2.eps, width=5in,angle=-90}}
\caption[]{Spectral energy distributions of six of the high-redshift objects
discussed in this paper. The 350$\,\mu$m fluxes, shown as squares, are
the measurements made with SHARC at the CSO. Fluxes taken from the
literature are shown as circles - see Tables 1 for references.
The observed fluxes at
$\lambda_{\rm obs}\geq 350\,\mu$m were approximated by grey body
spectra, shown as solid lines, assuming an emissivity index $\beta =
1.5$. The dust temperature results from a non-linear least-squares
fit to the data - see text.
\label{seds}}
\end{figure*}
A greybody was fit to the data points $\rm S_\nu$, for wavelengths of
350$\,\mu$m and longward, as a function of the rest frequency
$\nu=\nu_{\rm obs}/(1+z)$, of the form
\begin{equation}
\rm S_\nu\,=\,B_\nu \Omega [1-\exp(-\tau)]
{\rm ~with~ }\tau\,=\,(\nu/\nu_0)^\beta,
\end{equation}
where $\nu_0 = 2.4 \, {\rm THz}$ is the critical frequency at which
the source becomes optically thin and $\Omega$ is the solid angle of
emission. The shape of the fitted greybody is very weakly dependent on
the value of $\nu_0$ (\cite{Hughes}). The data were each weighted by
their statistical errors in the $\chi^2$ minimization. This yields
the dust temperature, dust mass (following Hildebrand (1983), using a
dust mass emission coefficient at $\nu_0$ of $\rm 1.9~m^2~kg^{-1}$),
and luminosity of the sources. When $\beta$ is considered as a free
parameter, we find that the average value of $\beta=1.5\pm0.2$ for the
detected sources. The fits shown in Figure~\ref{seds} assume an
emissivity index of $\beta = 1.5$. We estimated the $1\,\sigma$
uncertainty in the temperature by examining the $\chi_\nu^2$
hypersurface in the range $1\leq \beta\leq 2$, similarly to the method
of Hughes et al. (1993). To evaluate the uncertainties associated with
the mass and luminosity, derived from the fitted temperature and
$\beta=1.5$, we used the maximum and minimum values of the mass and
luminosity which are compatible with the data plus or minus the
statistical error. No lensing amplification was taken into account.
The temperature, dust mass and luminosities derived under these
assumptions are given in Table~\ref{detections}. Two of the sources
with upper limits have 1.25mm detections (\cite{Omont96b}), which,
together with the 350$\,\mu$m data, yields an upper limit to their
temperature. For Q$\,0842+3431$, we find that $\rm T_{dust}<40$K
while for BR$\,1117-1329$ a limit of $\rm T_{dust}<60K$ is found. If
the dust is at the temperature limit, these quasars have dust masses
$< 10^8 M_\odot$. For the other sources, an estimate of the maximum
luminosity has been given under the assumption that each object has a
temperature of 50$\,$K and an emissivity index of $\beta=1.5$.
The total luminosity is probably underestimated, since a large
luminosity contribution from higher temperature dust cannot be ruled
out for most sources. However, in the case of H$\,$1413+117 and
IRAS~F10214+4724, the available IRAS data allow us to fit an
additional warm component. For IRAS~F10214+4724, the cold component
model carries roughly 60\% of the total luminosity; in the case of
H$\,$1413+117, which has a hotter mid-IR spectrum, the total
luminosity is underestimated by a factor of 3. Under the assumption
that the majority of the luminosity is carried by the cold component
(Table~\ref{detections}), the median luminosity-to-mass ratio is
around 100 $L_\odot/M_\odot$, assuming a gas-to-dust ratio of $\sim
500$ similar to that of IRAS~F10214+4724 and H$\,$1413+117
(\cite{Downes92}; \cite{Barvainis95}) or of ultraluminous infrared
galaxies (ULIRGs), i.e. $\rm 540\pm290$ (\cite{Sanders91}). The peak
emission in the rest frame is found to be in the wavelength range $\rm
\lambda_{peak} \sim 60-80 \,\mu$m (Figure~\ref{seds}) implying dust
temperatures of 40-60$\,$K (Table~\ref{detections}). These
temperatures are nearly a factor of two lower than previously
estimated for ultraluminous sources (e.g. \cite{Chini94}). If the
temperature range we find is typical for the cold component of highly
redshifted objects, multi-band photometric studies in the
submillimeter/FIR, such as planned with FIRST, will provide reasonably
accurate redshift estimates for the sources detected in deep field
surveys.
The global star formation rate in each of the detected sources can be
estimated using the relation of Thronson \& Telesco (1986): ${\rm SFR}
\sim \Psi \, 10^{-10} L_{\rm FIR}/L_{\odot} h_{100}^{-2}\, M_{\odot}
\rm ~yr^{-1}$ with $\rm \Psi \sim\,$0.8-2.1. For our mean luminosity
of $1.7\times10^{13}\,h_{100}^{-2} \,L_\odot$, this yields a SFR of
$\sim 2000 \, M_{\odot} h_{100}^{-2} \rm ~yr^{-1}$ (uncorrected for
lensing) if all the submillimeter flux is from a starburst component.
If we assume a final stellar mass of $\sim2\times10^{12} M_\odot$, a
value appropriate to a giant elliptical like M87 (\cite{Okazaki84}),
then the timescale for formation in a single massive starburst is
$\simeq 10^9h_{100}^2\,$yr. Given the large mass of dust already
present in these quasars, a substantial amount of this star formation
must already have occurred. For the most distant quasars, where the
age of the universe is similar to the derived formation timescale,
this implies a very high redshift ($z\gtrsim5$) for the era of initial
star formation, in agreement with models of high redshift Lyman-alpha
emitters (\cite{Haiman98}).
\begin{center} {\it Acknowledgments} \end{center}
The CSO is funded by the NSF under contract AST96-15025. We thank
T.R. Hunter for help with the fitting/derivation programming and D.
Downes for helpful comments. One of us (P.C.) acknowledges
financial supports from INSU (Programmes Grands T\'elescopes
Etrangers) and PCMI.
|
\section{Introduction}
Since the discovery of giant magnetoresistance (GMR) phenomenon in
magnetic multilayers \cite{Grunberg} there has been a great deal of
interest in studying them by both theoretical and experimental
methods. The reason for the interest is, already partially realized,
possibility of practical applications as magnetic sensors, recording
heads and magnetic memory elements.
The standard system exhibiting the GMR is a trilayer (\emph{i.e.} two
magnetic layers separated by a non-magnetic spacer) with thickness of
the spacer chosen so as to produce antiferromagnetic coupling between
magnetic layers. While such a system, due to its simplicity, is
convenient for theoretical treatment, it presents some problems in
practical applications. The main problem is a high switching field
which is usually necessary to rotate the magnetizations (overcoming
antiferromagnetic coupling) and produce GMR. One way to deal with this
problem is to use somewhat more complex \emph{spin-engineered}
structures. Widely known structures of this type are spin-valve
systems \cite{Dieny}, in which one of the magnetic moments is fixed by
the strong exchange coupling due to an additional antiferromagnetic
layer (\emph{e.g.} MnFe or CoO). There exists however also a different
approach in which a system composed of three magnetic layers is used.
Two of them are strongly antiferromagnetically coupled, forming the so
called artificial antiferromagnetic subsystem (AAF) \cite{Berg}, and
the third one --- detection layer --- is only weakly coupled (or just
decoupled). Such a setup was proposed both for laboratory measurements
\cite{Parkin-Mauri,Bloeman} and more practically as angular velocity
meter \cite{Berg}. A similar system (superlattice with strong and weak
exchange couplings) was also studied theoretically (on \emph{ab
initio} level) \cite{SKFSUK}, the thicknesses involved were however
small due to numerical limitations.
The aim of the present paper is to perform thorough studies of
transport and magnetic properties of the systems in question and to
relate them with corresponding magnetic structure phase diagrams.
\section{The model and the method of calculations}
We consider a system consisting of three magnetic layers separated by
two non-magnetic spacers, \emph{i.e.} the structure of the \(
F_{1}/S_{1}/F_{2}/S_{2}/F_{3} \) type, where \( F_{i} \) stands for
ferro- and \( S_{i} \) for paramagnetic layer. In order to describe
collinear configurations we employ tight-binding hamiltonian with two,
hybridized bands and spin-dependent on-site potentials (see
Ref.~\cite{SKUK} for details). We restrict ourselves for simplicity to
the case of simple cubic structure. The following values of model
parameters have been chosen: \( E_{F}=0 \), \( t_{s}=-1 \), \(
t_{d}=-0.2 \), \( V_{sd}=1 \), \( \epsilon ^{s}_{i\sigma }=0 \), \(
\epsilon ^{d}_{i\uparrow }=-1 \) for all the layers, \( \epsilon
^{d}_{i\downarrow }=-0.2 \) within magnetic layers and \( -1 \)
elsewhere, where \( E_{F} \) is the Fermi energy, \( t_{\alpha } \)
(with \( \alpha \) being the band index --- \( s,\, d \)) are the
hopping integrals, \( V_{sd} \) is the \( s-d \) intra-atomic
hybridization and \( \epsilon ^{\alpha }_{i\sigma } \) are the on-site
potentials (where \( \sigma \) is \( \uparrow \) for majority- and \(
\downarrow \) for minority-spin carriers). The above set of parameters
enables us to mimic essential features of the electronic structure of
Co/Cu multilayers (\emph{i.e.} the spin-polarized density of states is
qualitatively reproduced --- in particular the majority \emph{d}-bands
in magnetic and \emph{}(non-polarized) \emph{d}-bands in paramagnetic
layers are matched perfectly which closely resembles the situation in
Co/Cu systems).
The conductance is computed from the Kubo formula with the help of
recursion Green function technique \cite{SKUK,Asano}. The only
difference in comparison with \cite{SKUK} is that hybridization in
lead wires (attached to the multilayer from both sides, for transport
calculations) has been taken into account this time. The GMR has been
defined as
\begin{equation}
\label{GMR}
\mathrm{GMR}=\frac{\Gamma ^{\uparrow \downarrow ;\downarrow }}{\Gamma
^{\uparrow \downarrow ;\uparrow }}-1,
\end{equation}
where the arrows show the orientations of magnetic moments. Note that
without the first magnetic layer this definition would be identical
with the usual one. Additionally the thermopower (TEP) has also been
calculated from the following formula (see eg. Ref.~\cite{Mott})
\begin{equation}
\label{onsanger}
S=-\frac{\pi ^{2}k^{2}_{B}T}{3\left| e\right| }\frac{d}{dE}\,
\mathrm{ln}\, \Gamma \left( E\right) .
\end{equation}
We define, as in Ref.~\cite{SKUK_thp}, the ``giant
magneto-TEP-effect'' GMTEP analogously to Eq.~(\ref{GMR}) (with \(
\Gamma \) replaced by \( S \)).
For studying the magnetization processes we employ the
phenomenological expression, not unlike the one introduced in
Ref.~\cite{Schmidt}, (\( \Theta _{i} \) is an angle between the
\emph{i}-th magnetic moment and the external field, B)
\begin{eqnarray}
E\left( \Theta _{1},\Theta _{2},\Theta _{3}\right) & = & -J_{12}\cos
\left( \Theta _{1}-\Theta _{2}\right) -J_{23}\cos \left( \Theta
_{2}-\Theta _{3}\right) -J_{13}\cos \left( \Theta _{1}-\Theta
_{3}\right) \nonumber \\
& & -B\sum ^{3}_{i=1}t_{i}\cos \left( \Theta _{i}\right) /t+\sum
^{3}_{i=1}E_{A}(\Theta _{i}),\label{phen_energy}
\end{eqnarray}
where the first three terms describe bilinear exchange coupling
between magnetic layers, the next three are Zeeman energy terms (\(
t_{i} \) being \( i \)-th layer thickness and \( t \) the overall
thickness of all the magnetic layers) and \( E_{A}(\Theta _{i}) \) is
the crystalline anisotropy, that is \( t_{i}K\sin ^{2}(2\Theta
_{i})/4t \) and \( -t_{i}D\cos ^{2}\Theta _{i}/t \) for cubic and
uniaxial case respectively. We assume that external magnetic field is
applied along the {[}10{]} in-plane crystallographic axis, which can
be either easy or hard axis depending on the sign of the anisotropy
constants. The magnetic moments are confined to the layers plane which
corresponds to the strong shape anisotropy. Expression
(\ref{phen_energy}) was then numerically minimized, with respect to
the \( \Theta _{i} \)-s, by taking, starting from initial
configuration, little steps in the direction opposite to the energy
gradient. All the extremal points found in this way were additionally
checked against the stability condition (\emph{i.e.} the positivity of
all the minors of \( M_{ij}=\partial ^{2}E/(\partial \Theta
_{i}\partial \Theta _{j}) \)) in order to eliminate saddle points.
Using Eq.~(\ref{phen_energy}) one can write (for \( B \) equal to 0)
\begin{eqnarray}
J_{12} & = & \frac{1}{4}\left[ E\left( \pi 00\right) +E\left( 0\pi
0\right) -E\left( 000\right) -E\left( 00\pi \right) \right]
,\nonumber \\
J_{23} & = & \frac{1}{4}\left[ E(00\pi )+E\left( 0\pi 0\right)
-E\left( 000\right) -E\left( \pi 00\right) \right] ,\label{joty}
\\
J_{13} & = & \frac{1}{4}\left[ E\left( 00\pi \right) +E\left( \pi
00\right) -E\left( 000\right) -E\left( 0\pi 0\right) \right]
.\nonumber
\end{eqnarray}
Therefore, having known the energies of collinear configurations from
the model calculations (based on the two-band tight-binding
hamiltonian), we are able to determine the exchange coupling
constants.
From now on we will be using reduced values of the magnetic field \(
b=B/|J_{12}| \) and the anisotropy constants \( k=K/|J_{12}| \) and \(
d=D/|J_{12}| \). We will also assume, if not stated explicitly
otherwise, the magnetic layers thicknesses to be 8, 3 and 3~ML
(monolayers), respectively, in order to keep the length ratios as in
Ref.~\cite{Bloeman}. The first spacer thickness will be set to 3~ML in
order to achieve the needed antiferromagnetic coupling between the
first two magnetic layers.
\section{Results}
Figure~\ref{exch_pl} presents the exchange coupling constants (\( J
\)-s) plotted against the thickness of the second spacer (\( ns_{2}
\)). As mentioned above, for the chosen thicknesses we got strong
antiferromagnetic coupling (\( J_{12} \)) between the first two
magnetic layers, while \( J_{23} \) and \( J_{13} \) oscillate around
zero. In all three cases the period of oscillations is about 3~ML,
which is close to the theoretically predicted value (2.8~ML) coming
from the stationary spanning vector \cite{Bruno,SKMZUK} placed at the
\( (0,\pi ) \) (and equivalent positions) in the two-dimensional
Brillouin zone. The second period (8~ML) originating from the hole
pocket placed at \( (\pi ,\pi ) \) does not seem to appear in the
present context. Note however that it can become visible under some
circumstances like in Ref.~\cite{SKUK} where the tunnelling
conductance has been considered. The GMR and \( J_{23} \) for the same
system have been plotted in Fig.~\ref{gmr_pl}. It can be noted that
GMR asymptotically tends to oscillate with the same period but in
opposite phase to \( J_{23} \). This is in agreement with our
previous findings \cite{SKMZUK}, but it is still not clear to what
extent this correlation is universal. The values of GMR are strongly
reduced in comparison with the trilayer case. This can be easily
understood if we note that, due to the fixed antiferromagnetic
alignment of the first two magnetic layers, there is no non-scattering
channel in any of the configurations involved (see Eq.~\ref{GMR}).
Figure~\ref{d-band_pl} where the on-site potentials for the
\emph{d}-bands (\( \epsilon ^{d}_{i\sigma } \)) have been
schematically plotted, shows that there exist at least two scattering
interfaces in each case. Basing on the number of interfaces one can
qualitatively predict that the \( \uparrow \downarrow ;\downarrow \)
down- and \( \uparrow \downarrow ;\uparrow \) up-spin electron
channels have the higher conductances than the remaining two. This is
indeed clearly visible in Fig.~\ref{gamma_pl} where the computed
conductances are shown. As already discussed there is no obvious
highest conductance channel. Instead, we have two higher and two lower
conducting channels close to each other within the pairs. As a
consequence the sign of GMR is determined by all the channels, and can
be changed by manipulating some parameters (\emph{eg.} thicknesses of
the layers). This is the case in Fig.~\ref{gmr1_pl} where we have
plotted the GMR and \( J_{23} \) for a system with thickness of the
second magnetic layer set to 5~ML (instead of 3~ML as in
Fig.~\ref{gmr_pl}). The GMR oscillations have a small but clearly
negative bias. The asymptotic opposite-phase correlation with respect
to \( J_{23} \) is again clear in this case.
The GMTEP, calculated for the same set of parameters as in
Fig.~\ref{gmr_pl}, has been plotted in Fig.~\ref{gmtep_pl}. In
agreement with the findings of Ref.~\cite{SKUK_thp}, the oscillations
are quite pronounced and have the same period as GMR but exhibit
negative bias. Asymptotically they seem to have roughly the same phase
as GMR.
For studying the magnetization processes we have chosen the parameters
as in Fig.~\ref{exch_pl} with the second spacer thickness (\( ns_{2}
\)) set to 7~ML (the second ferromagnetic maximum of \( J_{23} \)). As
already stated, two cases have been taken into account, \emph{i.e.}
the cubic and uniaxial anisotropies.
In the first case the anisotropy term (\( t_{i}k\sin ^{2}(2\Theta
_{i})/4t \)) gives rise to four potential wells placed at the
following in-plane crystallographic axes~: {[}10{]}, {[}01{]},
{[}\=10{]} and {[}0\=1{]} for \( k>0 \) and {[}11{]}, {[}1\=1{]},
{[}\=1\=1{]} and {[}\=11{]} for \( k<0. \)
Figure~\ref{phase-cubic_pl}a exemplifies some phase diagrams for
various initial configurations. We have chosen the simplest ones
\emph{i.e.} the collinear configurations with relative alignment of
magnetic moments favoured by interlayer exchange coupling. It is
however, in principal, possible to stabilize also non-collinear ones,
provided that the anisotropy is strong enough. The phase diagrams have
been obtained by taking subsequent scans along \( b \) for different
\( k \) values. Dotted horizontal lines are thus only guides to the
eye. The diagrams exhibit rich structure with a number of
configurations (phases) occurring during the magnetization process
(including the non-collinear configurations). The transition between
them can be either of first or second type, that is they manifest
themselves as discontinuities in magnetization or its first
derivative. Some exemplary hysteresis loops are presented in
Fig.~\ref{phase-cubic_pl}b. As expected stronger anisotropy produces a
richer structure. For positive, and sufficiently big, values of \( k
\) some flat regions, typical for exchange-biased systems, occur.
Note however that there is no \( \uparrow \downarrow ;\downarrow \) to
\( \uparrow \downarrow ;\uparrow \) transition in the first diagram of
Fig.~\ref{phase-cubic_pl}a (but see below).
For the case of uniaxial anisotropy (\( -t_{i}d\cos ^{2}(\Theta
_{i})/t \)) there exist only two potential wells, \emph{}that is
{[}10{]} and {[}\=10{]} for \( d>0 \) and {[}0\=1{]}, {[}01{]} for \(
d<0 \). The phase diagrams (Fig.~\ref{phase-uni_pl}a) are somewhat
less complicated now, due to the simpler energy landscape. The
saturation curve for the upper part of the first diagram (with the \(
\uparrow \downarrow ;\downarrow \) initial configuration) has been
obtained on analytical basis (\emph{i.e.} from the stability
condition) because of the weak stability of the \( \uparrow \downarrow
;\uparrow \) configuration (by which we mean that the existing energy
minimum is very shallow) which makes it difficult to perform reliable
numerical minimalization. The same precautions have been applied to
the first hysteresis loop in Fig.~\ref{phase-uni_pl}b. Note that this
time the flat regions are present already for small values of \( d \).
For \( d>0.51 \) there exists, in the first diagram, the above
mentioned \( \uparrow \downarrow ;\downarrow \) to \( \uparrow
\downarrow ;\uparrow \) transition. Only positive values of \( d \)
were presented since the opposite case is trivial --- there is
practically no hysteresis.
\section{Conclusions}
Within the microscopic two-band tight-binding model we have performed
the calculations of interlayer exchange coupling,
current-perpendicular-to-plane conductance and thermopower for a
system consisting of three magnetic layers, separated by paramagnetic
ones. With the thicknesses chosen so as to produce strong
antiferromagnetic coupling between the first two magnetic layers, we
found both the interlayer exchange coupling, GMR and GMTEP to
oscillate, as a function of the second spacer thickness, with the
period originating from one of the extremal spanning vectors of the
Fermi surface. Additionally, using the phenomenological approach, we
have computed magnetic phase diagrams and commented on their relevance
to the magnetoresistance. We found that the phase diagrams exhibit
rich structure and there are flat regions in hysteresis loops, typical
for exchange-biased spin-valves. It has been also found that in the
case of the systems under consideration, in contrast to conventional
trilayers, it is possible to obtain a negative (inverse) perpendicular
GMR by merely changing thicknesses of particular layers.
\section{Acknowledgments}
The KBN grants 2PO3B-118-14 and 2PO3B-117-14 are gratefully
acknowledged. We also thank Pozna\'n Computing and Networking Center
for the computing time.
|
\section{Introduction} \label{intro}
\setcounter{equation}{0}
The f\/ield equations of the general relativity theory, which
in the usual notation are written in the form
\begin{equation}
G_{\alpha \beta} = \kappa \,\; T_{\alpha \beta}\;, \label{ein}
\end{equation}
relate the geometry of the spacetime to its source. The general
relativity theory, however, does not prescribe
the various forms of matter, and takes over the energy-momentum
tensor $\,T_{\alpha \beta}\,$ from other branches of physics.
In this sense, general relativity (GR) is not a closed theory.
The separation between the gravitational f\/ield and its source
has been often considered as one undesirable feature of GR%
~\cite{Einstein56}~--~\cite{Salam80}.
Recently, Wesson and co-workers~\cite{Wesson90,WessonLeon92a}
have introduced a new approach to GR, in which
the matter and its role in the determination of the spacetime
geometry is given from a purely f\/ive-dimensional geometrical
point of view. In their f\/ive-dimensional (5D) version of general
relativity the f\/ield equations are given by
\begin{equation} \label{5DfeqsG}
\widehat{G}_{AB} = 0 \;.
\end{equation}
Henceforth, the f\/ive-dimensional geometrical objects are denoted by
overhats and Latin letters are 5D indices and run from $0$ to $4$.
In this new approach to GR the 5D vacuum f\/ield equations%
~(\ref{5DfeqsG}) give rise to both curvature and matter in 4D.
Indeed, it can be shown~\cite{WessonLeon92a} that it is always
possible to rewrite the f\/ifteen f\/ield equations~(\ref{5DfeqsG})
as a set of equations such that ten of which are precisely Einstein's
f\/ield equations~(\ref{ein}) in 4D with an {\em induced\/}
energy-momentum
\begin{eqnarray} \label{Tinduced}
\kappa \; T_{\alpha\beta} & = & \frac{\phi_{\alpha\,;\:\beta}}{\phi} -
\frac{\varepsilon}{2\,\phi^2}
\left\{ \frac{\phi^{*} \, g^{*}_{\alpha\beta}}{\phi}
- g^{**}_{\alpha\beta} + g^{\gamma\delta} \,
g^{*}_{\alpha\gamma} \, g^{*}_{\beta\delta}
- \frac{g^{\gamma\delta} \,
g^{*}_{\gamma\delta} \, g^{*}_{\alpha\beta}}{2}
\right. \nonumber \\
& &
+ \; \left.
\frac{g^{}_{\alpha\beta}}{4^{} {}_{} } \, \left[\,g^{*}{}^{\gamma\delta} \,
g^{*}_{\gamma\delta} + (g^{\gamma\delta_{}} \,g^{*}_{\gamma\delta})^2
\, \right] \,\right\} \;,
\end{eqnarray}
where the Greek letters denote 4D indices and run from $0$ to $3$,
$g_{44} \equiv \varepsilon\, \phi^2$ with $\varepsilon=\pm 1 $,
$\phi_\alpha \equiv \partial \phi / \partial x^\alpha$, a star
denotes $\partial / \partial x^4$, and a semicolon denotes the usual
4D covariant derivative. Obviously,
the remaining f\/ive equations (a wave equation and four conservation
laws) are automatically satisf\/ied by any solution of the 5D
vacuum equations~(\ref{5DfeqsG}).
Thus, not only the matter but also its role in the
determination of the geometry of the 4D spacetime can be considered
to have a f\/ive-dimensional geometrical origin. This approach
unif\/ies the gravitational f\/ield with its source (not just
with a particular f\/ield) within a purely 5D geometrical framework.
This 5D version of general relativity is often referred to as
induced matter gravity theory (IM gravity theory, for short).
The IM theory has become a focus of a recent research f\/ield%
~\cite{Overduin97}.
The basic features of the theory have been explored by Wesson and
others~\cite{Leon93}~--~\cite{Wesson96a}, whereas the implications
for cosmology and astrophysics have been investigated by a number
of researchers~\cite{Wesson92b}~--~\cite{Liu96b}.
For a fairly updated list of references on IM gravity theory
and related issues we refer the reader to Overduin and
Wesson~\cite{Overduin97}.
In general relativity, the causal structure of 4D spacetime
has locally the same qualitative nature as the f\/lat spacetime
of special relativity --- causality holds locally. The global
question, however, is left open and signif\/icant dif\/ferences can
occur. On large scale, the violation of causality is not excluded.
Actually, it has long been known that there are solutions to
the Einstein f\/ield equations which possess causal anomalies in
the form of closed timelike curves. The famous solution found
by G\"odel~\cite{Godel49} in 1949 might not be the f\/irst but
it certainly is the best known example of a cosmological
model which makes it apparent that general relativity,
as it is normally formulated, does not exclude the existence of
closed timelike world lines, despite its Lorentzian character
which leads to the local validity of the causality principle.
Owing to its striking properties G\"odel's model
has a well-recognized importance and has to a certain extent
motivated the investigations on rotating cosmological
G\"odel-type models and on causal anomalies in the
framework of general relativity~\cite{Som68}~--~\cite{Krasinski98}
and other theories of gravitation~\cite{Vaidya84}~--%
~\cite{FonsecaReboucas98}.
Two recent articles have been concerned with {\em five-dimensional\/}
G\"odel-type spacetimes. Firstly in Ref.~\cite{ReboucasTeixeira98a}
the main geometrical properties of f\/ive-dimensional
Riemannian manifolds endowed with a 5D counterpart of the 4D
G\"odel-type metric of general relativity were investigated.
Among several results, an irreducible set of isometrically
nonequivalent 5D homogeneous (locally) G\"odel-type
metrics were exhibited. Therein it was also shown that,
apart from the degenerated G\"odel-type metric, in all
classes of homogeneous G\"odel-type geometries there is
breakdown of causality. As no use of any particular f\/ield
equations was made in this f\/irst paper, its results hold for
any 5D G\"odel-type manifolds regardless of the underlying
5D Kaluza-Klein gravity theory. In the second article%
~\cite{ReboucasTeixeira98b} the classes of 5D G\"odel-type
spacetimes discussed in~\cite{ReboucasTeixeira98a}
were investigated from a more physical viewpoint.
Particularly, it was examined the question as to
whether the induced matter theory of gravitation permits the
family of noncausal solutions of G\"odel-type metrics studied
in~\cite{ReboucasTeixeira98a}. It was shown that the
IM gravity excludes this class of 5D G\"odel-type
non-causal geometries as solution to its f\/ield equations.
In both articles~\cite{ReboucasTeixeira98a,ReboucasTeixeira98b}
the 5D G\"odel-type family of metrics discussed is the simplest
5D class of geometries for which the section $u = \mbox{const}$
($u$ is the extra coordinate) is the 4D G\"odel-type metric of
general relativity. Actually the 5D G\"odel-type line element
of both papers does not depend on the f\/ifth coordinate
$u$, and therefore as regards to the IM theory a radiation-like
equation of state is an underlying assumption of both articles.
However, it is well know~\cite{Overduin97} that the dependence
of the 5D metric on the extra coordinate is necessary to ensure
that the 5D IM theory permits the induction of matter of a very
general type in 4D.
In this work, on the one hand, we shall examine the main
geometrical properties of a class of {\em generalized\/}
G\"odel-type geometries in which the 5D metric depends on the
f\/ifth coordinate, generalizing therefore the results found in
Ref.\ \cite{ReboucasTeixeira98a}. On the other hand,
we shall also investigate the question as to whether
the induced matter gravity theory, as formulated
by Wesson and co-workers~\cite{Wesson90,WessonLeon92a},
admits these generalized G\"odel-type metrics as solutions
to its f\/ield equations, thus also extending the
investigations of Ref.\ \cite{ReboucasTeixeira98b}.
The outline of this article is as follows.
In the next section we present a summary of some important
prerequisites for Section~3, where using the equivalence
problem techniques as formulated by Cartan~\cite{Cartan}
we derive the necessary and suf\/f\/icient conditions for local
homogeneity of this class of 5D generalized
G\"odel-type manifolds. In Section~3 we also exhibit an
irreducible set of isometrically nonequivalent homogeneous
generalized G\"odel-type metrics. In Section~4 we discuss
the integration of the Killing equations and present the Killing
vector f\/ields as well as the corresponding Lie algebra for
all homogeneous generalized G\"odel-type metrics.
In the last section we examine whether the IM f\/ield
equations permit solutions of this generalized
G\"odel-type class of geometries. The unique solution of
this type is found therein.
The question as to whether the IM version of general
relativity rules out the existence of closed
timelike curves of G\"odel type is also discussed
(Section~5) in connection with a recent paper by
Romero {\em et al.\/}~\cite{Romero96}.
\vspace{3mm}
\section{Prerequisites}
\label{prereq}
\setcounter{equation}{0}
The arbitrariness in the choice of coordinates in the metric
theories of gravitation gives rise to the problem of deciding
whether or not two manifolds whose metrics $g$ and $\tilde{g}$
are given explicitly in terms of coordinates, viz.,
\begin{equation}
ds^2 = g_{\mu \nu} \,dx^\mu \, dx^\nu \qquad \;
\mbox{and} \qquad \;
d\tilde{s}^2 = \tilde{g}_{\mu \nu}\,d\tilde{x}^\mu\,d\tilde{x}^\nu\:,
\end{equation}
are locally isometric. This is the so-called equivalence problem (see
Cartan~\cite{Cartan} for the local equivalence of $n$-dimensional
Riemannian manifolds, Karlhede~\cite{Karlhede80} and
MacCallum~\cite{MacCallumSkea94} for the special case $n=4$ of
general relativity).
\begin{sloppypar}
The Cartan solution~\cite{Cartan} to the equivalence problem
for Riemannian manifolds can be summarized as follows. Two
$n$-dimensional Lorentzian Riemannian manifolds ${\cal M}_n$ and
$\widetilde{\cal M}_n$ are locally equivalent if there exist coordinate
and generalized $n$-dimensional Lorentz transformations such that
the following {\em algebraic} equations relating the frame components
of the curvature tensor and their covariant derivatives:
\end{sloppypar}
\parbox{14cm}{\begin{eqnarray*}
R^{A}_{\ BCD} & = & \widetilde{R}^{A}_{\ BCD}\:, \nonumber \\
R^{A}_{\ BCD;M_1} & = & \widetilde{R}^{A}_{\ BCD;M_1}\:, \nonumber \\
R^{A}_{\ BCD;M_1 M_2} &=&\widetilde{R}^{A}_{\ BCD;M_1 M_2}\:,\nonumber \\
& \vdots & \nonumber \\
R^{A}_{\ BCD;M_1\ldots M_{p+1}} & = & \widetilde{R}^{A}_{\ BCD;M_1
\ldots M_{p+1}} \nonumber \\
\end{eqnarray*}} \hfill
\parbox{1cm}{\begin{eqnarray} \label{eqvcond} \end{eqnarray}}
are compatible as {\em algebraic} equations in
$\left( x^{\mu}, \xi^{A} \right)$. Here and in what follows we use
a semicolon to denote covariant derivatives. Note that $x^{\mu}$
are coordinates on the manifold ${\cal M}_n$ while $ \xi^{A}$
parametrize the group of allowed frame transformations
[$n$-dimensional generalized Lorentz group usually denoted%
~\cite{HawkingEllis73} by $O(n-1, 1)\,$]. Reciprocally, equations
(\ref{eqvcond}) imply local equivalence between the $n$-dimensional
manifolds ${\cal M}_n$ and $\widetilde{\cal M}_n$.
In practice, a f\/ixed frame is chosen to perform the calculations so
that only coordinates appear in the components of the curvature tensor,
i.e. there is no explicit dependence on the parameters $\xi^{A}$ of the
generalized Lorentz group.
Another important practical point to be considered, once one wishes
to test the local equivalence of two Riemannian manifolds, is that
before attempting to solve eqs.\ (\ref{eqvcond}) one can extract and
compare partial pieces of information at each step of dif\/ferentiation
as, for example, the number $\{t_{0},t_1, \dots ,t_{p}\}$ of
functionally independent functions of the coordinates $x^\mu$
contained in the corresponding set
\begin{equation} \label{CartanScl}
I_{p} = \{ R^{A}_{\ BCD} \,, \,R^{A}_{\ BCD;M_{1}} \,, \,
R^{A}_{\ BCD;M_1 M_2}\,,\,\ldots,\,R^{A}_{\ BCD;M_1 M_2\ldots M_{p}}\}\,,
\end{equation}
and the isotropy subgroup $\{H_{0}, H_1, \ldots ,H_{p}\}$ of the
symmetry group $G_r$ under which the set corresponding $I_p$ is
invariant. They must be the same for each step $q= 0, 1, \cdots ,p$
if the manifolds are locally equivalent.
In practice it is also important to note that in calculating the
curvature and its covariant derivatives, in a chosen frame, one
can stop as soon as one reaches a step at which the $p^{th}$
derivatives (say) are algebraically expressible in terms of the
previous ones, and the residual isotropy group (residual frame
freedom) at that step is the same isotropy group of the previous
step, i.e. $H_p = H_{(p-1)}$. In this case further dif\/ferentiation
will not yield any new piece of information.
Actually, if $H_p = H_{(p-1)}$ and, in a given frame, the $p^{th}$
derivative is expressible in terms of its predecessors, for any
$q > p$ the $q^{th}$ derivatives can all be expressed in terms
of the $0^{th}$, $1^{st}$, $\cdots$, $(p-1)^{th}$ derivatives%
~\cite{Cartan,MacCallumSkea94}.
Since there are $t_p$ essential coordinates, in 5D
clearly $5-t_p$ are ignorable, so the isotropy group will
have dimension $s = \mbox{dim}\,( H_p )$, and the group of
isometries of the metric will have dimension $r$ given by
(see Cartan~\cite{Cartan})
\begin{equation}
r = s + 5 - t_p \:, \label{gdim}
\end{equation}
acting on an orbit with dimension
\begin{equation}
d = r - s = 5 - t_p \:. \label{ddim}
\end{equation}
\vspace{3mm}
\section{Homogeneity and Nonequivalent Metrics}
\label{homoconds}
\setcounter{equation}{0}
The line element of the f\/ive-dimensional {\em generalized\/}
G\"odel-type manifolds ${\cal M}_5$ we are concerned with
is given by
\begin{equation} \label{ds2a}
d\hat{s}^{2} = dt^2 + 2\,H(x)\, dt\,dy - dx^2 - G(x)\,dy^2 -
\widetilde{F}^2(\tilde{u})\,(d\tilde{z}^2 + d\tilde{u}^2) \:,
\end{equation}
where $H(x)$, $G(x)$ and $\widetilde{F}(\tilde{u})$ are arbitrary
real functions.
By a suitable choice of coordinates the line element%
~(\ref{ds2a}) can be brought into the form
\begin{equation} \label{ds2}
d\hat{s}^{2} = [\,dt + H(x)\,dy\,]^2 - dx^2 - D^2(x)\,dy^2 -
F^2(u)\,\,dz^2 - du^2 \:,
\end{equation}
where $D^2(x) = G + H^2$ and $u$ clearly is a new f\/ifth coordinate.
At an arbitrary point of ${\cal M}_5$ one can choose the following set
of linearly independent one-forms $\widehat{\Theta}^A$:
\begin{equation} \label{lorpen}
\widehat{\Theta}^{0} = dt + H(x)\,dy\:, \: \quad
\widehat{\Theta}^{1} = dx\:, \: \quad
\widehat{\Theta}^{2} = D(x)\,dy\:, \:\quad
\widehat{\Theta}^{3} = F(u)\, dz \:, \: \quad
\widehat{\Theta}^{4} = du \:,
\end{equation}
such that the G\"odel-type line element (\ref{ds2}) can be
written as
\begin{equation} \label{ds2f}
d\hat{s}^2 = \widehat{\eta}^{}_{AB} \: \widehat{\Theta}^A \,\,
\widehat{\Theta}^B =
(\widehat{\Theta}^0)^2 - (\widehat{\Theta}^1)^2 -
(\widehat{\Theta}^2)^2 - (\widehat{\Theta}^3)^2 -
(\widehat{\Theta}^4)^2\:.
\end{equation}
Here and in what follows capital letters
are 5D Lorentz frame indices and run from 0 to 4; they are raised
and lowered with Lorentz matrices
$\widehat{\eta}^{AB} = \widehat{\eta}^{}_{AB} =
\mbox{diag} (+1, -1, -1, -1, -1)$,
respectively.
Using as input the one-forms (\ref{lorpen}) and the Lorentz frame
(\ref{ds2f}), the computer algebra package {\sc classi}%
~\cite{MacCallumSkea94,Aman87}, e.g., gives the following
nonvanishing Lorentz frame components $\widehat{R}_{ABCD}$ of
the curvature:
\begin{eqnarray}
\widehat{R}_{0101} &=& \widehat{R}_{0202}=- \frac{1}{4} \,
\left( \frac{H'}{D}\, \right)^2\:, \label{rie1st} \\
\widehat{R}_{0112} & =& \frac{1}{2}\, \left(\frac{H'}{D}\,\right)' \:,
\label{rie2nd} \\
\widehat{R}_{1212} &=& \frac{D''}{D}-\frac{3}{4}\,
\left( \frac{H'}{D}\,\right)^2 \label{rie3rd}\:, \\
\widehat{R}_{3434} &=& \frac{\ddot{F}}{F} \label{rielast} \;\,,
\end{eqnarray}
where the prime and the dot denote, respectively, derivative
with respect to $x$ and $u$.
For 5D (local) homogeneity from eq.~(\ref{ddim}) one must have
$t_q=0$ for $q=0, 1, \cdots\, p$, that is, the number of
functionally independent functions of the coordinates $x^\mu$ in
the set $I_p$ must be zero. Therefore, from eqs.%
~(\ref{rie1st})~--~(\ref{rielast}) we conclude that for 5D
homogeneity it is necessary that
\begin{eqnarray}
\frac{H'}{D} &=&\mbox{const} \equiv -\,2\,\omega \label{cond1} \:, \\
\frac{D''}{D}&=&\mbox{const} \equiv m^2 \label{cond2} \:, \\
\frac{\ddot{F}}{F}&=&\mbox{const} \equiv k \:. \label{cond3}
\end{eqnarray}
The above necessary conditions are also suf\/f\/icient for
5D local homogeneity. Indeed, under these conditions the
nonvanishing frame components of the curvature reduce to
\begin{eqnarray}
\widehat{R}_{0101} &=& \widehat{R}_{0202}=- \omega^2
\label{rieh1st} \:, \\
\widehat{R}_{1212} &=& m^2 - 3\,\omega^2 \label{rieh2nd} \:, \\
\widehat{R}_{3434} &=& k \label{riehlast} \:.
\end{eqnarray}
Following Cartan's method for the local equivalence,
we calculate the f\/irst covariant derivative of
the Riemann tensor. One obtains the following non-null
covariant derivatives of the curvature:
\begin{equation} \label{drieh}
\widehat{R}_{0112;1} = \widehat{R}_{0212;2}=
\omega\, (m^2 - 4\,\omega^2) \:.
\end{equation}
Clearly, regardless of the value of the constant $k\,$, the
f\/irst covariant derivative of the curvature is algebraically
expressible in terms of the Riemann tensor. Moreover, the number
of functionally independent functions of the coordinates
$x^\mu$ among the components of the curvature and its
f\/irst covariant derivative is zero ($t_0=t_1=0$).
As far as the dimension of the residual isotropy group is
concerned we distinguish three dif\/ferent classes of locally
homogeneous 5D generalized G\"odel-type curved manifolds,
according to the relevant parameters $m^2$, $\omega$ and
$k\,$, namely~\cite{foot1}
\begin{enumerate}
\item[1.] $\mbox{dim}\,(H_0) = \mbox{dim}\, (H_1)= 2\,$ when
\begin{enumerate}
\item[a)]
$\,\omega \not=0\,$, any real $k\,$, $\,m^2 \not=4\,\omega^2\,$ ;
\item[b)]
$\,\omega =0\,$, $k\not= 0\,$, $\,m^2 \not=0\,$ ;
\end{enumerate}
\item[2.] $\,\mbox{dim}\,(H_0) = \mbox{dim}\, (H_1)= 4\,$ when
\begin{enumerate}
\item[a)]
$\,\omega \not=0\,$, any real $k\,$, $\,m^2=4\,\omega^2\,$ ;
\item[b)]
$\,\omega =0\,$, $k=0\,$, $\,m^2 \not=0\,$ ;
\item[c)]
$\,\omega =0\,$, $k\not=0\,$, $\,m^2=0\,$ ;
\end{enumerate}
\item[3.] $\,\mbox{dim}\,(H_0) = \mbox{dim}\, (H_1)= 10\,$
when $\omega = k = m^2 = 0\,$.
\end{enumerate}
Thus, from eqs.\ (\ref{gdim}) and (\ref{ddim}) one f\/inds that
the locally homogeneous 5D generalized G\"odel-type
manifolds admit a (local) $G_r$, with either $r =7$, $r=9\,$,
or $r=15$ acting on an orbit of dimension $d = 5$, that is on
the manifold ${\cal M}_5$.
The above results can be collected together in the following
theorems:
\vspace{2mm}
\begin{theorem} \label{TheoHom}
The necessary and sufficient conditions for a five-dimensional
generalized G\"odel-type manifold to be locally homogeneous
are those given by equations (\ref{cond1})~--~(\ref{cond3}).
\end{theorem}
\begin{theorem} \label{EquivTheo}
\begin{sloppypar}
The five-dimensional homogeneous generalized G\"o\-del-type
manifolds are locally characterized by three independent real
parameters $\omega$, $k$ and $m^2\,$: identical triads
($\omega, k,\, m^2$) specify locally equivalent manifolds.
\end{sloppypar}
\end{theorem}
\begin{theorem} \label{GroupTheo}
The five-dimensional locally homogeneous generalized G\"o\-del-type
manifolds admit group of isometry $G_r$ with
\begin{enumerate}
\item[(i)]
$r=7\:$ if either of the above conditions (1.a) and (1.b) is fulfilled;
\item[(ii)]
$r=9\:$ if one of the above set of conditions (2.a), (2.b) and (2.c) is
fulfilled;
\item[(iii)]
$r=15\;$ if the above condition (3) is satisfied.
\end{enumerate}
\end{theorem}
We shall now focus our attention on the irreducible set of
isometrically nonequivalent homogeneous generalized
G\"odel-type metrics. These nonequivalent classes of
metrics can be obtained by a similar procedure to that used
by Rebou\c{c}as and Tiomno~\cite{Reboucas83}, namely by
integrating equations~(\ref{cond1})~--~(\ref{cond3}),
and eliminating through coordinate transformations the
non-essential integration constants taking into account
the relevant parameters according to the above
theorem~\ref{EquivTheo}.
For the sake of brevity, however, we shall only present
the irreducible classes without going into details of
calculations. It turns out that one ought to distinguish
six classes of metrics according to:
{\bf Class I} : $\,m^2 > 0\,$, any real $k\,$, $\,\omega \not=0 $.
The line element for this class of homogeneous generalized
G\"odel-type manifolds can always be brought [in cylindrical
coordinates $(r, \phi, z)$] into the form
\begin{equation} \label{ds2c}
d\hat{s}^{2}=[\,dt+H(r)\, d\phi\,]^{2} -D^{2}(r)\, d\phi^{2} -dr^{2}
-F^2(u)\,dz^{2} - du^2
\end{equation}
with the metric functions given by
\begin{eqnarray}
H(r) &=&\frac{2\,\omega}{m^{2}}\: [1 - \cosh\,(mr)]\;, \label{Hh} \\
D(r) &=& m^{-1}\, \sinh\,(mr) \label{Dh} \;,
\end{eqnarray}
\begin{eqnarray} \label{Ffun}
F\,(u) = \left\{ \begin{array}
{l@{\qquad \mbox{if} \qquad}l}
\alpha^{-1}\, \sin\, (\alpha\, u ) & k = - \alpha^2 < 0 \;, \\
u & k = 0 \;, \\
\alpha^{-1}\, \sinh\, (\alpha\,u) & k = \alpha^2 > 0 \;.
\end{array} \right.
\end{eqnarray}
According to theorem~\ref{GroupTheo} the possible isometry groups
for this class are either $G_7$ (for $m^2 \not= 4\,\omega^2$) or
$G_9$ (when $m^2 = 4\,\omega^2$), irrespective of the value
of $k\,$.
{\bf Class II} : $\,m^2 = 0\,$, any real $k\,$, $\,\omega \not=0 $.
The line element for this class can be brought into the form~(\ref{ds2c}),
with the metric function $F(u)$ given by~(\ref{Ffun}), but now the
functions $H(r)$ and $D(r)$ are given by
\begin{equation} \label{DHsr}
H(r) = - \,\omega\, r^{2} \: \qquad \mbox{and} \: \qquad D(r) = r \:.
\end{equation}
For this class from theorem~\ref{GroupTheo} there is a group $G_7$ of
isometries, regardless of the value of $k$.
{\bf Class III} : $\,m^{2} \equiv - \mu^{2} < 0\,$, any real $k\,$,
$\,\omega \not=0 $.
Similarly for this class the line element reduces to ~(\ref{ds2c}) with
$F(u)$ given by (\ref{Ffun}) and
\begin{eqnarray}
H(r) &=& \frac{2\,\omega}{\mu^{2}} \:[\cos\,(\mu r) - 1 ] \;, \label{Hc} \\
D(r) &=& \mu^{-1}\, \sin\,(\mu r)\;. \label{Dc}
\end{eqnarray}
{}From theorem~\ref{GroupTheo}, regardless the value of $k$ for this class
there is a group $G_7$ of isometries.
{\bf Class IV} : $\,m^{2} \not= 0\,$, any real $k\,$, and $\,\omega = 0$.
We shall refer to this class as degenerated G\"odel-type manifolds,
since the cross term in the line element, related to the
rotation $\omega$ in 4D G\"odel model, vanishes.
By a trivial coordinate transformation one can make $H = 0$
with $D(r)$ given, respectively, by (\ref{Dh})
or (\ref{Dc}) depending on whether $m^2>0$ or
$m^{2} \equiv - \mu^{2} < 0$. The function $F(u)$ depends on
the sign of $k$ and is again given by (\ref{Ffun}).
For this class according to theorem~\ref{GroupTheo} one
may have either a $G_7$ for $k\not=0$, or a $G_9$ for
$k=0\,$.
{\bf Class V} : $\,m^{2} = 0\,$, $k \not=0 \,$, and $\,\omega = 0$.
By a trivial coordinate transformation one can make $H=0\,$, $D = r\,$
and $F(u) = \alpha^{-1}\,\sin\, (\alpha\,u)\,$ or
$\,F(u) = \alpha^{-1}\sinh\, (\alpha\,u) \,$ depending on whether $\,k<0\,$
or $\,k>0\,$, respectively. From theorem~\ref{GroupTheo} there is
a group $G_9$ of isometries.
{\bf Class VI} : $\,m^{2} = 0\,$, $k=0\,$, and $\,\omega = 0$.
{}From (\ref{rieh1st})~--~(\ref{riehlast}) this corresponds to
the 5D f\/lat manifold. Therefore, one can make
$H=0\,$, $D(r)= r\,$ and $F(u) = u \,$.
Theorem~\ref{GroupTheo} ensures that there is a group $G_{15}$
of isometries.
\vspace{3mm}
\section{Killing Vector Fields}
\label{Killing}
\setcounter{equation}{0}
In this section we shall present the inf\/initesimal
generators of isometries of the 5D homogeneous generalized
G\"odel-type manifolds, whose line element (\ref{ds2c}) can
be brought into the Lorentzian form (\ref{ds2f}) with
$\widehat{\Theta}^A$ given by
\begin{equation} \label{lorpen1}
\widehat{\Theta}^{0} = dt + H(r)\,d\phi\:, \: \quad
\widehat{\Theta}^{1} = dr\:, \: \quad
\widehat{\Theta}^{2} = D(r)\,d\phi\:, \:\quad
\widehat{\Theta}^{3} = F(u)\,dz \:, \: \quad
\widehat{\Theta}^{4} = du \:,
\end{equation}
where the functions $H(r)\,$, $D(r)$ and $F(u)$ depend
upon the essential parameters $m^2\,$, $k\,$ and $\omega\,$
according to the above classes of locally homogeneous
manifolds.
Denoting the coordinate components of a generic Killing vector f\/ield
$\widehat{K}$ by $\widehat{K}^{u} \equiv (Q, R, S,\bar{Z},U)$,
where $Q, R, S, \bar{Z} $ and $U$ are functions of all
coordinates $t,r,\phi,z$, $u$, then the f\/ifteen Killing equations
\begin{equation} \label{killeqs}
\widehat{K}_{(A;B)} \equiv \widehat{K}_{A;B} + \widehat{K}_{B;A} = 0
\end{equation}
can be written in the Lorentz frame (\ref{ds2f})~--~(\ref{lorpen1})
as
\begin{eqnarray}
&T_t = 0 \:, \qquad T_{u} - U_t = 0 \label{um} \:, \\
&R_r = 0 \:, \qquad U_r + R_{u} = 0 \label{dois} \:, \\
&U_{u} = 0 \:, \label{tres} \\
&D\,(T_r - R_t) - H_r P = 0 \label{quatro} \:, \\
& D P_{u} + U_{\phi} - H U_t = 0 \label{cinco} \:, \\
&T_{\phi} + H_r R - D P_t = 0 \label{seis} \:, \\
&R_{\phi} - H R_t - D_r P + D P_r = 0 \label{sete} \:, \\
&P_{\phi} - H P_t + D_r R = 0 \label{oito} \:, \\
& T_z - F\,Z_t = 0 \label{nove} \:, \\
& F\, Z_r + R_z = 0 \label{dez} \:, \\
& Z_z + U\, F_u = 0 \label{onze} \:, \\
&U_z + F\, Z_{u} - Z\, F_u = 0 \label{doze} \:, \\
& D P_z + F\,( Z_{\phi} - H Z_t) = 0 \label{treze} \:,
\end{eqnarray}
where the subscripts denote partial derivatives, and where we
have made
\begin{equation} \label{pencor}
T \equiv H\,S + Q, \qquad P \equiv D\,S, \qquad \mbox{and} \qquad
Z \equiv F\, \bar{Z}
\end{equation}
to make easier the comparison and the use of the results obtained
in~\cite{Teixeira85}.
To this end we note that with the changes $u \rightarrow z$ and
$U \rightarrow Z$ the above equations (\ref{um})~--~(\ref{oito})
are formally identical to the Killing equations (4) to (11)
of~\cite{Teixeira85}. However, in the equations%
~(\ref{um})~--~(\ref{oito}) the functions $T, R, P, U$
depend additionally on the f\/ifth coordinate $u$. Taking into
account this similitude, the integration of the
Killing equations (\ref{um})~--~(\ref{treze}) can be obtained
in two steps as follows. First, by analogy with (4) to (11)
of Ref.~\cite{Teixeira85} one integrates (\ref{um})~--~(\ref{oito}),
but at this step instead of the integration constants one has
integration functions of the f\/ifth coordinate $u$.
Second, one uses the remaining eqs.\ (\ref{nove})~--~(\ref{treze})
to achieve explicit forms for these integration functions and
to obtain the last component $U$ of the generic Killing vector $K$.
We have used the above two-steps procedure to integrate the
Killing equations (\ref{um})~--~(\ref{treze}) for all class
of homogeneous generalized G\"odel-type manifolds.
However, for the sake of brevity, we shall only present the
Killing vector f\/ields and the corresponding Lie algebras without
going into details of calculations, which can be verif\/ied
by using, for example, the computer algebra program {\sc killnf},
written in {\sc classi} by {\AA}man~\cite{Aman87}.
{\bf Class I} : $\,m^2 > 0\,$, any real $k\,$, $\,\omega \not=0 $.
In the integration of the Killing equation for this general class
one is led to distinguish two dif\/ferent subclasses of solutions
depending on whether $m^2 \not= 4 \,\omega^2$ or $m^2 =4 \,\omega^2$.
We shall refer to these subclasses as classes Ia and Ib, respectively.
{\bf Class Ia} : $\,m^2 >0\,$, any real $k\,$, $\,m^2\not=4\,\omega^2$.
In the coordinate basis in which as (\ref{ds2c}) is given,
a set of linearly independent Killing vector f\/ields
$K_N$ ($N$ is an enumerating index) is given by
\begin{eqnarray}
K_1 &=&\partial_t \:, \qquad \quad
K_2 = \frac{2\,\omega}{m}\, \,\partial_t - m \,\partial_{\phi} \:,
\label{KIa1} \\
K_3 &=& -\,\frac{H}{D}\, \sin\phi\, \,\partial_t +\cos\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \sin\phi\, \,\partial_{\phi} \:,
\label{KIa2} \\
K_4 &=& -\,\frac{H}{D}\, \cos\phi\, \,\partial_t -\sin\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \cos\phi\, \,\partial_{\phi} \:,
\label{KIa3} \\
K_5 &=& \sin z\, \,\partial_u +\frac{F_u}{F}\,\cos z\, \,\partial_z \:,
\label{KIa4} \\
K_6 &=& \cos z\, \,\partial_u - \frac{F_u}{F}\,\sin z\, \,\partial_{z} \:,
\label{KIa5} \\
K_7 &=& \partial_{z} \;. \label{KIa6}
\end{eqnarray}
The Lie algebra has the following nonvanishing commutators:
\begin{eqnarray}
& \left[ K_2, K_3 \right] = - m\, K_4 \:, \qquad
\left[ K_2, K_4 \right] = m\, K_3 \:, \qquad
\left[ K_3, K_4 \right] = m\, K_2 \:, \\
&\left[ K_5, K_6 \right] = -\, k \, K_7 \:, \qquad
\left[ K_5, K_7 \right] = - K_6 \:, \qquad
\left[ K_6, K_7 \right] = K_5 \:.
\end{eqnarray}
Therefore the corresponding algebra is
${\cal L}_{Ia} = {\cal L}_k \oplus \tau \oplus so\,(2,1)$.
Here and in what follows the symbols $\oplus\,$
and $\,\mbox{$\subset \!\!\!\!\!\!+$}$ denote and direct and semi-direct sum of sub-algebras,
and the sub-algebra ${\cal L}_k$ is $so\,(3)$ for $k < 0 \,$,
$so\,(2,1)$ for $k > 0 \,$, and $t^2 \,\mbox{$\subset \!\!\!\!\!\!+$} \, so\,(2)\,$ for $k=0\,$.
For the present class ${\cal L}_k$ is generated by
$K_5, K_6\,$ and $K_7$, the symbol $\tau$ is associated to the time
translation $K_1$, and f\/inally the inf\/initesimal generators of
sub-algebra $so\,(2,1)$ are $K_2,\, K_3\,$ and $K_4$.
{\bf Class Ib} : $\,m^2 =4\, \omega^2\,$, any real $k\,$, $\,\omega\not=0$.
For this class the Killing vector f\/ields are
\begin{eqnarray}
K_1 &=&\partial_t \:, \quad \qquad
K_2 = \partial_t - m \,\partial_{\phi} \:,
\label{KIb1} \\
K_3 &=& -\,\frac{H}{D}\, \sin\phi\, \,\partial_t +\cos\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \sin\phi\, \,\partial_{\phi} \:, \label{KIb2} \\
K_4 &=& -\,\frac{H}{D}\, \cos\phi\, \,\partial_t -\sin\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \cos\phi\, \,\partial_{\phi} \:, \label{KIb3} \\
K_5 &=&-\,\frac{H}{D}\,\cos(mt+\phi)\,\,\partial_t
+\sin(mt+\phi)\,\,\partial_r
+\,\frac{1}{D}\, \cos(mt+\phi)\,\,\partial_{\phi} \:, \label{KIb4} \\
K_6 &=&-\,\frac{H}{D}\,\sin(mt+\phi)\,\,\partial_t
-\cos(mt+\phi)\,\,\partial_r
+\,\frac{1}{D}\, \sin(mt+\phi)\,\,\partial_{\phi} \:, \label{KIb5} \\
K_7 &=& \sin z\, \,\partial_u +\frac{F_u}{F}\,\cos z\, \,\partial_z \:,
\label{KIb6} \\
K_8 &=& \cos z\, \,\partial_u - \frac{F_u}{F}\,\sin z\, \,\partial_{z} \:,
\label{KIb7} \\
K_9 &=& \partial_{z} \:, \label{KIb8}
\end{eqnarray}
whose Lie algebra is given by
\begin{eqnarray}
&\left[ K_1, K_5 \right] = -m\, K_6 \:, \qquad
\left[ K_1, K_6 \right] = m\, K_5 \:, \qquad
\left[ K_2, K_3 \right] = - m\, K_4 \:, \\
&\left[ K_2, K_4 \right] = m\, K_3 \:, \qquad
\left[ K_3, K_4 \right] = m\, K_2 \:, \qquad
\left[ K_5, K_6 \right] = m\, K_1 \:, \\
&\left[ K_7, K_8 \right] = -\, k \, K_9 \:, \qquad
\left[ K_7, K_9 \right] = - K_8 \:, \qquad
\left[ K_8, K_9 \right] = K_7 \:.
\end{eqnarray}
So, the corresponding algebra for this case is
${\cal L}_{Ib} = {\cal L}_k \oplus so\,(2,1) \oplus so\,(2,1)$.
As in the previous class the sub-algebra ${\cal L}_k$ depends on
the sign of $k$, and here is generated by $K_7,K_8$ and $K_9$.
The two sub-algebras $so\,(2,1)$ are generated by the Killing vector
f\/ields $K_1, K_5, K_6$ and $K_2, K_3, K_4$.
{\bf Class II} : $\,m^2 = 0\,$, any real $k\,$, $\,\omega \not=0$.
For this class the Killing vector f\/ields turns out to be the
following:
\begin{eqnarray}
K_1 &=&\partial_t \:, \quad \qquad
K_2 = \partial_{\phi} \:, \label{KII1} \\
K_3 &=& -\,\omega\,r\, \sin\phi\, \,\partial_t -\cos\phi\, \,\partial_r
+\,\frac{1}{r}\, \sin\phi\, \,\partial_{\phi} \:, \label{KII2} \\
K_4 &=& -\,\omega\,r\, \cos\phi\, \,\partial_t +\sin\phi\, \,\partial_r
+\,\frac{1}{r}\, \cos\phi\, \,\partial_{\phi} \:, \label{KII3} \\
K_5 &=& \sin z\, \,\partial_u +\frac{F_u}{F}\,\cos z\, \,\partial_z \:,
\label{KII4} \\
K_6 &=& \cos z\, \,\partial_u - \frac{F_u}{F}\,\sin z\, \,\partial_{z} \:,
\label{KII5} \\
K_7 &=& \partial_{z} \:. \label{KII6}
\end{eqnarray}
The Lie algebra has the following nonvanishing commutators:
\begin{eqnarray}
&\left[ K_2, K_3 \right] = K_4 \:, \quad
\left[ K_2, K_4 \right] = - K_3 \:, \quad
\left[ K_3, K_4 \right] = 2\, \omega\, K_1 \:, \\
&\left[ K_5, K_6 \right] = -\, k \, K_7 \:, \qquad
\left[ K_5, K_7 \right] = - K_6 \:, \qquad
\left[ K_6, K_7 \right] = K_5 \:.
\end{eqnarray}
Therefore, the corresponding algebra for this case is
${\cal L}_{II} = {\cal L}_k \oplus {\cal L}_4$.
The sub-algebra ${\cal L}_4$ is generated by $K_1, K_2, K_3$
and $K_4$. This algebra ${\cal L}_4$ is soluble and
does not contain abelian 3D sub-algebras; it is classif\/ied
as type $III$ with $q=0$ by Petrov~\cite{Petrov69}.
The sub-algebra ${\cal L}_k$ is the same of the previous
classes and is generated by $K_5, K_6$ and $K_7$.
{\bf Class III} : $\,m^{2} \equiv - \mu^{2} < 0\,$, any real $k\,$,
$\,\omega \not=0 $.
For this class the set of linearly independent Killling vector f\/ields
we have found is given by
\begin{eqnarray}
K_1 &=&\partial_t \:, \quad \:
K_2 = \frac{2\,\omega}{\mu} \, \partial_t
+ \mu\, \partial_{\phi} \:, \label{KIII1} \\
K_3 &=& -\,\frac{H}{D}\, \sin\phi\, \,\partial_t
+\cos\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \sin\phi\, \,\partial_{\phi} \:, \label{KIII2} \\
K_4 &=& -\,\frac{H}{D}\, \cos\phi\, \,\partial_t
-\sin\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \cos\phi\, \,\partial_{\phi} \:, \label{KIII3} \\
K_5 &=& \sin z\, \,\partial_u +\frac{F_u}{F}\,\cos z\, \,\partial_z \:,
\label{KIII4} \\
K_6 &=& \cos z\, \,\partial_u - \frac{F_u}{F}\,\sin z\, \,\partial_{z} \:,
\label{KIII5} \\
K_7 &=& \partial_{z} \:. \label{KIII6}
\end{eqnarray}
The Lie algebra has the following nonvanishing commutators:
\begin{eqnarray}
&\left[ K_2, K_3 \right] = \mu \, K_4 \:, \qquad
\left[ K_2, K_4 \right] = -\mu \, K_3 \:, \qquad
\left[ K_3, K_4 \right] = \mu \, K_2 \:, \\
&\left[ K_5, K_6 \right] = -\, k \, K_7 \:, \qquad
\left[ K_5, K_7 \right] = - K_6 \:, \qquad
\left[ K_6, K_7 \right] = K_5 \:.
\end{eqnarray}
Thus, the corresponding algebra for this case is
${\cal L}_{III} = {\cal L}_k \oplus \tau \oplus so\,(3)$.
Here $\tau$ is associated to the Killing vector f\/ield $K_1$,
whereas to the sub-algebra $so\,(3)$ correspond $K_2$,
$K_3$ and $K_4$. Again ${\cal L}_k$ is generated by
$K_5, K_6$ and $K_7$.
{\bf Class IV} : $\,m^2 \not= 0\,$, any real $k\,$, $\,\omega=0 $.
In the integration of the Killing equation for this general class
one is led to distinguish two dif\/ferent subclasses according to
$k\not=0$ or $k=0$. We shall denote these subclasses
as classes IVa and IVb, respectively.
{\bf Class IVa} : $\,m^{2} \not= 0\,$, $k\not=0$, $\,\omega = 0$.
This class corresponds to the so-called degenerated
G\"odel-type manifolds. One obtains for this class the
following Killing vector f\/ields:
\begin{eqnarray}
K_1 &=&\partial_t \:, \qquad \quad K_2 = \partial_\phi \:,
\label{KIVa1} \\
K_3 &=& \cos\phi\, \,\partial_r-
\,\frac{D_r}{D}\, \sin\phi\, \,\partial_{\phi} \:,
\label{KIVa2} \\
K_4 &=& -\sin\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \cos\phi\, \,\partial_{\phi} \:,
\label{KIVa3} \\
K_5 &=& \sin z\, \,\partial_u +\frac{F_u}{F}\,\cos z\, \,\partial_z \:,
\label{KIVa4} \\
K_6 &=& \cos z\, \,\partial_u - \frac{F_u}{F}\,\sin z\, \,\partial_{z} \:,
\label{KIVa5} \\
K_7 &=& \partial_{z} \:, \label{KIVa6}
\end{eqnarray}
where $D(r) = (1/m)\, \sinh mr$ for $m^2>0\,$, or
$\,D(r) = (1/\mu)\, \sin \mu r$ for $m^2 \equiv - \mu^2 < 0 $,
and the function $F(u)$ for $k\not=0$ is given by~(\ref{Ffun}).
The Lie algebra has the following nonvanishing commutators:
\begin{eqnarray}
&\left[ K_2, K_3 \right] = K_4 \:, \qquad
\left[ K_2, K_4 \right] = - K_3 \:, \qquad
\left[ K_3, K_4 \right] = - m^2 \, K_2 \:, \\
&\left[ K_5, K_6 \right] = -\, k \, K_7 \:, \qquad
\left[ K_5, K_7 \right] = - K_6 \:, \qquad
\left[ K_6, K_7 \right] = K_5 \:,
\end{eqnarray}
where one should substitute $-m^2$ by $\mu^2$ if $m^2 < 0 $. So, the
corresponding Lie algebra is ${\cal L}_{IVa}={\cal L}_k \oplus
\tau \oplus {\cal L}_m$, where ${\cal L}_m$ is $so\,(2,1)$ for $m^2>0$,
and $so\,(3)$ for $m^2 = - \mu^2 <0$. The sub-algebra ${\cal L}_k$
(generated by $K_5, K_6$ and $K_7$) is $so\,(3)$ for $k < 0\,$,
and $so\,(2,1)$ for $k > 0 \,$. Again $\tau$ is associated to the
Killing vector f\/ield $K_1$.
{\bf Class IVb} : $\,m^{2} \not= 0\,$, $k=0$, $\,\omega = 0$.
We shall refer to this class as doubly-degenerated
G\"odel-type manifolds. One obtains for this class the
following Killing vector f\/ields:
\begin{eqnarray}
K_1 &=&\partial_t \:, \qquad \quad K_2 = \partial_\phi \:,
\label{KIVb1} \\
K_3 &=& \cos\phi\, \,\partial_r-
\,\frac{D_r}{D}\, \sin\phi\, \,\partial_{\phi} \:,
\label{KIVb2} \\
K_4 &=& -\sin\phi\, \,\partial_r
-\,\frac{D_r}{D}\, \cos\phi\, \,\partial_{\phi} \:,
\label{KIVb3} \\
K_5 &=& \sin z\, \,\partial_u +\frac{1}{u}\,\cos z\, \,\partial_z \:,
\label{KIVb4} \\
K_6 &=& \cos z\, \,\partial_u - \frac{1}{u}\,\sin z\, \,\partial_{z} \:,
\label{KIVb5} \\
K_7 &=& \partial_{z} \:, \label{KIVb6} \\
K_8 &=& u\,\sin z\, \,\partial_t + t \, \sin z\, \,\partial_u
+ \frac{1}{u}\,\,t\, \cos z \, \,\partial_z \:,
\label{KIVb7} \\
K_9 &=& u\,\cos z\, \,\partial_t + t\,\cos z\, \,\partial_{u}
- \frac{1}{u}\,\,t\, \sin z\, \,\partial_z \:,
\label{KIVb8}
\end{eqnarray}
where again $D(r) = (1/m)\, \sinh mr$ for $m^2>0\,$, or
$\,D(r) = (1/\mu)\, \sin \mu r$ for $m^2 \equiv - \mu^2 < 0 $.
The Lie algebra has the following nonvanishing commutators:
\begin{eqnarray}
&\left[ K_2, K_3 \right] = K_4 \:, \qquad
\left[ K_2, K_4 \right] = - K_3 \:, \qquad
\left[ K_3, K_4 \right] = - m^2 \, K_2 \:, \\
&\left[ K_5, K_7 \right] = - K_6 \:, \qquad
\left[ K_6, K_7 \right] = K_5 \:, \qquad
\left[ K_1, K_8 \right] = K_5 \:, \\
&\left[ K_1, K_9 \right] = K_6 \:, \qquad
\left[ K_5, K_8 \right] = K_1 \:, \qquad
\left[ K_6, K_9 \right] = K_1 \:, \\
&\left[ K_7, K_8 \right] = K_9 \:, \qquad
\left[ K_7, K_9 \right] = - K_8 \:, \qquad
\left[ K_8, K_9 \right] = - K_7 \:,
\end{eqnarray}
where one should substitute $-m^2$ by $\mu^2$ if $m^2 < 0 $.
So, the corresponding Lie algebra is
${\cal L}_{IVb} = t^3\, \mbox{$\subset \!\!\!\!\!\!+$}\, so\,(2,1) \oplus {\cal L}_m$, where
${\cal L}_m$ is generated by $K_2, K_3, K_4$, and is either
$so\,(2,1)$ or $so\,(3)$ depending on whether $m^2>0$ or
$m^2 = -\mu^2 <0$. The sub-algebra $t^3\,\mbox{$\subset \!\!\!\!\!\!+$}\, so\,(2,1)\,$
is generated by $K_1, K_5, K_6, K_7, K_8, K_9$.
{\bf Class V} : $\,m^{2} = 0\,$, $k\not=0$, $\,\omega = 0$.
A set of linearly independent Killing vector f\/ield for this
class is
\begin{eqnarray}
K_1 &=&\partial_t \:, \qquad \quad K_2 = \partial_\phi \:,
\label{KV1} \\
K_3 &=& \cos\phi\, \,\partial_r-
\,\frac{1}{r}\, \sin\phi\, \,\partial_{\phi} \:,
\label{KV2} \\
K_4 &=& -\sin\phi\, \,\partial_r
-\,\frac{1}{r}\, \cos\phi\, \,\partial_{\phi} \:,
\label{KV3} \\
K_5 &=& \sin z\, \,\partial_u +\frac{F_u}{F}\,\cos z\, \,\partial_z \:,
\label{KV4} \\
K_6 &=& \cos z\, \,\partial_u - \frac{F_u}{F}\,\sin z\, \,\partial_{z} \:,
\label{KV5} \\
K_7 &=& \partial_{z} \:, \label{KV6} \\
K_8 &=& r\,\sin \phi\, \,\partial_t + t \, \sin\phi\, \,\partial_r
+ \frac{1}{r}\,\,t\, \cos \phi \, \,\partial_\phi \:,
\label{KV7} \\
K_9 &=& r\,\cos \phi\, \,\partial_t + t\,\cos \phi\, \,\partial_r
- \frac{1}{r}\,\,t\, \sin \phi\, \,\partial_\phi \:,
\label{KV8}
\end{eqnarray}
where $F(u)$ depends upon the sign of $k$ and is given by eq.~(\ref{Ffun}).
The Lie algebra has the following nonvanishing commutators:
\begin{eqnarray}
&\left[ K_2, K_3 \right] = K_4 \:, \qquad
\left[ K_2, K_4 \right] = - K_3 \:, \qquad
\left[ K_5, K_6 \right] = -\, k \, K_7 \:, \\
&\left[ K_5, K_7 \right] = - K_6 \:, \qquad
\left[ K_6, K_7 \right] = K_5 \:, \qquad
\left[ K_1, K_8 \right] = - K_4 \:, \\
&\left[ K_1, K_9 \right] = K_3 \:, \qquad
\left[ K_4, K_8 \right] = - K_1 \:, \qquad
\left[ K_3, K_9 \right] = K_1 \:, \\
&\left[ K_2, K_8 \right] = K_9 \:, \qquad
\left[ K_2, K_9 \right] = - K_8 \:, \qquad
\left[ K_8, K_9 \right] = - K_2 \:.
\end{eqnarray}
So, the corresponding Lie algebra is
${\cal L}_{V} = t^3\, \mbox{$\subset \!\!\!\!\!\!+$}\, so\,(2,1) \oplus {\cal L}_k$, where
${\cal L}_k$ is generated by $K_5, K_6, K_7$, and is either
$so\,(2,1)$ or $so\,(3)$ depending on whether $k>0$ or
$k<0$. The sub-algebra $t^3\,\mbox{$\subset \!\!\!\!\!\!+$}\, so\,(2,1)\,$
is generated by $K_1, K_2, K_3, K_4, K_8, K_9$.
{\bf Class VI} : $\,m^{2} = 0\,$, $k= 0$, $\,\omega = 0$.
{}From (\ref{rieh1st})~--~(\ref{riehlast}) this case corresponds to
the 5D f\/lat manifold whose Lie algebra is
${\cal L}_{VI} = so\,(4,1)\,$ since it clearly has the well
known f\/ifteen Killing vector f\/ields, namely f\/ive translations,
four spacetime rotations, and six space rotations.
It is worth noting that none of the above Lie algebras
is semi-simple, but some of their sub-algebras are.
Besides, most of the simple sub-algebras are noncompact.
The 3D sub-algebra $so\,(3)$ present in all classes
is compact, though.
The number of Killing vector f\/ields we have found for each
of the above six classes makes explicit that the 5D locally
homogeneous generalized G\"odel-type manifolds admit a
group of isometry $G_7$ when
({\em 1a\/}):
$\: m^2 \not = 4\,\omega^2\,$, any real $k\,$, $\,\omega \not=0$,
or when
({\em 1b\/}):
$\,m^2 \not=0$, $k\not=0$, $\omega=0\,$.
Groups $G_9$ of isometry occur when
({\em 2a\/}):
$\:\,m^2 = 4\, \omega^2$, any real $k$, $\omega \not=0$, or
({\em 2b\/}):
$m^2 \not= 0$, $k = 0$, $\omega =0$, or when
({\em 2c\/}):
$m^2 = 0$, $k \not= 0$, $\omega =0$.
Clearly when $\: m^2 = \omega = k = 0\,$ there is $G_{15}$.
These possible groups are in agreement with theorem~\ref{GroupTheo}
of the previous section.
Actually the integration of the Killing equations constitutes a
dif\/ferent way of deriving that theorem. Furthermore, these
equations also show that the isotropy subgroup $H$ of $G_r$
is such that $\,\mbox{dim}\,(H) = 2\,$ when the above conditions
({\em 1a\/}) and ({\em 1b\/}) are satisf\/ied, while the conditions
({\em 2a\/}), ({\em 2b\/}) and ({\em 2c\/}) lead to
$\,\mbox{dim}\,(H) = 4\,$, also in agreement with the
previous section. Clearly $\,\mbox{dim}\,(H) = 10\,$
when $m^2 = \omega = k = 0$.
\vspace{3mm}
\section{Causal Anomalies and Final Remarks}
\label{Anom}
\setcounter{equation}{0}
In this section we shall initially be concerned with the problem of
causal anomalies in the generalized G\"odel-type manifolds.
Then we proceed by examining whether the IM gravity
allows solutions of generalized G\"odel-type metrics~(\ref{ds2c}).
Finally, we conclude by addressing to the general question as to whether
the IM gravity theory rules out the 4D noncausal G\"odel-type
solutions to Einstein's equations of general relativity.
In the f\/irst three of the six classes of homogeneous generalized
G\"odel-type manifolds we have discussed in Section~\ref{homoconds},
there are closed timelike curves.
Indeed, the analysis made in a previous paper%
~\cite{ReboucasTeixeira98a} can be easily extended to
the generalized 5D G\"odel-type manifolds of the present
article. To this end, we write the line element~(\ref{ds2c})
in the form
\begin{equation} \label{ds2cx}
ds^2=dt^2 + 2\,H(r)\, dt\,d\phi -dr^2 -G(r)\,d\phi^2
- F^2(u)\,dz^2 - du^2\,,
\end{equation}
where $G(r)= D^2 - H^2$ and $(r,\phi, z)$ are cylindrical
coordinates.
Now, the existence of closed timelike curves of the G\"odel-type depends
on the behavior of $G(r)$. Indeed, if $G(r) < 0$ for a certain
range of $r$ ($r_1 < r < r_2$, say), G\"odel's circles%
~\cite{Calvao88} $u,t,z,r =const$ are closed timelike curves.
Since one can always make $H=0$ for the generalized G\"odel-type
manifolds of classes IV, V and VI, then $G(r) > 0$ for all $r>0$. Thus
there are no closed timelike G\"odel's circles in these classes of
manifolds.
On the other hand, following the above-outlined reasoning it easy to
show (see~\cite{ReboucasTeixeira98a} for details) that for each of the
remaining three classes (Class I to Class III) one can always f\/ind a
critical radius $r_c$ such that for all $r > r_c$ one has $G(r) < 0$,
making clear that there are closed timelike curves in these
families of homogeneuous generalized G\"odel-type manifolds.
However, in what follows we shall show that these types of
noncausal {\em curved\/} manifolds are not permitted
in the context of the induced matter theory.
In the Lorentz frame $\widehat{\Theta}^A$ given by~(\ref{lorpen1})
the nonvanishing frame components of the Einstein tensor
$\widehat{G}_{AB}=\widehat{R}_{AB}-\frac{1}{2}\,R\,\widehat{\eta}_{AB}$
are
\begin{eqnarray}
\widehat{G}_{00} &=& - \,\frac{D''}{D} + \frac{3}{4}\,
\left( \frac{H'}{D}\,\right)^2
- \frac{\ddot{F}}{F} \label{ein00} \;, \\
\widehat{G}_{02} &=& \frac{1}{2} \, \left( \frac{H'}{D}\, \right)' \;,
\label{ein02} \\
\widehat{G}_{11} &=& \widehat {G}_{22}\; = \; \frac{1}{4} \,
\left( \frac{H'}{D}\, \right)^2
+ \frac{\ddot{F}}{F} \:, \label{ein11} \\
\widehat{G}_{33} &=& \widehat{G}_{44}\; = \; \frac{D''}{D} - \frac{1}{4}\,
\left( \frac{H'}{D}\,\right)^2 \label{ein33} \;,
\end{eqnarray}
where the prime and dot denote derivative with respect to $r$
and $u$, respectively.
The f\/ield equations~(\ref{5DfeqsG}) require that
$\widehat{G}_{02}=0$, which in turn implies that
\begin{equation} \label{G02}
\frac{H'}{D} = \mbox{const} \equiv - 2\,\omega\;.
\end{equation}
Inserting~(\ref{G02}) into~(\ref{ein11}), (\ref{ein33}) and (\ref{ein00})
one easily f\/inds that the IM f\/ield equations are fulf\/illed
if and only if the independent parameters $\omega$, $k$ and $m^2$
[ see eqs.~(\ref{cond1})~--~(\ref{cond2} ] vanish identically,
which leads to the only solution given by
\begin{equation} \label{sol}
H = a \;, \qquad
D = b\,r + c \;, \qquad \mbox{and} \qquad
F = \beta \, u + \gamma,
\end{equation}
where $a$, $b$, $c$, $\beta$, and $\gamma$ are arbitrary real
constants. However, these constants have no physical meaning, and
can be taken to be $a = c = \gamma = 0$ and $b=\beta= 1$ by a
suitable choice of coordinates.
Indeed, if one performs the coordinate transformations
\begin{eqnarray}
t & = & \bar{t} - \frac{a}{b}\,\, \bar{\phi}\:, \qquad \qquad
r = \bar{r} - \frac{c}{b}\:, \label{tr} \\
\phi & = & \frac{\bar{\phi}}{b}\, \:, \; \quad
z = \frac{\bar{z}}{\beta} \:, \qquad
u = \bar{u}- \frac{\gamma}{\beta} \:, \label{ppz}
\end{eqnarray}
the line element~(\ref{ds2cx}) becomes
\begin{equation} \label{ds2flat}
d\hat{s}^2=d\bar{t}^2 -d\bar{r}^2 -\bar{r}^2 \,d\bar{\phi}^2 -
d\bar{z}^2 -d\bar{u}^2\;,
\end{equation}
in which we obviously have $\,G(\bar{r})= \bar{r}^2 >0\,$ for
$\bar{r} \not=0$.
The line element (\ref{ds2flat}) corresponds to a
manifestly f\/lat 5D manifold, making it clear that the underlying
manifold can be taken to be the simply connected Euclidean
manifold $\R^5$, and therefore as $\,G(\bar{r})>0\,$
no closed timelike circles are permitted.
Furthermore the above results clearly show that the IM theory
does not admit any {\em curved\/} 5D G\"odel-type
metric~(\ref{ds2c}) as solution to its f\/ield
equations~(\ref{5DfeqsG}).
However, in a recent work Mc Manus~\cite{McManus94} has shown
that a one-parameter family of solutions of the f\/ield
equations~(\ref{5DfeqsG}) previously found by Ponce de Leon%
~\cite{Leon88} was in fact f\/lat in five dimensions.
And yet the corresponding 4D induced models were shown to be a
perfect f\/luid family of Friedmann-Robertson-Walker curved models
(see Refs.~\cite{Wesson96a,Wesson92c,Coley95} and also%
~\cite{Abolghasem96}~--~\cite{Liu98}, where other Riemann-f\/lat
solutions are also discussed).
Therefore a question which naturally arises here is whether the above
5D f\/lat metric, which is the only solution to the IM f\/ield
equations, can similarly give rise to any 4D {\em curved\/}
spacetime. However, from~(\ref{ds2flat}) one obviously
has that the corresponding 4D spacetime is nothing but
the Minkowski f\/lat space (this result can also
be derived by using a computer algebra package as, e.g.,
{\sc clas\-si}~\cite{Aman87,MacCallumSkea94} to calculate
the 4D curvature tensor for $m^2=\omega=0\,$).
In brief, the only solution of the IM f\/ield equations~(\ref{5DfeqsG})
of generalized G\"odel-type is the 5D f\/lat space~(\ref{sol}), which
give rise only to the 4D Minkowski (f\/lat) spacetime,
whose topology can be taken to be the simply connected Euclidean
$\R^5$, in which no closed timelike curves are permitted.
Although the above results can be looked upon as if the induced
matter theory works as an ef\/fective therapy for the causal
anomalies which arises when one starts from the specif\/ic
generalized 5D G\"odel-type family of metrics~(\ref{ds2cx}),
this does not ensure that the induced matter version
of general relativity is an ef\/f\/icient treatment for the causal
anomalies (solutions with closed timelike curves) in general
relativity as it has been conjectured in~\cite{ReboucasTeixeira98b}.
Actually, in a recent paper (which unfortunately has not been
initially noticed by Rebou\c{c}as and Teixeira%
~\cite{ReboucasTeixeira98b}) Romero {\em et al.}~\cite{Romero96}
(see also~\cite{Lidsey97}) have shown that the induced matter 5D
scheme is indeed general enough to locally generate all solutions
to 4D Einstein's f\/ield equations. This is ensured by a theorem
due to Campbell~\cite{Campbell26} which states that any analytic
$n$-dimensional Riemannian space can be locally embedded in a
$(n+1)$-dimensional Ricci-f\/lat space.
In our context this amounts to saying that there must exist
a f\/ive-dimensional Ricci-f\/lat space which locally gives
rise to the 4D G\"odel noncausal solution of Einstein's
equations of general relativity.
Thus, what still remains to be done regarding G\"odel-type
spaces is to f\/ind out this 5D Ricci-f\/lat
space which gives rise (locally) to the 4D G\"odel-type spacetimes
of general relativity.
To conclude it is worth stressing some features of the local
underlying embedding of the induced matter theory. Any
Riemann-f\/lat manifold obviously is also Ricci-f\/lat. The
reverse, however, does not necessarily holds, and one can have
Ricci-f\/lat spaces which are not Riemann-f\/lat.
For the generalized 5D G\"odel-type geometries we have discussed
in this paper the condition for Ricci-f\/latness
($\widehat{R}_{AB}=0$) necessarily leads to Riemann-f\/lat spaces.
Remarkably many solutions of the f\/ield equations~(\ref{5DfeqsG})
are indeed Riemann-f\/lat (see~\cite{Wesson96a,McManus94,Coley95}
and ~\cite{Leon88}~--~\cite{Liu98}).
{}From a purely mathematical 5D point of view all Riemann-f\/lat
spaces are locally equivalent (locally isometric).
However, from the viewpoint of the 5D induced matter gravity
all the above-referred 5D Riemann-f\/lat solutions give rise
to physically (and geometrically) distinct 4D spacetimes%
~\cite{Wesson96a,McManus94,Coley95}, ~\cite{Leon88}~--~\cite{Liu98}.
On the other hand, in the light of the equivalence problem
techniques we have discussed in Section~\ref{prereq}, these
5D Riemann-f\/lat examples also show that all 5D Cartan
scalars~(\ref{CartanScl}) can vanish identically, with or
without the vanishing of the corresponding (induced) 4D
Cartan scalars.
\vspace{3mm}
\section*{Acknowledgement} \label{acknowl}
\setcounter{equation}{0}
The authors are grateful to the scientif\/ic agency CNPq for
the grants under which this work was carried out.
M.J. Rebou\c{c}as thanks C. Romero, V.B. Bezerra and J.B.
Fonseca-Neto for motivating and fruitful discussions.
\vspace{1cm}
|
\section{Introduction}
In the last decade quantum computing has become an extremely active
research area. The initial idea that the simulation of quantum
physical systems might be out of reach for classical devices goes
back to Feynman\cite{Fey82}. He raised the possibility that
computational devices based on the principles of quantum mechanics
might be more powerful than classical ones. This challenge to the
quantitative version of the Church-Turing thesis, which asserts that
all physically realisable computational devices can be simulated
with
only a polynomial overhead by a probabilistic Turing machine, is
the driving force behind the study of quantum computers and
algorithms.
The first formal models of quantum computing, the quantum Turing
machine and quantum circuits were defined by
Deutsch\cite{Deu85,Deu89}. Yao has shown\cite{Yao93} that these
two models have polynomially equivalent computational power when
the circuits are uniform.
In a sequence of papers oracles have been exhibited
\cite{DJ92,BB92,BV97,Sim97},
relative to which quantum Turing machines are
more powerful than classical (probabilistic or non-deterministic) ones.
These results culminated in the seminal paper
of Shor\cite{Sho97} where he gave polynomial time quantum
algorithms for the factoring and the discrete logarithm problems.
A quantum circuit operates on $n$ quantum bits (qubits), where $n$
is some integer. The actual computation takes place in the Hilbert
space $\mathbb{C}^{\{0,1\}^n}$ whose computational basis consists of the
$2^n$ orthonormal vectors $\ket{i}$ for $i \in {\{0,1\}}^n.$
According to the standard model, during the computation the state
of the system is a unit length linear superposition of the basis
states. The computational steps of the system are done by quantum
gates which perform unitary operations and are local in the sense
that they involve only a constant number of qubits. At the end of
the computation a measurement takes place on one of the qubits.
This is a probabilistic experiment whose outcome can be $0$ or $1$,
and the probability of measuring the bit $b$ is the squared length
of the projection of the superposition to the subspace spanned by
the basis states that are compatible with the outcome. As a result
of a measurement, the state of the system becomes this projected
state.
The most convenient way to describe all possible operations on a
quantum register is in the formalism of `density matrices'. In this
approach, which differs from the Dirac notation, the quantum
operations are described by completely positive superoperators
(CPSOs) that act on matrices. These density matrices describe mixed
states (that is, classical probability distributions over pure
quantum states), and the CPSOs correspond exactly to all the
physically allowed transformations on them. Such a model of quantum
circuits with mixed states was described by Aharonov, Kitaev and
Nisan\cite{AKN98}, and we will adopt it here. The unitary quantum
gates of the standard model and measurements are special CPSOs.
CPSOs can be simulated by unitary quantum gates on a larger number
of qubits, and in \cite{AKN98} it was shown that the computational
powers of the two models are polynomially equivalent.
Unitary quantum gates for small number of qubits have been extensively
studied. One reason is that although quantum gates for up to three
qubits have already been built, constructing gates for large
numbers seems to be elusive. Another reason is that universal sets
of gates can be built from them, which means that they can simulate
(approximately) any unitary transformation on an arbitrary number
of qubits. The first universal quantum gate which operates on three
qubits was identified by Deutsch\cite{Deu89}. After a long sequence
of work on universal quantum gates
\cite{DiV95,Bar95,DBE95,Llo95,BBC95,Sho96,KLZ96,Kit97}, Boykin et
al.\cite{bmprv99} have recently shown that the set consisting of a
Hadamard gate, a ${\mathrm{c\text{-}NOT}}$ gate, and a phase rotation gate of angle
$\pi/4$ is universal. In order to form a practical basis for
quantum computation, a universal set must also be able to operate
in a noisy environment, and therefore has to be
fault-tolerant\cite{Sho96,AB97,Kit97,KLZ98}. The above set of three
gates has the additional advantage of also being fault-tolerant.
Experimental procedures for determining the properties of
quantum
``black boxes'' were given by Chuang and
Nielsen\cite{CN97} and
Poyatos, Cirac and Zoller\cite{PCZ97}, however these procedures
implicitly require apparatus that has already been tested and
characterized.
The idea of self-testing in quantum devices is
implicit in the work of Adleman, Demarrais and Huang\cite{ADH97}.
They have developed a procedure by which a quantum Turing machine
is able to estimate its internal angle by its own means under the
hypothesis that the machine is unitary. In the context of quantum
cryptography Mayers and Yao\cite{MY98} have designed tests for
deciding if a photon source is perfect. These tests guarantee that
if source passes them then it is adequate for the security of the
Bennett-Brassard\cite{BB84} quantum key distribution protocol.
In this paper we develop the theory of self-testing of quantum gates
by classical procedures. Given a CPSO $\super{G}$ for $n$ qubits,
and a family $\mathcal{F}$ of unitary CPSOs, we would like to
decide if $\super{G}$ belongs to $\mathcal{F}$.
Intuitively, a self-tester is a procedure that answers the question
``$\super{G}\in \mathcal{F}$ ?'' by interacting with the CPSO $\super{G}$
in a purely classical way. More precisely, it will be a probabilistic
algorithm that is able to access $\super{G}$ as a black box in the
following sense: it can prepare the classical states $w\in\{0,1\}^n$,
iterate $\super{G}$ on these states, and afterwards, measure in the
computational basis.
The access must be seen as a whole, performed
by a specific, experimental oracle for $\super{G}$: once the basis
state ${w}$ and the number of iterations $k$ have been specified,
the program in one step gets back one of the possible probabilistic
outcomes of measuring the state of the system
after $\super{G}$ is iterated $k$-times on ${w}$. The intermediate
quantum states of this process cannot be used by the program, which
cannot perform any other quantum operations either. For $0
\leq \delta_1 \leq
\delta_2$, such an algorithm will be a $(\delta_1,
\delta_2)$-tester for $\mathcal{F}$ if for every CPSO $\super{G}$,
whenever the distance of $\super{G}$ and $\mathcal{F}$ is at most
$\delta_1$ (in some norm), it accepts with high probability, and
whenever the same distance is greater than $\delta_2$, it rejects
with high probability, where the probability is taken over the
measurements performed by the oracle and by the internal coin
tosses of the algorithm.
Finally we will say that $\mathcal{F}$ is
{\em testable} if for every $\delta_2 > 0$, there exists $0 <
\delta_1 \leq \delta_2$ such that there exists a $(\delta_1,
\delta_2)$-tester for $\mathcal{F}$. These definitions can be
extended to several classes of CPSOs.
The study of self-testing programs is a well-established research
area which was initiated by the work of Blum, Luby and
Rubinfeld\cite{BLR93}, Rubinfeld\cite{Rub90},
Lipton\cite{Lip91}
and Gemmel
\mbox{et al.}\cite{GLR91}. The purpose of a self-tester for a function
family is to detect by simple means if a program which is
accessible as an oracle computes a function from the given family.
This clearly inspired the definition of our self-testers
which have the particularity that they should test quantum objects
that
they can access only in some particular way. The analogy
with self-testing does not stop with the definition. One of the
main tools in self-testing of function families is the
characterization of these families by robust properties.
Informally, a property is robust if whenever a function satisfies
the property approximately, then it is close to a function which
satisfies it exactly. The concept of robustness was introduced and
its implication for self-testing was first studied by Rubinfeld and
Sudan\cite{RS96} and by Rubinfeld\cite{Rub94}. It will play a
crucial role in our case.
We note in the Preliminaries that for any real $\phi$ the states
$\ket{1}$ and $e^{i\phi}\ket{1}$ are experimentally
indistinguishable. This implies that if we start by only
distinguishing the classical states $0$ and $1$ then there are
families of CPSOs which are indistinguishable
as well.
For example, let
$\super{H}$ be the well-known Hadamard gate, and let
$\super{H}_{\phi}$ be the same gate expressed in the basis
$(\ket{0}, e^{i \phi}\ket{1})$, for $\phi \in [0,2\pi)$.
Any experiment that starts in state $0$ or $1$ and uses only
$\super{H}$ will produce outcomes $0$ and $1$ with the same
probabilities as the same experiment with $\super{H_{\phi}}$.
Thus no experiment that uses this quantum gate alone can
distinguish it from all the other Hadamard gates. Indeed,
as stated later in Fact~\ref{conjugate_basis}, a family
$\mathcal{F}$ containing $\super{H}$ can only be tested if the
entire Hadamard family $\mathcal{H}=\{\super{H}_\phi:\phi\in
[0,2\pi)\}$ is included in $\mathcal{F}$.
The main result of this paper is \textbf{\reftheo{maintheo}} which states
that for several sets of unitary CPSOs,
in particular,
the Hadamard gates
family, Hadamard gates together with ${\mathrm{c\text{-}NOT}}$
gates, and
Hadamard gates with ${\mathrm{c\text{-}NOT}}$
and phase rotation gates of angle $\pm\pi/4$, are testable.
This last family is of particular importance since
every triplet in the family
forms a universal and fault-tolerant set of gates
for quantum computation\cite{bmprv99}.
For the proof we will define the notion of experimental equations
which are functional equations for CPSOs corresponding to the
properties of the quantum gate that a self-tester can approximately
test. These tests are done via the interaction with the
experimental oracle. The proof itself contains three parts. In
\textbf{Theorems \ref{general}},
\textbf{\ref{couple}}, and \textbf{\ref{cnot}} we will exhibit
experimental equations for the families of unitary CPSOs we want to
characterize. In \textbf{\reftheo{robustness}} we will show that
actually all experimental equations are robust;
in fact, the distance of a CPSO from
the target family is polynomially related to
the error tolerated in the experimental equations.
Finally
\textbf{\reftheo{robtotester}} gives self-testers for CPSO families
which are characterized by a finite set of robust experimental
equations.
In some cases, we are able to calculate explicitly the polynomial bound
in the robustness of experimental equations.
Such a result will be illustrated in \textbf{\reftheo{hadamardrob}}
for the equations characterizing the Hadamard family $\mathcal{H}$.
Technically, these results will be based on the representation of
one-qubit states and CPSOs in $\mathbb{R}^3$, where they are respectively
vectors in the unit ball of $\mathbb{R}^3$, and particular affine
transformations. This correspondence is known as the Bloch Ball
representation.
\section{Preliminaries}
\subsection{The quantum state}
A {\em pure state} in a quantum physical system is described by a
unit vector in a Hilbert space. In the {\em Dirac} notation it is
denoted by $\ket{\psi}$. In particular a {\em qubit} (a quantum
two-state system) is an element of the Hilbert space $\C^{\{0,1\}}$. The
orthonormal basis containing $\ket{0}$ and $\ket{1}$ is called the
{\em computational basis} of $\C^{\{0,1\}}$. Therefore a pure state
$\ket{\psi}\in\C^{\{0,1\}}$ is a {\em superposition} of the computational
basis states, that is, $\ket{\psi}=c_{0}\ket{0}+c_{1}\ket{1}$, with
$\abs{c_{0}}^{2}+\abs{c_{1}}^{2}=1$. A physical system which deals
with $n$ qubits is described mathematically by the
$2^n$-dimensional Hilbert space which is by definition
$\C^{\{0,1\}}\otimes\cdots\otimes\C^{\{0,1\}}$, that is, the $n^{\mbox{\footnotesize
th}}$ tensor power of $\C^{\{0,1\}}$. Let $N=2^n$. The computational basis
of this space consists of the $N$ orthonormal states $\ket{i}$ for
$0\leq i<N$. If $i$ is in binary notation $i_1 i_2\ldots i_n$, then
$\ket{i_1\ldots i_n}=\ket{i_1}\ldots\ket{i_n}$, where this is a
short notation for $\ket{i_1}\otimes\cdots\otimes\ket{i_n}$. All
vectors and matrices will be expressed in the computational basis.
The transposed complex conjugate $\ket{\psi}^\dag$ of $\ket{\psi}$
is denoted by $\bra{\psi}$. The inner product between $\ket{\psi}$
and $\ket{\psi'}$ is denoted by $\braket{\psi}{\psi'}$, and their
outer product by $\ketbra{\psi}{\psi'}$.
Quantum systems can also be in more general states than what
can be described by pure states. The most general states are
\emph{mixed states}, described by
a probability distribution over pure states. Such a mixture can be
denoted by $\{(p_{k},\ket{\psi_{k}}) : k\in\mathbb{N} \}$, where the system
is in the pure state $\ket{\psi_{k}}$ with probability $p_{k}$.
Different mixtures (even different pure states $\ket{\psi}$) can
represent the same physical system. This notational redundancy can
be avoided if we use the formalism of the density matrices.
A \emph{density matrix} that represents an $n$-qubit
state is an $N\times N$ Hermitian semi-positive matrix with trace
$1$. The pure state $\ket{\psi}$ in this representation is
described by the density matrix $\psi=\ketbra{\psi}{\psi}$, and a
mixture $\{(p_{k},\ket{\psi_{k}})
: \ k\in\mathbb{N}\}$ by the density matrix
$\psi=\sum_{k\in\mathbb{N}}p_{k}\ketbra{\psi_{k}}{\psi_{k}}$. For example,
the pure states $e^{i\gamma}\ket{\psi}$, for $\gamma\in[0,2\pi)$,
or the mixtures $\{(\mbox{$\frac{1}{2}$},\ket{0}),(\mbox{$\frac{1}{2}$},\ket{1})\}$ and
$\{(\mbox{$\frac{1}{2}$},\frac{\ket{0}+\ket{1}}{\sqrt{2}}),
(\mbox{$\frac{1}{2}$},\frac{\ket{0}-\ket{1}}{\sqrt{2}})\}$ have respectively the
same density matrix.
Since a density matrix is Hermitian semi-positive, its eigenvectors
are orthogonal and its eigenvalues are non-negative. Because
the density matrix has trace $1$, its eigenvalues sum to $1$.
Therefore a density matrix
represents the mixture of its orthonormal eigenvectors, where the
probabilities are the respective eigenvalues. Note that diagonal
density matrices correspond to a mixture over pure states
$\ket{i}$, for $0\leq i<N$. Density matrices that represent pure
states have a simple algebraic characterization: $\rho$ is a pure
state if and only if it has two eigenvalues, $0$ with multiplicity
$N-1$ and $1$ with multiplicity $1$, equivalently $\rho$ is a pure
state exactly when $\rho^2=\rho$.
A $2\times 2$ Hermitian matrix of unit trace is semi-positive if
and only if its determinant is between $0$ and $1/4$. Therefore in
the case of one qubit, any density matrix $\rho$ can be written as
$\rho = p\ketbra{0}{0}+(1-p)\ketbra{1}{1}
+\alpha\ketbra{1}{0}+{\alpha^*}\ketbra{0}{1}$, where $p\in[0,1]$,
and $\alpha$ is a complex number such that $\abs{\alpha}^2\leq
p(1-p)$. This density matrix will be denoted by $\rho(p,\alpha)$.
Remark that $\rho(p,\alpha)$ is a pure state exactly when
$\abs{\alpha}^2= p(1-p)$, that is, its determinant is $0$.
\subsection{Superoperators}
The evolution of physical systems is described by specific
transformations on density matrices, that is, on operators. A
{\em superoperator} for $n$ qubits is a linear transformation on
$\mathbb{C}^{N\times N}$. A {\em positive} superoperator (PSO) is a
superoperator that sends density matrices to density matrices. A
{\em completely positive} superoperator (CPSO) $\super{G}$ is a PSO
such that for all positive integers $M$,
$\super{G}\otimes\super{I}_{M}$ is also a PSO,
where $\super{I}_{M}$ is the identity on $\mathbb{C}^{M\times M}$.
CPSOs are exactly the physically allowed transformations on
density matrices. An example of a PSO for one qubit that is not a
CPSO is the transpose superoperator $\super{T}$ defined by
$\super{T}(\ketbra{i}{j})=\ketbra{j}{i}$, for $0\leq i,j\leq 1$.
Quantum~computation is based on the possibility of constructing
some particular CPSOs, {\em unitary} superoperators, which preserve
the set of pure states. These operators are characterized by
transformations from $\mathrm{U}(N)$, the set of $N\times N$ unitary
matrices. For any $A\in \mathrm{U}(N)$, we define a CPSO which maps a
density matrix $\rho$ into $A\rho A^\dag$. When the underlying
unitary transformation $A$ is clear from the context, by somewhat
abusing the notation, we will denote this CPSO simply by
$\super{A}$. If $\ket{\psi'}$ denotes $A\ket{\psi}$, then the
unitary superoperator $\super{A}$ maps the pure state $\psi$ to the
pure state $\psi'$. As was the case in the Dirac representation of
states, there is the same phase redundancy in the set of unitary
transformations $\mathrm{U}(N)$. If $A\in\mathrm{U}(N)$, then for all $\gamma\in
[0,2\pi)$, the transformations $e^{i\gamma} A$ are different,
however the corresponding superoperators are identical. We will
therefore focus on $\mathrm{U}(N)/\mathrm{U}(1)$.
\subsection{Measurements}
Measurements form another important class of (non-unitary) CPSOs.
They describe physical transformations corresponding to the
observation of the system. We will define now formally one of the
simplest classes of measurements which correspond to the
projections to elements of the computational basis.
A {\em Von Neumann measurement in the computational basis} of $n$
qubits is the $n$-qubit CPSO $\super{M}$ that, for every
density matrix
$\rho$, satisfies $\super{M}(\rho)_{i,i}=\rho_{i,i}$ and
$\super{M}(\rho)_{i,j}=0$, for $i\neq j$.
In the case of one qubit, the Von Neumann measurement in the
computational basis maps the density matrix $\rho(p,\alpha)$ into
$\rho(p,0)$. We will say that $p=\bra{0}\rho\ket{0}$ is the {\em
probability of measuring} $\ketbra{0}{0}$, and we will denote it by
$\PM{0}{\rho}$.
In general, a {\em Von Neumann measurement} of $n$ qubits in any
basis can be viewed as the Von Neumann measurement in the
computational basis preceded by some unitary superoperator.
\subsection{The Bloch Ball representation}
Specific for the one-qubit case, there is an isomorphism
between the group $\mathrm{U}(2)/\mathrm{U}(1)$ and the special rotation group $\mathrm{SO}(3)$,
the set of $3\times 3$ orthogonal matrices with determinant $1$.
This allows us to
represent one-qubit states as vectors in the unit ball of $\mathbb{R}^3$,
and unitary superoperators as rotations on $\mathbb{R}^3$.
We will now describe exactly this correspondence.
The {\em Bloch Ball} $\mathcal{B}$ (respectively {\em Bloch Sphere} $\mathcal{S}$)
is the unit ball (respectively unit sphere) of the Euclidean affine space
$\mathbb{R}^{3}$.
Any point $\ball{u}\in\mathbb{R}^{3}$ determines a vector
with the same coordinates which we will also denote by $\ball{u}$.
The inner product of $\ball{u}$ and $\ball{v}$ will be denoted by
$(\ball{u},\ball{v})$, and
their Euclidean norm by $\norm{\ball{u}}$.
Each point $\ball{u}\in\mathbb{R}^{3}$ can be also characterized by its
norm $r\geq 0$,
its latitude $\theta\in[0,\pi]$, and its longitude
$\phi\in[0,2\pi)$.
The {\em latitude} is the angle between the $z$-axis and the vector
$\ball{u}$, and
the {\em longitude} is the angle between the $x$-axis and the orthogonal
projection of $\ball{u}$ in the plane defined by $z=0$.
If $\ball{u}=(x,y,z)$, then
these parameters satisfy
$x=r\sin\theta\cos\phi$,
$y=r\sin\theta\sin\phi$ and $z=r\cos\theta$.
For every density matrix $\rho$ for one qubit there exists a
unique point
$\ball{\rho}=(x,y,z) \in\mathcal{B}$ such that
\begin{eqnarray*}
\rho
&=&\frac{1}{2}\dens{1+z}{x-i y}{x+i y}{1-z}.
\end{eqnarray*}
This mapping is a bijection that also obeys
\begin{eqnarray*}
\ball{\rho(p,\alpha)} & = & (2\Re(\alpha),2\Im(\alpha),2p-1).
\end{eqnarray*}
In this formalism, the pure states are nicely characterized in $\mathcal{B}$
by their norm.
\begin{fact}
A density matrix $\rho$ represents a pure state if and only if
$\ball{\rho}\in\mathcal{S}$, that is, $\norm{\ball{\rho}}=1$.
\end{fact}
Also, if $\theta\in[0,\pi]$ and $\phi\in[0,2\pi)$ are respectively
the latitude and the longitude of $\ball{\psi}\in\mathcal{S}$, then the
corresponding density matrix represents a pure state and satisfies
$\ket{\psi} =
\cos(\theta/2)\ket{0}+\sin(\theta/2)e^{i\phi}\ket{1}$.
Observe that the pure states $\ket{\psi}$ and $\ket{\psi^{\perp}}$
are orthogonal if and only if $\ball{\psi}=-\ball{\psi^\perp}$. We
will use the following notation for the six pure states along the
$x$, $y$ and $z$ axes: $\ket{\zeta_x^\pm} =
\frac{1}{\sqrt{2}}(\ket{0}\pm\ket{1})$, $\ket{\zeta_y^\pm} =
\frac{1}{\sqrt{2}}(\ket{0}\pm i \ket{1})$, $\ket{\zeta_z^+} = \ket{0}$, and
$\ket{\zeta_z^-} = \ket{1}$, with the respective coordinates $(\pm
1,0,0)$, $(0,\pm 1,0)$ and $(0,0,\pm 1)$ in $\mathbb{R}^3$.
For each CPSO $\super{G}$, there exists a unique affine
transformation $\affine{G}$ over $\mathbb{R}^{3}$, which
maps the ball $\mathcal{B}$ into $\mathcal{B}$ and
is such that, for all density matrices $\rho$,
$\affine{G}(\ball{\rho})=\ball{\super{G}(\rho)}$.
Unitary superoperators have a nice characterization
in $\mathcal{B}$.
\begin{fact}
The map between $\mathrm{U}(2)/\mathrm{U}(1)$ and $\mathrm{SO}(3)$, which sends $A$ to
$\affine{A}$, is an isomorphism.
\end{fact}
For $\alpha\in(-\pi,\pi]$, $\theta\in[0,\frac{\pi}{2}]$, and
$\phi\in[0,2\pi)$, we will define the unitary transformation
$R_{\alpha,\theta,\phi}$ over $\mathbb{C}^{2}$. If
$\ket{\psi}=\cos(\theta/2)\ket{0} +e^{i\phi}\sin(\theta/2)\ket{1}$
and $\ket{\psi^\perp}=\sin(\theta/2)\ket{0}
-e^{i\phi}\cos(\theta/2)\ket{1}$
then by definition $R_{\alpha,\theta,\phi}\ket{\psi}=\ket{\psi}$
and
$R_{\alpha,\theta,\phi}\ket{\psi^\perp}=e^{i\alpha}\ket{\psi^\perp}$.
If $\super{A}$ is a unitary superoperator then we have
$\super{A}=\super{R}_{\alpha,\theta,\phi}$ for some $\alpha$,
$\theta$, and $\phi$. In $\mathbb{R}^3$ the transformation
$\affine{R}_{\alpha,\theta,\phi}$ is the rotation of angle $\alpha$
whose axis cuts the sphere $\mathcal{S}$ in the points $\ball{\psi}$ and
$\ball{\psi^\perp}$. Note that for $\theta=0$ the CPSO
$\super{R}_{\alpha,0,\phi}$ does not depend on $\phi$. We will
denote this phase rotation by $\super{R}_{\alpha}$.
The affine transformation in $\mathcal{B}$
which corresponds to the Von Neumann measurement in the computational basis
is the
orthogonal projection to the $z$-axis.
Therefore it maps
$\ball{\rho}=(x,y,z)$ into $(0,0,z)$, the point which corresponds to the
density matrix $\frac{1+z}{2}\ketbra{0}{0}+\frac{1-z}{2}\ketbra{1}{1}$.
Thus $\PM{0}{\rho}=\frac{1+z}{2}$.
\subsection{Norm and distance}
Let $N=2^n$. We will consider the {\em trace norm} on $\mathbb{C}^{N\times
N}$ which is defined as follows: for all $V\in\mathbb{C}^{N\times
N}$,
$\normone{V}=\mathrm{Tr}\sqrt{V^{\dag}V}$. This norm has several advantages
when we consider the difference of density matrices. Given a Von
Neumann measurement, a density matrix induces a probability
distribution over the basis of the measurement. The trace norm of
the difference of two density matrices is the maximal variation
distance between the two induced probability distributions, over
all Von Neumann measurements. It also satisfies the following
properties.
\begin{fact}
For all density matrices $\rho(p,\alpha)$ and $\rho(q,\beta)$ for
one qubit we have:
\begin{equation*}
\begin{array}{rcccl}
\normone{\rho(p,\alpha)-\rho(q,\beta)}&=&
\norm{\ball{\rho(p,\alpha)}-\ball{\rho(q,\beta)}}
&=&2\sqrt{(p-q)^2+\abs{\alpha-\beta}^2}.
\end{array}
\end{equation*}
\end{fact}
\begin{fact}
For all $V \in\mathbb{C}^{N \times N}$ and $W\in\mathbb{C}^{M\times M}$ we have
$\normone{V\otimes W}=\normone{V}\normone{W}$ and $\abs{\mathrm{Tr}(V)}\leq\normone{V}$.
For density matrices $\rho$ it holds that $\normone{\rho}=1$.
\end{fact}
For $n$-qubit superoperators, the superoperator norm
associated to the trace norm is defined as
\begin{eqnarray*}
\normso{\super{G}}& = &
\sup\{
\normone{\super{G}(V)} : \normone{V}=1\}.
\end{eqnarray*}
This norm is always $1$ when $\super{G}$ is a CPSO.
The norm $\normso{\,}$ can be easily generalized for $k$-tuples
of superoperators by
$\normso{(\super{G}_1,\ldots,\super{G}_k)}=
\max(\normso{\super{G}_1},\ldots,\normso{\super{G}_k}).$
We will denote by $\dist_{\infty}$ the natural induced distance
by the norm $\normso{\,}$.
\section{Properties of CPSOs}
Here we will establish the properties of CPSOs that we will need
for the characterization of our CPSO families.
In this extended abstract we will
omit the proof of Lemma \ref{lemma1}, and the proof of Lemma
\ref{lemma2} will be in \refap{APL2}.
\begin{lemma}\label{lemma1}
Let $\super{G}$ be a CPSO for one qubit,
and let $\rho$ and $\tau$ be density
matrices for one qubit.
\begin{enumerate}
\item[(a)]
$\normone{\super{G}(\rho)-\super{G}(\tau)}\leq\normone{\rho-\tau}$.
\item[(b)] If $\super{G}$ is not constant and $\super{G}(\rho)$ is a pure
state
then $\rho$ is a pure state.
\end{enumerate}
\end{lemma}
An affine transformation of $\mathbb{R}^3$ is uniquely defined by the images
of four non-coplanar points. Surprisingly, if the transformation is
a CPSO for one qubit, the images of three points are
sometimes sufficient. The following will make this precise more generally
for $n$ qubits.
\begin{lemma}\label{lemma2}
Let $n\geq 1$ be an integer, and
let $\rho_1$, $\rho_2$, and $\rho_3$ be three distinct one-qubit
density matrices representing pure states,
such that the plane in $\mathbb{R}^3$
containing the points $\ball{\rho_1},\ball{\rho_2},\ball{\rho_3}$
goes through the center of $\mathcal{B}$. If $\super{G}$ is a CPSO for
$n$ qubits which acts as the identity on the set
$\{\rho_1,\rho_2,\rho_3\}^{\otimes n}$,
then $\super{G}$ is the identity mapping.
\end{lemma}
We also use the property that for CPSOs unitarity and invertibility
are equivalent (see e.g. \cite[Ch. 3, Sec. 8]{Pre98}).
\begin{lemma} \label{lemma1c}
Let $\super{G}$ be a CPSO for $n$ qubits.
If there exists a CPSO $\super{H}$ for $n$ qubits
such that $\super{H}\circ\super{G}$ is the
identity mapping, then $\super{G}$ is a unitary superoperator.
\end{lemma}
\section{Characterization}
\subsection{One-Qubit CPSO Families}
In this section, every CPSO will be for one qubit.
First we define the notion of experimental
equations, and then we show that several important
CPSO families
are characterizable by them.
An {\em experimental equation} in one variable
is a CPSO equation of the form
\begin{eqnarray}\label{expCPSOeq}
\PM{0}{\super{G}^{k}(\ketbra{b}{b})} & = & r,
\end{eqnarray}
where $k$ is a non-negative integer, $b\in\{0,1\}$, and $0\leq
r\leq 1$. We will call the left-hand side of the equation the {\em
probability term}, and the right-hand side the {\em constant term}.
The {\em size} of this equation is $k$. A CPSO $\super{G}$ will
``almost'' satisfy the equations if, for example, it is the result
of adding small systematic and random errors (independent of time)
to a CPSO that does.
For $\varepsilon\geq 0$, the CPSO
$\super{G}$ {\em $\varepsilon$-satisfies} \refeq{expCPSOeq} if
$\abs{\PM{0}{\super{G}^{k} (\ketbra{b}{b})}-r}\leq\varepsilon,$ and when
$\varepsilon=0$ we will just say that $\super{G}$ {\em satisfies}
\refeq{expCPSOeq}. Let $(E)$ be a finite set of experimental
equations. If $\super{G}$ $\varepsilon$-satisfies all equations in $(E)$
we say that $\super{G}$ $\varepsilon$-satisfies $(E)$. If some $\super{G}$
satisfies $(E)$ then $(E)$ is {\em satisfiable}. The set
$\{\super{G} :
\super{G}\mbox{ satisfies $(E)$}\}$
will be denoted by $\mathcal{F}_{(E)}$.
A family $\mathcal{F}$ of CPSOs is {\em characterizable}
if it is $\mathcal{F}_{(E)}$
for some finite set $(E)$ of experimental equations.
In this case we say that $(E)$ {\em characterizes} $\mathcal{F}$.
All these definitions
generalize naturally for $m$-tuples of CPSOs for
$m\geq 2$.
In what follows we will need
only the case $m=2$.
An {\em experimental equation} in two CPSO variables
is an equation of the form
\begin{eqnarray*}\label{expCPSOeq2}
\PM{0}{\super{F}^{k_{1}}\circ\super{G}^{l_1}\circ\cdots\circ
\super{F}^{k_{t}}\circ\super{G}^{l_{t}}(\ketbra{b}{b})} & = & r,
\end{eqnarray*}
where $k_1,\ldots,k_t,l_1,\ldots,l_t$ are non-negative integers,
$b\in\{0,1\}$, and $0\leq r\leq 1$.
We discuss now the existence of finite sets of experimental
equations in one variable that characterize unitary superoperators,
that is, the operators $\super{R}_{\alpha,\theta,\phi}$, for
$\alpha\in(-\pi,\pi]$, $\theta\in[0,\pi/2]$, and $\phi\in[0,2\pi)$.
First observe that due to the restrictions of experimental
equations, there are unitary superoperators that they cannot
distinguish.
\begin{fact} \label{conjugate_basis}
Let $\alpha\in[0,\pi]$, $\theta\in[0,\pi/2]$, and
$\phi_1,\phi_2\in[0,2\pi)$ such that $\phi_1\neq\phi_2$.
Let $(E)$ be a finite set of experimental
equations in $m$ variables. If $(\super{R}_{\alpha,\theta,\phi_1},
\super{G}_2,\ldots,\super{G}_m)$ satisfies $(E)$ then
there exist $\super{G}'_2,\ldots,\super{G}'_m$ and
$\super{G}''_2,\ldots,\super{G}''_m$ such that
$(\super{R}_{-\alpha,\theta,\phi_1},
\super{G}'_2,\ldots,\super{G}'_m)$
and $(\super{R}_{\alpha,\theta,\phi_2},
\super{G}''_2,\ldots, \super{G}''_m)$ both satisfy $(E)$.
\end{fact}
In the Bloch Ball formalism this corresponds to the following
degrees of freedom in the choice of
the orthonormal basis of $\mathbb{R}^3$.
Since
experimental
equations contain exactly the states $\ketbra{0}{0}$ and
$\ketbra{1}{1}$
there is no freedom in the choice of the $z$-axis, but there
is complete freedom in the choice of the $x$ and $y$ axes.
The indistinguishability of the latitude $\phi$ corresponds
to the freedom of choosing the oriented $x$-axis, and the
indistinguishability of the sign of $\alpha$ corresponds
to the freedom of choosing the orientation of the $y$-axis.
We introduce the following notations.
Let $\mathcal{R}_{\alpha,\theta}$ denote the superoperator family
$\{\super{R}_{\pm\alpha,\theta,\phi} : \phi\in[0,2\pi)\}$.
For $\phi\in[0,2\pi)$,
let the $\neg_\phi$ transformation be defined by
$\neg_\phi\ket{0} = e^{i\phi}\ket{1}$
and $\neg_\phi(e^{i\phi}\ket{1})=\ket{0}$,
and recall that the Hadamard transformation
$H_\phi$ obeys $H_{\phi}\ket{0}=(\ket{0}+e^{i\phi}\ket{1})/\sqrt{2}$
and $H_{\phi}(e^{i\phi}\ket{1})=(\ket{0}-e^{i\phi}\ket{1})/\sqrt{2}$.
Observe that $\super{H}_\phi=\super{R}_{\pi,\pi/4,\phi}$
and $\super{\neg}_\phi=\super{R}_{\pi,\pi/2,\phi}$, for
$\phi\in[0,2\pi)$.
Finally let $\mathcal{H}=
\{\super{H}_\phi : \phi\in[0,2\pi)\}$, and
$\mathcal{N}=\{\super{\neg}_\phi : \phi\in[0,2\pi)\}$.
Since the sign of $\alpha$ cannot be determined, we will assume
that $\alpha$ is in the interval $[0,\pi]$. We will also consider
only unitary superoperators such that $\alpha/\pi$ is rational.
This is a reasonable choice since these superoperators form a dense
subset of all unitary superoperators. For such a unitary
superoperator, let $n_\alpha$ be the smallest positive integer $n$
for which $n\alpha=0 \mod{2\pi}$. Then either $n_\alpha=1$, or
$n_\alpha\geq 2$ and there exists $t\geq 1$ which is coprime with
$n_\alpha$ such that $\alpha=(t/n_\alpha)2\pi$. Observe that the
case $n_\alpha=1$ corresponds to the identity superoperator.
Our first theorem shows that almost all families
$\mathcal{R}_{\alpha,\theta}$ are characterizable by
some finite set of experimental equations.
In particular $\mathcal{H}$ is characterizable.
\begin{theorem}\label{general}
Let $(\alpha,\theta)\in(0,\pi]\times(0,\pi/2]\backslash\{(\pi,\pi/2)\}$
be such that
$\alpha/\pi$ is rational.
Let $z_k(\alpha,\theta)=\cos^{2}\theta+\sin^{2}\theta\cos(k\alpha)$.
Then the following experimental
equations characterize $\mathcal{R}_{\alpha,\theta}$:
\begin{equation*}\label{generaltest1}
\begin{array}{rcll}
\PM{0}{\super{G}^{n_\alpha}(\ketbra{1}{1})}= 0
& \textrm{ and } &
\PM{0}{\super{G}^{k}(\ketbra{0}{0})} = \mbox{$\frac{1}{2}$}+\mbox{$\frac{1}{2}$} z_{k}(\alpha,\theta),
& k\in\positive{n_\alpha}.
\end{array}
\end{equation*}
\end{theorem}
\noindent{\bf Proof: }
First observe that every CPSO in $\mathcal{R}_{\alpha,\theta}$
satisfies the equations of the theorem since the $z$-coordinate of
$\ball{\super{R}_{\alpha,\theta,\phi}^k(\ketbra{0}{0})}$ is
$z_k(\alpha,\theta)$ for every $\phi\in[0,2\pi)$. Let $\super{G}$
be a CPSO which
satisfies these equations. We will prove that
$\super{G}$ is a unitary superoperator. Then,
\reffact{unicityofangles} implies that
$\super{G}\in\mathcal{R}_{\alpha,\theta}$.
Since $z_1(\alpha,\theta)\neq \pm 1$, we know
$\super{G}(\ketbra{0}{0})\not\in\{\ketbra{0}{0},\ketbra{1}{1}\}$.
Observing that $\super{G}^{n_\alpha}(\ketbra{0}{0})=\ketbra{0}{0}$,
\reflemma{lemma1}(b) implies that $\super{G}(\ketbra{0}{0})$ is a
pure state. Thus $\ketbra{0}{0}$, $\ketbra{1}{1}$, and
$\super{G}(\ketbra{0}{0})$ are distinct pure states, and since
$\super{G}^{n_\alpha}$ acts as the identity on them, by
\reflemma{lemma2} it is the identity mapping. Hence by
\reflemma{lemma1c} $\super{G}$ is a unitary superoperator.
\qed
\begin{fact}\label{unicityofangles}
Let $\alpha\in(0,\pi]$,
$\theta\in(0,\pi/2]$,
$\alpha'\in(-\pi,\pi]$, $\theta'\in(0,\pi/2]$
be such that $\alpha/\pi$ is rational.
If $z_k(\alpha,\theta)=
z_k(\alpha',\theta')$, for
$k\in\positive{n_\alpha}$, then $\abs{\alpha'}=\alpha$
and $\theta'=\theta$.
\end{fact}
The remaining families $\mathcal{R}_{\alpha,\theta}$ for which
$\alpha/\pi$ is rational are
$\{\super{R}_{-\alpha},\super{R}_\alpha\}$, for $\alpha\in[0,\pi]$,
and $\mathcal{N}$. Let us recall that $\super{M}$ is the CPSO which
represents the Von Neumann measurement in the computational basis.
Since $\super{M}$ satisfies exactly the same equations as
$\super{R}_{\pm\alpha}$, and $\super{\neg}_0\circ\super{M}$
satisfies exactly the same equations as $\super{\neg}_\phi$, for
any $\phi\in[0,2\pi)$, these families are not characterizable by
experimental equations in one variable. Nevertheless it turns out
that together with the family $\mathcal{H}$ they become
characterizable. This is stated in the following theorem whose
proof is omitted.
\begin{theorem}~\label{couple}
The family
$\{(\super{H}_\phi,\super{\neg}_{\phi}) : \phi\in[0,2\pi)\}
\subset \mathcal{H}\times\mathcal{N}$ is characterized by
the experimental equations in two variables $(\super{F},\super{G})$:
\begin{equation*}
\left\{{\begin{array}{lll}
\PM{0}{\super{F}(\ketbra{0}{0})}=\mbox{$\frac{1}{2}$},\!\!&
\PM{0}{\super{F}^2(\ketbra{0}{0})}=1,\!\!&
\PM{0}{\super{F}^2(\ketbra{1}{1})}=0,\!\!\\
\vspace{-.1cm}
& & \\
\PM{0}{\super{G}(\ketbra{0}{0})}=0,\!\!&
\PM{0}{\super{G}(\ketbra{1}{1})}=1,\!\!& \\
\vspace{-.1cm}
& & \\
\PM{0}{\super{F}\circ \super{G}^2\circ \super{F}(\ketbra{0}{0})}=1,\!\!&
\PM{0}{\super{F}\circ \super{G}\circ \super{F}(\ketbra{0}{0})}=1.&\!\!\\
\end{array}}\right.
\end{equation*}
If $\alpha/\pi$ is rational, then the family
$\mathcal{H}\times\{ \super{R}_{\pm\alpha}\}$
is characterized by the experimental equations
in two variables $(\super{F},\super{G})$:
\begin{equation*}
\left\{{\begin{array}{lll}
\PM{0}{\super{F}(\ketbra{0}{0})}=\mbox{$\frac{1}{2}$},\!\!&
\PM{0}{\super{F}^2(\ketbra{0}{0})}=1,\!\!&
\PM{0}{\super{F}^2(\ketbra{1}{1})}=0,\!\!\\
\vspace{-.1cm}
& & \\
\PM{0}{\super{G}(\ketbra{0}{0})}=1,\!\!&
\PM{0}{\super{G}(\ketbra{1}{1})}=0,\!\!& \\
\vspace{-.1cm}
& & \\
\PM{0}{\super{F}\circ \super{G}^{n_\alpha}\circ
\super{F}(\ketbra{0}{0})}=1,\!\!&
\PM{0}{\super{F}\circ \super{G}\circ
\super{F}(\ketbra{0}{0})}=\mbox{$\frac{1}{2}$}+\mbox{$\frac{1}{2}$}\cos\alpha.\!\!& \\
\end{array}}\right.
\end{equation*}
\end{theorem}
\subsection{Characterization of $\super{{\mathrm{c\text{-}NOT}}}$ gates}
In this section we will extend our theory of characterization
of CPSO families for several qubits.
In particular, we will show that the family of $\super{{\mathrm{c\text{-}NOT}}}$ gates
together with
the family $\mathcal{H}$ is characterizable.
First we need some definitions.
For every $\phi\in[0,2\pi)$, we define ${\mathrm{c\text{-}NOT}}_\phi$ as
the only unitary transformation over $\mathbb{C}^4$ satisfying
${\mathrm{c\text{-}NOT}}_{\phi}(\ket{0}\ket{\psi})=\ket{0}\ket{\psi}$
and ${\mathrm{c\text{-}NOT}}_{\phi}\ket{1}\ket{\psi}=\ket{1}\neg_\phi\ket{\psi}$, for
all $\ket{\psi}\in\mathbb{C}^2$.
We extend the definition of the experimental equation for CPSOs
given in~\refeq{expCPSOeq2} for $n$ qubits.
It is an equation of the form
\begin{eqnarray}\label{expCPSOeq3}
\PM{v}{\super{F}^{k_{1}}\circ\super{G}^{l_1}\circ\cdots\circ
\super{F}^{k_{t}}\circ\super{G}^{l_{t}}(\ketbra{w}{w})}& = & r,
\end{eqnarray}
where in addition to the notation of~\refeq{expCPSOeq2}
$v,w\in\{0,1\}^n$, and $\mathrm{Pr}^{v}$ is the probability of
measuring $\ketbra{v}{v}$. For the variables $\super{F}$ and
$\super{G}$ of~\refeq{expCPSOeq3}, we also allow both the tensor
product of two CPSO variables and the tensor product of a CPSO
variable with the identity. We now state the characterization.
\begin{theorem}~\label{cnot}
The family
$\{(\super{H}_\phi,\super{{\mathrm{c\text{-}NOT}}}_{\phi}) : \phi\in[0,2\pi)\}$
is characterized by
the experimental equations in two variables $(\super{F},\super{G})$:
\begin{equation*}
\left\{{\begin{array}{llll}
\PM{0}{\super{F}(\ketbra{0}{0})}=\mbox{$\frac{1}{2}$},\!\!&
\PM{0}{\super{F}^2(\ketbra{0}{0})}=1,\!\!&
\PM{0}{\super{F}^2(\ketbra{1}{1})}=0,\!\!& \\
\vspace{-.1cm}
&&& \\
\PM{00}{\super{G}(\ketbra{00}{00})}=1,\!\!&
\PM{01}{\super{G}(\ketbra{01}{01})}=1,\!\!&
\PM{11}{\super{G}(\ketbra{10}{10})}=1,\!\!&
\PM{10}{\super{G}(\ketbra{11}{11})}=1,\!\!\\
\vspace{-.1cm}
&&& \\
\multicolumn{2}{l}{
\PM{00}{(\super{I}_2\otimes \super{F})\circ \super{G}\circ
(\super{I}_2\otimes \super{F})(\ketbra{00}{00})}=1,\!\!}&
\multicolumn{2}{l}{
\PM{10}{(\super{I}_2\otimes \super{F})\circ \super{G}\circ
(\super{I}_2\otimes \super{F})(\ketbra{10}{10})}=1,\!\!}\\
\vspace{-.1cm}
&&& \\
\multicolumn{2}{l}{
\PM{00}{(\super{F}\otimes\super{I}_2)\circ \super{G}^2\circ
(\super{F}\otimes\super{I}_2)(\ketbra{00}{00})}=1,\!\!}&
\multicolumn{2}{l}{
\PM{01}{(\super{F}\otimes\super{I}_2)\circ \super{G}^2\circ
(\super{F}\otimes\super{I}_2)(\ketbra{01}{01})}=1,\!\!}\\
\vspace{-.1cm}
&&& \\
\multicolumn{2}{l}{
\PM{00}{(\super{F}\otimes\super{F})\circ \super{G}\circ
(\super{F}\otimes\super{F})(\ketbra{00}{00})}=1.\!\!}&&
\end{array}}\right.
\end{equation*}
\end{theorem}
\noindent{\bf Proof: }
Let $\super{F}$ and $\super{G}$ satisfy these equations.
By \reftheo{general}, with $\alpha=\pi$
and $\theta=\pi/4$, the first three equations imply that
$\super{F}=\super{H}_\phi$, for some $\phi\in[0,2\pi)$.
Using \reflemma{lemma2}, the remaining equations imply
that $\super{G}^2=\super{I}_4$,
and it follows from \reflemma{lemma1c}
that $\super{G}$ is a unitary CPSO. A straightforward
verification then shows that indeed $\super{G}=\super{{\mathrm{c\text{-}NOT}}_\phi}$.
\qed
\section{Robustness}
In this section we introduce the notion of robustness for
experimental equations which will be the crucial ingredient for
proving self-testability. In this extended abstract we will deal
only with the case of experimental equations for one qubit and in
one variable. {From} now on $(E)$ will always denote a set of such
equations. Similar results can be obtained for several qubits and
several variables.
\begin{definition}
Let $\varepsilon,\delta\geq 0$, and let $(E)$ be
a finite satisfiable set of experimental equations.
We say that $(E)$ is {\em ($\varepsilon,\delta$)-robust}
if whenever a CPSO $\super{G}$
$\varepsilon$-satisfies $(E)$, we have
$\dist_{\infty}(\super{G},\mathcal{F}_{(E)})\leq\delta$.
\end{definition}
When a CPSO family is characterized by a
finite set of experimental equations $(E)$, one would like to prove
that $(E)$ is robust. The next theorem shows that this is always
the case.
\begin{theorem}\label{robustness}
Let $(E)$ be a finite satisfiable set of experimental
equations. Then there exists an integer $k\geq 1$ and a real $C>0$
such that for all $\varepsilon\geq 0$,
$(E)$ is ($\varepsilon,C\varepsilon^{1/k}$)-robust.
\end{theorem}
\noindent{\bf Proof: }
We will use basic notions from algebraic geometry for which we refer
the reader for example to~\cite{br90}.
In the proof, $\mathbb{C}$ is identified with $\mathbb{R}^2$. Then the set $K$ of
CPSOs for one qubit is a real compact semi-algebraic set. Suppose
that in $(E)$ there are $d$ equations. Let $f:K\rightarrow\mathbb{R}$ be
the function that maps the CPSO $\super{G}$ to the maximum of the
magnitudes of the difference between the probability term and the
constant term of the $i^{\mbox{\scriptsize th}}$ equation in
$(E)$, for $i=1,\ldots,d$. By definition of $f$, we get
$f^{-1}(0)=\mathcal{F}_{(E)}$. Moreover, $f$ is a continuous
semi-algebraic function, since it is the maximum of the magnitudes
of polynomial functions in the (real) coefficients of $\super{G}$.
Let $g:K\rightarrow\mathbb{R}$ defined in $\super{G}$ by
$g(\super{G})=\dist_{\infty}(\super{G},\mathcal{F}_{(E)})$. Since $K$ is a
compact semi-algebraic set, $g$ is a continuous semi-algebraic
function. Moreover, for all $\super{G}\in K$, we have
$f(\super{G})=0$ if and only if $g(\super{G})=0$. Then
\reffact{semi-alg} concludes the proof.
\qed
For a proof of the following fact,
see for example~\cite[Prop.~2.3.11]{br90}.
\begin{fact}[Lojasiewicz's inequality]\label{semi-alg}
Let $X\subseteq\mathbb{R}^m$ be a compact semi-algebraic set.
Let $f,g:X\rightarrow\mathbb{R}$ be continuous semi-algebraic functions.
Assume that for all $x\in X$, if
$f(x)=0$ then $g(x)=0$.
Then there exists an integer $k\geq 1$ and a real $C>0$ such that,
for all $x\in X$, $\abs{g(x)}^k\leq C\abs{f(x)}$.
\end{fact}
In some cases we can explicitly compute the constants $C$ and $k$
of \reftheo{robustness}.
We will illustrate these techniques with the equations in
\reftheo{general} for the case $\alpha=\pi$ and $\theta=\pi/4$. Let
us recall that these equations characterize the set $\mathcal{H}$.
\begin{theorem}\label{hadamardrob}
For every $0 \leq \varepsilon\leq 1$, the following equations are
($\varepsilon,4579\sqrt{\varepsilon}$)-robust:
\begin{eqnarray*}\label{approxhadamard}
\PM{0}{\super{G}(\ketbra{0}{0})}=\mbox{$\frac{1}{2}$},&
\PM{0}{\super{G}^2(\ketbra{0}{0})}=1,
\textrm{ and} &
\PM{0}{\super{G}^2(\ketbra{1}{1})}=0.
\end{eqnarray*}
\end{theorem}
The proof of this theorem will be given in \refap{APTH}.
\section{Quantum Self-Testers}
In this final section we define formally our testers
and establish the relationship between robust equations
and testability.
Again, we will do it here only for the case of one qubit and one
variable.
Let $\super{G}$ be a CPSO.
The {\em experimental oracle} $\oracle{\super{G}}$ for $\super{G}$
is a probabilistic procedure.
It takes inputs from $\{0,1\}\times\mathbb{N}$
and generates outcomes from the set $\{0,1\}$ such that
for every $k\in\mathbb{N}$,
\begin{eqnarray*}
\Pr[\oracle{\super{G}}(b,k)=0]
& = &
\PM{0}{\super{G}^{k}(\ketbra{b}{b})}.
\end{eqnarray*}
An oracle program $T$ with an experimental oracle
$\oracle{\super{G}}$ is a program denoted by
$T^{\oracle{\super{G}}}$ which can ask queries
from the experimental oracle
in the following sense:
when it presents a query $(b,k)$ to the oracle,
in one computational step it receives the probabilistic outcome of
$\oracle{\super{G}}$ on it.
\begin{definition}
Let $\mathcal{F}$ be a family of CPSOs, and
let $0\leq\delta_1\leq\delta_2<1$.
A {\em $(\delta_1,\delta_2)$-tester for} $\mathcal{F}$
is a probabilistic oracle program $T$
such that for every CPSO $\super{G}$,
\begin{itemize}
\item if $\dist_{\infty}(\super{G},\mathcal{F})\leq\delta_1$
then
$\Pr[T^{\oracle{\super{G}}}\mbox{ says { \tt PASS}}]\geq 2/3$,
\item if $\dist_{\infty}(\super{G},\mathcal{F})>\delta_2$ then
$\Pr[T^{\oracle{\super{G}}}\mbox{ says { \tt FAIL}}]\geq 2/3$,
\end{itemize}
where the probability is taken over the probability distribution of the
outcomes
of the experimental oracle and the internal coin tosses of the program.
\end{definition}
\begin{theorem}\label{robtotester}
Let $\varepsilon,\delta>0$, and let $(E)$ be a satisfiable set of $d$ experimental
equations such that the size of every equation is at most $k$.
If $(E)$ is ($\varepsilon,\delta$)-robust then there exists an
$(\varepsilon/(3k),\delta)$-tester for $\mathcal{F}_{(E)}$
which makes $O(d\ln (d)/\varepsilon^2)$ queries.
\end{theorem}
\textsc{Sketch of proof.}
We will describe a probabilistic oracle program $T$. Let
$\super{G}$ be a CPSO. We can suppose that for every equation in
$(E)$, $T$ has a rational number $\tilde{r}$ such that
$\abs{\tilde{r}-r}\leq\varepsilon/6$, where $r$ is the constant term of
the equation. By sampling the oracle $\oracle{\super{G}}$, for
every equation in $(E)$, $T$ obtains a value $\tilde{p}$ such that
$\abs{\tilde{p}-p}\leq\varepsilon/6$ with probability at least $1-1/(3d)$,
where $p$ is the probability term of the equation. A standard
Chernoff bound argument shows that this is feasible with
$O(\ln(d)/\varepsilon^2)$ queries for each equation. If for every equation
$\abs{\tilde{p}-\tilde{r}}\leq 2\varepsilon/3$, then $T$ says {\tt PASS},
otherwise $T$ says {\tt FAIL}. Using the robustness of $(E)$ and
\reflemma{lemmar}, one can verify that $T$ is a
$(\varepsilon/(3k),\delta)$-tester for $\mathcal{F}_{(E)}$.\qed
\begin{lemma}\label{lemmar}
Let $(E)$ be a finite satisfiable set of experimental
equations such that the size of every equation is at most $k$,
and let $\super{G}$ be a CPSO. For every $\varepsilon\geq 0$,
if $\dist_{\infty}(\super{G},\mathcal{F}_{(E)})\leq\varepsilon$
then $\super{G}$ ($k\varepsilon$)-satisfies $(E)$.
\end{lemma}
Our main result is the consequence of Theorems \ref{general},
\ref{couple}, \ref{cnot},
\ref{robustness}, \ref{hadamardrob}, \ref{robtotester},
and the many-qubit generalizations of them.
\begin{theorem}\label{maintheo}
Let $\mathcal{F}$ be one of the following families~:
\begin{itemize}
\item $\mathcal{R}_{\alpha,\theta}\ $ for
$(\alpha,\theta)\in(0,\pi]\times(0,\pi/2]\backslash\{(\pi,\pi/2)\}$
where $\alpha/\pi$ is rational,
\item $\{(\super{H}_\phi,\super{\neg}_{\phi}) : \phi\in[0,2\pi)\}$,
\item $\mathcal{H}\times\{\super{R}_{\pm\alpha}\}\ $ for $\alpha/\pi$
rational,
\item $\{(\super{H}_\phi,\super{{\mathrm{c\text{-}NOT}}}_{\phi})
: \phi\in[0,2\pi)\}$,
\item $\{(\super{H}_\phi,\super{R}_{s\pi/4},\super{{\mathrm{c\text{-}NOT}}}_{\phi})
: \phi\in[0,2\pi),s=\pm 1\}$.
\end{itemize}
Then there exists an integer $k\geq 1$ and a real $C>0$ such that,
for all $\varepsilon>0$, $\mathcal{F}$ has an $(\varepsilon,C\varepsilon^{1/k})$-tester
which makes $O(1/\varepsilon^2)$ queries.
Moreover, for every
$0< \varepsilon\leq 1$, $\mathcal{H}$ has an $(\varepsilon/6,4579\sqrt{\varepsilon})$-tester
which makes $O(1/\varepsilon^2)$ queries.
\end{theorem}
Note that each triplet of the last family forms a
universal and fault-tolerant set of quantum gates\cite{bmprv99}.
\section{Acknowledgements}
We would like to thank Jean-Benoit Bost,
St\'ephane Boucheron, Charles Delorme,
St\'ephane Gonnord,
Lucien Hardy, Richard Jozsa, and Vlatko Vedral for several
useful discussions and advice.
This work has been supported by C.E.S.G., Wolfson College Oxford,
Hewlett-Packard, European TMR Research Network ERP-4061PL95-1412,
the Institute for Logic, Language and Computation in Amsterdam,
ESPRIT Working Group RAND2 no. 21726, NSERC, British-French
Bilateral Project ALLIANCE no. 98101, and the
Quantum Information Theory programme of the European Science Foundation.
|
\section{Introduction}
The main goal of the present paper is two-fold. First we extend
the theory of toroidal embeddings introduced by Kempf,
Knudsen, Mumford, and Saint-Donat to the class of toroidal
varieties with stratifications. Second we give a proof of the following weak factorization conjecture as an application and illustration of the theory.
\begin{conjecture} {\bf The Weak Factorization conjecture}
\begin{enumerate}
\item Lef $f:X{{-}{\to}} Y$ be a
birational map of
smooth complete varieties
over an algebraically closed field of characteristic zero, which is an
isomorphism over an open set $U$. Then $f$ can be factored
as
$$X=X_0\buildrel f_0 \over {{-}{\to}} X_1
\buildrel f_1 \over {{-}{\to}} \ldots \buildrel f_{n-1} \over
{{-}{\to}} X_n=Y ,$$
where each $X_i$ is a smooth complete variety and $f_i$ is a blow-up
or blow-down at a smooth center which is an isomorphism
over $U$.
\item Moreover, if $X\setminus U$ and $Y\setminus U$ are divisors
with normal crossings, then each $D_i:=X_i\setminus U$
is a divisor with normal crossings and $f_i$ is a blow-up
or blow-down at a smooth center which has normal
crossings with components of $D_i$.
\end{enumerate}
\end{conjecture}
This theorem extends a theorem of Zariski, which states that
any birational map between two smooth complete
surfaces can be factored into a succession of blow-ups at
points followed by a succession of blow-downs at points.
A stronger version of the above theorem, called the strong factorization
conjecture remains open.
\begin{conjecture} {\bf Strong factorization conjecture}.
Any birational map $f:X\mathrel{{-}\,{\to}}Y$ of smooth
complete varieties can be factored into a succession of blow-ups
at
smooth centers followed by a succession of
blow-downs at
smooth centers.
\end{conjecture}
One can find the statements of both conjectures in many papers. Hironaka \cite{Hironaka1} formulated the strong
factorization conjecture. The weak factorization
problem was stated by Miyake and Oda \cite{Miyake-Oda}. The toric versions of the
strong and weak factorizations were also conjectured by
Miyake and Oda \cite{Miyake-Oda} and are called the strong and weak Oda
conjectures. The $3$-dimensional toric version of the weak form was
established by Danilov \cite{Danilov2} (see also Ewald \cite{Ewald}).
The weak toric conjecture
in arbitrary dimensions was proved in \cite{Wlodarczyk1} and later
independently by Morelli \cite{Morelli1}, who proposed a proof of the
strong factorization conjecture
(see also Morelli\cite{Morelli2}).
Morelli's
proof of the weak Oda conjecture was completed, revised and
generalized to the toroidal case by Abramovich, Matsuki and Rashid
in \cite{Abramovich-Matsuki-Rashid}. A gap in Morelli's proof of the strong Oda conjecture,
which was
not noticed in \cite{Abramovich-Matsuki-Rashid}, was later
found by K. Karu.
The local version of the strong factorization problem was posed
by Abhyankar in dimension 2 and by Christensen in general,
who has solved it for 3-dimensional toric varieties \cite{Christensen}.
The local version of the weak factorization problem (in
characteristic 0) was
solved by Cutkosky in \cite{Cutkosky1}, who also
showed that Oda's strong conjecture implies the local
version of the strong conjecture for proper\ birational
morphisms \cite{Cutkosky2} and proved the local strong factorization conjecture
in dimension 3 (\cite{Cutkosky2}) via Christensen's theorem.
Shortly after the present proof was found, another proof
of the weak factorization conjecture was conceived
by
Abramovich, Karu, Matsuki and the author \cite{AKMW}. Unlike the proof in \cite{AKMW} the present proof does not refer to the weak factorization theorem
for toric varieties.
Another
application of the theory of stratified toroidal varieties is
given in Section \ref{se: resolution}, where we show
the existence of a
resolution of
singularities of toroidal varieties in arbitrary characteristic by
blowing up ideals determined by
valuations (see Theorem \ref{th: resolution}).
All schemes and varieties in the present paper are
considered over an algebraically closed field $K$.
The assumption of characteristic 0 is needed for the
results in Sections \ref{se: construction} and \ref{se: factorization} only, where we use Hironaka's canonical
resolution of singularities and canonical principalization for the second part of the theorem (see Hironaka \cite{Hironaka1},
Villamayor \cite{Villamayor} and Bierstone-Milman \cite{Bierstone-Milman}).
\section{Main ideas}\label{se: main ideas}
\subsection{Toroidal embeddings} \label{se: toroidal embeddings}
The theory of toroidal embeddings was introduced and developed
by Kempf, Knudsen, Mumford and Saint-Donat in \cite{KKMS}.
Although it was originally conceived for the purpose of
compactifying symmetric spaces and certain moduli spaces it has been successfully applied to
various problems concerning resolution of singularities
and factorization of morphisms (see
\cite{KKMS}, \cite{Abramovich-de-Jong}, \cite{Abramovich-Karu}). It is an
efficient tool in many situations, turning complicated or
even "hopeless" algebro-geometric problems into relatively "easy"
combinatorial problems. Its significance lies in
a simple combinatorial description of
the varieties considered carrying rich and precise information about
singularities and stratifications.
By {\it a
toroidal embedding} (originally a
toroidal embedding without self-intersections) we mean a variety $X$ with an open subset
$U$ with the following property: for any $x\in X$ there is an open
neighborhood $U_x$ and an \'etale morphism
$\phi : U_x\to X_{\sigma_x}$ into a toric variety $X_{\sigma_x}$
containing a torus $T$ such that $U_x\cap U=\phi^{-1}(T)$.
Such a morphism is called a {\it chart}.
Let $D_i, i\in I$, be the irreducible
components of the complement divisor $D=X\setminus U$. We
define a stratification $S$ on $X$
with strata $s\in S$ such that either $\overline{s}$ is an
irreducible component of $\bigcap_{i\in
J} D_i$ for some $J\subset I$ or $\overline{s}=X$. The strata are naturally ordered
by the following relation: $s\leq s'$ iff the closure $\overline{s}$
contains $s'$.
In particular
$s=\overline{s}\setminus\bigcup_{s'>s}\overline{s'}$. Any
chart $\phi : U_x\to X_{\sigma_x}$ preserves strata: as a
matter of fact, all strata $s\subset U_x$ for $s\in S$ are
preimages of $T$-orbits on $X_{\sigma_s}$. In fact a toroidal
embedding can be defined as a stratified variety such that
for any point $x\in X$ there exists a neighborhood $U_x$
of $x$ and an \'etale morphism $\phi : U_x\to X_{\sigma_x}$
preserving strata.
The cone ${\sigma_x}$ in the above definition can be
described using Cartier divisors. Let $x$ belong to a
stratum $s\in S$.
We associate with $s\in S$
the following data (\cite{KKMS}):
${\bf M}^s$: the group of Cartier divisors in $
\operatorname{Star}(s,S):=\bigcup_{s'\leq s}{s'}$, supported in $\operatorname{Star}(s,S)\setminus U$,
${\bf N}_s:={\rm Hom}({\bf M}_s,{\mathbb{Z}})$,
${\bf M}_+^s\subset {\bf M}^s$: effective Cartier divisors,
$\sigma_s\subset {\bf N}_{\mathbb{R}}^s$: the dual of ${\bf M}_+^s$.
This gives a correspondence
between strata $s\in S$ and cones. If $s'\leq s$ then
$\sigma_{s'}$ is a face of $\sigma_{s}$.
By glueing cones $\sigma_s$ along their subfaces we
construct a {\it
conical complex} $\Sigma$. In contrast to in the standard theory of
complexes two faces might share a few common faces, not
just one.
For any stratum $s\in S$ the integral vectors in the relative interior of
$\sigma_s$ define monomial valuations on $X_{\sigma_s}$.
These valuations induce, via
the morphisms $\phi$ monomial
valuations centered at $\overline{s}$ (independent of charts).
The natural class of
morphisms of toroidal embeddings are called
{\it toroidal morphisms} and are exactly those which are locally
determined by charts $\phi$ and toric morphisms:
A birational morphism of toroidal
embeddings $f: (Y,U) \to (X, U)$ is {\it toroidal} if
for any $x\in s\subset X$ there exists a chart $x\in U_x\to
X_{\Delta^\sigma}$ and a subdivision
$\Delta^\sigma$ of $\sigma_s$ and a
fiber square of morphisms of stratified varieties
\[\begin{array}{rcccccccc}
&& & (U_x,U_x\cap U) & \longrightarrow & (X_{\sigma_s},T)&&&\\
&&&\uparrow f & & \uparrow &&& \\
U_x \times_{X_{\sigma_s}}
X_{\Delta^\sigma} &&\simeq& (f^{-1}(U_x),f^{-1}(U_x)\cap U)
& \rightarrow &
(X_{\Delta^\sigma} , T)&&&
\end{array}\]
Note that
toroidal morphisms are well defined and do not depend upon
the choice of the charts.
This fact can be described nicely using
the following Hironaka condition:
For any points $ x, y$ which are in the same stratum every
isomorphism $ \alpha : \widehat{X}_x\to
\widehat{X}_y $ preserving stratification
can be lifted
to an isomorphism $\alpha':Y\times_X\widehat{X}_x
\to Y\times_X\widehat{X}_y$ preserving stratification.
Birational toroidal morphisms with fixed target are in bijective correspondence with subdivisions of
the associated complex.
\subsection{Stratified toroidal varieties}
\label{sse: stratified toroidal}
A more general class of objects occurring naturally in
applications is the class of stratified toroidal varieties
defined as follows.
A {\it stratified toric variety} is a toric variety with an
invariant equisingular stratification. The important
difference between toric varieties and
stratified toric varieties is that the latter come with a
stratification which may be coarser than the one given by
orbits. As a consequence, the
combinatorial object associated to a stratified toric
variety, called a {\it semifan}, consists of those faces of
the fan of the toric variety
corresponding to strata. The faces which do not
correspond to strata are
ignored (see Definition \ref{de: embedded}).
By a {\it stratified toroidal variety} we mean a stratified
variety $(X,S)$ such that for any $x\in s$, $s\in S$ there
is an \'etale map called a {\it chart}
$\phi_x:U_x\to X_{\sigma}$
from an open neighborhood to a stratified toric variety such that all strata in $U_x$ are preimages of
strata in $X_{\sigma}$ (see Definition
\ref{de: stratified toroidal}). A toroidal embedding is a particular
example of a stratified toroidal variety if we consider the
stratification described in section \ref{se: toroidal embeddings}
(see also Definition \ref{de: toroidal embeddings}).
Each chart associates with a
stratum $s$ the {\it semicone} $\sigma$ i.e the semifan consisting of
faces of $\sigma$
corresponding to strata in ${\rm Star(s,S)}$ for the
corresponding stratum $s$.
If one chart associates to a stratum a semicone $\sigma_1$, and
another associates $\sigma_2$ then Demushkin's theorem says that
$\sigma_1\simeq\sigma_2$.
This gives us the
correspondence between strata and cones with semifans.
Consequently, if $s\leq s'$, then
$\sigma$ is a face of $\sigma'$ and the semicone
$\sigma$ is a subset of the semicone
$\sigma'$. By glueing
the semicones $\sigma$ and $\sigma'$ along the common faces
$\sigma''$
we obtain the so called {\it semicomplex} $\Sigma$
associated to the stratified toroidal variety. The
semicomplex is determined uniquely up to isomorphism.
Its faces of
the semicomplex are indexed by the stratification. If a
stratified toroidal variety is a toroidal embedding then
the associated semicomplex is the usual conical complex.
Note that in contrast in the case of toroidal
embeddings only some integral vectors $v$ in
faces of this semicomplex determine unique valuations on $X$ which
do not depend upon a chart. These vectors are called {\it stable}
and the valuations they induce are also called {\it stable}
(see Definitions \ref{de: stable valuations}, \ref{de: stable vectors}). The stable vectors form a subset in
the support of the semicomplex,
which we call the {\it stable support} (see Definition \ref{de:
stable support}).
In particular, let $X$ be a smooth surface and $S$ be its
stratification consisting of a point $p\in X$ and its complement. The associated
semicomplex consists of the regular cone ${\bf Q}_{\geq 0}\cdot
e_1+{\bf Q}_{\geq 0}\cdot e_2$ corresponding to the point $p$ and of the
vertex corresponding to a big stratum. The only stable
valuation is the valuation of the point $p$ which corresponds to the
unique (up to proportionality) stable vector $e_1+e_2$.
The notion of stable support plays a key role in the theory of stratified
toroidal varieties. The stable support has many nice properties.
It intersects the relative interiors of all faces of
the semicomplex and this intersection is polyhedral. The relative interiors of faces which are not
in the semicomplex do not intersect the stable support.
Consequently, all stable valuations
are centered at the closures of strata. The part of the stable support
which is contained in a face of the semicomplex is
convex and is a union of relative interiors of a finite
number of polyhedral cones. Also some ''small'' vectors
like those which are in the parallelogram determined by ray
generators belong to the stable support. If we consider the
canonical resolution of toric singularities then all divisors
occurring in this resolution determine stable valuations.
In section \ref{se: comparison}, where we discuss another
approach to the weak factorization conjecture in \cite{AKMW}, we also show a method of
computing the stable support which is based on the idea of
torification of Abramovich and de Jong \cite{Abramovich-de-Jong}.
The stability condition for vectors in a face $\sigma$ of
the semicomplex $\Sigma$ can be expressed in
terms of invariance under the action of the group $G^{\sigma}$ of
automorphisms of a scheme $\widehat{X}_\sigma$
associated to this face (see Definition \ref{de: stable
valuations}) (see
section \ref{se: -semicomplexes}).
The vector $v$ is stable if the valuation determined by this
vector on $\widetilde{X}_\sigma$ is $G^{\sigma}$-invariant.
(Definition \ref{de: stable vectors}).
Birational toroidal mophisms or simply {\it toroidal
morphisms} between stratified toroidal varieties
are those induced locally by toric morphisms via toric
charts and satisfying the Hironaka condition (see Definition
\ref{de: canonical stratification}).
The Hironaka condition implies that toroidal morphisms, which
are a priori induced by toric charts, in fact do not depend upon the charts
and depend eventually only on the subdivisions of the
semicomplex associated to the original stratified toroidal variety.
This leads to a $1-1$ correspondence between toroidal
morphisms and certain subdivisions,
called {\it canonical}, of the semicomplex associated to the
original variety
(see Theorems \ref{th: modifications} and \ref{th: correspondence}).
The canonical subdivisions are
those satisfying the following
condition: all "new" rays of the subdivision are in the stable
support (see Proposition
\ref{pr: simple}). It is not difficult to see the necessity
of the condition since the "new" rays correspond to
invariant divisors which determine invariant valuations.
It is more complicated to show the sufficiency of the
condition.
Let $(Y,R)\to (X,S)$ be a toroidal morphism of stratified
toroidal varieties associated to the canonical subdivision
$\Delta$ of the semicomplex $\Sigma$ associated to
$(X,S)$. Then the semicomplex $\Sigma_R$ associated to
$(Y,R)$ consists of all faces in $\Delta$ whose relative
interiors intersect the stable support of $\Sigma$ (see
Theorem \ref{th: correspondence}).
Toroidal morphisms between stratified toroidal varieties
generalize those defined for toroidal embeddings. In that
case, all subdivisions are canonical and we get the
correspondence between toroidal embeddings and subdivisions
of associated semicomplexes mentioned in Section
\ref{se: toroidal embeddings} above.
In our considerations we
require some condition, called orientability, of compatibility of charts on
stratified toroidal varieties. This condition is
analogous to the orientability of charts on an oriented
differentiable manifold. It says that two \'etale charts
$\phi: U\to X_{{\sigma_s}}$ and $\phi': U\to X_{{\sigma_s}}$
define an automorphism
$\widehat{\phi}'\circ\widehat{\phi}^{-1}$ in
the identity component of the group of automorphisms of
$\operatorname{Spec}(\widehat{{\mathcal O}}_{X_{\sigma}})$
(see Definition \ref{de:
orientation}). In the differential setting, the group of
local automorphisms consists of two components. In the
algebraic situation the number of components depends upon
the singularities. It is finite for toric singularities.
Smooth toroidal varieties are oriented, since
the group of automorphisms of the completion of a regular local
ring is connected.
Let $X$ be a toroidal variety with isolated
singularity $x_1\cdot x_2=x_3\cdot x_4$ or equivalently
after some change of coordinates $y_1^2+y_1^2+
y_3^2+y_4^2=0$. In this case the group of automorphisms of
the
formal completion of $X$ at the singular point consists of two
components. Consequently,
there are two orientations on $X$ (for more details see Example
\ref{ex: isolated}).
Toroidal embeddings are oriented stratified toroidal
varieties (see Example \ref{ex: toroidal embeddings}).
Stratified toroidal varieties are orientable
(see Section \ref{se: existence}).
\subsection{Toroidal varieties with torus action}
Let $\Gamma$ denote any algebraic subgroup of an algebraic
torus $T:=K^*\times\ldots\times K^*$.
If $\Gamma$ acts on a stratified toric variety via a group
homomorphism $\Gamma \to T$, then we require
additionally, that the stabilizers at closed
points in a fixed stratum $s$ are all the same and equal to a group
$\Gamma_s$ associated to $s$, and there is a
$\Gamma_s$-equivariant isomorphism of the local rings at these points.
A {\it $\Gamma$-stratified toroidal variety}
is a stratified toroidal variety with a $\Gamma$-action which
is locally $\Gamma$-equivariantly \'etale isomorphic to a
$\Gamma$-stratified toric variety. We associate with it a
$\Gamma$-semicomplex which is the semicomplex
$\Sigma$
with
associated groups $\Gamma_\sigma\subset \Gamma$ acting on $X_{\sigma}$.
$\Gamma$-semicomplexes determine local projections
$\pi:\sigma\to\sigma^\Gamma$ defined by good
quotients $X_{\sigma_s}\to
X_{\sigma_s^\Gamma}:=X_{\sigma}/\Gamma_\sigma$. These local
projections are coherent in the sense that they commute
with face restrictions and subdivisions.
All the definitions in this category are adapted from the
theory of stratified toroidal varieties. We additionally require that
charts, morphisms and automorphisms occurring in the definitions
are $\Gamma_\sigma$ or $\Gamma$-equivariant.
\subsection{Birational cobordisms, Morse theory and
polyhedral cobordisms of Morelli} The theory of birational
cobordisms, which
was developed in \cite{Wlodarczyk2},
was inspired by Morelli's proof of the weak
factorization theorem for toric varieties,
where the notion of combinatorial cobordism was introduced (see
\cite{Morelli1}).
By a {\it birational cobordism} between birational
varieties $X$ and $X'$ we understand a variety $B$
with the action of the multiplicative group $K^*$ such that
the "lower boundary" $B_-$ of $B$ (resp. the "upper boundary" $B_+$)
is an open subset which consists of orbits with no limit at
$0$ (resp. $\infty$).
The cobordant varieties $X$ and $X'$ are isomorphic
to the spaces of orbits (geometric quotients) $B_-/K^*$ and
$B_+/K^*$ respectively. In differential cobordism the
action of the $1$-parameter group of diffeomorphisms $G\simeq
({\bf R},+)\simeq ({\bf R^*},\cdot)$, defined by
the gradient field of a Morse function gives us an
analogous interpretation of Morse theory. The bottom and top
boundaries determined by a Morse function are isomorphic to
the spaces of all orbits with no limit at $-\infty$ and
$+\infty$ respectively (or $0$ and $+\infty$
in multiplicative notation). The critical points of the
Morse function are the fixed points of the action. ''Passing
through'' these points induces simple birational
transformations analogous to spherical transformations in
differential geometry. If $B$ is a smooth cobordism, then
these birational transformations locally replace one weighted
projective space with a complementary weighted projective
space. These birational transformations can be nicely described
using the language of toric varieties and associated fans.
The semiinvariant parameters in the neighbouhood of a fixed
point provide a chart which is a locally analytic
isomorphism of the tangent
spaces with the induced linear action. Hence locally, the
cobordism is isomorphic to an affine space with a linear
hyperbolic action. Locally we can identify $B$ with an
affine space ${\bf A}^n$, which is a toric variety
corresponding to a regular cone $\sigma$ in the vector space
$N^{\bf Q}$. The action of $K^*$ determines a $1$-parameter
subgroup, hence an integral vector $v\in N^{\bf Q}$. The sets $B_-$
and $B_+$
correspond to the sets $\sigma_+$ and $\sigma_-$ of all
faces of $\sigma$ visible from above or
below (with respect to the direction of $v$), hence to the
''upper'' and ''lower'' boundaries of $\sigma$. The vector $v$
defines a projection $\pi: N^{\bf Q}\to N^{\bf Q}/{\bf Q}v$. The quotient spaces
$B_-/K^*$, $B_+/K^*$ correspond to two ''cobordant'' (in the
sense of Morelli) subdivisions
$\pi(\sigma_+)$ and $\pi(\sigma_-)$ of the cone $\pi(\sigma)$.
The problem lies in the fact that the fans $\pi(\sigma_+)$,
$\pi(\sigma_-)$ are singular (which means that they are not spanned by a
part of an integral basis). Consequently, the corresponding
birational transformation is a composition of weighted
blow-ups and blow-downs between singular varieties.
The process of resolution of
singularities of the quotient spaces is called
$\pi$-desingularization and can be achieved locally by
a combinatorial algorithm. The difficulty is now in patching
these local $\pi$-desingularizations. This problem can be
solved immediately once the theory of stratified toroidal
varieties (with $K^*$-action) is established.
\subsection{Sketch of proof}
Let $B=B(X,X')$ be a smooth cobordism between smooth
varieties $X$ and $X'$. Let $S$ be the stratification determined by
the isotropy groups. Then $(B,S)$ is
a stratified toroidal variety with $K^*$-action. We can
associate with it a $K^*$-semicomplex $\Sigma$.
The $K^*$-semicomplex $\Sigma$ determines for any
face ${{\sigma_s}}\in\Sigma$ a projection
$\pi_\sigma: \sigma\to \sigma^\Gamma$ determined by the good
quotients
$X_{\sigma}\to X_{\sigma}/\Gamma_\sigma=X_{\sigma}^\Gamma$.
We apply a combinatorial algorithm - the so
called $\pi$-desingularization Lemma of Morelli - to
the semicomplex associated to the cobordism. The algorithm
consists of starf subdivisions at stable vectors. Such
subdivisions are canonical. As a result we obtain a
semicomplex with the property that the projections of all
simplices are either regular (nonsingular) cones or cones
of smaller dimension (than the dimension of the projected cones)
(see Lemma \ref{le: pi-lemma} for details).
The semicomplex corresponds to a $\pi$-regular toroidal
cobordism, all of whose
open affine fixed point free subsets have smooth geometric
quotients. The existence of such a cobordism
easily implies the weak factorization theorem. (The
blow-ups, blow-downs and flips induced by elementary
cobordisms are regular (smooth)). Since each flip is a
composition of a blow-up and a blow-down at a smooth center we
come to a factorization of the map $X{-}{\to}X'$ into
blow-ups and blow-downs at smooth centers (see Proposition
\ref{pr: regular factorization}).
\section{Preliminaries}
\subsection{ Basic notation and terminology}
Let $N\simeq
{\bf Z}^k$ be a lattice contained in
the vector space $N^{\bf Q}:=N\otimes {\bf
Q}\supset N$.
By a {\it cone} in this paper
we mean a convex set $\sigma = {\bf Q}_{\geq 0}\cdot
v_1+\ldots+{\bf Q}_{\geq 0}\cdot v_k\subset N^{\bf Q}$.
By abuse of language we shall speak of a {\it cone $\sigma$ in a
lattice $N$}. To avoid confusion we
shall sometimes write $(\sigma, N)$ for the cone $\sigma$
in $N$. For a cone $\sigma$ in $N$ denote by $N_\sigma:=N\cap\operatorname{lin}(\sigma)$ the
sublattice generated by $\sigma$. Then by
$$\underline{\sigma}:=(\sigma,N_\sigma)$$
we denote the corresponding cone in $N_\sigma$. (Sometimes the cones $\sigma$ and
$\underline{\sigma}$ will be identified).
We call a vector $v\in N$ {\it primitive} if it generates
sublattice ${\bf Q}_{\geq 0}v_i\cap N$.
Each ray $\rho\in N^{\bf Q}$ contains a unique primitive
ray ${\operatorname{prim}}(\rho)$.
If $v_1,\ldots,v_k$ form a
minimal set of primitive vectors generating $\sigma\subset N^{\bf Q}$
in the above sense, then we write $$\sigma =\langle
v_1,\ldots,v_k\rangle .$$ If
$\sigma$ contains no line we call it {\it strictly convex}. All
cones considered in this paper are strictly convex. For any
$\sigma$ denote by ${\rm lin}(\sigma)$ the linear span of $\sigma$. For any
cones $\sigma_1$ and $\sigma_2$ in $N$ we write $$\sigma
=\sigma_1+ \sigma_2$$ if $\sigma=\{v_1+v_2\mid v_1\in
\sigma_1 ,v_2\in
\sigma_2\}$, and
$$\sigma =\sigma_1\oplus \sigma_2$$ if
${\rm lin}(\sigma_1)\cap {\rm lin}(\sigma_2)=\{0\}$ and for any $v\in
\sigma\cap N$ there exist $v_1\in \sigma_1\cap N$ and
$v_2\in \sigma_2\cap N$ such that $v=v_1+v_2$.
For any
cones $\sigma_1$ in $N_1$ and $\sigma_2$ in $N_2$ we
define the cone $\sigma_1\times\sigma_2$ in $N_1\times N_2$
to be
$$\sigma_1\times\sigma_2:=\{(v_1,v_2)\mid v_i\in\sigma_i\quad \mbox{for} \quad i=1,2 \}.$$
We say that a cone
$\sigma$ in $N$ is {\it regular} if there exist vectors
$e_1,..,e_k\in N$ such that
$$\sigma =\langle e_1\rangle \oplus \ldots\oplus \langle e_k\rangle. $$
A cone $\sigma$ is {\it simplicial} if
$\sigma =\langle v_1,\ldots,v_k\rangle $ is generated by linearly
independent vectors.
We call $\sigma$
{\it indecomposable} if it cannot be
represented as $\sigma =\sigma'\oplus \langle e \rangle $ for
some nonzero vector $e\in N$.
\begin{lemma}
Any cone $\sigma$ in $N$ is uniquely
represented as $$\sigma ={\rm sing}(\sigma )\oplus \langle e_1,\ldots,e_k\rangle,
$$
where $\langle e_1,\ldots,e_k\rangle $ is a regular cone and
${\rm sing}(\sigma )$ is the maximal indecomposable face of
$\sigma$.\qed
\end{lemma}
By ${\operatorname{int}}(\sigma)$ we denote the relative interior of $\sigma$.
For any simplicial cone
$\sigma =\langle v_1,\ldots,v_k\rangle $ in $N$ set $${\rm par}(\sigma ):=\{ v\in
\sigma\cap N_{\sigma}\mid v=\alpha_1v_1+\ldots+\alpha_kv_k,
\mbox{where}\,\, 0\leq\alpha_i< 1\},$$
$$\overline{{\rm par}(\sigma )}:=\{ v\in
\sigma\cap N_{\sigma}\mid v=\alpha_1v_1+\ldots+\alpha_kv_k,
\mbox{where}\,\, 0\leq\alpha_i\leq 1\}.$$
For any simplicial cone $\sigma =\langle v_1,\ldots,v_k\rangle $ in $N$, by
$\det(\sigma) $ we mean $\det(v_1,\ldots,v_k)$, where all
vectors are considered in some
basis of $N\cap {\rm lin}(\sigma)$ (a change
of basis can only change the sign of the determinant).
\begin{definition} \label{de: minimal}
A {\it minimal
generator} of a cone $\sigma$ is a vector not contained in a
one dimensional face of $\sigma$ and which cannot be
represented as the sum of two nonzero integral vectors in
$\sigma$. A {\it minimal
internal vector} of a cone $\sigma$ is a vector in ${\rm
int}(\sigma)$ which cannot be
represented as the sum of two nonzero integral vectors in
$\sigma$ such that at least one of them belongs to ${\rm
int}(\sigma)$.
\end{definition}
Immediately from the definition we get
\begin{lemma} \label{le: minimal0} Any minimal generator $v$ of $\sigma$
is a minimal internal vector of the face
$\sigma_v$ of $\sigma$, containing $v$ in its relative interior.\qed
\end{lemma}
\begin{lemma} For any simplicial $\sigma$ each vector
from ${\rm par}(\sigma)$ can be represented as a nonnegative
integral combination of minimal generators.
\end{lemma}
\qed
\subsection{Toric varieties}
\begin{definition} (see \cite{Danilov1},
\cite{Oda}).\label{de: fan} By a {\it fan}
$\Sigma $ in
$N$ we mean a finite collection of finitely
generated strictly convex cones $\sigma$ in $N$ such
that
$\bullet$ any face of a cone in $\Sigma $ belongs to $\Sigma$,
$\bullet$ any two cones of $\Sigma $ intersect in a common face.
By the {\it support} of the fan we mean the union of all
its faces,
$|\Sigma|=\bigcup_{\sigma\in \Sigma}\sigma$.
If $\sigma$ is a face of $\sigma'$ we shall write $\sigma\preceq\sigma'$.
If $\sigma\preceq\sigma'$ but $\sigma\neq\sigma'$
we shall write $\sigma\prec\sigma'$.
For any set of cones $\Sigma$ in $N$ by $\overline{\Sigma}$
we denote the set $\{\tau \mid \tau\prec \tau'\quad \mbox{for some}\quad\tau'\in\Sigma\}$
\end{definition}
\begin{definition}\label{de: star} Let $\Sigma$ be a fan
and $\tau \in \Sigma$. The {\it star} of the
cone $\tau$ and the {\it closed star} of $\Sigma$ are
defined as follows:
$${\rm Star}(\tau ,\Sigma):=\{\sigma \in \Sigma\mid
\tau\preceq \sigma\},$$
$$\overline{{\rm Star}}(\tau ,\Sigma):=\{\sigma \in
\Sigma\mid \sigma'\preceq \sigma \, \mbox{for some} \, \sigma'\in
{\rm Star}(\tau ,\Sigma)\},$$
\end{definition}
\begin{definition}\label{de: product} Let $\Sigma_i$ be a
fan in $N_i$ for $i=1,2$. Then the {\it product} of
$\Sigma_1$ and $\Sigma_2$ is a fan $\Sigma_1\times\Sigma_2$
in $N_1^{\bf Q}\times N_2^{\bf Q}$ defined as follows:
$$\Sigma_1\times\Sigma_2:=\{\sigma_1\times
\sigma_2 \mid \sigma_1\in\Sigma_1, \sigma_2\in\Sigma_2\}.$$
\end{definition}
To a fan $\Sigma $ there is associated a toric variety
$X_{\Sigma}\supset {\bf }T$, i.e. a normal variety on which a torus
$T$ acts effectively with an open dense orbit
(see \cite{KKMS}, \cite{Danilov2}, \cite{Oda}, \cite{Fulton}). To
each cone $\sigma\in \Sigma$
corresponds an open affine invariant subset
$X_{\sigma}$ and its unique closed orbit $O_{\sigma}$. The
orbits in the
closure of the orbit $O_\sigma$ correspond to the cones of
${\rm Star}(\sigma ,\Sigma)$.
Denote by $$M:={\rm Hom}_{alg.gr.}(T,K^*)$$ \noindent the lattice of
group homomorphisms to $K^*$, i.e. characters of $T$. Then
the dual lattice $N=
{\rm Hom}_{alg.gr.}(K^*,T)$ can be
identified with the lattice of $1$-parameter subgroups of $T$. Then
vector space $M^{\bf Q}:=M\otimes{\bf Q}$ is dual to $N^{\bf Q}:=
N\otimes{\bf Q}$. Let $(v,w)$ denote the relevant pairing for $v\in N,w\in M$
For any $\sigma\subset N^{\bf Q}$ we denote by
$$\sigma^\vee:=\{m\in M \mid (v,m) \geq 0 \,\, {\rm
for\,\,
any} \,\,\, m\in \sigma\}$$
\noindent the set of integral vectors of the dual cone to $\sigma$. Then the
ring of regular functions $K[X_\sigma]$ is $K[\sigma^\vee]$.
Each vector $v\in N$ defines a linear function on $M$ which
determines a
valuation ${\rm val}(v)$ on
$X_{\Sigma}$.
For any regular function $f=\sum_{w\in M} a_wx^w\in K[T]$ set
$${\operatorname{val}}(v)(f):=\min\{(v,w)\mid a_w\neq 0\}.$$
Thus $N$
can be perceived as the lattice of all $T$-invariant
integral valuations of the function field of $X_{\Sigma}$.
For any $\sigma\subset N^{\bf Q}$ set
$$\sigma^\perp:=\{m\in M \mid (v,m) = 0 \,\, {\rm
for\,\,
any} \,\,\, m\in \sigma\}.$$
\noindent The latter set represents all invertible characters
on $X_\sigma$. All noninvertible characters are $0$ on $O_\sigma$.
The ring of regular functions on $O_\sigma\subset X_\sigma$ can be
written as $K[O_\sigma]=K[\sigma^\perp]\subset
K[\sigma^\vee]=K[\sigma^\perp]\otimes K[\underline{\sigma}]$.
Thus
$K[X_\sigma]=K[O_\sigma][\underline{\sigma}^\vee]$
Let $T_\sigma\subset T$ be the subtorus corresponding the
sublattice $N_\sigma:=\operatorname{lin}(\sigma)\cap N$ of $N$. Then by
definition, $T_\sigma$ acts trivially on
$K[O_\sigma]=K[\sigma^\perp]=K[X_\sigma]^{T_\sigma}$. Thus
$T_\sigma=\{t\in T\mid tx=x, x\in O_\sigma\}$.
This leads us to the lemma
\begin{lemma} \label{le: und} Any toric variety
$X_\sigma$ is isomorphic to $X_{\underline{\sigma}}\times
O_\sigma$, where $O_\sigma\simeq T/T_\sigma$.
\qed
\end{lemma}
For any $\sigma\in\Sigma$
the closure $\overline{O}_\sigma$ of the orbit
$O_\sigma\subset X_\Sigma$ is a toric variety
with the
big torus $T/T_\sigma$. Let $\pi: N\to N/N_\sigma$ be the natural projection.
Then $\overline{O}_\sigma$ corresponds to the fan $\Sigma':=\{
\frac{\tau+\operatorname{lin}(\sigma)}{\operatorname{lin}(\sigma)}\mid \tau \in {\rm Star}(\tau ,\Sigma)\}$
in $N/N_\sigma$.
\bigskip
\subsection {Morphisms of toric varieties}
\begin{definition}(see \cite{KKMS}, \cite{Oda},
\cite{Danilov2}, \cite{Fulton}). A
{\it birational toric morphism} or simply a {\it toric morphism} of toric
varieties $X_\Sigma \to X_{\Sigma'}$ is a
morphism identical on $T\subset X_\Sigma, X_{\Sigma'}$
\end{definition}
\begin{definition} (see \cite{KKMS}, \cite{Oda},
\cite{Danilov2}, \cite{Fulton}).
A {\it subdivision} of a fan
$\Sigma$ is a fan $\Delta$ such that $|\Delta|=|\Sigma|$
and any cone $\sigma\in
\Sigma $ is a union of cones $\delta\in
\Delta$.
\end{definition}
\begin{definition}\label{de: star subdivision} Let $\Sigma$ be a fan and
$\varrho$ be a ray passing in the
relative interior of $\tau\in\Sigma$. Then the {\it star
subdivision} $\varrho\cdot\Sigma$ of $\Sigma$ with respect to
$\varrho$ is defined to be
$$\varrho\cdot\Sigma=(\Sigma\setminus {\rm Star}(\tau ,\Sigma) )\cup
\{\varrho+\sigma\mid \sigma\in \overline{\rm Star}(\tau
,\Sigma)\setminus {\rm Star}(\tau
,\Sigma)\}.$$ If $\Sigma$ is regular, i.e. all its cones are
regular, $\tau=\langle v_1,\ldots,v_l\rangle $ and
$\varrho=\langle v_1+\ldots+v_l\rangle $
then we call the star
subdivision $\varrho\cdot\Sigma$ {\it regular}.
\end{definition}
\begin{proposition} (see \cite{KKMS}, \cite{Danilov2},
\cite{Oda}, \cite{Fulton}). Let
$X_\Sigma$ be a toric variety. There is a 1-1 correspondence
between subdivisions of the fan $\Sigma$ and proper toric
morphisms $X_{\Sigma'} \to X_{\Sigma}$.\qed
\end{proposition}
\begin{remark} Regular star subdivisions from
\ref{de: star subdivision} correspond to blow-ups of smooth varieties
at closures of orbits (\cite{Oda}, \cite{Fulton}). Arbitrary
star subdivisions correspond to blow-ups of some ideals
associated to valuations (see Lemma \ref{le: blow-up valuation}).
\end{remark}
\subsection{Toric varieties with $\Gamma$-action }
In further considerations let $\Gamma$ denote any algebraic
subgroup of an algebraic torus $T=K^*\times\ldots\times K^*$.
By $X/\Gamma$ and $X//\Gamma$ we denote respectively geometric and good
quotients. If $\Gamma$ acts on an algebraic variety $X$ and
$x\in X$ is a closed point, then $\Gamma_x$ denotes the
isotropy group of $x$.
\begin{definition} \label{de: G-indecomposable} Let
$\Gamma$ act on $X_{\sigma}\supset T$ via
a group homomomorphism $\Gamma\to T$.
Set $$\Gamma_\sigma:=\{g\in\Gamma\mid g(x)=x \quad \mbox{
for\,\, any} \quad x \in O_\sigma\}
.$$ We shall write $$\sigma=\sigma'\oplus^\Gamma\langle
e_1,\ldots,e_k \rangle$$ \noindent if
$\sigma=\sigma'\oplus\langle
e_1,\ldots,e_k \rangle$, $\langle
e_1,\ldots,e_k \rangle$ is regular and $\Gamma_{\sigma}=\Gamma_{\sigma'}$.
We say that $\sigma$ is
{\it $\Gamma$-indecomposable} if it cannot be
represented as $\sigma=\sigma'\oplus^\Gamma\langle
e_1,\ldots,e_k \rangle$ (or equivalently $X_{\sigma}$ is not
of the form $X_{\sigma'}\times {\bf A}^k$ with $\Gamma_\sigma$ acting
trivially on ${\bf A}^k$). By
$${\rm sing}^{\Gamma}(\sigma)$$ we mean the maximal
$\Gamma$-indecomposable face of $\sigma$.
\end{definition}
If $\Gamma$ acts on a toric vatiety via a
group homomomorphism $\Gamma\to T$ we shall speak of a
{\it toric action} of $\Gamma$.
\subsection{Demushkin's Theorem}
For any algebraic variety $X$ and its (in general
nonclosed) point $x\in X$ we
denote by
$\widehat{{\mathcal O}}_{X,x}$ the completion of the local
ring ${{\mathcal O}}_{X,x}$ at the maximal ideal of $x$. We also set
$$\widehat{X}_x:={\rm Spec}(\widehat{{\mathcal O}}_{X,x}).$$
For the affine toric variety $X_\sigma$ define
$$\widehat{X}_\sigma:= {\rm Spec}(\widehat{{\mathcal O}}_{X_\sigma,O_\sigma}).$$
We shall use the following Theorem of
Demushkin (\cite{Demushkin}):
\begin{theorem}\label{th: Dem0} Let $\sigma$ and
$\tau$ be
two cones of maximal dimension in isomorphic lattices
$N_\sigma\simeq N_\tau$.
Then the following conditions are equivalent:
\begin{enumerate}
\item $\sigma \simeq \tau$.
\item $\widehat{X}_{\sigma} \simeq \widehat{X}_{\tau}$.
\end{enumerate}
\end{theorem}
\noindent{\bf Proof.} For the proof see \cite{Demushkin} or
the proof of
\ref{le: Dem2}. \qed
The above theorem can be formulated for affine toric
varieties with $\Gamma$-action.
\begin{theorem}\label{th: Dem00} Let $\sigma$ and
$\tau$ be
two cones of maximal dimension in isomorphic lattices
$N_\sigma\simeq N_\tau$.
Let $\Gamma$ act on $X_{\sigma}\supset{
T}_\sigma$ and on $X_{\tau}\supset{
T}_\tau$ via group homomorpisms $\Gamma\to T_\sigma$ and
$\Gamma\to T_\tau$.
Then the following conditions are equivalent:
\begin{enumerate}
\item There exists an isomorphism of cones $\sigma \simeq
\tau$ inducing a $\Gamma$-equivariant isomorphism of toric
varieties.
\item There exists a $\Gamma$-equivariant isomorphism
$\widehat{X}_{\sigma} \simeq \widehat{X}_{\tau}$.
\end{enumerate}
\end{theorem}
\noindent{\bf Proof.} For the proof see the proof of
\ref{le: Dem2} or
\cite{Demushkin} . \qed
\begin{definition} Let $\Gamma$ act on $X_{\sigma}\supset{
T}_\sigma$ and on $X_{\tau}\supset{
T}_\tau$ via group homomorpisms $\Gamma\to T_\sigma$ and
$\Gamma\to T_\tau$.
We say that cones $\sigma$ and
$ \tau$ are {\it $\Gamma$-isomorphic} if there exists
an isomorphism of cones $\sigma \simeq
\tau$ inducing a $\Gamma$-equivariant isomorphism of toric
varieties.
\end{definition}
The above theorem can be formulated as follows:
\begin{theorem} \label{th: Dem}
Let $\sigma$ and
$\tau$ be
two cones in isomorphic lattices
$N_\sigma\simeq N_\tau$.
Let $\Gamma$ act on $X_{\sigma}\supset{
T}_\sigma$ and on $X_{\tau}\supset{
T}_\tau$ via group homomorpisms $\Gamma\to T_\sigma$ and
$\Gamma\to T_\tau$. Assume that $\Gamma_{\sigma}=\Gamma_{\tau}$.
Then the following conditions are equivalent:
\begin{enumerate}
\item ${\rm sing}^\Gamma(\sigma)$ and ${\rm
sing}^\Gamma(\tau)$ are $\Gamma_\sigma$-isomorphic.
\item
For any closed points $x_\sigma\in O_{\sigma}$ and $x_\tau\in
O_{\tau}$, there exists a $\Gamma_\sigma$-equivariant
isomorphism of the local rings ${{\mathcal O}}_{X_{\sigma},x_\sigma}$
and ${{\mathcal O}}_{X_{\tau},x_\tau}$.
\item
For any closed points $x_\sigma\in O_{\sigma}$ and $x_\tau\in
O_{\tau}$, there exists a $\Gamma_\sigma$-equivariant
isomorphism of the completions of the local rings $\widehat
{{\mathcal O}}_{X_{\sigma},x_\sigma}$ and $\widehat{{\mathcal O}}_{X_{\tau},x_\tau}$ .
\end{enumerate}
\end{theorem}
\noindent{\bf Proof.} $(1)\Rightarrow(2)$ Write $\sigma={\rm
sing}^\Gamma(\sigma)\oplus^\Gamma {\rm r}(\sigma)$ and $\tau={\rm
sing}^\Gamma(\tau)\oplus^\Gamma {\rm r}(\tau)$, where
${\rm r}(\sigma)$, ${\rm r}(\tau)$ denote regular cones. By Lemma
\ref{le: und},
$X_\sigma=X_{\underline{\sigma}}\times O_\sigma=X_{\underline{\rm
sing}^\Gamma(\sigma)}\times X_{\underline{{\rm r}(\sigma)}}\times O_\sigma$,
where $\Gamma_\sigma$ acts trivially on $X_{\underline{{\rm r}(\sigma)}}\times O_\sigma$.
Thus there exists a $\Gamma_\sigma$-equivariant isomorphism
${{\mathcal O}}_{X_{\sigma},x_\sigma}\simeq{{\mathcal O}}_{X_{\underline{\rm
sing}^\Gamma(\sigma)}\times X_{{\rm r'}(\sigma)}}
$, where ${\rm r'}(\sigma)$ is a regular cone of dimension
$\dim(N_\sigma)-\dim({\operatorname{sing}^\Gamma(\sigma)})$ and
$\Gamma_\sigma$ acts trivially on $X_{{\rm r'}(\sigma)}$.
Analogously ${{\mathcal O}}_{X_{\tau},x_\tau}\simeq{{\mathcal O}}_{X_{\underline{\rm
sing}^\Gamma(\tau)}\times X_{{\rm r'}(\tau)}}\simeq {{\mathcal O}}_{X_{\underline{\rm
sing}^\Gamma(\sigma)}\times X_{{\rm r'}(\sigma)}}={{\mathcal O}}_{X_{\sigma},x_\sigma}
$.
The implication $(2)\Rightarrow(3)$
is trivial.
For the proof of $(3)\Rightarrow (1)$ we find a regular
cone ${\rm re}(\sigma)$ in $N_\sigma$ such that
$\sigma':= \sigma\oplus {\rm re}(\sigma)$
is of maximal dimension in $N_\sigma$. Since $\Gamma_\sigma$
acts trivially on $O_\sigma$ and on $O_\tau$ we have
$\sigma':= \sigma\oplus^{\Gamma_\sigma} {\rm re}(\sigma)$.
Analogously $\tau':= \tau\oplus^{\Gamma_\tau}{\rm re}(\tau)$.
By Theorem \ref{th: Dem00}, $\sigma'$ and $\tau'$ are
$\Gamma_\sigma$-isomorphic.
Consequently,
${\rm sing}^\Gamma({\sigma})=
{\rm sing}^{\Gamma_\sigma}({\sigma'})\simeq {\rm
sing}^{\Gamma_\tau} ({\tau'})={\rm
sing}^\Gamma({\tau})$. \qed
Theorem \ref{th: Dem} allows us to assign a singularity type to
any closed point $x$ of a toric variety $X$ with a
toric action of the group $\Gamma$:
\begin{definition} \label{de: singularity type}
By the {\it singularity type} of
a point $x$ of a toric variety $X$ we mean the function
$$\operatorname{sing}(x):=\underline{\operatorname{sing}(\sigma_x)},$$
\noindent
where $\sigma_x$ is a cone of maximal dimension such that
$\widehat{X}_x\simeq\widehat{X}_{\sigma_x}$.
By the {\it singularity type} of
a point $x$ of a toric variety $X$ with a
toric action of group $\Gamma$ we mean
$$\operatorname{sing}^\Gamma(x):=(\Gamma_x,\underline{\operatorname{sing}^{\Gamma_x}(\sigma_x)}),$$
\noindent
where $\sigma_x$ is a cone of maximal dimension with toric action of
$\Gamma_x$ on $\widehat{X}_{\sigma_x}$ and such that there exists a
$\Gamma_x$-equivariant
isomorphism
$\widehat{X}_x\simeq\widehat{X}_{\sigma_x}$.
\end{definition}
\section{Stratified toric varieties and semifans}
\subsection{Definition of a stratified toric variety}
In this section, we give a combinatorial description of stratified
toric varieties in terms of so called (embedded) semifans.
\begin{definition}\label{de: stratification}
Let $X$ be a noetherian scheme $X$ over ${\rm Spec}(K)$. A
{\it stratification\/} of $X$ is a decomposition of $X$ into a
finite collection $S$ of pairwise disjoint locally closed irreducible
smooth subschemes $s \subset X$, called strata, with the following
property: For every $s \in S$, the closure $\overline{s} \subset X$
is a union of strata.
\end{definition}
\smallskip
\begin{definition}
Let $S$ and $S'$ be two stratifications of $X$.
We say that $S$ is {\it finer} than $S'$ if any stratum
in $S'$ is a union of strata in $S$. In this case we shall
also call $S'$ {\it coarser} than $S$.
\end{definition}
\noindent \begin{definition}\label{de: stratified toric} \enspace
Let $X$ be a toric variety with big torus $T \subset X$.
A {\it toric stratification\/} of $X$ is a stratification $S$ of
$X$ consisting of $T$-invariant strata $s$ such that for any two
closed points $x,x' \in s$ their local rings ${{\mathcal O}}_{X,x}$ and
${{\mathcal O}}_{X,x'}$ are isomorphic.
\end{definition}
\begin{definition}\label{de: stratified toric2}
Let $X$ be a toric variety with a toric action of $\Gamma$.
We say that a toric stratification $S$ of $ X$ is {\it
compatible with the action of $\Gamma$} if
\begin{enumerate}
\item All points in the same stratum $s$ have the
same isotropy group $\Gamma_s$.
\item For any two closed points $x, y$ from one
stratum $s$ there exists a $\Gamma_s$-equivariant
isomorphism $\alpha:\widehat{X}_x\to\widehat{X}_y$,
preserving all strata.
\end{enumerate}
\end{definition}
\smallskip
If $S$ is a toric stratification of $X$, then we shall also speak of a
{\it stratified toric variety} $(X,S)$. If $X$ is a toric
variety with a toric stratification , compatible with
$\Gamma$, then we shall also speak of a
{\it $\Gamma$-stratified toric variety} $(X,S)$.
The combinatorial objects we shall use
in this context are the following:
\begin{definition} \label{de: embedded} \enspace
An {\it embedded semifan\/} is a subset $\Omega \subset \Sigma$ of a
fan $\Sigma$ in a lattice $N$ such that for every $\sigma
\in \Sigma$ there is an $\omega(\sigma) \in \Omega$ satisfying
\begin{enumerate}
\item $\omega(\sigma) \preceq \sigma$ and any other
$\omega \in \Omega$ with $\omega \preceq \sigma$ is a face of
$\omega(\sigma)$,
\item $\sigma = \omega(\sigma) \oplus {\rm r}(\sigma)$ for some
regular cone ${\rm r}(\sigma) \in \Sigma$.
\end{enumerate}
A {\it semifan\/} in a lattice $N$ is a set $\Omega$ of cones in $N$
such that the set $\Sigma$ of all faces of the cones of $\Omega$ is a
fan in $N$ and $\Omega \subset \Sigma$ is an embedded semifan.
\end{definition}
\smallskip
\begin{definition} \label{de: G-semifan} An embedded
semifan $\Omega\subset\Sigma$ with a toric action of
$\Gamma$ on $X_\Sigma$ will be called an
{\it embedded $\Gamma$-semifan} if for every
$\sigma\in\Sigma$, $\sigma=\omega(\sigma)\oplus^\Gamma {\rm r}(\sigma)$.
\end{definition}
\begin{remark}
A semifan or an embedded semifan can be viewed as a $\Gamma$-semifan
or an embedded $\Gamma$-semifan with trivial group $\Gamma$.
\end{remark}
Some examples are discussed at the end of this section. The main
statement of this section says that stratified toric varieties are
described by embedded semifans:
\smallskip
\begin{proposition} \label{le: semifans correspondence} \enspace
Let $\Sigma$ be a fan in a lattice $N$, and let $X$ denote the
associated toric variety with a toric action of $\Gamma$.
There is a canonical 1-1 correspondence
between the toric stratifications of $X$ compatible withn the
action of $\Gamma$ and the embedded
$\Gamma$-semifans $\Omega \subset \Sigma$:
\begin{enumerate}
\item If $S$ is a toric stratification of $X$, compatible
with the action of $\Gamma$, then the
corresponding embedded $\Gamma$-semifan $\Omega \subset \Sigma$ consists of
all those cones $\omega \in \Sigma$ that describe the big orbit of
some stratum $s \in S$.
\item If $\Omega \subset \Sigma$ is an embedded $\Gamma$-semifan, then
the strata of the associated toric stratification $S$ of $X$ arise
from the cones of $\Omega$ via
$$ \omega \mapsto {\rm strat}(\omega)
:= \bigcup_{\omega(\sigma) = \omega} O_\sigma. $$
\end{enumerate}
\end{proposition}
\smallskip
\noindent{\bf Proof.}
$(1)\Rightarrow (2)$ Since strata of $S$ are $T$-invariant and
disjoint, each
orbit $O_\tau$ belongs to a
unique stratum $s$. Let $\omega \in \Omega$ describe the
big open orbit of $s$. Then $O_\tau$ is contained in the
closure of $O_{\omega}$. Hence $\omega$ is a face of $\tau$.
By Definition {de: embedded}, $\Gamma_\tau=\Gamma_\omega$
and there exists
a $\Gamma_\tau$-equivariant isomorphism of the local
rings
${{\mathcal O}}_{X_\tau,x}$ and ${{\mathcal O}}_{X_\omega,y}$ of two
points $x\in O_{\tau}$ and $
y\in O_{\omega} $.
By Theorem \ref{th: Dem} and since
$\operatorname{sing}^\Gamma(\tau)\supset\operatorname{sing}^\Gamma(\omega)$, we infer that
$\operatorname{sing}^\Gamma(\tau)=\operatorname{sing}^\Gamma(\omega)$. Hence
$\omega=\operatorname{sing}^\Gamma(\omega)\oplus^\Gamma {\rm r}(\omega)$ and
$\tau=\operatorname{sing}^\Gamma(\tau)\oplus^\Gamma {\rm r}(\tau)=
\omega\oplus^\Gamma {\rm r'}(\tau)$,
where ${\rm r}(\tau)={\rm r}(\omega)\oplus {\rm r'}(\tau)$.
$(2)\Rightarrow (1)$. By definition
$\{\sigma\in\Sigma\mid \omega(\sigma)=\omega)\}=
{\rm Star}(\omega,{\Sigma})\setminus \bigcup_{\omega\prec
\omega' \in \Omega}
{\rm Star}(\omega', {\Sigma})$.
Hence all the defined subsets
${\operatorname{strat}}(\omega)$ are
locally closed. The closure of each subset
${\operatorname{strat}}(\omega)$ corresponds to ${\rm
Star}(\omega,\Sigma)$ and hence it is a union of
the sets ${\operatorname{strat}}(\omega')$, where $\omega\prec\omega'$.
Since $\tau=\omega\oplus^\Gamma {\rm r}(\tau)$ we have $\Gamma_\tau=\Gamma_\omega$.
Each subset ${\operatorname{strat}}(\omega)$ is a toric
variety with a fan
$\Sigma':=\{\frac{\tau+\operatorname{lin}(\omega)}{\operatorname{lin}(\omega)}|\omega(\tau)=\omega\}$
in $(N^Q)':=N^Q/\operatorname{lin}(\omega)$.
Since
$\tau=\omega\oplus^\Gamma {\rm r}(\tau)$ the cone
$\frac{\tau+\operatorname{lin}(\omega)}{\operatorname{lin}(\omega)}$ is isomorphic to
the regular cone ${\rm r}(\tau)$ in $(N^Q)'$.
Thus the strata
${\operatorname{strat}}(\omega)$ is smooth.
The local rings of closed points of the the strata
${\operatorname{strat}}(\omega)$ have the same isotropy group $\Gamma_\omega$.
Moreover if $O_\tau\subset{\operatorname{strat}}(\omega)$ then
$\operatorname{sing}^\Gamma(\tau)=\operatorname{sing}^\Gamma(\omega)$.
By Theorem \ref{th: Dem}, we conclude that there exists
a $\Gamma$-equivariant isomorphism of the local rings of any
two points $x\in O_{\tau}$ and $
y\in O_{\omega} $.
\qed
As a corollary from the above we obtain the following lemmas:
\begin{lemma} \label{le: strata}
${\rm strat}(\omega)=\overline{{\rm
strat}(\omega)}\setminus\bigcup_ {\omega'\prec\omega}
\overline{ {\rm strat}(\omega')}$. \qed
\end{lemma}
\begin{lemma} \label{le: strata2} Let $S_1$ and $S_2$ be two
toric stratifications on a toric variety $X$ corresponding to
two embedded semifans $\Omega_1,\Omega_2\subset\Sigma$.
Then $S_1$ is coarser than $S_2$ iff $\Omega_1\subset \Omega_2$.
\end{lemma}
\noindent{\bf Proof} $(\Rightarrow)$ Assume that $S_1$ is coarser
than $S_2$. Let $\omega\in\Omega_1$ be the cone
corresponding to a stratum $s_1\in S_1$. The closure $\overline{s_1}$
is a union of strata in $S_2$.
There is a generic stratum $s_2\in S_2$ contained in
$\overline{s_1}$. Then $\overline{s_1}=\overline{s_2}$ and
$\omega\in\Omega_2$ corresponds to the stratum $s_2$.
$(\Leftarrow)$ Now assume that $\Omega_1\subset \Omega_2$.
Then the closure of any stratum $s_1$ equals
$\overline{s_1}=\overline{\omega_1}$, where
$\omega_1 \in \Omega_1$ is the cone
corresponding to $s_1$. But then
$\overline{\omega_1}=\overline{s_2}$, where $s_2\in S_2$ is the stratum
corresponding to $\omega_1\in\Omega_2$. Thus
$\overline{s_1}=\overline{s_2}$ and it suffices to apply
Lemma \ref{le: strata}. \qed
\begin{remark}\enspace
Let $(X,S)$ be the stratified toric variety arising from a semifan
$\Omega \subset \Sigma$. The above correspondence between the cones of
$\Omega$ and the strata of $S$ is order reversing in the sense that
$\omega \preceq \omega'$ iff ${\rm strat}(\omega')$ is contained in
the closure of ${\rm strat}(\omega)$.
\end{remark}
\smallskip
\noindent\begin{lemma}\label{le: stratified toric}
Let $\Omega\subset\Sigma$ be an embedded $\Gamma$-semifan.
Then all $\Gamma$-indecomposable faces
of $\Sigma$ are in $\Omega$.
\end{lemma}
\noindent{\bf Proof.}
For a $\Gamma$-indecomposable face $\sigma$ write
$\sigma=\omega(\sigma)\oplus^\Gamma {\rm r}(\sigma)$. This
implies $\sigma=\omega(\sigma)\in\Omega$. \qed
We conclude the section with some examples. The first two examples
show that for any fan there are a maximal and a minimal embedded
semifan.
\smallskip
\begin{example}\label{ex: orbits} \enspace
If $\Sigma$ is a fan in a lattice $N$, then $\Sigma \subset
\Sigma$ is an embedded semifan. The corresponding stratification of
the toric variety $X$ associated to $\Sigma$ is the decomposition of
$X$ into the orbits of the big torus $T \subset X$. The
orbit stratification is the finest among all toric
stratifications of $X$.
\end{example}
\smallskip
\begin{example} \label{ex: sing} \enspace
For a fan $\Sigma$ in a lattice $N$ and a toric action of
$\Gamma$ on $X_\Sigma$ let
${\rm Sing}^\Gamma(\Sigma)$ denote the set of all maximal indecomposable
parts ${\rm sing}^\Gamma(\sigma)$, where $\sigma \in \Sigma$.
Then ${\rm Sing}^\Gamma(\Sigma) \subset \Sigma$ is an embedded
$\Gamma$-semifan, and for any other embedded
$\Gamma$-semifan $\Omega \subset \Sigma$
one has ${\rm Sing}^\Gamma(\Sigma) \subset \Omega$. The toric stratification
corresponding to ${\rm Sing}^\Gamma(\Sigma) \subset \Sigma$ is the
stratification by singularity type on the toric variety
$X_\Sigma$ with the toric action of $\Gamma$. The toric stratification
corresponding to ${\rm Sing}(\Sigma) \subset \Sigma$ is the
coarsest among all toric stratifications of $X$. The toric stratification
corresponding to ${\rm Sing}^\Gamma(\Sigma) \subset \Sigma$
is the
coarsest among all toric stratifications of $X$ which are
compatible with the action of $\Gamma$.
\end{example}
\bigskip
\begin{example} \label{ex: affine} Let ${\bf A}^2\supset K^*\times K^*$
be a toric variety with the strata $s_0:=\{(0,0)\}$,
$s_1:=K^* \times \{0\}$, and $s_2:={\bf A}^2\setminus (s_0\cup
s_1)$. Then ${\bf A}^2$ with the above stratification is a
stratified toric variety.
We have $\sigma_{0}=\langle(1,0),(0,1)\rangle$,
$\sigma_{1}=\langle(0,1)\rangle,$ $\sigma_{2}=\{(0,0)\},$
$\overline{\sigma_{0}}=\{\langle(1,0),(0,1)\rangle\}$,
$\overline{\sigma_{1}}=\{\langle(0,1)\rangle\}$,
$\overline{\sigma_{2}}=\{(0,0)\}, \langle(1,0)\rangle\}$.
\end{example}
\section{Stratified toroidal varieties} \label{se: stratified toroidal}
\subsection {Definition of a stratified toroidal variety}
\begin{definition} (see also \cite{Danilov1}). Let $X$ be a variety
or a noetherian scheme over $K$. $X$ is called {\it
toroidal} if for any $x\in X$ there exists an open affine
neighborhood $U_x$ and an \'etale morphism $\phi_x:U_x\rightarrow
X_{\sigma_x}$ into some affine toric variety
$X_{\sigma_x}$.
\end{definition}
\begin{definition} \label{de: gsmoth} Let $\Gamma$ act on
noetherian schemes $X$ and $Y$. We say that a $\Gamma$-equivariant
morphism $f: Y \to X$ is $\Gamma$-{\it smooth} (resp. $\Gamma$-{\it \'etale})
if there is a smooth (resp. \'etale) morphism
$f': Y' \to X'$ of varieties with trivial action of
$\Gamma$ and
$\Gamma$-equivariant fiber square
\[\begin{array}{rccccccc}
&X \times_{X'}Y' & \simeq &Y&\buildrel f \over \rightarrow
&X&& \\
&&& \downarrow & &\downarrow&& \\
&&&Y' & \buildrel f' \over \rightarrow &X' \\
\end{array}\]
\end{definition}
\begin{lemma} \label{le: gsmoth} Let $Y\to X$ be a
$\Gamma$-smooth morphism. If the good quotient $X//\Gamma$ exists then
$Y//\Gamma$ exists and $Y=Y//\Gamma\times_{X//\Gamma}X$.
\end{lemma}
\noindent{\bf Proof.}
Let $Y_0:=Y'\times_X{X//\Gamma}$. Then
$Y=Y_0\times_{X//\Gamma}X$ and $Y//\Gamma=Y_0$. \qed
\begin{definition} \label{de: G-toroidal variety} Let $X$ be a
toroidal variety or a toroidal scheme. We say that an action
of a group
${\Gamma}$ on $X$ is {\it toroidal} if for any
$x\in X$ there exists an open
$\Gamma$-invariant neighborhood $U$ and a
$\Gamma$-equivariant, $\Gamma_x$-smooth,
morphism $U\rightarrow X_\sigma$, into a toric variety
$X_\sigma$ with a toric action of
$\Gamma\supset \Gamma_x$.
\end{definition}
\begin{definition} \label{de: stratified toroidal}
Let $X$ be a
toroidal variety or a toroidal scheme. We say that a
stratification $S$ of $X$ is {\it toroidal} if
any point $x$ in a stratum $s \in S$ admits
an open neighborhood $U_x$ and an \'etale morphism
$\phi_x: U_x\rightarrow
X_{\sigma}$ into a stratified toric variety such that $s\cap U_x$ equals $\phi_x^{-1}(O_{\sigma})$
and the intersections $s'\cap U$, $s'\in S$, are precisely the inverse
images of strata in $X_\sigma$.
\end{definition}
\begin{definition}\label{de: G-stratified toroidal}
Let $X$ be
toroidal variety or a toroidal scheme with a toroidal
action of $\Gamma$. A toroidal stratification $S$ of $X$ is
said to be {\it compatible with the action of $\Gamma$} if
\begin{enumerate}
\item Strata of $S$ are invariant with respect to the
action of $\Gamma$.
\item All points in one stratum $s$ have the
same isotropy group $\Gamma_s$.
\item For any point $x\in s$ there exists a $\Gamma$-invariant
neighborhood $U$ and a $\Gamma$-equivariant,
$\Gamma_s$-smooth morphism
$U \rightarrow
X_\sigma$ into a $\Gamma$-stratified toric variety, such
that $s\cap U_x$ equals $\phi_x^{-1}(O_{\sigma})$
and the intersections $s'\cap U$, $s'\in S$, are precisely the inverse
images of strata in $X_\sigma$.
\end{enumerate}
\end{definition}
If $X$ is a toroidal variety or a toroidal scheme with a toroidal
stratification $S$ then we shall speak of a {\it stratified
toroidal variety} (resp. a {\it stratified
toroidal scheme}). If $X$ is a toroidal variety or a
toroidal scheme with a
toroidal action of $\Gamma$ and a toroidal
stratification $S$ compatible with the action of $\Gamma$
then we shall speak of a {\it $\Gamma$-stratified
toroidal variety} (resp. {\it $\Gamma$-stratified
toroidal scheme}).
If $(X,S)$ is a $\Gamma$-stratified toroidal variety then for any
stratum $s\in S$ set $\Gamma_s:=\Gamma_x$, where $x\in s$.
Note that if $s\leq s'$ then $\Gamma_s\subseteq \Gamma_{s'}$.
\begin{remark}
A stratified toroidal variety can be considered as a
$\Gamma$-stratified toroidal variety with trivial action of
the trivial group $\Gamma=\{e\}$.
\end{remark}
A simple generalization of Example \ref{ex: sing} is the following lemma:
\begin{lemma} \label{le: singularity type} Let $X$ be a toroidal
variety with a toroidal action of $\Gamma$. Let ${\rm Sing}^\Gamma(X)$ be the
stratification determined by the singularity type ${\rm
sing}^\Gamma(x)$ (see Definition \ref{de: singularity
type}) . Then ${\rm Sing}^\Gamma(X)$ is a toroidal
stratification compatible with the action of $\Gamma$.\qed
\end{lemma}
\begin{definition}\label{de: toroidal embeddings} (see also \cite{KKMS}).
A {\it toroidal embedding} is a
stratified toroidal variety such that any point
$p$ in a stratum $s \in S$ admits an open neighborhood
$U_p$ and an \'etale morphism
of stratified varieties $\phi_p: U_p \rightarrow
X_{\sigma}$ into a toric variety with orbit stratification
(see Example \ref{ex: orbits}).
A {\it toroidal embedding} with $\Gamma$-action is a
$\Gamma$-stratified toroidal variety which is a toroidal
embedding.
\end{definition}
\begin{remark} The definition of a toroidal embedding differs from the original
definition in \cite{KKMS}. It is equivalent to the definition of a
toroidal embedding without self-intersections
(see Section \ref{se: toroidal embeddings}).
\end{remark}
\subsection{Existence of invariant stratifications on smooth
varieties with $\Gamma$-action}
\noindent
\begin{lemma} \label{le: existence2}
\begin{enumerate}
\item Let $X$ be a smooth variety with $\Gamma$-action.
Then there
exists a toroidal stratification $\operatorname{Sing}^\Gamma(X)$ of $X$
which is compatible with the action of $\Gamma$.
\item Let $X$ be a smooth variety with $\Gamma$-action and $D$
be a $\Gamma$-invariant divisor with simple normal crossings. Let
$S_D$ be the stratification determined by the components of
the divisor $D$. Set
$$\operatorname{sing}^{\Gamma,D}(x):=(\operatorname{sing}^{\Gamma}(x), D(x)),$$ \noindent
where
$D(x)$ is the set of components of $D$ passing through $x\in X$
Then $\operatorname{sing}^{\Gamma, D}$ determines a toroidal stratification
$\operatorname{Sing}^\Gamma(X,D)$ compatible with the action of $\Gamma$. In particular all the
closures of strata from $\operatorname{Sing}^\Gamma(X,D)$
have normal crossings with components of $D$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} For any $x$ we can find $\Gamma$-semiinvariant
parameters $u_1,\ldots,u_k$ describing the orbit
$\Gamma\cdot x$ in
some $\Gamma$-invarinat neighborhood $U_x$ of $x$.
Additionally in (2) we require that each component of $D$
through $x\in X$ is described by one of the parameters in
$U_x$. We define a smooth morphism $\phi_x:U_x\to {\bf A}^k$ by
$\phi_x(y)=(u_1(y),\ldots,u_k(y))$. The action of $\Gamma$ on $X$
defines the action of $\Gamma$ on ${\bf A}^k$ with standard
coordinates be $\Gamma$-semiinvariant with corresponding
weights. Then $\operatorname{Sing}^\Gamma({\bf A}^k,D_A)$, where
$D_A=\overline{\phi(D)}$,
defines a $\Gamma$-invariant stratification on ${\bf A}^k$,
whose strata are determined by standard coordinates.
Therefore $\operatorname{Sing}^\Gamma(X,D)$ is also a stratification on
$X$ which is locally a pull-back of $\operatorname{Sing}^\Gamma({\bf A}^k,D_A)$.
The morphism $\phi_x$ is $\Gamma$-equivariant. Denote
by $u_1,\ldots,u_k,\ldots,u_n$ all local $\Gamma_x$-semiinvariant
parameters at $x$. Let $\widetilde{\phi}_x:U_x\to {\bf A}^n$ be the \'etale
$\Gamma_x$-equivariant morphism
defined by ${\widetilde{\phi}_x(y)}:=(u_1(y),\ldots,u_n(y))$. By
Luna's fundamental lemma (see Luna \cite{Luna}) we can find
a $\Gamma_x$-invariant neighborhood $U'_{x}\subset U_x$ for
which the induced quotient morphism
${\widetilde{\phi}_x}/\Gamma_x: U'_x//\Gamma_x \rightarrow
{\bf A}^n//\Gamma_x$ is \'etale
and $U'_x\simeq U'_x//\Gamma_x \times_{{\bf A}^n//\Gamma_x}
{{\bf A}^n}$. By Sumihiro \cite{Sumihiro} we can find a
$\Gamma$-invariant open affine
$V\subset \Gamma\cdot U'$ such that
$V\simeq V//\Gamma_x \times_{V//\Gamma_x}
{{\bf A}^n}$. Let $p: {\bf A}^n\to {\bf A}^k$ denote the standard projection on the first coordinates.
Then $p\widetilde{\phi}_x: V\to {\bf A}^k$ is a
$\Gamma$-equivariant $\Gamma_x$-smooth morphism.
By definition $(X,\operatorname{Sing}^\Gamma(X,D))$ is a $\Gamma$-stratified toroidal variety.
\qed
\subsection{Conical semicomplexes}
Here we generalize the notion of a semifan. For this we first have to
figure out those semifans that describe affine stratified toric
varieties. In analogy to usual cones and fans we shall denote them
by small Greek letters $\sigma, \tau$, etc.:
\smallskip
{\bf Definition.}\enspace
Let $N$ be a lattice. A {\it semicone\/} in $N$ is a semifan $\sigma$
in $N$ such that the support $| \sigma |$ of $\sigma$ occurs as an
element of $\sigma$.
\smallskip
The {\it dimension\/} of a semicone is the dimension of its
support. Moreover, for an injection $\imath \colon N \to N'$ of
lattices, the {\it image\/} $\imath(\sigma)$ of a semicone $\sigma$ in
$N$ is the semicone consisting of the images of all the elements of
$\sigma$.
Note that every cone becomes a semicone by replacing it with the set
of all its faces. Moreover, every semifan is a union of maximal
semicones. Generalizing this observation, we build up semicomplexes
from semicones:
\smallskip
\begin{definition}\label{de: conical semicomplex} \enspace
Let $\Sigma$ be a finite collection of semicones $\sigma$ in
lattices $N_{\sigma}$ with $\dim(\sigma) = \dim(N_{\sigma})$.
Moreover, suppose that there is a partial ordering ``$\le$'' on
$\Sigma$.
We call $\Sigma$ a {\it semicomplex\/} if for any pair $\tau \le
\sigma$ in $\Sigma$ there is an associated linear injection
$\imath^{\sigma}_{\tau} \colon N_{\tau} \to N_{\sigma}$ such that
$\imath^{\sigma}_{\tau}(N_{\tau}) \subset N_{\sigma}$ is a saturated
sublattice and
\begin{enumerate}
\item $\imath^{\sigma}_{\tau} \circ
\imath^{\tau}_{\varrho}(\varrho) =
\imath^{\sigma}_{\varrho}(\varrho)$,
\item $\imath^{\sigma}_{\varrho}(|\varrho|) =
\imath^{\sigma}_{\tau}(|\tau|)$ implies $\varrho = \tau$,
\item $\sigma = \bigcup_{\tau \le \sigma}
\imath^{\sigma}_{\tau}(|\tau|)=\bigcup_{\tau \le \sigma}
\imath^{\sigma}_{\tau}(\tau)$.
\end{enumerate}
\end{definition}
\smallskip
As a special case of the above notion, we recover back the notion of a
(conical) complex introduced by Kempf, Knudsen, Mumford and Saint-Donat:
\begin{definition}\label{de: conical complex} \enspace
A {\it complex\/} is a semicomplex $\Sigma$ such that every $\sigma
\in \Sigma$ is a fan.
\end{definition}
\begin{definition}\label{de: conical semicomplex2}
A semicomplex $\Sigma$ is called a {\it $\Gamma$-semicomplex} if
there is a collection of algebraic groups
$\Gamma_\sigma\subset T_\sigma$, where $\sigma\in \Sigma$, such that
\begin{enumerate}
\item Each $\sigma\in \Sigma$ is a $\Gamma_\sigma$-semicone.
\item For any $\tau\leq \sigma$, $\Gamma_{\tau}\subset\Gamma_\sigma$ and
the morphism $\imath^{\sigma}_{\tau}$ induces a
$\Gamma_{\tau}$-equivariant morphism $X_{\tau}\to
X_{\sigma}$. Moreover $\Gamma_{\tau}=(\Gamma_{\sigma})_{\tau}=\{g\in\Gamma_\sigma\mid gx=x, x\in O_\tau\}$.
\end{enumerate}
\end{definition}
\begin{remarks}
\begin{enumerate}
\item Each semicomplex can be considered as a
$\Gamma$-semicomplex with trivial groups $\Gamma_\sigma$
\item From now on we shall often identify vectors of $\sigma$
with their images under $\imath^{\sigma}_{\tau}$ if $\tau\leq
\sigma$, and simplify
the notation replacing $\imath^{\sigma}_{\tau}$ with set-theoretic inclusions.
\item
In what follows we shall write $\sigma\preceq
\sigma'$ means that $\sigma$ is a face of the cone
$\sigma'$, while $\sigma\leq
\sigma'$ to mean that $\sigma$ is a face of the semicone $\sigma'$.
\item
In the case when $\Sigma$ is a conical complex the
relations $\leq$ and $\preceq$ coincide.
\item Note that cones in a semicomplex intersect along a union of
common faces.
\end{enumerate}
\end{remarks}
Denote by ${\operatorname{Aut}}(\sigma)$ the group of automorphisms
of the $\Gamma_\sigma$-semicone $\sigma$.
\begin{lemma} \label{le: s2} For any $\Gamma$-semicomplex $\Sigma$ and any
$\varrho\leq \tau\leq \sigma$ in $\Sigma$, there is an automorphism \\
$\alpha_{\varrho}\in{\operatorname{Aut}}(\varrho)$ such that
$\imath^{\sigma}_{\tau} \circ
\imath^{\tau}_{\varrho} =\imath^{\sigma}_{\varrho} \alpha_\varrho.$
\end{lemma}
\noindent{\bf Proof.} By definition, the maps $\imath^{\sigma}_{\tau} \circ
\imath^{\tau}_{\varrho} $ and
$\imath^{\sigma}_{\varrho}$ are both linear isomorphisms of $\varrho$
onto the image
$\imath^{\sigma}_{\tau} \circ
\imath^{\tau}_{\varrho}(\varrho)=
\imath^{\sigma}_{\varrho}(\varrho)$. Thus $\alpha_{\varrho}:=
{(\imath^{\sigma}_{\varrho})}^{-1}\imath^{\sigma}_{\tau} \circ
\imath^{\tau}_{\varrho}$.
\qed
\begin{remarks} \begin{enumerate}
\item It follows from Lemma \ref{le: s2} that vectors in
$\sigma$ are in general defined up
to automorphisms from ${\operatorname{Aut}}(\sigma)$. Hence the notion of support of a
semicomplex as a
topological space which is the totality of such vectors is
not well defined.
However if we consider, for instance, vectors invariant
under all such automorphisms then the relevant topological
space can be constructed (see notion of {\it stable
support} \ref{de: stable support}) and plays the key role
for semicomplexes and their subdivisions.
\item In further considerations we shall introduce the
notion of {\it oriented semicomplex} (see Definition
\ref{de: oriented semicomplex}) which allows a
smaller group of automorphisms ${\operatorname{Aut}}(\sigma)^0
\subset {\operatorname{Aut}}(\sigma)$. Consequently, the corresponding stable
support of an oriented semicomplex is essentially larger. This
allows one to perform more subdivisions and carry out certain
important birational transformations.
\end{enumerate}
\end{remarks}
\noindent \begin{lemma}\label{le: semic}
A $\Gamma$-semifan $\Sigma$ in $N$ determines
a $\Gamma$-semicomplex $\Sigma^{\rm semic}$ where each cone
$\sigma\in\Sigma$ determines the semicone consisting of all faces
$\tau\leq\sigma$ and $\Gamma_\tau=\{g\in \Gamma\mid gx=x, x\in O_\tau\}$.\qed
\end{lemma}
\begin{definition}\label{de: semicomplexes isomorphism}
By an {\it isomorphism} of two $\Gamma$-semicomplexes $\Sigma\to
\Sigma'$ we mean a face bijection
$\Sigma\ni\sigma\mapsto \sigma'\in\Sigma'$ such that
$\Gamma_\sigma=\Gamma_{\sigma'}$, along with a collection of
face $\Gamma_\sigma$-isomorphisms $j_\sigma:\sigma\to\sigma'$, such that
for any $\tau\leq \sigma$, there is a $\Gamma_\tau$-automorphism
$\beta_\tau$ of $\tau$ such that $j_{\sigma}
\imath^\sigma_\tau=\imath^{\sigma'}_{\tau'}j_{\tau}\beta_\tau$.
\end{definition}
\bigskip
\subsection{Inverse systems of affine algebraic groups}
\begin{definition}\label{de: affine proalgebraic group}
By an {\it affine proalgebraic group} we mean an affine
group scheme that is the limit of
an inverse system $(G_i)_{i\in
{\bf N}}$ of affine algebraic groups and
algebraic group homomorphisms $\phi_{ij}:G_i\rightarrow
G_j$, for $j\geq i$.
\end{definition}
\begin{lemma} \label{le: epimorphisms} Consider the natural morphism
$\phi_i:G\to G_i$. Then $H_i:=\phi_i(G)$ is an algebraic subgroup
of $G_i$, all induced morphisms $H_j\to H_i$ for $i\leq j$
are epimorphisms and
$G=\lim_{\leftarrow}G_i=\lim_{\leftarrow}H_i$. In
particular $K[H_i]\subset K[H_{i+1}]$ and $K[G]=\bigcup K[H_i]$.
\end{lemma}
\noindent {\bf Proof.} The set $H'_i:=\bigcap_{j>i}
\phi_{ji}(G_j)$ is an intersection of algebraic subgroups
of $G_i$. Hence it is an algebraic subgroup of $G_i$. Note
that the induced homomorphisms $\phi^H_{ij}:H'_j\to H'_i$ are
epimorphisms and
that $G=\lim_{\leftarrow}G_i=\lim_{\leftarrow}H'_i$. Consequently,
$\phi_i(G)=H_i=H'_i$. \qed
\begin{lemma}\label{le: K-points}
The set $G^K$ of $K$-rational points of $G$ is an
abstract group which is the inverse limit
$G^K=\lim_{\leftarrow}G^K_i$ in the category of abstract groups.
\end{lemma}
\noindent {\bf Proof.} By the previous lemma we can assume
all morphisms $G_i\to G_j$ to be epimorphisms.
Any $K$-rational point $x$ in $G$ is mapped to $K$-rational points $x_i$ in
$G_i$. This gives an abstract group homomorphism
$\phi: G^K\to\lim_{\leftarrow}G^K_i$.
We also have $K[G]=\bigcup K[G_i]$. We have to show that any
point $x$ of $\lim_{\leftarrow}G^K_i$ determines a unique
point $y$
in $G^K$. The point $x$ determines a sequence of
maximal ideals $m_i\in K[G_i]$ such that $m_i\subset
m_{i+1}$. Let $m:= \bigcup m_i$. Then $m\cap K[G_i]=m_i$.
Let $f\in K[G]$. Then $f\in K[G_i]$ for some $i$ and $f-k
\in m_i$ for some $k\in K$. Hence $f\in k+m$, which means
that $K[G]/m=K$. The ideal $m$ defines a $K$-rational point $y$ on
$G^K$ which is mapped to $x$. Note that the point $y$ is
unique since there is a unique ideal $m$ for which $m\cap K[G_i]=m_i$
. \qed
By abuse of notation we shall identify $G$ with $G^K$.
\bigskip
\subsection{Group of automophisms of the completion of
a local ring}
Let $X$ be a stratified noetherian scheme over $\operatorname{Spec}(K)$.
Let $S$ denote the stratification and $x\in X$ be
a closed $K$-rational point. Let $\widehat{{\mathcal O}}_{X,x}$
denote the completion of the local ring of $x$ on
the scheme $X$. Let $u_1,\ldots u_k\in
\widehat{{\mathcal O}}_{X,x}$ generate the maximal ideal
$m_{x,X}\subset\widehat{{\mathcal O}}_{X,x}$. Then $\widehat{{\mathcal O}}_{X,x}=K[[u_1,\ldots,u_k]]/I$,
where $I$ is the defining ideal. Set $\widehat{X}_x:={\rm
Spec}(\,\widehat{{\mathcal O}}_{X,x})$ and $X^{(n)}_x:=
{\rm Spec}({{\mathcal O}}_{X,x}/m_{x,X}^{ n+1})$ for $n\in{\bf N}$.
Let ${\rm Aut}(\widehat{X}_x, S)$ (respectively ${\rm
Aut}(X^{(n)}_x, S)$) denote the automorphism group of
$\widehat{X}_x$ (resp. $X^{(n)}_x$) preserving all closures
$\overline{s}\ni x$ of strata $s\in S$. (resp. preserving all subschemes
\\${\rm Spec}({{\mathcal O}}_{X,x}/(I_{\overline{s}}+m_{x,X}^{n+1}))$, for all $s\in
S$)
If $\Gamma$ acts on $X$ denote by ${\rm
Aut}^\Gamma(\widehat{X}_x,S)$ (respectively ${\rm
Aut}^\Gamma(X^{(n)}_x, S)$) the group of all
$\Gamma_x$-equivariant atomorphisms of $(\widehat{X}_x,S)$
(resp. $(X^{(n)}_x, S)$).
In further considerations set
$G:={\rm Aut}(\widehat{X}_x,S)$ (resp. $G:={\rm
Aut}^\Gamma(\widehat{X}_x,S))$ in the case of the action of $\Gamma$, and $G_n:={\rm Aut}(X^{(n)}_x,S)$ (resp. $G_n:={\rm
Aut}^\Gamma(X^{(n)}_x,S)$).
\begin{lemma} \label{le: group action}
\begin{enumerate}
\item $G_n$ is an algebraic group
acting on $X^{(n)}_x$. That is there exists a co-action
$$\Phi_n^*: {{\mathcal O}}_{X,x}/m_{x,X}^{n+1}\to
K[G_n]\otimes
{{\mathcal O}}_{X,x}/m_{x,X}^{n+1}$$ defining
the action morphism
$\Phi_n: G_n\times X^{(n)}\to X^{(n)}$\\
and an action automorphism
$\Psi_n:=
\pi_n\times\Phi_n:
G_n\times X^{(n)}\to G_n\times X^{(n)}$\\
where $\pi_n$ is the projection on $G_n$.
\item The $G_n$ form an inverse system
of algebraic groups and $G=\lim_{\leftarrow} G_n$ is a proalgebraic group.
\item The co-actions in (1) define a ring homomorphism
$ \Phi^*:\widehat{{\mathcal O}}_{X,x} \to
K[G][[\widehat{{\mathcal O}}_{x,X}]]$\\
and a \underline{formal action morphism}
$ \Phi:
G\widehat{\times}\widehat{X}_x:=\operatorname{Spec} K[G][[\widehat{{\mathcal O}}_{X,x}]]\to
\widehat{X}_x$\\
such that for any $g\in G^K$ the restriction
$
\Phi_{|\{g\}\widehat{\times}\widehat{X}_x}:
\{g\}\widehat{\times}\widehat{X}_x\to \widehat{X}_x$ is
given by the action of $g$.
\item The automorphisms $\Psi_n$ of $X^{(n)}$
define a \underline{ formal action automorphism}
$ \Psi$ of \\
\centerline{$G\widehat{\times}\widehat{X}_x=\lim_{\leftarrow}G\times X^{(n)}$}
such that for any $g\in G^K$ the restriction
$
\Psi_{|\{g\}\widehat{\times}\widehat{X}_x}:
\{g\}\widehat{\times}\widehat{X}_x\to
\{g\}\widehat{\times}\widehat{X}_x$ is the automorphism
given by the action of $g$.
\item Let $u_1,\ldots,u_k$ denote the local
parameters on $\widehat{X}_x$. There exist regular functions
$g_{\alpha\beta}$ generating $K[G]$ such
that the action is given by $u^{\alpha}\mapsto \sum g_{\alpha\beta}u^\beta$, where $\alpha,\beta\in{\mathbb{Z}}_{\geq 0}^n$ are multiindices.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} (1) Write
$K[X^{(n)}_x]=K[u_1,\ldots,u_k]/(I+m_{x}^{n+1})$.
Set $W^n:=K[u_1,\ldots,u_k]/(m_{x}^{n+1})$,\\
$I^n:=(I+m_{x}^{n+1})/m_{x}^{n+1}$, $I_s^n:=(I_s+m_{x}^{n+1})/m_{x}^{n+1}$. \\
Define the product $W^n\times W^n \to W^n$ to be the bilinear
map for which
$u^{\alpha_1}\cdot u^{\alpha_2}=u^{\alpha_1+\alpha_2}$. Set \\
$$H_n:=\{g\in{\rm Gl}(W^n)\mid
g(u^{\alpha_1}\cdot u^{\alpha_2})=g(u^{\alpha_1})\cdot g(u^{\alpha_2})\}.$$
$H_n$ is an algebraic subgroup of ${\rm Gl}(W^n)$.
Then $$G_n=\{g\in H^n\mid g(I^n)=I^n, g(I_s^n)=I_s^n,
\quad \mbox{ for all}\quad
s\in S, ga=ag \quad \mbox{ for all}\quad
a\in \Gamma_x\}$$ \noindent is
algebraic. The embedding $G_n\subset
{\rm Gl}(W^n)$ determines the natural co-action $K[W^n]\to
K[G_n]\otimes K[W_n]$, $$u^\alpha\mapsto
\sum_{\beta \in {\bf Z}_{\geq 0}^n}g_{\alpha\beta}u^\beta,$$ which factors to
$K[X^{(n)}_x]\to K[G_n]\otimes K[X^{(n)}_x]$.
(2) If the morphism $X^{(n)}_x\hookrightarrow X^{(n+1)}_x$ commutes with
automorphisms ,i.e. if $\Phi$ is an automorphism of
$X^{(n+1)}_x$, then $\Phi_{|X^{(n)}}:X^{(n)}\to X^{(n)}\subset
X^{(n+1)}_x$ is an automorphism of $X^{(n)}$. This defines
the morphisms
${\rm Aut}(X^{n+1}_x,S)\to
{\rm Aut}(X^{(n)}_x,S)$.
The $K$-rational points of the proalgebraic group $G=\lim_{\leftarrow}
G_n$ can be identified, by Lemma
\ref{le: K-points}, with ($\Gamma_x$-equivariant)
automorphisms of $\widehat{X}_x$ preserving strata.
(3) The morphisms $\widehat{{\mathcal O}}_{X,x} \to
K[G_n][[{{\mathcal O}}_{x,X}]]/m^n_{x,X}\to
K[G][[\widehat{{\mathcal O}}_{X,x}]]/m^n_{x,X}$ determine \\
\centerline{$\widehat{{\mathcal O}}_{X,x}\to
\lim_{\leftarrow} K[G][[\widehat{{\mathcal O}}_{X,x}]]/m^n_{x,X}$}
(4) and (5) follow from (1) \qed
\begin{lemma}\label{le: group ideal} Let $G=\lim_{\leftarrow}G_i$ be
a connected
proalgebraic group acting on $\widehat{X}_x$. Let
$\Phi:G\widehat{\times}\widehat{X}_x\to\widehat{X}_x$ be
the action morphism and
$\pi:G\widehat{\times}\widehat{X}_x\to\widehat{X}_x$ be the
standard projection. Then for any
ideal $I$ on $\widehat{X}_x$ the following conditions are
equivalent:
\begin{itemize}
\item $I$ is invariant with respect to the action of the
abstract group $G^K$
\item $\Phi^*(I)=\pi^*(I)$
\end{itemize}
\end{lemma}
\noindent {\bf Proof.} It suffices to prove the equivalence
of the conditions
for any scheme $X^{(n)}_x$ and ideal $I_n=I\cdot K[X^{(n)}_x]$. In this
situation the assertion reduces to the well known case of an
algebraic action of $G_n$ on $X^{(n)}_x$. If $\Phi^*(I_n)=\pi^*(I)$ then for any $g\in
G_n^K$ and $f\in I_n $, we have $g\cdot
f=\Phi^*_g(f)\in\pi_g^*(I_n)=I_n$. Thus $I_n$ is $G_n^K$-invariant.
Now suppose $I_n$ is $G_n^K$-invariant. Let $\Phi^*(f)=\sum
f_ig_i$ for $f\in I_n$. We can assume that $g_i\in K[G_n]$
are linearly independent. Hence we
can find elements $g^i\in G^K_n$ such that
$v_i:=(g_i(g^1),\ldots, g_i(g^l))$ are linearly independent
vectors for $i=1,\ldots, l$. Then we find coefficients
$\alpha_{ij}$ such that $\sum_j \alpha_{ij}v_j=(0,\ldots,1_j,\ldots,0)$.
Then $f_i=\sum_j \alpha_{ij}{f_i}g_i(g^j)=\sum_j \alpha_{ij}(g^j\cdot f_i)
\in I_n$. Thus for any $f\in I_n$, $\Phi^*(f)\in \pi^*(I_n)$, which gives
$\Phi^*(I)\subset \pi^*(I)$.
Denote by ${\rm inv}$ the automorphism of $G\widehat{\times}\widehat{X}_x$ induced
by taking the inverse $g\mapsto g^{-1}$. Then $\pi^*(I)=\Psi^{-1*}\Phi^*(I)\subset
\Psi^{-1*}\pi^*(I)=\Psi^{*}{\rm inv}^*\pi^*(I)=\Psi^{*}\pi^*(I)=\Phi^{*}(I)$.
\qed
\bigskip\goodbreak
Let $L$ denote an algebraically closed field containing $K$.
Set $$ X^{(n)L}_x:=X^{(n)}_x\times_{\operatorname{Spec}{K}}\operatorname{Spec}{L},$$
$$\widehat{X}^L_x:=\lim_{\leftarrow}X^{(n)L}_x=\operatorname{Spec}{L[[{\mathcal O}_{X,x}]]}$$
Let ${\rm Aut}^\Gamma_L(X^{(n)L}_x,S)$ denote the group of
$\Gamma_x$-equivariant
$L$-automorphisms of $X^{(n)L}_x$ preserving the relevant
ideals of strata and $G:={\rm Aut}_L(\widehat{X}^L_x,S)$ denote the group of
$L$-automorphisms of $\widehat{X}^L_x$.
\begin{lemma}\begin{enumerate}\label{le: base}
\item ${\rm Aut}^\Gamma_L(X^{(n)L}_x,S)=
{\rm Aut}^\Gamma_K(X^{(n)}_x,S))\times_{\operatorname{Spec}{K}}\operatorname{Spec}{L}$.
\item
${\rm Aut}^\Gamma_L(\widehat{X}^L_x,S)={\rm
Aut}^\Gamma_K(\widehat{X}_x,S)\times_{\operatorname{Spec}{K}}\operatorname{Spec}{L}$.
\item Let $G\subset {\rm Aut}_K(\widehat{X}_x,S)$ be a
proalgebraic subgroup.
An ideal $I\subset \widehat{{\mathcal O}}_{X,x}$ is
$G$-invariant iff
$I\cdot L[[{\mathcal O}_{X,x}]]$ is $G\times_{\operatorname{Spec}{K}}\operatorname{Spec}{L}$-invariant. \qed
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) and (2) follow from the construction of
${{\operatorname{Aut}}}^\Gamma_K(\widehat{X}_x,S)$.
(3)($\Rightarrow$) It follows from Lemma
\ref{le: group ideal} that $I\subset \widehat{{\mathcal O}}_{X,x}$ is
$G$-invariant iff
$\Phi^*(I)= \pi^*(I)$ or
$\Psi^*(\frac{I+m_x^n}{m_x^n})=
\pi^*(\frac{I+m_x^n}{m_x^n})$.
Let $\Psi^L$ and
$\pi$ be pull-backs of the morphisms $\Psi$ and
$\pi^L$ under ${\operatorname{Spec}{L}}\to \operatorname{Spec}{K}$. Then
$\Psi^{L*}(\frac{I+m_x^n}{m_x^n} \cdot L)=
\pi^{L*}(\frac{I+m_x^n}{m_x^n}\cdot L)\subset L\otimes_{K}\widehat{{\mathcal O}}_{X,x}/m_x^n$,
which shows , by Lemma \ref{le: group ideal}, that
$I\cdot L[[{\mathcal O}_{X,x}]]$ is
$G\times_{\operatorname{Spec}{K}}\operatorname{Spec}{L}$-invariant.
($\Leftarrow$). Assume the latter
holds then $\Psi^{L*}(\frac{I+m_x^n}{m_x^n}\cdot L)=
\pi^{L*}(\frac{I+m_x^n}{m_x^n}\cdot L)\subset
L\otimes_{K}\widehat{{\mathcal O}}_{X,x}/m_x^n$.
Let $\operatorname{Gal}(L/K)$ be the Galois group of the extension $K\subset
L$. Both homomorphisms $\pi^{L*}$ and
$\Psi^{L*}$ are $\operatorname{Gal}(L/K)$-equivariant.
Considering $\operatorname{Gal}(L/K)$-invariant elements gives $\Psi^*(\frac{I+m_x^n}{m_x^n})=
\pi^*(\frac{I+m_x^n}{m_x^n})$.\qed
\begin{example}\label{ex: differential} Let $\phi_n:{\rm Aut}(\widehat{X}_x,S)\to {\rm Aut}(X^{(n)}_x,
S)$ denote the natural morphisms. For $n=1$ we get
the differential mapping:
$$d=\phi_1:{\rm Aut}(\widehat{X}_x, S) \longrightarrow
{\rm Aut}(X^{(1)}_x,S)\subset
{\rm Gl}({\rm Tan}_{X,x})).$$
\end{example}
\bigskip
\subsection{Proof of Demushkin's Theorem}
In what follows we shall use the following
generalization of
Theorem \ref{th: Dem0}.
\begin{lemma}\label{le: Dem2}
Let $\sigma$ and $\tau$ be
two $\Gamma$-semicones in isomorphic lattices $N_\sigma \simeq N_\tau$.
Then there exists a $\Gamma$-equivariant isomorphism
$\widehat{X}_{\sigma}
\simeq \widehat{X}_{\tau}$ preserving strata iff
there exists an isomorphism of $\Gamma$-semicones
$\sigma \simeq \tau$.
\end{lemma}
For a $\Gamma$-semicone $\sigma$ in a lattice $N_\sigma$
denote by
${\rm Aut}(\widehat{X}_{\sigma})$ the group of all $\Gamma$-equivariant automorphisms
of $\widehat{X}_\sigma$ which preserve strata defined by
faces of the $\Gamma$-semicone $\sigma$.
\begin{lemma}(Demushkin \cite{Demushkin}, also Gubeladze
\cite{Gubeladze}) \label{le: Dem4}
Let $\sigma$ and $\tau$ be
two $\Gamma$-semicones in isomorphic lattices $N_\sigma \simeq N_\tau$.
\begin{enumerate}
\item
The torus $T_\sigma$ associated to $
\widehat{X}_{\sigma}$ determines a maximal torus in the proalgebraic group
$G:={\rm Aut}(\widehat{X}_{\sigma})$.
\item
Let $d:G\to \operatorname{Gl}({\operatorname{Tan}}_{X_\sigma,O_\sigma})$ be the differential morphism as
in Example
\ref{ex: differential}. Then $d(T_\sigma)$ is a
maximal torus in $d(G)$.
\item
Any $\Gamma$-equivariant isomorphism $\widehat{\phi}:\widehat{X}_{\sigma}
\simeq \widehat{X}_{\tau}$, preserving strata
determines an action of the torus $T_\tau$ on
$\widehat{X}_{\sigma}$ and a group
embedding $T_\tau\hookrightarrow G$.
If $d(T_\sigma)=d(T_\tau)$ then
$T_\sigma$ and $T_\tau$ are conjugate in $G$.
\item Any two tori $T_\sigma, T_\tau\subset G$ which are
determined by $\widehat{\phi}$, are conjugate in $G$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.}
(1) The torus $T_\sigma$ is maximal in
$G$ since its centralizer consists of
$T_\sigma$-equivariant automorphisms of
$\widehat{X}_\sigma$, preserving strata
and therefore it coincides
with $T_\sigma$ ($T_\sigma$-equivariant
automorphisms multiply characters by constants and are
defined by elements of $T_\sigma$); (see also Step 4 in \cite{Demushkin}).
(2) By Lemma \ref{le: epimorphisms} we can write $G=\lim_{\leftarrow} G_i$, where
all homomorphisms $G_i\to G_j$ are epimorphisms.
We show that the relevant images of
$T_\sigma$ are maximal tori in
all groups $G_i$. Otherwise we find maximal
tori $T_{i}\subset G_i$ containing the images of $T_\sigma$,
such that $\lim_{\leftarrow} T_{i}$ is a maximal torus in
$G$ containing $T_\sigma$ as a proper\ subgroup (see
\cite{Borel} Proposition 11.14(1));
(see also step 9 in \cite{Demushkin}).
(3) Let $T_\sigma\subset G$ be the maximal torus associated to
$\widehat{X}_{\sigma}$. Let $u_1,\ldots,u_k$ be characters of
$T_\sigma$ generating the maximal ideal of $O_\sigma$ and
$x_1,\ldots,x_k$ be semiinvariant functions determined by the action of
$T_\sigma$,
induced on the tangent space ${\rm Tan}_{O_{\sigma},X_{\sigma}}$.
Characters of $T_\sigma$ define the
natural $T_\sigma$-equivariant embedding
$\phi_{\sigma}:\widehat{X}_{\sigma}
\rightarrow {\rm Tan}_{O_{\sigma},X_{\sigma}}$
into the tangent space induced by the ring epimorphism
$\phi^*_{\sigma}:K[[x_1,\ldots,x_k]]\to
K[[u_1,\ldots,u_k]]$ sending $x_i$ to $u_i$.
The morphism $\phi_{\sigma}$ determines the
isomorphism $\widehat{\phi}_{\sigma}:
\widehat{X}_{\sigma}\rightarrow
\widehat{\phi}_{\sigma}(\widehat{X}_{\sigma})$ preserving strata corresponding to
the isomorphism $\widehat{\phi}^*_{\sigma}:K[[x_1,\ldots,x_k]]/I \to
K[[u_1,\ldots,u_k]]$ where the ideal $I$ is generated by all
elements $x^\alpha-x^\beta$ where $x^\alpha, x^\beta$ are
monomials on which $T_\sigma$ acts with the same weights.
If $d(T_\sigma)=d(T_\tau)$ then
$\phi_{\sigma}(\widehat{X}_{\sigma})=\phi_{\tau}(\widehat{X}_{\tau})=\operatorname{Spec} K[[x_1,\ldots,x_k]]/I $.
Consequently, $\beta:=\phi_{T_\tau}^{-1}\phi_{T_\sigma}$
is a $T_\sigma=d(T_\sigma)$-equivariant
isomorphism $\widehat{X}_{\sigma}\to
\widehat{X}_{\tau}$ preserving strata.
It follows that $\beta$ determines a conjugating automorphism with
$T_\sigma=\beta T_\tau\beta^{-1}$.
(4) By Borel's theorem (\cite{Borel}) and (2), the tori $d(T_\sigma)$
and $d(T_\tau)$ are conjugate in $d(G)$. Consequently, we
can find a torus $T$ conjugate to $T_\sigma$
and such that $d(T_\tau)=d(T)$. By (3) $T_\tau$ and $T$
are conjugate. \qed
\bigskip
\noindent {\bf Proof of Lemma \ref{le: Dem2}.}
Let
$\widehat{\phi}: \widehat{X}_{\sigma}\rightarrow
\widehat{X}_{\tau}$ be a $\Gamma$-equivariant isomorphism
preserving strata. By Lemma \ref{le: Dem4} there is a
$\Gamma$-equivariant automorphism
$\alpha$ of $\widehat{X}_\sigma$ for which ${
T}_\sigma=\alpha{
T}_\tau\alpha^{-1}$.
Then $T_\sigma$ acts on $\widehat{X}_{\tau}$ as a big torus and
the $\Gamma$-equivariant isomorphism
$\widehat{\phi}\circ\alpha^{-1}:\widehat{X}_{\sigma}\rightarrow
\widehat{X}_{\tau}$ is $T_\sigma$-equivariant and
preserves strata. Such an isomorphism yields an
isomorphism of the associated $\Gamma$-semicones. \qed
\subsection{Singularity type of nonclosed points}
\bigskip
For a ring $L$ and a cone $\sigma$ in $N_\sigma$
denote by
$L[[{\sigma}^\vee]]$ the completion of the ring $L[{\sigma}^\vee]$
at the ideal $m_\sigma$ of $O_\sigma$.
\begin{lemma}\label{le: decomposition} Let $L$ be a ring
containing $K$, and $\sigma$ and
$\tau$ be $\Gamma$-semicones in isomorphic lattices
$N_\sigma\simeq N_\tau$.
Let $m_\sigma, m_\tau$ denote the
ideals of $O_{\sigma}$ and $O_{\tau}$ respectively.
The action of $\Gamma$ on $K[[{\sigma}^\vee]]$ and $K[[{\tau}^\vee]]$
induces an action on
$L[[{\sigma}^\vee]]$ and $L[[{\tau}^\vee]]$ which is
trivial on $L$.
Any $\Gamma$-equivariant isomorphism
$\phi: L[[{\sigma}^\vee]]\simeq L[[{\tau}^\vee]]$
for which $\phi(m_\sigma)=m_\tau$ can be decomposed as
$\phi=\phi_0\phi_1$, where $\phi_0$ is a $\Gamma$-equivariant automorphism of
$L[[{\tau}^\vee]]$ identical on monomials and $\phi_1$
is a $\Gamma$-equivariant $L$-isomorphism.
\end{lemma}
\noindent {\bf Proof.}
(1) The restriction of $\phi$ to ${L}$
is a ring monomorphism
$\phi_{|{L}}:{L}\to
{L}[[{\tau}^\vee]]$. This ring homomorphism
can be extended to a ring homomorphism
$\phi_0:{L}[[{\tau}^\vee]]\to
{L}[[{\tau}^\vee]]$ by sending monomials
identically to the same monomials, i.e.
$\phi_0(\sum_{\alpha\in{\tau}^\vee}
a_\alpha x^\alpha)=\sum_{\alpha\in{\tau}^\vee}
\phi_0(a_\alpha) x^\alpha$, where $ \phi_0(a_\alpha)=
\sum_{\beta\in{\tau}^\vee} a_{\alpha\beta} x^\beta$.
Note that in the expression $\sum_{\alpha\in{\tau}^\vee}
\phi_0(a_\alpha) x^\alpha$, the coefficient of $x^\alpha$
is a finite sum since there are finitely many
possibilities to express $\alpha$ as $\alpha'+\beta$ where
$\alpha', \beta\in {\tau}^\vee$. Thus the above expression defines a
formal power series and consequently $\phi_0$ is a ring homomorphism.
Then $\phi_{0|L}$ determines an automorphism ${\phi}_1: {L}\to
{L}[[{\tau}^\vee]]/m_\tau={L}$. Let
${\phi}_2$ denote the
automomorphism of ${L}[[{\tau}^\vee]]$ induced
by the ${\phi}_1$ and identical on monomials.
The composition
$\psi=\phi_2^{-1}\phi_0$ is a ring endomorphism which as a
$K$-linear transformation can be written in the form
$\psi=id+p$, where $p(m^i)\subset m^{i+1}$. Such a linear
transformation is invertible (with inverse
$id-p+p^2-\ldots$), hence $\psi$ is a linear isomorphism and ring
isomorphism. This implies that $\phi_0$ is an
automorphism and $\phi_1:=\phi_0^{-1}\phi$ is an ${L}$-isomorphism.
Note that the above morphisms are $\Gamma$-equivariant.
\qed
\bigskip
The following is a generalization of the Demushkin Lemma.
\begin{lemma}\label{le: Dem3} Let $\sigma$ and
$\tau$ be $\Gamma$-semicones in isomorphic lattices
$N_\sigma\simeq N_\tau$, and $L$ be a field
containing $K$.
Consider the induced action of $\Gamma$ on
$L[[{\sigma}^\vee]]$ and $L[[{\tau}^\vee]]$ which is
trivial on $L$.
There exists a $\Gamma$-equivariant isomorphism
$L[[{\sigma}^\vee]]\simeq L[[{\tau}^\vee]]$ over $K$, preserving strata iff
$\sigma$ and $\tau$ are $\Gamma$-isomorphic.
\end{lemma}
\noindent{\bf Proof.} Denote by $\phi$
an isomorphism
$L[[{\sigma}^\vee]]
\buildrel \phi \over \simeq L[[{\tau}^\vee]]$.
By Lemma \ref{le: decomposition}, $\phi=\phi_0\phi_1$, where
$\phi_1$ is an ${L}$-isomorphism.
Tensoring with the algebraic closure $\overline{L}$ of $L$
and
taking completion determines an $\overline{L}$-isomorphism
$\overline{\phi}_1: \overline{L}[[{\sigma}^\vee]]
\simeq \overline{L}[[{\tau}^\vee]]$.
It suffices to apply Theorem \ref{th: Dem} to the above
$\overline{L}$-isomorphism
Note that the above isomorphisms are $\Gamma$-equivariant.
\qed
This lemma allows us to extend the notion of singularity type to
any nonclosed points on toric varieties $X_{\Delta}$ and
even on some other toroidal schemes.
\begin{lemma} \label{le: 1} Let $\Delta$ be a subdivision
of $\sigma$ and
$\widehat{X}_{\Delta}:=X_{\Delta}\times_{X_{\sigma}}\widehat{X}_{\sigma}$
be a toroidal scheme with an action of a group $\Gamma\subset T_\sigma$.
Let
$p\in Y=\widehat{X}_{\Delta}$ be a point which need not
be closed. Set
$\Gamma_p:=\{g\in\Gamma\mid g(p)=p, g_{|K_p}={\rm
id}_{|K_p}\}$.
\begin{enumerate}
\item Let $O_{\tau,Y}\in Y$ denote the locally closed
subscheme defined by a toric orbit ${O}_\tau\subset
X_{\Delta}$, where $\tau\in\Delta$. Then $\widehat{{\mathcal O}}_{O_{\tau},Y}\simeq
K_{O_{\tau}}[[\underline{\tau}^\vee]]$, where $K_{O_{\tau}}$ is the
residue field of ${O_{\tau}}$.
\item There is a $\Gamma_p$-equivariant isomorphism
$\widehat{{\mathcal O}}_{p,Y}\simeq K_p[[\sigma_p^\vee]]$,
where $K_{p}$ is the
residue field of ${p}$ and $\sigma_p$ is a
uniquely determined cone. Moreover
$\operatorname{sing}^\Gamma(p):=(\Gamma_p,
\underline{\operatorname{sing}^{\Gamma_p}(\sigma_p)})=\operatorname{sing}^\Gamma(\tau):=(\Gamma_\tau,
\underline{\operatorname{sing}^{\Gamma_\tau}(\tau)}) $, where $O_{\tau,Y}\in Y$
is the minimal orbit scheme
containing $p$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1)
Let $X_{\tau,Y}:=X_{\tau}\times_{X_\sigma}\widehat{X_\sigma}\subset
Y$. Then $K[X_{\tau,Y}]=K[\tau^\vee]\otimes_
{K[\sigma^\vee]}K[[\sigma^\vee]]$. Let $T_\tau\subset T$ be
the torus
corresponding to the sublattice $N_\tau:= N\cap\operatorname{lin}(\sigma)$.
Then $T_\tau$ acts on $K[X_{\tau,Y}]$ with nonnegative
weights. Therefore the subring of ${T_\tau}$-invariant
functions $K[X_{\tau,Y}]^{T_\tau}$ in $K[X_{\tau,Y}]$
equals
$K[\tau^\perp]\otimes_{K[\sigma^\vee\cap
\tau^\perp]}K[[\sigma^\vee\cap
\tau^\perp]]$. The ideal $I=I_{O_{\tau,
Y}}\subset K[X_{\tau,Y}]$ of the orbit
$O_{\tau, Y}$ is generated by all monomials with positive weights
in the set $\sigma^\vee+\tau^\vee\subset
\tau^\vee=\underline{\tau}^\vee\oplus \tau^\perp$ and
consequently it is generated by
$\underline{\tau}^\vee\setminus \{0\}$.
Thus $K[X_{\tau,Y}]^{T_\tau}\simeq K[O_{\tau,Y}]$ and
$\lim_{\leftarrow}K[X_{\tau,Y}]/I^k\simeq
K[O_{\tau,Y}][[\underline{\tau}^\vee]]$.
This gives $K[\widehat{Y}_{O_{\tau, Y}}]
\simeq K(O_{\tau,Y})[[\underline{\tau}^\vee]]$.
(2) Let $O_{\tau,Y}\in Y$
be the minimal orbit scheme
containing $p$. First we prove that
$\Gamma_{O_\tau}=\Gamma_p$. We have the obvious inclusion
$\Gamma_{O_\tau}\subset \Gamma_p$. Now if $g\in \Gamma_p$ then
for any $f\in K[{O_\tau}]$, $(g(f)-f)\in I_p$, where
$I_p\subset K[{O_\tau}]$ describes $p$. Since
$\Gamma_p=\Gamma_{p'}$ for any $p'\in\Gamma\cdot p$, it follows that
$(g(f)-f)\in \bigcap_{h\in T} h\cdot I_p =\{0\}$.
The ideal $I_p\subset {{\mathcal O}}_{p,O_{\tau}}$ of
$p$ is generated by local parameters
$u_1,\ldots,u_l$. Then there are $\Gamma_p$-equivariant
isomorphisms
$\widehat{{\mathcal O}}_{p,O_{\tau}}\simeq K_p[[u_1,\ldots,u_l]]$ and
$\widehat{{\mathcal O}}_{p,Y}\simeq K_p[[u_1,\ldots,u_l]][[\tau^\vee]]=
K_p[[(\tau\oplus^{\Gamma_p} \langle e_1,\ldots,e_l \rangle)^\vee]] $
where $\widehat{{\mathcal O}}_{p,O_{\tau}}=\widehat{{\mathcal O}}_{p,Y}^{T_\tau}$.
Therefore $\sigma_p=\tau\oplus^{\Gamma_p} \langle e_1,\ldots,e_l
\rangle$ and $\underline{\operatorname{sing}^{\Gamma_p}}(\sigma_p)=
\underline{\operatorname{sing}^{\Gamma_\tau}(\tau)}$.
\qed
By
Lemmas \ref{le: Dem3} and by \ref{le: 1} the singularity type
of a nonclosed point $p$,
$ {\rm sing}(p):={\rm sing}(\sigma_p)$ (resp. $ {\rm sing}^{\Gamma_p}(p):=
(\Gamma_p, \underline{\rm sing}^{\Gamma_p}(\sigma_p))$, is
uniquely determined.
Moreover singularity type determines a stratification ${\rm
Sing}(Y)$ (resp. ${\rm
Sing}^\Gamma(Y)$) on $Y$ such that all points in the same stratum have the
same singularity type. This yields
\begin{lemma}\label{le: singularity type2} Let $\Gamma$
act on
$\widehat{X}_{\Delta}:=X_{\Delta}\times_{X_{\sigma}}\widehat{X}_{\sigma}$
There is a
stratification ${\rm Sing}^\Gamma(\widehat{X}_{\Delta})$ of
$\widehat{X}_{\Delta}$ which is determined by singularity
type and therefore preserved by any $\Gamma$-equivariant
automorphism of $\widehat{X}_{\Delta}$. \qed
\end{lemma}
\bigskip
\subsection{Semicomplex associated to a stratified toroidal
variety}
For a $\Gamma$-semicone $\sigma$ we denote by $X_\sigma$ the
associated stratified toric variety.
\begin{definition} \label{de: associated semicomplex}
Let $(X,S)$ be a $\Gamma$-stratified toroidal
variety. We say that a semicomplex $\Sigma$ is {\it associated\/} to
$(X,S)$ if there is a bijection $\Sigma \to S$ with the following
properties: Let $\sigma \in \Sigma$ map to
$s={\operatorname{strat}}_X(\sigma) \in S$. Then any $x \in
s$ admits an open $\Gamma$-invariant neighborhood
$U_\sigma \subset X$ and a $\Gamma$-equivariant $\Gamma_s$-smooth morphism
$ \phi_\sigma \colon U_\sigma \to X_{\sigma}$ of stratified varieties such that $s
\cap U$ equals $\phi_\sigma^{-1}(O_{\sigma})$ and the intersections $s' \cap
U$, $s' \in S$, are precisely the inverse images of the strata of
$X_{\sigma}$ and the action of $\Gamma$
on $X_{\sigma}$ extends the action of
$\Gamma_s$.
We call the smooth morphisms $U_\sigma\to X_{\sigma}$ from the above definition
{\it charts}. A collection of charts
satisfying the conditions from the above definition is
called an {\it atlas}.
\end{definition}
\begin{remarks}
\begin{enumerate}
\item Different
$(X,S)$ may have the same associated semicomplex $\Sigma$.
Smooth varieties endowed with the trivial stratification have the
associated semicomplex consisting of one zero cone.
\item
The action of $\Gamma_s=\Gamma_\sigma$ is fixed for
$\sigma$ while the action of $\Gamma$ on $X_\sigma$ depends upon charts.
\end{enumerate}
\end{remarks}
\noindent
\begin{lemma}\label{le: associated semicomplex} For any
$\Gamma$-stratified toroidal variety $(X,S)$ there
exists a unique (up to isomorphism) associated $\Gamma$-semicomplex
$\Sigma$. Moreover $\tau\leq\sigma$ iff
$\overline{{\operatorname{strat}}_X(\tau)}\supset {\operatorname{strat}}_X(\sigma)$.
$(X,S)$ is a toroidal embedding iff
$\Sigma$ is a complex.
\end{lemma}
\noindent {\bf Proof } First we assign to any stratum $s$
a semicone $\sigma$.
By Definition \ref{de: stratified toroidal} there is a
$\Gamma_s$-smooth morphism
${\phi}_\sigma: U_\sigma \rightarrow X_{\sigma}$ into a
stratifed toric variety $(X_\sigma,S_\sigma)$ preserving strata.
Note that $\Gamma_s$ acts trivially on $O_\sigma$. So we
can assume that $\sigma=\underline{\sigma}$ is of maximal dimension in
$N_\sigma$ by composing $\phi$, if necessary, with a suitable projection.
We define the $\Gamma_s$-semicone $\sigma$ to be the $\Gamma_s$-semifan
associated to the $\Gamma_s$-stratified toric variety $(X_{\sigma},
S_{\sigma})$.
Then there is an open subset $U$ of $U_\sigma$
intersecting $s$ and a { $\Gamma_s$-\'etale} morphism
$\phi:=\phi_\sigma\times\psi: U \to X_{\sigma}\times {\bf A}^k=X_{\tau}$
preserving strata,
where $\tau={\sigma\times \langle
e_1,\ldots, e_{{\rm dim}(s)}}\rangle $,
and $\psi: U\to {\bf A}^k$ is a morphism defined by local
parameters $u_1,\ldots,u_k$ on $s$. The stratification of
$X_\tau$ is defined by the embedded semifan $\sigma\subset
\overline{\tau}$, consisting of the faces of the semicone $\sigma$ in the
lattice $N_\tau$.
Then
$\widehat{\phi}: \widehat{X}_x \to
\widehat{X}_\tau$ is a $\Gamma_s$-equivariant isomorphism
preserving strata.
By Lemma \ref{le: Dem2}
the embedded $\Gamma_s$-semifan $\sigma\subset \tau$ and
the $\Gamma_s$-semicone $\sigma$
are determined uniquely up to isomorphism.
We write $s={\operatorname{strat}}_X(\sigma)$ and define $\Gamma_\sigma:=\Gamma_s$.
Assume that the closure of a stratum $s={\operatorname{strat}}_X(\sigma)$
contains a stratum $t={\operatorname{strat}}_X(\tau)$.
Let $\phi_\tau: U_\tau
\to X_\tau$ denote a chart associated with $\tau$.
Then the stratum $s\cap U_\tau$
and the strata of $U_\tau$ having $s\cap U_\tau$ in their closure determine a
semicone $\sigma_s\subset \overline{\tau}$. It follows from the
uniqueness of $\sigma$ that there is a
saturated embedding $i^\tau_{\sigma}$ of the semicone
$\sigma$ into the semicone $\tau$ with the image
$i^\tau_{\sigma}(\sigma)=\sigma_s$. Then we shall write $\sigma\leq\tau$.
The $\Gamma$-semicomplex $\Sigma$ is defined as the collection of
the $\Gamma_\sigma$-semicones $\sigma$, and the saturated face
inclusions $i^\tau_{\sigma}$ for $\sigma\leq \tau$.
Now let $\Sigma$ and $\Sigma'$ be two semicomplexes
associated to $(X,S)$. By uniqueness there are
isomorphisms of semicones $j_\sigma: \sigma\to \sigma'$ (see Definition
\ref{de: semicomplexes isomorphism}). These isomorphisms
induce an isomorphism of semicomplexes.
The second part follows from the fact that
locally toroidal embeddings correspond to fans consisting
of all faces of some cone.
\qed
\begin{lemma} \label{le: asemicomplex} Let $(X_\Sigma,S)$
be a $\Gamma$-stratified
toric variety corresponding to an embedded
$\Gamma$-semifan
$\Omega\subset\Sigma$. Then $(X_\Sigma,S)$ is a $\Gamma$-stratified
toroidal variety with the associated
$\Gamma$-semicomplex $\Omega^{\rm semic}$.
There is an atlas $${\mathcal U}^{\rm can}(X_\Sigma,S)={\mathcal U}(\Sigma,\Omega)$$ on
$(X_\Sigma,S)$ defined as follows: For any $\sigma$ in $N$ such that
$\omega(\sigma)=\omega$ in $N_\omega$
there is a chart
$\phi_\sigma: X_{\sigma}\to X_{\omega}$ given
by any projection $\pi^\omega_\sigma: \sigma
\to \omega$ such that $\pi^\omega_{\sigma|\omega}={\operatorname{id}}_{|\omega}$.
\end{lemma}
\noindent {\bf Proof.} Follows from Proposition \ref{le: semifans
correspondence} and from the definition of the associated
semicomplex.\qed
\subsection{Local properties of orientation}
\begin{definition}\label{de: orientation}
We shall call a proalgebraic group
$G$ {\it connected} if
it is a connected affine scheme.
For any proalgebraic group $G=\lim_{\leftarrow}G_i$, denote by
$G^0$ its maximal
connected proalgebraic subgroup $G^0=\lim_{\leftarrow}G^0_i$.
\end{definition}
\begin{lemma} $G=\lim_{\leftarrow}G_i$ is connected if each
$G_i$ is irreducible.
\end{lemma}
\noindent{\bf Proof.}
By Lemma \ref{le: epimorphisms} we can assume all morphisms $G_i\to
G_j$ to be epimorphisms
and $K[G]=\bigcup K[G_i]$.
If $g_1,g_2\in K[G]$ and $g_1\cdot g_2=0$ then $g_1,g_2\in
K[G_i]$ for some $i$ and $g_1$ or $g_2$ equals zero. \qed
\begin{lemma} \label{le: images} Let
$G=\lim_{\leftarrow}G_i$ be a connected proalgebraic group. Then the image
$\phi_i(G)\subset G_i$ of the
natural homomorphism $\phi_i:G\to G_i$ is connected. \qed
\end{lemma}
\begin{definition}\label{de: the same orientation2} Let $X$
be a stratified (resp. $\Gamma$-stratified) toroidal
scheme over $K$. We say
that an automorphism (resp. $\Gamma_x$-equivariant automorphism)
$\phi$ of $X$ {\it preserves
orientation} at a $K$-rational point $x=\phi(x)$ if
it induces an automorphism $\widehat{\phi}\in
{\operatorname{Aut}}(\widehat{X}_x,S)^0$ (resp. $\widehat{\phi}\in
{\operatorname{Aut}}^\Gamma(\widehat{X}_x,S)^0$).
\end{definition}
\begin{definition}\label{de: the same orientation}
We say that two \'etale morphisms (resp. $\Gamma$-\'etale
morphisms) of stratified toroidal
schemes over $K$,
$f_1,f_2:(X,S)\to (Y, T)$, {\it determine the same
orientation} at a closed $K$-rational point
$x\in X$ if $f_1(x)=f_2(x)$ and
$(\widehat{f}_2)^{-1}\circ \widehat{f_1}\in {\rm
Aut}(\widehat{X}_x,S)^0$
(resp. $(\widehat{f}_2)^{-1}\circ \widehat{f_1}\in {\rm
Aut}^\Gamma(\widehat{X}_x,S)^0$).
\end{definition}
In further considerations we shall consider the case of
$\Gamma$-stratified toroidal schemes. The case of
stratified toroidal schemes corresponds to the situation
when $\Gamma$ is a trivial group.
\begin{definition} Let $(X,S)$ is a $\Gamma$-stratified toroidal
scheme and $x\in X$ be a $K$-rational point. We call
functions $y_1,\ldots,y_k$ {\it locally toric parameters}
if $y_1=\phi^*_x(u_1),\ldots, y_k=\phi^*_x(u_k)$,
where $u_1,\ldots,u_k$ are semiinnvariant generators at the
orbit point $O_\sigma$ on
a $\Gamma$-stratified toric variety $X_{\sigma}$, and
$\psi_x: U_x\rightarrow
X_{\sigma}$ is a $\Gamma_x$-\'etale morphism from an open
$\Gamma_x$-invariant
neighborhood $U_x$ of $x$ such that
$\psi_x(x)=O_\sigma$
and the intersections $s'\cap U$, $s'\in S$, are precisely the inverse
images of strata of $X_\sigma$.
\end{definition}
\begin{lemma} Locally toric parameters exist for any
$K$-rational point of a $\Gamma$-stratified toroidal
scheme $X$.
\end{lemma}
\noindent{\bf Proof} Let $s$ be the stratum through $x$.
By Definition \ref{de: G-stratified
toroidal} there is a
$\Gamma_x$-\'smooth morphism from an open
neighborhood $U_x$ of $x$ such that
$s=\phi_x^{-1}(O_\sigma)$
and the intersections $s'\cap U$, $s'\in S$, are precisely the inverse
images of strata of $X_\sigma$. Let $x_1,\ldots,x_k$ be
local paramters of $s$ at $x$. Set $g:U\to {\bf A}^1$,
where $g(x)=(x_1,\ldots,x_k)$. Then the morphism
$\psi_x:=\phi_x\times g$ is $\Gamma_x$-\'etale and defines
locally toric parameters at $x$. \qed
\begin{lemma} \label{le: loc} Let $(X,S)$ be a $\Gamma$-stratified
toroidal scheme and $x\in X$ be a $K$-rational point.
Let $y_1,\ldots,y_n$ be locally toric
paramters at $x\in X$.
The ideals of closures of
the strata $I_s\subset {{\mathcal O}}_{x,X}$
are generated by subsets of $\{u_1,\ldots,u_k\}$.
\end{lemma}
\noindent{\bf Proof.} It suffices to show the lemma for characters
generating $\sigma^\vee$ on a $\Gamma$-stratified toric
variety $X_\sigma$. Let $T$ be the big torus on $X_\sigma$.
Each stratum $s$ is ${T}$-invariant irreducible
hence contains a ${\bf T}$- orbit $O_\tau$, where
$\tau\preceq \sigma$. We conclude that
$\overline{s}=\overline{O_\tau}$. But then
$I_{\overline{O_\tau}}\subset K[X_\sigma ]$ is generated by
functions corresponding to those generating functionals of
$\sigma^\vee$ which are not zero on $\tau$.\qed
\begin{lemma} \label{le: smooth}\begin{enumerate}
\item The group ${\rm Aut}^\Gamma(\widehat{X}_x,S)$
is connected
for any smooth $\Gamma$-stratified toroidal
scheme $(X,S)$ over $\operatorname{Spec}(K)$ and any
$K$-rational point $x\in X^\Gamma$.
\item
Any two $\Gamma$-\'etale morphisms $g_1: (X,S)\to (Y,R)$,
$g_2: (X,S)\to (Y,R)$
between smooth $\Gamma$-stratified toroidal schemes such that $g_1(x)=g_2(x)$ determine the same orientation at $x$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} (1) We can replace $X$ by ${\bf A}^k$ since
$(\widehat{X}_x,S)\simeq (\widehat{{\bf A}^k}_0,S_A)$ for a
toric stratification $S_A$ on ${\bf A}^k$. Let $m$ be the
maximal ideal of $0\in {\bf A}^k$.
The automorphism $g\in G:={\rm
Aut}^\Gamma(\widehat{{\bf A}^k}_0,S_A)$ is defined by
$\Gamma$-semiinvariant functions
$g^*(x_1),\ldots,g^*(x_k)$, where $x_1,\ldots, x_k$ are the
standard coordinates on ${\bf A}^k$ such that the
$\Gamma$-weights of $g^*(x_i)$ and
$x_i$, where $i=1,\ldots,k$, are equal. There is a
birational map $\alpha: {\bf A}^1\to G$ defined by
$$\alpha(z):=(x_1,\ldots,x_k)\mapsto
((1-z)g^*(x_1)+zx_1,\ldots,(1-z)g^*(x_k)+zx_k).$$
Note that $\alpha(z)$
defines automorphisms for the open subset $U$ of ${\bf A}^1$,
where the linear parts of coordinates of $\alpha(z)$ are
linearly independent. By Lemma \ref{le: loc} the closure
of each stratum $s$
is described by a subset of $\{x_1,\ldots,x_k\}$. Since $g$
preserves strata, $\overline{s}$ is described by the
corresponding subset of $\{g^*(x_1),\ldots,g^*(x_k)\}$.
Thus the corresponding coordinates of $\alpha(z)$, $z\in
U$, belong to the ideal $I_{\overline{s}}$. Since they are
linearly independent of order $1$ they generate the ideal
$\frac{I_{\overline{s}}}{{I_{\overline{s}}}\cdot m}$ and
by
the Lemma of Nakayama
they generate the ideal $I_{\overline{s}}$. Therefore the
automorphisms $\alpha(z)$, $z\in
U$, preserve strata.
The morphism $\alpha$ ''connects'' the identity $\alpha(1)={\rm
id}$ to an arbitrary
element $\alpha(0)=g\in {\rm Aut}^\Gamma(\widehat{X}_x,S)^0 $.
(2) Follows from (1). \qed
\begin{lemma} \label{le: extensions}
Let $f: (X,S)\to (Y,R)$ be a $\Gamma$-smooth morphism of relative
dimension $k$ between $\Gamma$-stratified toroidal schemes. Let $x\in X^\Gamma$ be
a closed $K$-rational point. Let $\Gamma$ act trivially on
${\bf A}^k$ and $g_1,g_2:
X\to {\bf A}^k$ be any two $\Gamma$-equivariant morphisms such that $g_1(x)=g_2(x)=0$ and $f\times
g_i: (X,S)\to (Y,R) \times {\bf A}^k$ are $\Gamma$-\'etale for $i=1,2$.
Then $f\times
g_1$ and $f\times g_2$ determine the same orientation at
$x$.
\end{lemma}
\begin{definition}\label{de: extensions}
We shall call such a morphism $f\times g$ an {\it
\'etale extension} of $f$ and denote it by $\widetilde{f}$.
\end{definition}
\noindent {\bf Proof of \ref{le: extensions}}. Let $y=f(x)$ and
$y_1,\ldots,y_l$ be locally toric parameters at $y\in Y$. Let
$x_1,\ldots,x_k$ be standard coordinates at $0\in {\bf A}^k$. Set
$v_i=f^*(y_i)$ for $i=1,\ldots,l$ and $w^1_j=g_1^*(x_j)$,
$w^2_j=g_2^*(x_j)$ for $j=1,\ldots,k$. Then
$v_1,\ldots,v_l,w^1_1,\ldots,w^1_k$ and
$v_1,\ldots,v_l,w^2_1,\ldots,w^2_k$ are locally toric parameters at $x$. The
automorphism $\widehat{(f\times
g_1)}_x\circ \widehat{(f\times g_2)}_x^{-1}: \widehat{X}_x\to
\widehat{X}_x$ maps $w^1_i$ to $w^2_i$.
We can find a linear
automorphism $\alpha_1\in \{{\rm id_v}\} \times \operatorname{Gl}(k)$, which
preserves $v$-coordinates and
acts nontrivially on $w^2_i$-coordinates so that
$\alpha_1^*(v_i)=v_i$ and $\alpha_1^*(w^2_i)=w^1_i+z_i$,
where $z_i$ are some functions from the ideal
$(w^2_1,\ldots,w^2_k)^2+ (v_1,\ldots,v_l)$. Clearly
$\alpha_1\in {\rm Aut}^\Gamma(\widehat{X}_x,S)^0$.
Now consider the morphism ${\bf A}^1\to
{\rm Aut}^\Gamma(\widehat{X}_x,S)^0$ such that $ {\bf A}^1\ni t \mapsto
\phi_t$, where $\phi^*_t(v_i)=v_i$
and $\phi^*_t(w^1_j)=w^1_j+t\cdot z_i$. This shows that
$\phi_1\in{\rm Aut}^\Gamma(\widehat{X}_x,S)^0$. Finally $(\widehat{f\times
g_1}_x)\circ (\widehat{f\times g_2}_x)^{-1} =
\phi^{-1}_1\circ\alpha_1\in {\rm Aut}^\Gamma(\widehat{X}_x,S)^0$. \qed
\bigskip
\noindent \begin{definition}
We say that two $\Gamma$-smooth morphisms $f_1,f_2:(X,S)\to (Y,R)$
of dimension $k$
{\it determine the same orientation} at a closed $K$-rational point $x\in X^\Gamma$
if $f_1(x)=f_2(x)$ and
there exist \'etale extensions
$e{f_1},\widetilde{f_2}: (X,S)\to (Y,R)\times {\bf A}^k$
which determine the same orientation at
$x$.
\end{definition}
\begin{lemma}\label{le: group of automorphisms}
Let $f:(X,S)\to (Y,R)$ be a $\Gamma$-smooth morphism of $\Gamma$-stratified toroidal
schemes over $K$. Let $x\in X^\Gamma$ and $y=f(x)\in Y^\Gamma$ be
$K$-rational points and $s=f^{-1}(y)\in
S$ be the stratum through $x$. Let $y_1,\ldots,y_l$ be
locally toric $\Gamma_x$-semiinvariant parameters at $y$ and
$v_1=f^*(y_1),\ldots,v_l=f^*(y_l), w_1,\ldots,w_k$ be locally toric
$\Gamma_x$-semiinvariant parameters at $x$.
Set $R:=K[[w_1,\ldots,w_k]]$.
Then
$\widehat{{\mathcal O}}_{X,x}=R[[v_1,\ldots,v_l]]/I$, where $I$ is the ideal span by binomial relations in $v_i$.
${\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)$ is a proalgebraic group of
$\Gamma_x$-equivariant automorphisms
preserving strata and the subring $R$,
and the monomorphism
$\beta:{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)\to {\rm
Aut}^\Gamma(\widehat{X}_x,S)$ induces an isomorphism $$\overline{\beta}:{\rm
Aut}^\Gamma_{R}(\widehat{X}_x,S)/{\rm
Aut}^\Gamma_{R}(\widehat{X}_x,S)^0 \to
{\rm Aut}^\Gamma(\widehat{X}_x,S)/{\rm Aut}^\Gamma(\widehat{X}_x,S)^0.$$
\end{lemma}
\bigskip
\noindent
{\bf Proof.} It suffices to construct a surjective morphism $\alpha:{\rm Aut}^\Gamma(\widehat{X}_x,S)\to
{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)$ with
connected fibers such that $\alpha\circ\beta={\operatorname{id}}_{{\rm
Aut}^\Gamma_{R}(\widehat{X}_x,S)}$.
Any automorphism $\phi$ from ${\rm
Aut}^\Gamma(\widehat{X}_x,S)$ maps $w_1,\ldots,w_k,v_1,\ldots,v_l$ to
$w'_1,\ldots,w'_k,v'_1,\ldots,v'_l$. We put
$\alpha(\phi)^*(w_i)=w_i$,
$\alpha(\phi)^*(v_j)=v'_j$. Then $\alpha(\phi)$ is an endomorphism
since $w_i$ are algebraically independent of $v_j$. It is an
automorphism since we can easily define an inverse homomorphism.
The fiber $\alpha^{-1}(\alpha(\phi))$ consists of all elements of the
type $\phi=\alpha(\phi)\circ\phi_1$, where $\phi_1^*(w_i)=w'_i$ and
$\phi_1^*(v'_j)=v'_j$.
It is connected: for
any two elements $\phi=\alpha(\phi)\circ\phi_1$ and $
\phi'=\alpha(\phi)\circ\phi'_1$ we can define a rational map
$\Phi:{\bf A}^1\to {\rm
Aut}^\Gamma(\widehat{X}_x,S)$ by
$\Phi(t)^*(v'_j)=v'_j$, $\Phi(t)^*(w_i)=tw_i+(1-t)w'_i$.
The latter
is defined on an open subset $U\subset {\bf A}^1$ containing $0$ and $1$.
\qed
\begin{lemma}\label{le: sections}
Let $\phi_i: (X,S)\to (Y,R)$ for $i=1,2$ be two
$\Gamma$-smooth morphisms of $\Gamma$-stratified toroidal schemes such that
$\phi_1(x)=\phi_2(x)=y\in Y$ for a $K$-rational point $x\in
X^\Gamma$ and
strata in $S$ are preimages of
strata in $T$ and $\phi^{-1}_1(y)=\phi^{-1}_2(y)$. Assume
there exists a smooth scheme $V$ with a trival action of
$\Gamma$ and a $\Gamma$-equivariant morphism $g:
X\to V$ such that $\phi_i\times
g: (X,S)\to (Y,R) \times V$ are $\Gamma$-smooth. Define
$X':=g^{-1}(g(x)), S':=\{s\cap X'\mid s\in S\}$.
Then $\phi_1$ and $\phi_2$ determine the
same orientation iff their restrictions $\phi'_i: (X',S')\to
(Y,R)$ do.
\end{lemma}
\noindent{\bf Proof.}
We can assume that $g$ is $\Gamma$-\'etale by replacing, if necessary, $(Y,R)$ with
$(Y,R)\times {\bf A}^m$, where $m=\dim(X)-\dim(Y)-\dim(V)$,
and $\phi_i: (X,S)\to (Y,R)$ and $\phi_i\times
g: (X,S)\to (Y,R) \times V$ with $\Gamma$-smooth morphisms $\widetilde{\phi_i}:
(X,S)\to (Y,R)\times {\bf A}^m$ and $\Gamma$-\'etale morphisms $\widetilde{\phi_i}\times
g: (X,S)\to (Y,R)\times {\bf A}^m \times V$.
Let $y=f(x)$ and
$y_1,\ldots,y_l$ be locally toric $\Gamma$-semiinvariant parameters at $y\in Y$. Let
$x_1,\ldots,x_k$ be locally toric parameters at $p:=g(x)\in V$. Set
$v^1_i=\phi_1^*(y_i)$, $v^2_i=\phi_2^*(y_i)$ for
$i=1,\ldots,l$ and
$w_j=g^*(x_j)$ for $j=1,\ldots,k$. Then
$v^1_1,\ldots,v^1_l,w_1,\ldots,w_k$ and
$v^2_1,\ldots,v^2_l,w_1,\ldots,w_k$ are locally toric parameters at $x$. The
automorphism $\alpha:=\widehat{(\phi_1\times
g)}_x^{-1}\circ \widehat{(\phi_2\times g)}_x: \widehat{X}_x\to
\widehat{X}_x$ maps the first set of
parameters onto the second one. Thus $\alpha$ belongs to the
group of automorphisms ${\rm
Aut}^\Gamma_{R}(\widehat{X}_x,S)$ preserving $R=K[[w_1,\ldots,w_k]]$.
The restriction of each automorphism from ${\rm
Aut}^\Gamma_{R}(\widehat{X}_x,S)$ to
$X'={\rm Spec}(\widehat{{\mathcal O}}_{X,x}/(w_1,\ldots,w_k))$ is an
automorphism. On the other hand we can write
$\widehat{{\mathcal O}}_{X,x}=\widehat{{\mathcal O}}_{X',x}[[w_1,\ldots,w_k]]$.
Hence each automorphism in ${\rm
Aut}^\Gamma_{K}(\widehat{X'_x},S')$ determines an automorphism in
${\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)$. We come to
a natural epimorphism of proalgebraic
groups:
$${\rm res}: {\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)\to
{\rm Aut}^\Gamma_{K}(\widehat{X'_x},S')
.$$
The kernel of ${\rm res}$ is a proalgebraic
group $H$
consisting of all automorphisms
$\beta\in
{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)$ which can
be written in the form $\beta(v^1_i)=v^1_i+r_i$, $\beta(w_j)=w_j$, where
$r_i\in (w_1,\ldots,w_k)\cdot\widehat{{\mathcal O}}_{X,x}$. For any fixed $\beta$ and $t\in
{\bf A}^1$ yield
$r_i^t:=r_i(v^1_1,\ldots,v^1_l,t\cdot w_1,\ldots,t\cdot w_k)$ and
$\beta^t(v^1_i)=v^1_i+r^t_i$, $\beta^t(w_j)=w_j$. This
gives a morphism $t:{\bf A}^1\to H$ such that
$t(1)=\beta^1=\beta$ and $t(0)={\rm id}$. Consequently, $H$ is
connected and $$res^{-1}({\rm Aut}^\Gamma_{K}(\widehat{X'}_x,S')^0=
H\cdot({\rm Aut}^\Gamma_{K}(\widehat{X'}_x,S')^0=
{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)^0.$$
By the
above and Lemma \ref{le: group of automorphisms} the homomorphisms
$$ {\rm Aut}^\Gamma_{K}(\widehat{X}_x,S)
\buildrel\beta\over\longleftarrow
{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)
\buildrel {\rm res}\over\longrightarrow
{\rm Aut}^\Gamma_{K}(\widehat{X'}_x,S')$$
define isomorphisms
$$ {\rm Aut}^\Gamma_{K}(\widehat{X}_x,S)/
{\rm Aut}^\Gamma_{K}(\widehat{X}_x,S)^0
\buildrel \overline{\beta}^{-1} \over \longrightarrow
{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)/{\rm Aut}^\Gamma_{R}(\widehat{X}_x,S)^0
\buildrel\overline{\rm res}_1\over\longrightarrow
{\rm Aut}^\Gamma_{K}(\widehat{X'}_x,S')/{\rm
Aut}^\Gamma_{K} (\widehat{X'}_x,S')^0.$$
Finally we see that $\widehat{(\phi_1\times
g)}_x\circ \widehat{(\phi_2\times g)}_x^{-1}\in {\rm Aut}^\Gamma_{K}(\widehat{X}_x,S)^0$ iff
$\widehat{\phi'_1}_x\circ \widehat{\phi'_2}_x^{-1}\in {\rm Aut}^\Gamma_{K}(\widehat{X'}_x,S')^0$
\qed
\subsection{Orientation on stratified toroidal varieties}
\begin{definition} \label{de: orientation3}
Let $(X,S)$ be a
$\Gamma$-stratified toroidal variety (or $\Gamma$-stratified toroidal scheme)
with an associated $\Gamma$-semicomplex
$\Sigma$. Let $\tau\leq \sigma$ and $x\in
s$. Let $\imath^\sigma_\tau: \tau\to\sigma$ denote the standard
embedding
and $\phi_{\sigma}: U_{\sigma}\to X_{\sigma}$ be
a chart. For any $\sigma'\preceq \sigma$ in $N_\sigma$
such that $\omega(\sigma')=\tau$ in $N_{\tau}$ we denote
by $\phi^{\sigma'}_{\sigma}:
U_\sigma^{\sigma'}:=\phi^{-1}_{\sigma} (X_{\sigma'})\to
X_{\sigma'}$ the restriction of $\phi_{\sigma}$ to $U_\sigma^{\sigma'}$.
Let $\pi^\tau_{\sigma'}:X_{\sigma'}\to X_{\tau}$
denote the toric morphism induced by
any projection
$\overline{\pi}^\tau_{\sigma'}:
{\sigma'}\to \tau$ such that
$\overline{\pi}^\tau_{\sigma'}\circ \imath^{\sigma}_{\tau}= {\operatorname{id}}_{|\tau}$.
\begin{enumerate}
\item
We say that the
$\Gamma$-stratified
toroidal variety $(X,S)$ with atlas ${\mathcal U}$ is {\it oriented}
if for any two charts $\phi_i: U_i\to X_{\sigma_i}$, where $i=1,2$, and
any $\sigma'_i\preceq\sigma_i$
such that $\omega(\sigma'_i)=\sigma \leq\sigma_i$
the $\Gamma_\sigma$-smooth morphisms
$\pi^\sigma_{\sigma'_i}\phi^{\sigma'_i}_{\sigma_i}:
U_{\sigma_i}^{\sigma'_i}\to X_{\sigma}$
determine the same orientation at
any $x\in U_{\sigma_1}^{\sigma'_1}\cap U_{\sigma_2}^{\sigma'_2}
\cap {\operatorname{strat}}_X(\sigma)$.
\item
Let $(X,S)$ be a $\Gamma$-stratified
toroidal scheme with atlas ${\mathcal U}$ which
contains a reduced subscheme $W$ of finite type over $K$.
We say that
$(X,S)$ is {\it oriented} along $W$ if for any two
charts $\phi_i: U_i\to X_{\sigma_i}$, where $i=1,2$, and
any $\sigma'_i\preceq\sigma_i$
such that $\omega(\sigma'_i)=\sigma \leq\sigma_i$
the $\Gamma_\sigma$-smooth morphisms
$\pi_{\sigma'_i}\phi_{\sigma_i}^{\sigma'_i}:
U_{\sigma_i}^{\sigma'_i}\to X_{\sigma}$
determine the same orientation at
any $x\in U_{\sigma_1}^{\sigma'_1}\cap U_{\sigma_2}^{\sigma'_2}
\cap {\operatorname{strat}}_X(\sigma) \cap W$.
\item We shall call such $W$ a {\it $K$-subscheme}.
\end{enumerate}
\end{definition}
The following lemma is a generalization of Lemma \ref{le: asemicomplex}
\begin{lemma} \label{le: asemicomplex2} Let $(X_\Sigma,S)$
be a $\Gamma$-stratified toric variety corresponding to an embedded
semifan $\Omega\subset \Sigma$. Then $(X_\Sigma,S)$ is an oriented
$\Gamma$-stratified toroidal variety with the associated
oriented $\Gamma$-semicomplex $\Omega^{\rm semic}$ and atlas
${\mathcal U}^{\rm can}(\Omega,\Sigma)$.
\qed
\end{lemma}
\begin{remark} By Lemma \ref{le: extensions} the above
definition does not depend upon the choice of the projection $\pi_\sigma$.
\end{remark}
\begin{definition} Let $(X,S)$ be a $\Gamma$-stratified
toroidal variety (respectively a $\Gamma$-stratified
toroidal scheme with a $K$-subscheme $W$) with two atlases ${\mathcal U}_1$ and ${\mathcal U}_2$
such that $(X,S,{\mathcal U}_1)$ and $(X,S,{\mathcal U}_2)$ are oriented
(resp. oriented along W).
Then ${\mathcal U}_1$ and ${\mathcal U}_2$ on
$(X,S)$ are
{\it compatible} (resp. {\it compatible along $W$}) if
$(X,S,{\mathcal U})$, where ${\mathcal U}:={\mathcal U}_1\cup{\mathcal U}_2$, is oriented
(resp. oriented along $W$).
\end{definition}
\begin{lemma}\label{le: induce}
Let $f:(X,S)\to (Y,R) $ be a $\Gamma$-smooth morphism of $\Gamma$-stratified
toroidal schemes such that the strata in $S$ are preimages of
strata in $R$ and all strata in $R$ are dominated by strata
in $S$. Assume that $(Y,R,{\mathcal U})$ with associated
$\Gamma$-semicomplex $\Sigma$ is oriented along a
$K$-subscheme $W$. Define $f^*({\mathcal U}):= \{\phi f\mid \phi \in
{\mathcal U}\}$. Then $(X,S,f^*({\mathcal U}))$ with the associated
$\Gamma$-semicomplex $\Sigma$ is oriented along any
$K$-subscheme $W'\subset f^{-1}(W)$.
\end{lemma}
\noindent{\bf Proof.} Let $\phi:U\to X_\sigma$ be a chart
on $Y$. Then $\phi$ is $\Gamma$-equivariant,
$\Gamma_\sigma$-smooth morphism such that the intersections
$r\cap U$ are inverse images of strata of $X_\sigma$. Then
$\phi f:\phi^{-1}(U)\to X_\sigma$ has exactly the same
properties. Since $X_\sigma//\Gamma_\sigma$ exists it
follows from Lemma \ref{le: gsmoth} that $\phi f$ is $\Gamma_\sigma$-smooth.
\qed
\begin{lemma}\label{le: sections2}
Let $\phi_i: (X,S)\to (Y,R)$ for $i=1,2$ be two
$\Gamma$-smooth morphisms of $\Gamma$-stratified toroidal schemes such that
the strata in $S$ are preimages of
strata in $R$. Assume $(Y,R,{\mathcal U})$ is oriented
and
there exists a $\Gamma$-equivariant morphism $g:
X\to V$ into a smooth scheme with trivial action of
$\Gamma$, such that $\phi_i\times
g: (X,S)\to (Y,R) \times V$ are $\Gamma$-smooth. Set
$X':=g^{-1}(g(x)), S':=S\cap X'$. Let $W\subset X'$ be a $K$-subscheme.
Let $\psi_i: (X',S')\to
(Y,R)$ denote the restrictions of $\psi_i$.
Then $\phi^*_1({\mathcal U}) $ and $\phi^*_2({\mathcal U})$ are compatible on
$(X,S)$ along $W$ iff
$\psi^*_1({\mathcal U}) $ and $\psi^*_2({\mathcal U})$ are compatible on
$(X',S')$ along $W$.
\end{lemma}
\noindent{\bf Proof.} Follows from Lemma \ref{le: sections}.\qed
\begin{example} \label{ex: toroidal embeddings}
{\it The group of automorphisms of the formal completion of a
toroidal embedding at a closed point is connected}.
The formal completion of a
toroidal embedding at a closed point is isomorphic to
$\widehat{X}_{\sigma\times {\operatorname{reg}}(\sigma)}$, where
${\operatorname{reg}}(\sigma)$ is a regular cone and strata are defined by all
faces of $\sigma$.
Let $x_1,\ldots, x_k$ be semiinvariant coordinates on
$\widehat{X}_\sigma$ corresponding to characters
$m_1,\ldots,m_k$. Denote by $x_{k+1},\ldots,z_{k+l}$ the standard
coordinates on $X_{{\operatorname{reg}}(\sigma)}$ corresponfing to
characters $m_{k+1},\ldots,m_{k+l}$.
Let $\phi$ be an automorphism of
$\widehat{X}_{\sigma\times{\operatorname{reg}}(\sigma)}$ which preserves the torus
orbit stratification.
Since all the divisors in the orbit
stratification are preserved, the divisors
of $x_i$, where $i=1,\ldots,k$, are also preserved, so we have
$\phi^*(x_i)=x_is_i$,
where
$s_i=a_{i0}+a_{i\alpha}x^\alpha+\ldots$ is invertible
and $\alpha=(\alpha_1,\ldots,\alpha_{k+l})$
denote multiindices.
By Lemma \ref{le: extensions},
we may assume that $\phi^*(x_i)=x_i$ for $i=k+1,\ldots,l$.
An integral vector
$v:=(v_1,\ldots,v_{l+k})\in{\operatorname{int}}(\sigma\times {\operatorname{reg}}(\sigma))$ defines
the $1$-parameter
subgroup $t\mapsto t^v$ of the ''big'' torus $T$.
For any $t\in T$ let $t^v$ denote the automorphism defined as
$$x_i\mapsto x_i\cdot t^{\langle v,m_i\rangle}.$$
consider the morphism
$\phi_t:=t^{-v}\phi
t^v$. In
particular $\phi_1=\phi$ and $$x_i\circ\phi_t=
x_i(a_{i0}+t^{\alpha_1m_1v_1+\ldots+\alpha_{l+k}m_{l+k}v_{l+k}}
a_{i\alpha}x^\alpha+\ldots),
$$ \noindent where $\alpha_1m_1v_1+\ldots+\alpha_{l+k}m_{l+k}v_{l+k}=
\langle v, \alpha_1m_1+\ldots+\alpha_{l+k}m_{l+k} \rangle \geq
0$. Then $\phi_0$
is the well defined automorphism $x_i\mapsto a_{i0}x_i$. The latter
automorphism belongs to the torus $T$. Since we can ''connect'' any
automorphism $\phi$ with an element of the torus the
connectedness of the group of automorphisms follows.
{\it Any toroidal embedding is an oriented
stratified variety}.
By Lemma \ref{le: sections} and
by the connectedness of the automorphism group any two charts
determine the same orientation.
\end{example}
\begin{example} \label{ex: isolated} Let $X:=\{x\in {\bf A}^4\mid x_1x_2=x_3x_4\}$ be a
toric variety with isolated singularity. Then $X$ is a
stratified toroidal variety with the stratification
consisting of the singular point $p$ and of its complement.
The blow-up of the ideal of the point is a resolution of singularities
with the exceptional divisor determining the valuation. Consequently
the
valuation determined by the point is preserved by all
automorphisms in ${\operatorname{Aut}}(\widehat{X}_x,S)$
. We
have the natural homomorphism
$d:{\operatorname{Aut}}(\widehat{X}_x,S)\to {\rm Gl}({\operatorname{Tan}}_{X,x})$. The kernel of
$d$ is a connected proalgebraic group (see Lemma \ref{le:
group structure}(3)). The image
$d({\operatorname{Aut}}(\widehat{X}_x,S))$ consists of linear automorphisms
preserving the ideal of $x_1x_2-x_3x_4$, that is, multiplying the
polynomial by a nonzero constant. Hence
$d({\operatorname{Aut}}(\widehat{X}_x,S))=K^*\cdot O$, where $K^*$ acts by
multiplying the coordinates by $t\in K^*$ and $O$ is the group of
linear automorphisms preserving $x_1x_2-x_3x_4$. By the linear
change of coordinates $x_1=y_1-iy_2$, $x_2=y_1+iy_2$,
$x_3=y_3-iy_4$, $x_4=y_3+iy_4$ we transform the polynomial into
$y_1^2+y_2^2+y_3^2+y_4^2$ (${\rm char}(K)\neq 2$). This shows that $O$ is conjugate
to the group of orthogonal matrices with $K$-rational coefficients $O(4,K)$.
$O(4,K)$ consists of two components with $O(4,K)^0=SO(4,K)$.
Therefore ${\operatorname{Aut}}(\widehat{X}_x,S)=ker(d)\cdot K^*\cdot O$ consists
of two components. Fix any isomorphism $\phi:X\to
X_\sigma$, which can be considered as a chart.
An oriented atlas consists of charts
compatible with $\phi$. Consequently, there are two
orientations corresponding to any two incompatible
isomorphisms $\phi:X\to
X_\sigma$.
\end{example}
\subsection{Subdivisions of oriented semicomplexes}
\begin{definition} \label{de: oriented semicomplex} For any
$\Gamma$-semicone $\sigma$ denote by ${\rm
Aut}(\sigma)^0$ the group of all its automorphisms
which define the $\Gamma$-equivariant automorphisms of
${X}_{\sigma}$ preserving orientation.
By an {\it oriented semicomplex} (resp. {\it oriented
$\Gamma$-semicomplex}) we mean a semicomplex (resp.
$\Gamma$-semicomplex) $\Sigma$ for
which for any $\sigma\leq\tau\leq\gamma$ there is
$\alpha_\sigma\in {\rm Aut}(\sigma)^0$ for which
$\imath^\gamma_{\tau}\imath^\tau_{\sigma}=\imath^{\gamma}_\sigma\alpha_\sigma$.
\end{definition}
\begin{definition}\label{de: semicomplexes isomorphism2}
By an {\it isomorphism} of two oriented $\Gamma$-semicomplexes $\Sigma\to
\Sigma'$ we mean a bijection of faces
$\Sigma\ni\sigma\mapsto\sigma'\in\Sigma'$ along with a collection of
face isomorphisms $j_\sigma:\sigma\to\sigma'$
such that
for any $\tau\leq \sigma$, there is an automorphism
$\beta_\tau\in {\operatorname{Aut}}(\tau)^0$ such that
$j_{\sigma}\imath^\sigma_{\tau}=\imath^{\sigma'}_{\tau'}j_{\tau}\beta_\tau$.
\end{definition}
\begin{remarks}
\begin{enumerate}
\item An oriented semicomplex can be viewed as an oriented
$\Gamma$-semicomplex
with trivial groups $\Gamma_\sigma$.
\item
We will show that the $\Gamma$-semicomplex associated
to an oriented $\Gamma$-stratified toroidal variety is oriented (Lemma
\ref{le: orientation4}).
\item The description of the group
${\operatorname{Aut}}(\sigma)^0$ is given in Lemma \ref{le: orientation group}.
\end{enumerate}
\end{remarks}
\begin{lemma}
Let $\Delta$ be a subdivision of a fan $\Sigma$ and
$\sigma\in \Sigma$. Then \\$\Delta{|\sigma}:=\{\delta\in
\Delta \mid \delta\subset\sigma\}$ is a subdivision of
${\sigma}$.\qed
\end{lemma}
\bigskip
\noindent \begin{definition} \label{de: subdivision of
an oriented semicomplex} A {\it subdivision } of an oriented
$\Gamma$-semicomplex $\Sigma$ is a collection $\Delta$ of fans
$\Delta^\sigma$ in $N_\sigma$ where $\sigma\in \Sigma$
such that
\begin{enumerate}
\item For any $\sigma\in \Sigma$, $\Delta^\sigma$ is a
subdivision of $\overline{\sigma}$ which is
${\operatorname{Aut}}(\sigma)^0$-invariant.
\item For any $\tau\leq \sigma$,
$\Delta^{\sigma}|\overline{\tau}= \imath^{\sigma}_\tau(\Delta^\tau)$.
\end{enumerate}
\end{definition}
\bigskip
\begin{remarks}\begin{enumerate}
\item By abuse of terminology we shall understand by
a subdivision of a semicone $\sigma$ a subdivision of the
fan $\overline{\sigma}$.
\item A subdivision of an
oriented semicomplex can be viewed as a subdivision of
an oriented $\Gamma$-semicomplex
with trivial groups $\Gamma_\sigma$.
\item By definition vectors in faces
$\sigma$ of an oriented $\Gamma$-semicomplex $\Sigma$ are defined up
to automorphisms from ${\operatorname{Aut}}(\sigma)^0$. Consequently, the
faces of subdivisions $\Delta^\sigma$ are defined up
to automorphisms from ${\operatorname{Aut}}(\sigma)^0$ and in
general don't give a structure of semicomplex. Only
invariant faces of $\Delta^\sigma$ may define a
semicomplex.
\item The condition on $\Delta^\sigma$ to be
${\operatorname{Aut}}(\sigma)^0$-invariant is for {\it canonical}
subdivisions replaced with a somewhat stronger condition
of similar nature which says
that the induced morphism
$\widehat{X}_{\Delta^\sigma}:={X}_{\Delta^\sigma}\times_{{X}_{\sigma}}
\widehat{X}_{\sigma}\to \widehat{X}_{\sigma}$
is ${\operatorname{Aut}}^\Gamma(\widehat{X}_{\sigma})^0$-equivariant.
The latter condition can be translated into the
condition that all "new" rays of the subdivision are in the {\it stable support} of $\Sigma$.
\end{enumerate}
\end{remarks}
\begin{definition}\label{le: starsub}
Let $\Delta$ be a subdivision of an oriented $\Gamma$-semicomplex
$\Sigma$. Let $\varrho_\sigma$ be an ray passing through
the interior of the cone $\sigma\in\Sigma$ such that defining a collection of
rays $\varrho:=\{\varrho_\tau\in\tau\mid\tau\geq\sigma\}$
such that for any $\tau\geq\sigma$, the ray
$\varrho_\tau=\imath^\tau_\sigma(\varrho_\sigma)$ is
${\operatorname{Aut}}(\tau)^0$-invariant .
By the {\it star
subdivision} of $\Delta$ at $\rho$ we mean the subdivision
$$\varrho\cdot\Delta:=\{\varrho_\tau\cdot\Delta^\tau \mid \tau\geq\sigma\}\cup\{\Delta^\tau \mid
\tau\not\geq\sigma\}.$$
\end{definition}
\bigskip
\subsection{Toroidal morphisms of stratified toroidal varieties}
\begin{definition}\label{de: toroidal modification} Let $(X, S)$ be a
a stratified toroidal variety
(resp. a $\Gamma$-stratified toroidal variety)
with an atlas ${{\mathcal U}}$. We say that $Y$ is a {\it toroidal
modification} of $(X, S)$ if
\begin{enumerate}
\item There is given a proper morphism $f:Y\to X$ (resp.
proper $\Gamma$-equivariant morphism) such that
for any $x\in s={\operatorname{strat}}_X(\sigma)$ there exists a chart $x\in U_\sigma\to
X_{\sigma}$, a subdivision $\Delta^\sigma$ of $\sigma$,
and a fiber square
\[\begin{array}{rcccccccc}
&& & U_\sigma & \buildrel \phi_\sigma \over\longrightarrow & X_{\sigma}&&&\\
&&&\uparrow f & & \uparrow &&&\\
U_\sigma \times_{X_{\sigma}}
X_{\Delta^\sigma} &&\simeq& f^{-1}(U_\sigma)
&\buildrel \phi^f_\sigma \over \longrightarrow &
X_{\Delta^\sigma} &&& \\
\end{array}\]
\item
For any point $ x$ in a stratum $s$ every
automorphism (resp. $\Gamma_s$-equivariant automorphism)
$\alpha$ of $\widehat{X}_x$ preserving strata and orientation
can be lifted
to an automorphism (resp. $\Gamma_s$-equivaiant automorphism)
$\alpha'$ of $Y\times_X\widehat{X}_x$.
\end{enumerate}
\end{definition}
\begin{definition}
\label{de: canonical stratification}
Let $Y$ be a toroidal modification of a stratified toroidal
variety $(X,S)$.
By a {\it canonical stratification $R$} on $Y$ we mean
the finest stratification on $Y$ satisfying the conditions:
\begin{enumerate}
\item Strata of $R$ are mapped onto strata of $S$.
\item
For any chart $\phi_\sigma: U_\sigma\to X_\sigma$ on $(X,S)$ there is an
embedded subdivision
$\Omega^\sigma\subset\Delta^\sigma$ such that the strata of
$ R\cap f^{-1}(U_\sigma)$ are defined as inverse images
$(\phi^f_\sigma)^{-1}({\operatorname{strat}}(\omega))$ of strata on
$(X_{\Delta^\sigma}, S_{\Omega^\sigma})$.
\item
For any point $x$ in a stratum $s$ every
$\Gamma_s$-equivariant automorphism $\alpha$ of
$\widehat{X}_x$ preserving strata and
orientation
can be lifted
to an $\Gamma_s$-equivariant automorphism $\alpha'$ of
$Y\times_X\widehat{X}_x$ preserving strata.
\end{enumerate}
\end{definition}
If $Y$ is a toroidal modification of $(X,S)$ and $R$ is a
canonical stratification on $Y$ then
we shall also speak of a {\it toroidal
morphism} $(Y,R)\to (X,S)$ of stratified toroidal varieties.
\begin{remarks} \begin{enumerate}
\item The definition of a toroidal morphism of
stratified toroidal varieties is a generalization of
the definition of a toroidal morphism of toroidal embeddings.
\item The condition (2) of Definition
\ref{de: toroidal modification} is similar to
Hironaka's ''allowability'' condition, used by Mumford in the theory of toroidal
embeddings (see \cite{KKMS} and section
\ref{se: toroidal embeddings}). It means that a toroidal morphism, which
by condition (1) is induced locally by charts, in fact
does not depend on the charts i.e.,
if a morhism which satisfies
the Hironaka condition is induced locally by the diagram (1) for
some chart, then it is also induced locally by any other
chart and the given subdivision
(see Lemma \ref{le: isomorphisms}).
\item In the above definition $\Omega^\sigma$ consists of those
faces of $\Delta^\sigma$ whose relative interior intersect the {\it
stable support} (see Lemma \ref{le: stab-subdivisions}).
\item Only some special subdivisions $\Delta^\sigma$ of $\sigma$
define (locally)
toroidal modifications (see example \ref{ex: example}).
These subdivisions, called {\it
canonical}, induce modifications which are independent
of the charts. They are characterized by the property that
"new" rays of the subdivision are contained in the stable
support (Proposition \ref{pr: simple}).
\end{enumerate}
\end{remarks}
\begin{example} \label{ex: example}
Let $X= {\bf A}^2$ and $S$ be the stratification on $X$ that
consists of the point $s_0=(0,0)$ and its complement $s_1=
{\bf A}^2\setminus \{(0,0)\}$. Then $(X,S)$ is an oriented stratified
toroidal variety corresponding to the semicomplex $\Sigma=
\{\sigma_{0},\sigma_{1}\}$
where $\sigma_{0}=\langle(1,0),(0,1)\rangle$, $\sigma_{1}=\{(0,0)\}$.
\begin{itemize}
\item Let $Y\to X$ be a toric morphism corresponding to
the normalized blow-up of $I=(x^2,y)$. This morphism of
toric varieties corresponds to the subdivision
$\Delta:=\langle (1,2)\rangle\cdot\Sigma$ of $\Sigma$.
Therefore condition (1) of Definition \ref{de: toroidal
modification} is satisfied
for the identity chart.
The condition (2) of the definition is not satisfied. The
automorphism $\alpha$ defined by $\alpha(x)=y$, $\alpha(y)=x$
doesn't lift.
\item Let $Y\to X$ be a blow-up of $I=(x,y)$. Then $Y$ is a
toric variety and let $S_Y$ be
the stratification consisting of $s_1$ and the toric orbits
in the exceptional curve ${\bf P}^1$: $s_2=\{0\}$,
$s_3=\{\infty\}$ and $s_4={\bf
P}^1\setminus\{0\}\setminus\{\infty\}$. Then $Y$ is a
toroidal \underline{modification} of $(X,S)$. The
automorphism $\alpha$ defined by $\alpha(x)=y$, $\alpha(y)=x$
lifts to the automorphism permuting $s_2$ and $s_3$. Therefore
$(Y,S_Y)\to (X,S)$ is NOT a toroidal \underline{morphism}.
\item Let $Y\to X$ be a blow-up of $I=(x,y)$ and $S_Y$ be
the stratification consisting of the exceptional curve $s_2=D$
and its complement $s_1$. Then $(Y,S_Y)\to (X,S)$ is a
toroidal morphism corresponding to the
subdivision $\Delta$ of $\Sigma$, where
$\Delta=\{\langle(1,0),(1,1)\rangle, \langle(1,1)\rangle,
\langle(1,1),(0,1)\rangle, \{(0,0)\}$. The stratification
$S_Y$ is described by the semicomplex
$\Omega=\{\langle (1,1) \rangle,\{(0,0)\}\}\subset
\Delta$. After blow-up some faces "disappear" from
$\Omega\neq \Delta$. The remaining faces of
$\Omega$ are ${\operatorname{Aut}}(\sigma)^0$-invariant.
\end{itemize}
\end{example}
\begin{lemma} \label{le: fibers} Let $ Y\to X$ be a
toroidal modification of an oriented
$\Gamma$-stratified toroidal variety $(X,S)$. Then all
fibers $f^{-1}(x)$, where $x\in s$, are isomorphic.
\end{lemma}
\noindent{\bf Proof}. It follows from condition (1) of
Definition \ref{de: toroidal modification} that all fibers
$f^{-1}(x)$, where $x\in s={\operatorname{strat}}_X(\sigma)\cap U_\sigma$, are
isomorphic for any chart $U_\sigma \to X_\sigma$. \qed
\begin{lemma} \label{le: dominate} Let $ Y\to X$ be a
toroidal modification of an oriented
$\Gamma$-stratified toroidal variety $(X,S)$. Then the exceptional divisors
of the morphism $f: Y\to X$ dominate strata.
\end{lemma}
\noindent{\bf Proof.} Suppose an exceptional divisor $D$
does not dominate a stratum. The generic point of $f(D)$ is in a
stratum $s$ and $\dim(f(D)) < \dim(s)$. By Lemma \ref{le: fibers},
$s$ is not a generic open
stratum in $X$. The dimension of a generic fiber $f^{-1}(x)$,
where $x\in f(D)$, is greater than or equal to
$\dim(D)-\dim(f(D))=n-1-\dim(f(D)$. The dimension of a
generic fiber $f^{-1}(x)$, where $x\in s$, is at most
$\dim(f^{-1}(s))-\dim(s)< n-1-\dim(f(D)$. This contradicts
Lemma \ref{le: fibers}. \qed
\begin{definition}\label{de: vertices}
If $\Sigma$ is a fan then we denote by ${\operatorname{Vert}}(\Sigma)$ the set of all 1-dimensional
faces (rays) in $\Sigma$.
\end{definition}
\begin{lemma}
Let $\sigma=\tau_1\oplus\tau_2$ and $\Delta$
be a subdivision of $\tau_1$. Then \\
$\Delta\oplus \tau_2:=\{\delta\oplus
\sigma
\mid \delta\in \Delta,\,\, \sigma\preceq \tau_2 \}$
is a subdivision of $\tau$ called the \underline{joint}
of $\Delta$ and $\tau_2$. \qed
\end{lemma}
\begin{lemma} \label{le: sum} Let $\Omega\subset\Sigma$ be an
embedded $\Gamma$-semifan. Let $\Delta$ be a subdivision of
$\Sigma$ such that any ray $\varrho\in{\operatorname{Vert}}(\Delta\setminus\Sigma)$
passes through the interior of
$\omega\in\Omega$. Then for any cone
$\sigma=\omega(\sigma)\oplus {\rm r}(\sigma)$ in $\Sigma$,
$\Delta|{\sigma}=(\Delta|\omega(\sigma))\oplus {\rm r}(\sigma)$.
\end{lemma}
\noindent{\bf Proof.}
Since $\omega:=\omega(\sigma)$ is a maximal face of $\sigma$
which belongs to $\Omega$, all ''new'' rays which are
contained in $\sigma$ are in
$\omega$ and therefore the maximal cones in
$\Delta|\sigma$ are of the form
$\tau=\omega\oplus {\rm r}(\sigma)$ and $\tau\subset \omega$.
This shows that $\Delta_1:=\Delta|{\sigma}$ is a subfan of
$\Delta_2:=(\Delta|\omega)\oplus {\rm r}(\sigma)$. By properness we get
$|\Delta_1|=|\Delta_2|$ and hence $\Delta_1=\Delta_2$. \qed
\begin{lemma}. \label{le: iso} Let $\psi_1: Y_1\to X$ and
$\psi_2: Y_2\to X$ be proper
birational morphisms of normal noetherian schemes.
\begin{enumerate}
\item Suppose $\widehat{X}_x\times_{X}Y_1$ and
$\widehat{X}_x\times_{X}Y_2$ are isomorphic over
$\widehat{X}_x$ for a
$K$-rational point $x\in X$. Then there exists an open
neighborhood $U$ of $x$ such that the proper birational
map $\psi_1^{-1}(U)\to \psi_2^{-1}(U)$ is an isomorphism.
\item Suppose $\widehat{X}_x\times_{X}Y_1\to
\widehat{X}_x\times_{X}Y_2$ is a proper morphism over
$\widehat{X}_x$ for a
$K$-rational point $x\in X$. Then there exists an open
neighborhood $U$ of $x$ such that the proper birational
map $\psi_1^{-1}(U)\to \psi_2^{-1}(U)$ is a morphism.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof} (1) Suppose that the sets
$\psi_1^{-1}(U)$ and $\psi_2^{-1}(U)$
are not isomorphic for any $U$. Then the closed subset of $Y_2$ where
$\psi_1\psi_2^{-1}$ is not an isomorphism interesect
the fiber $\psi_2^{-1}(x)$.
Let $Y_3$ be a component in $Y_1\times_{X}Y_2$ which
dominates $X$. Since all varieties are normal,
at least one of the proper morphisms $Y_3\to Y_i$ is not an
isomorphism for $i=1,2$ over $\psi_i^{-1}(U)$ for any $U\ni x$. Then it
collapses a curve to a point over $x$.
Let $\widehat{X}_3$ be a component in
$\widehat{X}_1\times_{\widehat{X}_x}\widehat{X}_2$ which
dominates $\widehat{X}_x$. The morphisms $\widehat{X}_3\to
\widehat{X}_i$ for $i=1,2$ are pull-backs of the morphisms
$Y_3\to Y_i$ via \'etale morphisms. This shows that they are
not both isomorhisms, and neither is the induced birational map
$\widehat{X}_1\to\widehat{X}_2$, which contradicts the
assumption.
(2) The same reasoning. \qed
\begin{lemma}\label{le: fiber square}
Let $\phi_\sigma:U_{\sigma}\to
X_{\sigma}$ denote a chart on an oriented
$\Gamma$-stratified toroidal variety $(X,S)$.
Let $\sigma'\preceq\sigma$ denote a face for which
$\omega(\sigma')=\varrho\leq\sigma$.
Consider the smooth morphism
$\psi_\sigma:=\pi^\varrho_{\sigma'}\phi^{\sigma'}_{\sigma} :
U_{\sigma'} \to X_\varrho$ induced by the chart $\phi_\sigma$.
Let $\Delta^\sigma$ be a subdivision of $\sigma$ and
$\Delta^\varrho:=\Delta^\sigma|\varrho$ be a subdivision of $\varrho$.
Suppose
${f}_\sigma:
\widetilde{U}_{\sigma}:=U_{\sigma}\times_{X_{\sigma}} X_{\Delta^\sigma}\to
U_{\sigma}$ satisfies Condition (2) of Definition \ref{de: toroidal modification}.
Then the following diagram consists of
two fiber squares of smooth morphisms:
\[\begin{array}{rcccccccc}
&& & U_\sigma^{\sigma'} & \longrightarrow & X_{\sigma'}=X_{\varrho\oplus
{\rm r}(\sigma')}& \longrightarrow &X_{\varrho} &\\
&&&\uparrow f & & \uparrow &&\uparrow&\\
U_\sigma:=U\times_{X_{\sigma'}} X_{\Delta^\sigma}
&&\simeq& f^{-1}(U_\sigma^{\sigma'})
& \rightarrow &
X_{\Delta^\sigma|\sigma'}=X_{\imath^\sigma_\varrho(\Delta^{\varrho})\oplus
{\rm r}(\sigma')} &
\rightarrow& X_{\Delta^\varrho}& \\
\end{array}\]
\end{lemma}
\noindent{\bf Proof}
It follows from Lemma \ref{le: dominate} that
the assumption of Lemma \ref{le: sum} is satisfied for $\Delta^\sigma$.
By Lemma \ref{le: sum},
${\Delta^\sigma|\sigma'}=
{\Delta^{\sigma}|\varrho\oplus {\rm r}(\sigma')}=\imath^\sigma_\varrho(\Delta^{\varrho})\oplus {\rm r}(\sigma')$.
\qed
\begin{lemma}\label{le: isomorphisms}
Let $\phi_\sigma:U_{\sigma}\to
X_{\sigma}$ and $\phi_\tau:U_{\tau}\to
X_{\tau}$ denote two charts on an oriented
$\Gamma$-stratified toroidal variety $(X,S)$.
Let $\Delta^{\sigma}$ be a subdivision of $\sigma$ and $\Delta^{\tau}$ denote
a subdivision of $\tau$.
Assume that for any $\varrho\leq \sigma, \tau$,
$\Delta^\sigma|\varrho=\Delta^\tau|\varrho$.
Suppose
${f}_\sigma:
\widetilde{U}_{\sigma}:=U_{\sigma}\times_{X_{\sigma}} X_{\Delta^\sigma}\to
U_{\sigma}$ and ${f}_\tau:
\widetilde{U}_{\tau}:=U_{\tau}\times_{X_{\tau}}X_{\Delta^\tau}\to
U_{\tau}$ satisfy Condition (2) of Definition \ref{de: toroidal modification}.
Then ${\phi}_\sigma^{-1}(U)$
and ${\phi}_\tau^{-1}(U)$
are isomorphic over $U:=U_{\sigma}\cap U_{\tau}$.
\end{lemma}
\noindent{\bf Proof.}
Suppose that the relevant
sets are not isomorphic over $\operatorname{Spec}({{\mathcal O}}_x)$ for $x\in
{\operatorname{strat}}_X(\varrho)\cap U$. Then there are faces
$\sigma'\preceq\sigma$ and $\tau'\preceq\tau$ such that
${\omega(\sigma')}={\omega(\tau')=\varrho}$.
Set by ${\operatorname{reg}}(\varrho)=\langle e_1,\ldots,e_{\dim({\operatorname{strat}}_X(\varrho))}\rangle$.
The diagram from Lemma \ref{le: fiber square} gives the following fiber square diagram of
\'etale extensions:
\[\begin{array}{rcccccccccc}
& U & \longrightarrow &X_{\varrho\times reg(\varrho)}&\,&& & U & \longrightarrow &X_{\varrho\times reg(\varrho)}&\\
&\uparrow f_\sigma &&\uparrow&\,&&&\uparrow f_\tau &&\uparrow&\\
&f_\sigma^{-1}(U)&\longrightarrow&
X_{\Delta\times reg(\varrho)}&\,&& &f_\tau^{-1}(U)
& \rightarrow &X_{\Delta\times {\operatorname{reg}}(\varrho)}& \\
\end{array}\]
\noindent where the horizontal morphisms are \'etale.
Consider the following fiber square diagram:
\[\begin{array}{rcccccc}
&\widehat{X}\times_U U_\sigma&\buildrel \alpha' \over
\longrightarrow&
\widehat{X}\times_U U_\tau &\buildrel
\widehat{\psi}'_{\sigma} \over \longrightarrow&
{X}_{\Delta^\varrho\times {\operatorname{reg}}(\varrho)}
\times_{{X}_{\varrho\times {\operatorname{reg}}(\varrho)}}
\widehat{X}_{\varrho\times {\operatorname{reg}}(\varrho)}\\
&\downarrow&&\downarrow&&\downarrow &\\
&\widehat{X}_x&\buildrel\alpha \over
\longrightarrow&\widehat{X}_x&\buildrel
\widehat{\psi}_{\sigma} \over \longrightarrow&
\widehat{X}_{\varrho\times {\operatorname{reg}}(\varrho)},
\end{array}\]
\noindent where $\widehat{\psi}_{\sigma}$ and
$\widehat{\psi}_{\tau}$ are isomorphisms of the completions of the local schemes
induced by \'etale extensions of the smooth morphisms
${\psi}_{\sigma}$ and ${\psi}_{\tau}$,
$\alpha$ is an isomorphism such that
$\widehat{\psi}_{\tau}:=\widehat{\psi}_{\sigma}\circ
\alpha$, and $\widehat{\psi}'_{\sigma}$,
$\widehat{\psi}'_\tau$ and $\alpha'$ are liftings
of $\widehat{\psi}_{\sigma}$,
$\widehat{\psi}_\tau$ and $\alpha$.
It follows from
the diagram that $\alpha$ is an automorphism of
$\widehat{X}_x$ which can be lifted to an automorphism
$\alpha'$ of $\widehat{X}_1$. Note that $\alpha$
preserves strata and orientation. It follows that
$\widehat{X}\times_U U_\sigma$ and $\widehat{X}\times_U U_\tau$.
are isomorphic.
By Lemma \ref{le: iso}, $\widetilde{U}_\sigma$ and
$\widetilde{U}_\tau$ are isomorphic over a neighborhoood of
$x\in X$, which
contradicts the choice of $x\in X$. \qed
\bigskip
\subsection{Canonical subdivisions of oriented semicomplexes}\label{se: -semicomplexes}
As a corollary from the proof of Lemma \ref{le:
isomorphisms} we obtain the following lemma.
\begin{lemma}\label{le: tilde} Let $(X,S)$ be an oriented
$\Gamma$-stratified toroidal
variety of dimension $n$ with associated
oriented $\Gamma$-semicomplex $\Sigma$ and let $f:Y\to (X,S) $ be a toroidal modification.
Let $x$ be a
closed point in stratum ${\operatorname{strat}}_X(\sigma)\in S$, $\phi_\sigma: U\to
X_{\sigma}$ be a chart of a neighborhood $U$ of $x$
and $\Delta^\sigma$ be a subdivision of $\sigma$ for which
there is a fiber square
\[\begin{array}{rcccccc}
&& & U & \buildrel \phi_\sigma \over \longrightarrow &X_{\sigma} &\\
&&&\uparrow f &&\uparrow&\\
&&&f^{-1}(U)
& \buildrel \phi^f_\sigma \over \longrightarrow &X_{\Delta^\sigma}&, \\
\end{array}\]
\noindent where the horizontal morphisms are $\Gamma_\sigma$-smooth.
Set $${\operatorname{reg}}(\sigma):=\langle
e_1,\ldots,e_{n-\dim(N_\sigma)}\rangle=\langle
e_1,\ldots,e_{\dim({\operatorname{strat}}_X(\sigma))}\rangle.$$
Then
\begin{enumerate}
\item
there is a fiber square of \'etale extensions
\[\begin{array}{rcccccc}
&& & U & \buildrel \widetilde{\phi}_\sigma \over \longrightarrow &X_{\widetilde{\sigma}} &\\
&&&\uparrow f &&\uparrow&\\
&&&f^{-1}(U)
&\buildrel \widetilde{\phi}^f_\sigma \over \longrightarrow
&X_{\widetilde{\Delta}^\sigma},& \\
\end{array}\]
\noindent where the horizontal morphisms are $\Gamma_\sigma$-\'etale and where
$$\widetilde{\sigma} := \overline{\sigma}\times
{\operatorname{reg}}(\sigma),\quad\mbox{and} \quad \widetilde{\Delta}^\sigma:=\Delta^\sigma\times {\operatorname{reg}}(\sigma).
$$
\item $X_{\widetilde{\sigma}}$ is a
$\Gamma_\sigma$-stratified toric variety with the strata
described by embedded $\Gamma_\sigma$-semifan
$\sigma\subset \widetilde{\sigma}$. Moreover the strata on
$U$ are exactly inverse images of strata of $X_{\widetilde{\sigma}}$.
\item There is a fiber square of isomorphisms
\[\begin{array}{rcccccc}
&& & \widehat{X}_x & \buildrel \widehat{\phi}_\sigma \over \simeq &{\widetilde{X}_{\sigma}} &\\
&&&\uparrow \widehat{f}_x &&\uparrow&\\
&&&Y\times_X \widehat{X}_x
& \buildrel \widehat{\phi}^f_\sigma \over \simeq &{\widetilde{X}_{\Delta^\sigma}} & \\
\end{array}\]
where
$$\widetilde{X}_{\sigma}:=
\widehat{X}_{\widetilde{\sigma}},\quad \mbox{and} \quad
\widetilde{X}_{\Delta^\sigma}:=
X_{\widetilde{\Delta^\sigma}}\times_{{X}_{
\widetilde{\sigma}}}
\widetilde{X}_{\sigma}.$$
\item $\widetilde{X}_{\sigma} $ is a
$\Gamma_\sigma$-stratified toroidal scheme with the strata
described by embedded $\Gamma_\sigma$-semifan
$\sigma\subset \widetilde{\sigma}$. The isomorphism
$\widehat{\phi}_\sigma$ preserves strata.
\item
The morphism $\widehat{f}_x:Y\times_X \widehat{X}_x\to
\widehat{X}_x$ is
${\rm Aut}^\Gamma(\widehat{X}_x,S)^0$-equivariant
.
\item
The morphism $\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma} $ is
${\rm Aut}(\widetilde{X}_{\sigma})^0$-equivariant.
\qed
\end{enumerate}
\end{lemma}
\begin{remark}
In what follows we assume that $n$ in the definition of $\widetilde{\sigma}$,
$\widetilde{\Delta}$, $\widetilde{X}_{\sigma}$, and
$\widetilde{X}_{\Delta}$ denotes the dimension of relevant
stratified toroidal varieties. If the
semicomplexes are not associated to stratified toroidal
varieties then $n$ will denote some, a priori given,
natural number, satisfying the condition that
all simplices and semicomplexes have dimension
$\leq n$. Thus the definition of $\widetilde{\sigma}$,
$\widetilde{\Delta}$, $\widetilde{X}_{\sigma}$, and
$\widetilde{X}_{\Delta}$ will make sense for any
cones
$\sigma$ and any fans $\Sigma$ considered.
\end{remark}
\begin{lemma} \label{le: isomorphisms2}
Let $\Sigma$ be an oriented $\Gamma$-semicomplex and
$\sigma\in \Sigma$.
Let $\phi_1, \phi_2: (U, S)\to (X_{\sigma},S_{\sigma})$ be smooth
morphisms of $\Gamma$-stratified toroidal schemes which determine the same
orientation at a closed point $x\in {\operatorname{strat}}_X(\sigma)$. Let $\Delta^\sigma$
be a subdivision of $\sigma$ such that $\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ is ${\rm
Aut}(\widetilde{X}_{\sigma},S_{\sigma})^0$-equivariant.
Let
$\widetilde{U}_i=U\times_{X_{\sigma}}X_{\Delta^\sigma}$
denote the fiber products for $\phi_i$, where $i=1,2$. Then
$\widetilde{U}_i$ are isomorphic over some neighborhood
$V\subset U$ of $x\in X$.
\end{lemma}
\noindent {\bf Proof.} Identical with the proof of Lemma \ref{le:
isomorphisms}.\qed
We shall assign to the faces of an oriented
$\Gamma$-semicomplex $\Sigma$ the collection of
connected proalgebraic groups $$G^{\sigma}=
{\rm
Aut}(\widetilde{X}_{\sigma})^0.$$
\begin{definition} \label{de: canonical} A proper
subdivision $\Delta=\{\Delta^\sigma\mid\sigma\in\Sigma\}$ of an oriented $\Gamma$-semicomplex $\Sigma$ is
called {\it canonical} if
for any $\sigma \in \Sigma$,
$G^{\sigma}$ acts
on $\widetilde{X}_{\Delta^\sigma}$ (as an abstract group) and the morphism
$\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma} $ is $G^{\sigma}$-equivariant.
\end{definition}
\bigskip
\begin{remark} This technical definition is replaced by
a simple combinatorial definition by using the notion of
stable support (see section \ref{se: simple}). Also
$\widetilde{X}_{\sigma}$ and
$\widetilde{X}_{\Delta^\sigma}$ are replaced with
$\widehat{X}_{\sigma}$ and
$\widehat{X}_{\Delta^\sigma}$ (Proposition \ref{pr: ssimple}).
\end{remark}
\bigskip
\subsection{Correspondence between canonical subdivisions
of semicomplexes and toroidal modifications}
\begin{theorem}\label{th: modifications} Let $(X,S)$ be
an oriented $\Gamma$-stratified toroidal variety
with the associated oriented $\Gamma$-semicomplex
$\Sigma$. There exists a 1-1 correspondence
between the toroidal modifications $Y$ of $(X,S)$ and
the canonical subdivisions $\Delta$ of $\Sigma$.
\begin{enumerate}
\item
If $\Delta$ is a canonical subdivision of $\Sigma$ then
the
toroidal modification associated to it is defined locally by
\[\begin{array}{rcccccccc}
&& & U_\sigma & \longrightarrow & X_{\sigma}&&&\\
&&&\uparrow f & & \uparrow &&&\\
U_\sigma \times_{X_{\sigma}}
X_{\Delta^\sigma} &&\simeq& f^{-1}(U_\sigma)
& \rightarrow &
X_{\Delta^\sigma} &&& \\
\end{array}\]
\item
If $Y^1\to X$,
$Y^2\to X$ are toroidal modifications associated to
canonical subdivisions $\Delta_{1}$ and $\Delta_{2}$ of $\Sigma$ then
the natural birational map $Y^1\to Y^2$ is a morphism iff
$\Delta_{1}$ is a subdivision of $\Delta_{2}$.
\end{enumerate}
\end{theorem}
\noindent{\bf Proof}.
We shall construct the canonical
subdivision $\Delta$ of $\Sigma$ associated to a given toroidal
modification $Y$.
By
Definition \ref{de:
toroidal modification} for any point $x$ there is a chart
and a subdivision $\Delta^\sigma$ of
$\sigma$ and the fiber square stated in property (1).
For two charts $\phi_{\sigma,1}: U_{\sigma,1}\to
X_\sigma$ and $\phi_{\sigma,2}: U_{\sigma,2}\to
X_\sigma$ we
may get two subdivisions $\Delta^\sigma_1$ and
$\Delta^\sigma_2$ of $\sigma$.
These two charts induce by Lemma \ref{le: tilde}
isomorphisms $\widetilde{\phi}_{\sigma,i}: {\widehat{X}_x}\to
\widetilde{X}_\sigma$ and their liftings
$\widetilde{\phi}^f_i: Y\times_{X}{\widehat{X}_x}\simeq
\widetilde{X}_{\Delta^\sigma_1}\simeq
\widetilde{X}_{\Delta^\sigma_2}$. The isomorphisms
$\widetilde{\phi}_{\sigma,i}$ define an automorphism
$\alpha:=\widetilde{\phi}_{\sigma,2}\widetilde{\phi}^{-1}_{\sigma,1}$
of $\widetilde{X}_\sigma$
preserving orientation and strata, and its lifting isomorphism
$\alpha_Y:=\widetilde{\phi}^f_{\sigma,2}({\widetilde{\phi}_{\sigma,1}}^{f})^{-1}:
\widetilde{X}_{\Delta^\sigma_1}\to
\widetilde{X}_{\Delta^\sigma_2}$. On the other hand
$\alpha$ lifts to an automorphism $\alpha^f$ of
$\widetilde{X}_{\Delta^\sigma_1}\simeq
Y\times_{X}{\widehat{X}_x}$. Finally the isomorphism
$\alpha'_Y\alpha_Y(\alpha^f)^{-1} :\widetilde{X}_{\Delta^\sigma_1}\to
\widetilde{X}_{\Delta^\sigma_2}$ is a lifting of the identity
morphism on $\widetilde{X}_\sigma$. Thus it is an identity
morphism on $\widetilde{X}_{\Delta^\sigma_1}=
\widetilde{X}_{\Delta^\sigma_2}$ and
$\Delta^\sigma_1 = \Delta^\sigma_2$. This implies
the uniqueness of $\Delta^\sigma$.
By Lemma \ref{le: tilde}(6) , $i_x:\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma} $ is
${\rm Aut}(\widetilde{X}_{\sigma})^0$-equivariant.
Thus the fan $\Delta^\sigma$ is ${\operatorname{Aut}}(\sigma)^0$-invariant.
It follows from the uniqueness that if $\tau\leq \sigma$
then
$\Delta^\sigma{|\tau}=\imath^\sigma_\tau(\Delta^\tau)\simeq
\Delta^\tau$. Therefore the
subdivisions $\Delta^\sigma$ define the canonical
subdivision $\Delta$ of $\Sigma$.
Now let $\Delta$ be a canonical subdivision of the oriented
$\Gamma$-semicomplex $\Sigma$.
We shall construct
the toroidal modification of $(X,S)$ associated to $\Delta$.
For any chart $U_{\sigma,a}\to X_{\sigma}$ set
$$\widetilde{U}_{\sigma,a}:=U_{\sigma,a}
\times_{X_{\sigma}}X_{\Delta^\sigma}$$
For any two charts $\phi_{\sigma,a}: U_{\sigma,a}\to X_{\sigma}$ and
$\phi_{\tau,b}:U_{\tau,b}\to X_{\tau}$ set
$$\widetilde{U}_{\sigma,\tau,a,b} :=
(U_{\sigma,a}\cap U_{\tau,b})\times_{X_{\sigma}}X_{\Delta^\sigma},$$
\noindent where the fiber product is defined for the
restriction of $\phi_{\sigma,a}: U_{s,a}\to X_{\sigma}$ to
$U_{\sigma,a}\cap U_{\tau,b}$.
It follows from Lemma \ref{le: isomorphisms} that
$\widetilde{U}_{\sigma,\tau,a,b}$ is
isomorphic over $U_{\sigma,a}\cap U_{\tau,b}$ to
$\widetilde{U}_{\tau,\sigma, b,a}$.
Thus we can glue all the sets $\widetilde{U}_{\sigma,a}$
along $\widetilde{U}_{\sigma,\tau,a,b}$ and get a separated
variety $Y$. Note that the action of $\Gamma$ lifts to any
subset $\widetilde{U}_{\sigma,a}$ and hence to an action
on the whole $Y$.
A chart $\phi_\sigma:U_\sigma\to
X_{\sigma}$ defines a $\Gamma_\sigma$-equivariant isomorphism
$\widehat{X}_x\buildrel \alpha_1\over
\to\widetilde{X}_{\sigma}$.
For any $x\in{\operatorname{strat}}_X(\sigma)$ any automorphism
$\alpha$ of $\widehat{X}_x\simeq \widetilde{X}_{\sigma}$
preserving strata and orientation can be lifted
to a $\Gamma_\sigma$-equivariant automorphism $\alpha'$
of $\widehat{X}_y\times_X Y\simeq \widetilde{X}_{\Delta^\sigma}$.
(1) Follows from the construction of a toroidal modification
from a canonical subdivision.
(2) Let $Y^1\to X$,
$Y^2\to X$ be toroidal modifications associated to
canonical subdivisions $\Delta_1$ and $\Delta_2$.
If $\Delta_1$ is a subdivision of $\Delta_2$ then
the natural birational map $Y^1\to Y^2$ is a morphism since
it is a morphism for each chart. On the other hand if
$Y^1\to Y^2$ is a morphism then $Y^1\times_{X}{\widehat{X}_x}\simeq
\widetilde{X}_{\Delta^\sigma_{1}}\to Y^2\times_{X}{\widehat{X}_x}\simeq
\widetilde{X}_{\Delta^\sigma_{2}}$ is a toric morphism. By
Lemma \ref{le: iso} the latter morphism is induced by a
toric morphism ${X}_{\Delta^\sigma_{1}}\to
X_{\Delta^\sigma_{2}}$. Hence ${\Delta^\sigma_{1}}$ is a
subdivision of ${\Delta^\sigma_{2}}$ for each $\sigma \in \Sigma$
and consequently $\Delta_1$ is a subdivision of
$\Delta_2$. This completes the proof of Theorem \ref{th: modifications}. \qed
\section{Stable valuations}
For simplicity we consider
only valuations with integral values.
\subsection{Monomial valuations}
\begin{definition}
Let $R$ be a local noetherian
ring, $u_1,\ldots,u_k$ be generators of its maximal ideal,
and let $a_1,\ldots,a_k$ be nonnegative integers. For any $a\in
{\bf Z}$ set
$$J_a:=(u^{i_1,\ldots,i_k}\mid i_1a_1+\ldots+i_ka_k\geq a)\subset R,$$
\noindent where ${i_1,\ldots,i_k}$, is a
multiindex in ${\bf Z}_{\geq 0}^k$.
We call a valuation $\nu$ {\it monomial} with basis $u_1,\ldots,u_k$
and weights $a_1,\ldots,a_k$ if for any $f\in R\setminus\{0\}$,
$$\nu(f)={\rm max}\{a\in {\bf Z}\mid f\in J_a\}.$$
\end{definition}
\begin{lemma} \label{le: monomial0} Let $\nu$ be a
nonnegative monomial valuation of a local noetherian
ring $R$. Then for any $a\in {\bf Z}$, $I_{\nu,a}=J_a$. \qed
\end{lemma}
\begin{lemma} \label{le: monomial} Let $\nu$ be a
nonnegative valuation of a local noetherian
ring $R$,
and $u_1,\ldots,u_k$ be generators of the maximal
ideal of $R$. Let $\nu_0$ be a monomial valuation with basis
$u_1,\ldots,u_k$ such that $\nu_0(u_i)=\nu(u_i)\geq 0$. Then
$\nu_0(f)\leq \nu(f)$ for any
$f\in R$.
\end{lemma}
\noindent{\bf Proof.} $I_{\nu_0,a}:=\{f\in R\mid
\nu_0(f)\geq a\}=J_{a}\subset I_{\nu,a}.$
\qed
\begin{lemma} \label{le: monomial2} Let $\nu$ be a
nonnegative monomial valuation of a local noetherian
ring $R$ with basis $u_1,\ldots,u_k$.
Let $g$ be an automorphism of $R$ such that
$\nu(g(u_i))\geq \nu(u_i)$ for any $i=1,\ldots,k$.
Then $\nu(g(f))=\nu(u_i)$ for any $f\in R$.
\end{lemma}
\noindent{\bf Proof.} Note that
$\nu(g(u^{i_1,\ldots,i_k}))\geq
\nu(g(u^{i_1,\ldots,i_k}))$. By Lemma \ref{le: monomial0},
$g(I_{\nu,a})= g(J_a)\subset I_{\nu,a}$.
If $g(I_{\nu,a})\subset\neq I_{\nu,a}$ then we obtain an
infinite ascending chain of ideals $I_{\nu,a}\subset
g^{-1}(I_{\nu,a})\subset g^{-2}(I_{\nu,a})\subset\ldots$. \qed
\subsection{Toric and locally toric valuations}
Let $X$ be an
algebraic variety and
$\nu$ be a valuation of the field $K(X)$ of rational functions. By the valuative criterion of
separatedness and properness the valuation ring of $\nu$
dominates the local ring of a uniquely determined point (in general nonclosed)
$c_{\nu}$ on a complete variety $X$. (If $X$ is not
complete such a point may not exist). We call the
closure of $c_\nu$ the {\it center of the valuation } $\nu$ and
denote it by ${\operatorname{Z}}(\nu)$ or ${\operatorname{Z}}(\nu,X)$. For any $x\in {\operatorname{Z}}(\nu)$
and $a\in {\bf Z}_{\geq 0}$,
$$I_{\nu,a,x}:=\{f\in {{\mathcal O}}_{X,x}\mid
\nu(f)\geq a\}$$ \noindent is an ideal in ${{\mathcal O}}_{X,x}$.
For fixed $a$ these ideals define a coherent
sheaf of ideals ${{\mathcal I}}_{\nu,a}$ supported at ${\operatorname{Z}}(\nu)$.
\begin{lemma} \label{le: center}
Let $v$ be an integral vector in the support of the fan
$\Sigma$. Then the toric valuation ${\operatorname{val}}(v)$ is centered on
$\overline{O}_\sigma$, where $\sigma$ is the cone whose
relative interior contains $v$.
\end{lemma}
\noindent{\bf Proof.} Let $I_{O_\sigma}\subset K[X_\sigma]$
be the ideal of the orbit $O_\sigma$. Let $f=\sum a_ix^{m_i}$,
$m_i\in\sigma^\vee$, be a regular function on $X_\sigma$.
Then ${\operatorname{val}}(v)(f)>0$ iff $(v,m_i)>0$ for all $a_i\neq 0$.
But $(v,m_i)\geq 0$ by definition and $(v,m_i)=0$ iff
$m_i\in\sigma^\perp$. Thus we have $(v,m_i)> 0$ iff
$m_i\in\sigma^\vee\setminus \sigma^\perp$. Finally
$I_{O_\sigma}=\{f\in K[X_\sigma]\mid {\operatorname{val}}(v)(f)>0\}$ and the
valuation ring dominates the local ring of $O_\sigma$.\qed
Let $L$ be a ring containing $K$, $\sigma$ be a cone of the
maximal dimension in $N_\sigma$ and $v$ be an integral
vector of $\sigma$.
Toric valuation ${\operatorname{val}}(v)$ can be defined on any ring
$L[[\sigma^\vee]]$ and its arbitrary subring $R$ containing
$L[\sigma^\vee]$. For any $f=\sum a_ix^{m_i}$, set
$${\operatorname{val}}(v)(f)=\min\{(m_i,v)\mid a_i\neq 0\}.$$
The valuation ${\operatorname{val}}(v)$ of $R$ will be sometimes denoted by
${\operatorname{val}}(v,R)$ or ${\operatorname{val}}(v,\operatorname{Spec}(R))$.
\begin{lemma} \label{le: invar} Let $\sigma$ be a cone of a
maximal dimension and $T_\sigma$ denote the
''big torus'' acting on $\widehat{X}_\sigma$. Let $\mu$ be a
$T_\sigma$-invariant valuation on $ \widehat{X}_\sigma$. Then
there exists an integral vector $v\in\sigma$ such that
${\operatorname{val}}(v)=\mu$.
\end{lemma}
\noindent{\bf Proof.} The valuation $\mu$ defines a linear
function on $\sigma^\vee$ corresponding to an integral vector $v\in\sigma$.
Since $\mu$ is $T_\sigma$ invariant so are the ideals
$I_{\mu,a}$. Therefore the ideals $I_{\mu,a}$ are generated
by characters $x^m$, where $(m,v)\geq a$. Consequently,
$I_{\mu,a}= I_{{\operatorname{val}}(v),a}$ and $\mu={\operatorname{val}}(v)$.\qed
\begin{lemma} \label{le: ind} Let $\sigma$ be a cone of the
maximal dimension in a lattice $N_\sigma$ and $v\in\sigma$ be its
integral vector. Let $R$ be a
subring of $K[\widehat{X}_\sigma]$ that contains
$K[{X}_\sigma]$. Then for any $a\in {\bf Z}$ we have
$I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot R=I_{{\operatorname{val}}(v,R),a}$
\end{lemma}
\noindent{\bf Proof.} Denote by $R'$ the localization of $R$
at the maximal ideal of $O_\sigma$.
Then
$I_{{\operatorname{val}}(v,R'),a}=I_{{\operatorname{val}}(v,\widehat{X}_\sigma),a}\cap R'=
I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot K[\widehat{X}_\sigma]\cap R'$.
Since $K[\widehat{X}_\sigma]$ is faithfully flat over $R'$
we obtain $I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot
K[\widehat{X}_\sigma]\cap R'= I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot
R'$. Thus we have $I_{{\operatorname{val}}(v,R'),a}=I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot
R'$. Then $I_{{\operatorname{val}}(v,R),a}=I_{{\operatorname{val}}(v,R'),a}\cap R=
I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot
R' \cap R=I_{{\operatorname{val}}(v,{X}_\sigma),a}\cdot
R$.
\qed
\begin{lemma} \label{le: ind2} Let $X_\Sigma$ be a toric
variety with a toric action of $\Gamma$. Let $f: U\to X_\Sigma$ be a
smooth $\Gamma$-equivariant morphism. Let $v\in {\operatorname{int}}(\sigma)$, where $\sigma\in
\Sigma$, be an integral vector. Assume that the inverse
image of $\overline{O}_\sigma$ is irreducible. Then there
exists a $\Gamma$-invariant valuation $\mu$ on $U$, such that $
{{\mathcal I}}_{\mu,a}=f^{-1}({{\mathcal I}}_{{\operatorname{val}}(v),a})\cdot{{\mathcal O}}_{U}$.
\end{lemma}
\noindent{\bf Proof.} Let $x$ belong to the support ${\operatorname{supp}}
(f^{-1}({{\mathcal I}}_{{\operatorname{val}}(v),a})\cdot{{\mathcal O}}_{U})=f^{-1}(\overline{O}_\sigma)$.
Then $f(x)\in O_\tau$, where $\tau\geq\sigma$. Let $\pi:
X_\tau\to X_{\underline{\tau}}$ denote the natural
projection. Then ${{\mathcal I}}_{{\operatorname{val}}(v,X_\tau),a}=
\pi^{-1}({{\mathcal I}}_{{\operatorname{val}}(v,X_{\underline{\tau}},a})\cdot{{\mathcal O}}_{X_\tau}$.
Thus it suffices to prove the lemma replacing $X_\Sigma$
with $X_{\underline{\tau}}$ and $U$ with the inverse image
$U'$ of $X_\tau$. Let
$U_x\subset U'$ be an open neighborhood of $x$
for which there exists
an \'etale extension $\widetilde{f}: U_x\to
X_{\underline{\tau}\times {\operatorname{reg}}(\underline{\tau})}$.
It defines an isomorphism $\widehat{f}: \widehat{X}_x\simeq
\widehat X_{\underline{\tau}\times
{\operatorname{reg}}(\underline{\tau})}$. The toric valuation
${\operatorname{val}}(v,X_{\underline{\tau}})$ defines the toric valuation ${\operatorname{val}}(v,X_{\underline{\tau}\times
{\operatorname{reg}}(\underline{\tau})})$ and the corresponding valuation
$\overline{\mu}_x$ on $\widehat{X}_x$. By Lemma \ref{le: ind} , $\overline{\mu}_x$ determines the valuation
${\mu}_x$ of ${{\mathcal O}}_{X,x}$ such that
$({{\mathcal I}}_{\mu_x,a})_x=f^{-1}({{\mathcal I}}_{{\operatorname{val}}(v),a})\cdot{{\mathcal O}}_{X,x}$.
Thus $\mu_x$ is a valuation on $U_x$ supported on the
irreducible set $\overline{f^{-1}(O_\sigma)}$. Note that
both sheaves of ideals
${{\mathcal I}}_{\mu_x,a}$ and
$f^{-1}({{\mathcal I}}_{{\operatorname{val}}(v),a})\cdot{{\mathcal O}}_{X}$ are equal in some
open neighborhood $V_x$ of $x$. Therefore the valuation $\mu_y$
is the same for all closed point $y\in
\overline{f^{-1}(O_\sigma)}\cap V_x$. Since
$\overline{f^{-1}(O_\sigma)}$ is irredducibe the valuations
$\mu_x$ are the
same for all $x\in \overline{f^{-1}(O_\sigma)}$ and define
a unique valuation $\mu$. Since the sheaves of ideals
$f^{-1}({{\mathcal I}}_{{\operatorname{val}}(v),a})\cdot{{\mathcal O}}_{X}={{\mathcal I}}_{\mu,a}$ are
$\Gamma$-invariant $\mu$ is also $\Gamma$-invariant.\qed
\bigskip
Let $f:X\rightarrow Y$ be a dominant morphism and $\nu$ be a
valuation of $K(X)$. Then $f_*(\nu)$ denotes the valuation
which is the restriction
of $\nu$ to $K(Y)\simeq f^*(K(Y))\subset K(X)$.
Let $\nu$
be a valuation on $X$ and ${{\mathcal I}}_{\nu,a}$ be the associated sheaves of ideals.
Let $f:Y\to X$ be a
morphism for which $f^{-1}({{\mathcal I}}_{\nu,a})\cdot{{\mathcal O}}_{Y}$
determine a unique
valuation $\mu$ on $Y$ such that for sufficiently divisible
integer $a$,
${{\mathcal I}}_{\mu,a}=f^{-1}({{\mathcal I}}_{\nu,a})\cdot{{\mathcal O}}_{Y}$. We
denote this valuation by $\mu=f^{*}(\nu)$.
\begin{definition} By a {\it locally toric valuation} on an $\Gamma$-toroidal
variety $X$ we mean a valuation $\nu$ of $K(X)$ such that for any
point $x\in {\operatorname{Z}}(\nu,X) $ there exists a
$\Gamma$-equivarinat $\Gamma_x$-smooth morphism $\phi: U\to
X_\sigma$ from an open $\Gamma$-invariant neighborhood $U$
of $x$ to a toric variety $X_\sigma$ with a toric action of
$\Gamma$ such that $\mu=\phi^*({\operatorname{val}}(v))$, where $v$ is an
integral vector of $\sigma$.
If $\mu$ is a locally toric valuation on $X$ then by
$\mu_{|\widehat{X}_x}$ we denote the induced valuation on
$\widehat{X}_x\simeq\widehat{X}_\sigma$.
\end{definition}
\begin{definition} \label{de: blow}
By the
{\it blow-up} ${\operatorname{bl}}_{\nu}(X)$ of $X$ at a locally toric
valuation $\nu$ we mean the normalization of
$$\operatorname{Proj}({{\mathcal O}\oplus {\mathcal I}}_{\nu,1}\oplus {{\mathcal I}}_{\nu,2}\oplus\ldots),$$
\end{definition}
\begin{lemma} For any natural $l$, $\operatorname{Proj}({{\mathcal O}\oplus {\mathcal I}}_{\nu,1}\oplus
{{\mathcal I}}_{\nu,2}\oplus\ldots)=\operatorname{Proj}({{\mathcal O}\oplus
{\mathcal I}}_{\nu,l}\oplus {{\mathcal I}}_{\nu,2l}\oplus\ldots)$
.\qed
\end{lemma}
Denote by ${\operatorname{bl}}_{{\mathcal J}}(X)\to X$ the
blow-up of any coherent sheaf of
ideals ${{\mathcal J}}$.
\noindent \begin{lemma}\label{le: blow-up valuation} Let $X_\Sigma$ be a toric variety
associated to a fan $\Sigma$ in $N$
and $v\in |\Sigma|\cap N$. Then ${\operatorname{bl}}_{{\operatorname{val}}(v)}(X_\Sigma)$ is the toric
variety associated to the subdivision $\langle
v\rangle\cdot\Sigma$. Moreover for any sufficiently
divisible integer $d$, ${\operatorname{bl}}_{{\operatorname{val}}(v)}(X_\Sigma)$ is the normalization of
the blow-up of ${{\mathcal I}}_{\nu,d}$.
\end{lemma}
\noindent {\bf Proof.} It follows from Definition \ref{de: blow} that
${\operatorname{bl}}_{{\operatorname{val}}(v)}(X_\Sigma)$
is a normal toric variety. By Lemma \ref{le: center}, the sheaf of ideals
${{\mathcal I}}_{{\operatorname{val}}(v),d}$ is supported on $\overline{O}_\sigma$,
where $v\in {\operatorname{int}}(\sigma)$, $\sigma\in\Sigma$.
By \cite{KKMS}, Ch.I, Th. 9, it determines a convex piecewise
linear function $f:={\operatorname{ord}}_{{{\mathcal I}}_{{\operatorname{val}}(v),d}}:|\Sigma|\to {\bf R}$ such that
that $f=0$ on any $\tau\in\Sigma$ that does not contain $v$, and
$f(u)={\rm min}\{ (m,u)\mid m \in \tau^{\vee}\cap M, (m,v) \geq
d\}$ for any $\tau$ containing $v$. Assume that $\tau$
contains $v$ and
let $\tau'$ be a face
of $\tau$ that does not contain $v$. Let $m_{\tau',\tau} \in \tau^\vee$ be a
nonnegative integral functional on $\tau$ which equals $0$
exactly on $\tau'$. Assume that $d$ divides all $(m_{\tau',\tau},v)$.
Then $f_{|\tau'+ \langle v \rangle} =
\frac{d}{(m_{\tau',\tau},v)}\cdot m_{\tau',\tau}$.
Consequently, ${{\mathcal I}}^k_{{\operatorname{val}}(v),d}={{\mathcal I}}_{{\operatorname{val}}(v),kd}$ and
${\operatorname{bl}}_{{\operatorname{val}}(v)}(X_\Sigma)={\operatorname{bl}}_{{{\mathcal I}}_{{\operatorname{val}}(v),d}}(X_\Sigma)$.
By \cite{KKMS} Ch II Th. 10, the blow-up of ${{\mathcal I}}_{{\operatorname{val}}(v),d}$ corresponds to
the minimal subdivision $\Sigma'$ of $\Sigma$ such that $f$ is
linear on each cone in $\Sigma'$. By the above,
$$\Sigma'=\Sigma \setminus {\rm Star}(\sigma,\Sigma)
\cup \bigcup_{\tau'\in \overline{{\rm
Star}(\sigma,\Sigma)}\setminus {{\rm
Star}(\sigma,\Sigma)}}
\tau'+ \langle v\rangle =
\langle v \rangle\cdot\Sigma.$$ \qed
\begin{proposition}\label{pr: blow} For any locally toric
valuation $\nu$ on $X$ there exists an
integer $d$ such that
$\bullet$ ${\operatorname{bl}}_\nu(X)={\operatorname{bl}}_{{{\mathcal I}}_{\nu,d}}(X)$.
$\bullet$ The valuation $\nu$ is induced by an irreducible $Q$-Cartier
divisor on ${\operatorname{bl}}_\nu(X)$.
\end{proposition}
\noindent{\bf Proof.} By quasicompactness of $X$ one can
find a finite open affine covering $U_i$ of $X$ of charts
$\phi_i: U_i\to X_{\sigma_i}$ such that the valuation $\nu$
on each $U_i\subset X$ is equal to $\phi^*_i({\operatorname{val}}(v_i)$. It
follows from Lemma \ref{le: blow-up valuation} that for any
$i$ there
exists $d_i$ that
${\operatorname{bl}}_\nu(U_i)={\operatorname{bl}}_{{{\mathcal I}}_{\nu,d_i}}(U_i)$. It suffices to
take $d$ equal the product of all $d_i$.
\qed
\bigskip
\subsection{Stable vectors and stable valuations}
\begin{definition} \label{de: stable valuations} Let
$(X,S)$ be an oriented
$\Gamma$-stratified toroidal variety.
A locally toric $\Gamma$-invariant valuation $\nu$
is {\it stable} on $X$
if
\begin{enumerate}
\item ${\operatorname{Z}}(\nu,X)=\overline{s}$ for some stratum $s \in S$.
\item For any $x\in \overline{s}$, $\nu_{|\widehat{X}_x}$ is invariant with
respect to any $\Gamma_s$-equivariant automorphism of
$(\widehat{X}_x,S)$,
preserving orientation.
\end{enumerate}
Let $f:Y\to (X,S)$ be a toroidal modification of an
oriented $\Gamma$-stratified toroidal variety.
Then a valuation $\nu$ on $Y$ is called $X$-{\it stable}
if $f_*(\nu)$ is stable on $(X,S)$.
\end{definition}
\begin{definition} \label{de: stable vectors} Let $\Sigma$ be an
oriented $\Gamma$-semicomplex.
A vector $v_0\in {\rm
int}({\sigma}_0)$, where $\sigma_0\in \Sigma$,
is $\Sigma$-{\it stable} on $\Sigma$ if for any $\tau\geq
\sigma_0$ the corresponding
valuation ${\rm val}(\imath^\tau_{\sigma_0}(v))$ on $\widetilde{X}_{\tau}$
is $G^\tau$-invariant. A vector $v\in \sigma$, where $\sigma\in\Sigma$ is $\Sigma$-{\it
stable} if there is a stable vector $v_0\in {\rm
int}({\sigma_{0}})$, where $\sigma_0\leq \sigma$, for which
$v=\imath^\sigma_{\sigma_0}(v_0)$.
\end{definition}
Let $\Sigma$ be an oriented $\Gamma$-semicomplex. For any $\sigma \in \Sigma$ set
$${\rm stab}(\sigma):=
\{a\cdot v\mid v\in\sigma, \,\,\,\,
v \, \mbox{
\,is a
stable vector}\,\, , a\in {\bf Q}_{\geq 0}\}.$$
\begin{lemma}\label{le: glueing}
\begin{enumerate}
\item Stable vectors $v\in\sigma$ are ${\operatorname{Aut}}(\sigma)^0$-invariant.
\item
For any $\varrho\geq \tau \geq \sigma$ and any vector $v\in {\rm
stab}(\sigma)$,
$\imath^\varrho_{\tau}\imath^\tau_{\sigma}(v)=\imath_\sigma^\varrho(v)$.
\item
$\imath_\sigma^{\tau}({\rm stab}(\sigma))
=\imath_\sigma^{\tau}(\sigma)\cap {\rm stab}(\tau)$ for any
$\tau\geq\sigma$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}(1) Follows from definition.
(2) By definition there is $v_0\in
{\operatorname{int}}(\sigma_{0})$ for which $v=\imath^\sigma_{\sigma_0}(v_0)$. First we prove
that for any $\tau\geq \sigma$, we have
$\imath^\tau_{\sigma}\imath^\sigma_{\sigma_0}(v_0)=\imath^\tau_{\sigma_0}(v_0)$.
By Definition \ref{de: oriented semicomplex}
there is an automorphism $\alpha_{0}\in{\operatorname{Aut}}(\sigma_{0})^0$
which preserves the orientation
such that $\imath^\tau_{\sigma}\imath^\sigma_{\sigma_0}
=\imath^\tau_{\sigma_0}\alpha_{0}$.
By Definition \ref{de: stable vectors} the
valuation ${\operatorname{val}}(v_0)$ is $G^{\sigma_0}$-invariant on
$\widetilde{X}_{\sigma_0}$ and hence $v_0$ is
preserved by $\alpha_{0}$. This gives
$\imath^\tau_{\sigma}\imath^\sigma_{\sigma_0}(v_0)=\imath^\tau_{\sigma_0}\alpha_{0}(v_0)=
\imath^\tau_{\sigma_0}(v_0)$
.
Now let $\varrho\geq\tau\geq\sigma$.
By the above we have
$\imath^\varrho_{\sigma}\imath^\sigma_{\sigma_0}(v_0)=\imath^\varrho_{\sigma_0}(v_0)$
Then $\imath^\varrho_{\tau}\imath^\tau_{\sigma}(v)=\imath^\varrho_{\tau}\imath^
\tau_{\sigma_0}(v_0)=\imath^\varrho_{\sigma_0}(v_0)$.
Finally
$\imath^\varrho_{\sigma_0}(v_0)=\imath^\varrho_{\sigma}\imath^{\sigma}_{\sigma_0}(v_0)
=\imath^\varrho_{\sigma}(v)$.
(3)
If $v$ is a stable vector on
$\sigma$ then by Definition \ref{de: stable vectors}
there is $\sigma_0\leq \sigma$ and a stable vector $v_0\in {\operatorname{int}} (\sigma_{0})$ such
that $v=\imath^{\sigma}_{\sigma_0}(v_0)$. Then for $\tau\geq\sigma$,
$\imath^\tau_{\sigma_0}(v_0)=\imath^\tau_{\sigma}\imath^\sigma_{\sigma_0}(v_0)=
\imath^\tau_{\sigma}(v)$ is stable on
$\tau$.
If $w\in \imath^\tau_{\sigma}(\sigma)\cap {\rm stab}(\tau)$ then
there is $\sigma_0\leq \sigma$ and a stable vector $v_0\in {\operatorname{int}} (\sigma_{0})$ such
that
$w=\imath^\tau_{\sigma_0}(v_0)=\imath^\tau_{\sigma}\imath^\sigma_{\sigma_0}(v_0)=
\imath^\tau_{\sigma}(v)$,
where $v=\imath^\sigma_{\sigma_0}(v_0)$.
\qed
\begin{example}
Let $\Sigma$ be a regular semicomplex and
$ \sigma=\langle e _1,\ldots, e_k\rangle \in \Sigma$. Then
$e:=e_1+\ldots+e_k$ is stable since it corresponds to the
valuation of the stratum ${\operatorname{strat}}_X(\sigma)$ on $\widetilde{X}_{\sigma}$. The subset $s$ defines a smooth
subvariety in a smooth neighborhood and therefore it
determines a valuation. If we blow-up the subvariety ${\operatorname{strat}}_X(\sigma)$ then this valuation
coincides with the one defined by the exceptional divisor.
\end{example}
\begin{example} \label{ex: affine2} Consider the stratified
toroidal variety ${\bf A}^2$ from Example \ref{ex: affine}. The valuation of the point $(0,0)$ corresponds to the
vector $(1,1)\in N$, the valuation of $\overline{s} =
{\bf A}^1\times \{0\}$ corresponds to the vector $(0,1)\in N$.
Then $|{\rm Stab(\Sigma)}|=\langle(1,1),(0,1)\rangle$. The valuations from
$\langle(1,0),(0,1)\rangle \setminus \langle(1,1),(0,1)\rangle$ are not stable: Let
$\widetilde{{\bf A}^2}\to {\bf A}^2$ be
the blow-up of the point $(0,0)\in {\bf A}^2$. The valuations from
$\langle(1,0),(0,1)\rangle \setminus \langle(1,1),(0,1)\rangle$
are centered on $\widetilde{{\bf A}^2}$ at $l_2\cap D$ or $l_2$ (using notation from
Example \ref{ex: affine}) and the automorphism
$\phi:{\bf A}^2\to {\bf A}^2$,
$\phi(x_1,x_2)=(x_1+x_2,x_2)$, preserves strata
and moves the subvarieties $l_2$ and $l_2\cap D$. In particular it changes the relevant valuations.
\end{example}
\begin{lemma} \label{le: stable valuations}
Let $(X,S)$ be a $\Gamma$-stratified toroidal
variety with the associated oriented $\Gamma$-semicomplex $\Sigma$.
The following conditions are equivalent:
\begin{enumerate}
\item $\nu$ is stable on $X$.
\item There exists a stable vector $v\in {\rm
int}({\sigma})$, where $\sigma\in \Sigma$,
such that for any $\tau\geq \sigma$ and any chart $\phi:U\to X_{\tau}$
we have
$\nu = {\phi}^*({\rm {val}}(\imath^\tau_{\sigma}(v)))$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} Let $x$ denote a closed point of
$\overline{{\operatorname{strat}}_X(\sigma)}$. Then $x\in
{\operatorname{strat}}_X(\tau)$, where $\tau\geq \sigma$.
The valuation $\nu$ on $X$ determines a $G^\tau$-invariant valuation on
$\widehat{X}_x\simeq\widetilde{X}_{\tau}$ which is
in particular a toric valuation so, by Lemmas \ref{le:
invar} and \ref{le: center}
it corresponds to a unique $v\in {\operatorname{int}}(\sigma)$. Thus
$v$ is stable and ,by Lemma \ref{le: ind2}, $\nu = {\phi}^*({\rm {val}}(v))$.
Let $v$ be a stable vector. The valuation ${\rm val}(v)$
determines a monomial valuation
$\nu:=\widehat{\phi}^*({\rm val}(v))$ on
$\widehat{X}_x$, where $x\in \overline{{\operatorname{strat}}_X(\sigma)}\cap
U$. Since ${\operatorname{val}}(v)$ is
$G^\tau$-invariant on
$\widetilde{X}_{\tau}$ for any $\tau\geq \sigma$
we see that $\nu_{|\widehat{X}_x}$ is invariant with
respect to any $\Gamma_x$-equivariant automorphism
preserving strata and orientation. Moreover
$\nu_{|\widehat{X}_x}$ and $\nu$ do not depend on $\phi$.\qed
\begin{lemma}\label{le: locally monomial}
Let $f: Y\to X$ be a toroidal modification of a $\Gamma$-stratified toroidal
variety $(X,S)$ associated to a canonical subdivision
$\Delta$ of $\Sigma$.
The following conditions are equivalent:
\begin{enumerate}
\item $\nu$ is $X$-stable on $Y$.
\item There exists a $\Sigma$-stable vector $v\in {\rm
int}({\sigma})$, where $\sigma\in\Sigma$,
such that for any $\tau\geq \sigma$ and morphism $\phi^f_\tau:
f^{-1}(U_\tau) \to X_{\Delta^\tau}$, induced
by a chart $\phi_\tau:U_\tau\to X_{\tau}$
we have
$\nu = \phi^{f*}_\tau({\rm {val}}(v))$.
\end{enumerate}
\end{lemma}
\noindent
{\bf Proof.} Let $ y\in {\operatorname{Z}}(\nu,Y)$, $x=f(y)\in X$ and
$\phi:U\to X_{\sigma}$ be a chart at $x$. We can extend
this $\Gamma^\sigma$-smooth morphism to an \'etale morphism
$\widetilde{\phi}:U\to X_{\widetilde{\sigma}}$.
By Lemma \ref{le: stable valuations} we have
$\nu = \widetilde{\phi}^*({\rm {val}}(v))$ for some stable
vector $v\in\sigma$.
Consider the fiber squares
\[\begin{array}{rccccccccc}
&f^{-1}(U) & \buildrel \widetilde{\phi}_f \over \longrightarrow &
X_{\widetilde{\Delta}}&&&&Y\times\widehat{X}_x
& \buildrel \widehat{\phi}_f \over \longrightarrow &
\widetilde{X}_{{\Delta}}\\
&\downarrow f& &\downarrow i
&&&&\downarrow \widehat{f}& &\downarrow \widehat{i}\\
&{U}
&\buildrel \widetilde{\phi}\over \longrightarrow &
{X_{\widetilde{\sigma}}}&&&&\widehat{X}_x
&\buildrel \widehat{\phi}\over \longrightarrow &
\widetilde{X}_{\sigma},
\end{array}\]
\noindent where $\widetilde{\phi}^f_\sigma$ is the induced \'etale morphism and $i$ is a
toric morphism.
Then $\widehat{\phi}$ as well as $\widehat{\phi^f_\sigma}$ are
isomorphisms
and we have
$\nu=
\widehat{f}_*^{-1}\widehat{\phi}^*i_*({\rm val}(v))=
\widehat{\phi}_f^*({\rm val}(v))$ on $Y\times_X\widehat{X}_x$.
Consequently
$\nu=
f_*^{-1}\widetilde{\phi}^*i_*({\rm val}(v))=
\widetilde{\phi}_f^*({\rm val}(v))$
since the above valuations coincide on $f^{-1}(U)$. \qed
\begin{lemma} \label{le: minimal vectors2} Let $\sigma$ and
$\tau$ be $\Gamma$-semicones in isomorphic lattices
$N_\sigma\simeq N_\tau$, and $L$ be a field
containing $K$.
Consider the induced action of $\Gamma$ on
$L[[{\sigma}^\vee]]$ and $L[[{\tau}^\vee]]$ which is
trivial on $L$.
Let $\psi: L[[\sigma^\vee]]\simeq L[[{\tau}^\vee]]$ be a
$\Gamma$-equivariant isomorphism over $K$.
Let $v_\sigma\in {\operatorname{int}}(\sigma) $ be a minimal internal vector of
$\sigma$. Then
there exists a minimal internal vector $v_\tau\in {\rm
int}(\tau)$ such
that $\psi_*({\rm val}(v_\sigma))={\rm val}(v_\tau)$.
\end{lemma}
\noindent{\bf Proof.} Note that ${\rm val}(v_\sigma)(f_\sigma)> 0$ and
$\psi_*({\rm val}(v_\sigma))(f_\tau)> 0$, for
any functions $f_\sigma\in m_\sigma$, $f_\tau\in m_\tau$
from the maximal ideals
$m_\sigma\subset L[[\sigma^\vee]]$ and $m_\tau\subset
L[[\tau^\vee]]$ respectively.
The dual cone $\tau^{\vee}$ of regular characters
defines a subgroup in the
multiplicative group of rational
functions. The valuation $\psi_*({\rm val}(v_\sigma))$ determines
a group homomorphism of that subgroup into ${\bf Z}$ and
hence defines
a linear function on $\tau^{\vee}$ corresponding to a vector
$v_\tau \in {\operatorname{int}}(\tau) $. We have to show that
${\operatorname{val}}(v_\tau)=\psi_*({\rm val}(v_\sigma))$ (it is not clear whether
$\psi_*({\rm val}(v_\sigma))$ is monomial with respect to characters). By
Lemma \ref{le: monomial}, $\psi_*({\rm val}(v_\sigma))\geq {\rm
val}(v_\tau)$. Let $v_\sigma'\in {\operatorname{int}}(\sigma)$ correspond to the
linear function determined by $\psi^{-1*}({\rm val}(v_\tau))$
on the cone
$\sigma^{\vee}$. Then
${\rm val}(v_\sigma)=\psi^{-1*}\psi_*({\rm val}(v_\sigma))\geq
\psi^{-1*}({\rm val}(v_\tau))\geq
{\rm val}(v'_\sigma)$. Finally, ${\rm val}(v_\sigma)\geq {\rm
val}(v'_\sigma)$. By the
minimality of $v_\sigma$ it follows that $v_\sigma=v'_\sigma$ and
${\rm val}(v_\sigma)=\psi^{-1*}({\rm val}(v_\tau))={\rm val}(v'_\sigma)$, or equivalently
$\psi^{*}({\rm val}(v_\sigma))={\rm val}(v_\tau)$.
We need to show that $v_\tau$ is a minimal internal vector.
Suppose it isn't. Write $v_\tau=v_{\tau 0}+v_{\tau 1}$, where
$v_{\tau0}\in {\rm int(\tau)}$ and $v_{\tau 1}\in {\rm \tau}$.
By the
above find $v_{\sigma 0}\in {\rm int(\sigma)}$
such that
${\rm val}(v_{\sigma 0})=\psi^{-1*}({\rm val}(v_{\tau 0}))$.
Then
${\rm val}(v_\sigma)=\psi^{-1*}({\rm val}(v_\tau)) \geq \psi^{-1*}({\rm
val}(v_{\tau 0}))= {\rm val}(v_{\sigma 0})$. By the minimality of $v_\sigma$, we
get $v_\sigma=v_{\sigma 0}$ and $v_\tau=v_{\tau 0}$.
\qed
\begin{lemma} \label{le: Hilbert} Let $\sigma$ be a
$\Gamma$-semicone and
$v\in \sigma$ be a vector such
that for any
$\phi\in G^{\sigma}$ there exists $v_\phi\in \sigma$ such that
$\phi^*({\operatorname{val}}(v))={\operatorname{val}}(v_\phi)$
. Then ${\operatorname{val}}(v)$ is $G^{\sigma}$-invariant.
\end{lemma}
\noindent {\bf Proof.}
Set $X:=\widetilde{X}_\sigma,\,\, S:=S_{\widetilde{\sigma}} ,\,\,
x:= O_{\widetilde{\sigma}}$. Let $W$ denote the set of all
vectors ${v_\phi} \in \sigma$ for all $\phi\in {{\rm
Aut}}(X,S)$. For any natural $a$, the ideals
$I_{{\operatorname{val}}(v_\phi),a}:=\{f\in\widehat{{{\mathcal O}}}_{X,x}\mid
{\operatorname{val}}(v_\phi)(f)\geq a\}$ are
generated by monomials. They define the same Hilbert-Samuel
function
$k\mapsto \dim_K(K[X]/(I+m^k))$, where $m
$ denotes the maximal ideal. It follows that the set $W$ is finite.
On the other hand
since the ideals $I_{{\operatorname{val}}(v_\phi),a}$
are
generated by monomials they can be distinguished by
the ideals ${\rm gr}(I_{{\operatorname{val}}(v_\phi),a})$ in the graded ring
$${\rm gr}(\widehat{{{\mathcal O}}}_{X,x_{\sigma}})=
\widehat{{{\mathcal O}}}_{X,x}/m_{x,X}\oplus
m_{x,X_{\sigma}}/m_{x, X_{\sigma}}^2\oplus\ldots,$$ \noindent where
$m_{x,X}\subset \widehat{{{\mathcal O}}}_{X,x}$ is the
maximal ideal of the point $x$.
Let
$d:{{\rm Aut}}(\widetilde{X}_{\sigma},
S_{\sigma})^0\to\operatorname{Gl}({\operatorname{Tan}}_{X,x})$ be the differential morphism (see Example
\ref{ex: differential}). Then the connected algebraic group $d({{\rm Aut}}(\widetilde{X}_{\sigma},
S_{\sigma})^0)$ acts algebraically on
the connected component of the Hilbert scheme of graded ideals with fixed Hilbert polynomial
(see Example \ref{ex: differential}) . In
particular it acts trivially on the finite subset ${\rm gr}(I_{{\operatorname{val}}(v_\phi),a})$,
and consequently ${{\rm Aut}}(\widetilde{X}_{\sigma},
S_{\sigma})^0$
preserves all
$I_{{\operatorname{val}}(v_\phi),a}$.\qed
\begin{lemma} \label{le: divisors} Let $\sigma$ be a
$\Gamma$-semicone and
$\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ be a $G^{\sigma}$-equivariant morphism.
Then all its
exceptional divisors are $G^{\sigma}$-invariant.
\end{lemma}
\noindent {\bf Proof.} Any automorphism $g\in G^{\sigma}$ maps an
exceptional divisor $D$ to another exceptional divisor
$D'$. Hence the valuation ${\operatorname{val}}_D$ defined by $D$ satisfies
the condition of the previous lemma, and therefore it is $G^{\sigma}$-invariant.
\qed
\begin{definition} Let $\sigma$ be a $\Gamma$-semicone.
A valuation ${\operatorname{val}}(v)$, where $v\in
\sigma$, is {\it $G^{\sigma}$-semiinvariant} if for any $\phi\in
G^{\sigma}$ there exist $v'\in \sigma$ such that $\phi^*({\operatorname{val}}(v))={\operatorname{val}}(v')$.
\end{definition}
By abuse of terminology a vector $v\in \sigma$ will be
called
{\it $G^{\sigma}$-semiinvariant} (resp. {\it
$G^{\sigma}$-invariant}) if the corresponding valuation
${\operatorname{val}}(v)$ is $G^{\sigma}$-semiinvariant (resp. $G^{\sigma}$-invariant).
\begin{definition} Let $\Sigma$ be an oriented $\Gamma$-semicomplex.
$v\in {\rm
int}(\sigma)$ is {\it semistable} if for any $\tau\geq
\sigma$, ${\operatorname{val}}(v)$ is $G^\tau$-semiinvariant on $\widetilde{X}_\tau$.
\end{definition}
Let ${\operatorname{Vert}}(\Sigma)$ denote the set of all $1$-dimensional
rays in the fan $\Sigma$.
\begin{definition} Let $\sigma$ be a
$\Gamma$-semicone and
$\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ be a $G^{\sigma}$-equivariant
morphism. A cone $\delta\in \Delta^\sigma$ is {\it
$G^\sigma$-invariant}
if
\begin{enumerate}
\item there is $\tau\leq\sigma$ such that
${\operatorname{int}}(\delta)\subset {\operatorname{int}}(\tau)$,
\item $\overline{O}_\delta\subset
\widetilde{X}_{\Delta^\sigma}$ is $G^\sigma$-invariant.
\end{enumerate}
\end{definition}
\begin{definition} Let
$\Delta$ be a
canonical subdivision of an oriented $\Gamma$-semicomplex
$\Sigma$. A cone $\delta\in \Delta^\sigma$, $\sigma\in\Sigma$ is called a {\it $\Sigma$-stable
face} of $\Delta$ if
\begin{enumerate}
\item
${\operatorname{int}}(\delta)\subset {\operatorname{int}}(\sigma)$,
\item $\imath^\tau_{\sigma}(\delta)\in \Delta^\tau$ is
$G^\tau$-invariant for any
$\tau\geq \sigma$.
\end{enumerate}
\end{definition}
\begin{lemma} \label{le: minimal vectorss} Let
Let $\sigma$ be a
$\Gamma$-semicone and
$\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ be a $G^{\sigma}$-equivariant
morphism.
\begin{enumerate}
\item All $G^{\sigma}$-semiinvariant vectors $v\in
\sigma$ are $G^{\sigma}$-invariant.
\item ${\operatorname{Vert}}(\Delta^\sigma)\setminus
{\operatorname{Vert}}(\overline{\sigma})$ are $G^{\sigma}$-invariant.
\item Let $\delta$ be a $G^\sigma$-invariant face of
$\Delta^\sigma$. Then any minimal
internal vector $v\in {\operatorname{int}}(\delta)$ are $G^{\sigma}$-invariant.
\item Let $\delta$ be a
\underline{$\Gamma$-indecomposable} face of $\Delta^\sigma$.
Then $\delta$ is $G^\sigma$-invariant and all its
minimal internal vectors $v$ are $G^{\sigma}$-invariant.
In particular all minimal generators of
any face $\delta\in\Delta^\sigma$ are $G^{\sigma}$-invariant.
\item Let $\delta\in \Delta^\sigma$ be a minimal face such that $
{\operatorname{int}}(\delta)\subset {\operatorname{int}}(\tau)$, where $\tau\neq
\{0\}$. Then $\delta$ is $G^{\sigma}$-invariant and
${\operatorname{int}}(\delta)$ contains a $G^{\sigma}$-invariant
vector.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) Follows from Lemma \ref{le: Hilbert}.
(2) Follows from Lemma \ref{le: divisors}.
(3) By Lemma \ref{le: minimal vectors2}, $v$ is
$G^{\sigma}$-semiinvariant.
By (1), $v$ is $G^{\sigma}$-invariant.
(4) Let $\tau\preceq\sigma$ be a face of $\sigma$ such
that ${\operatorname{int}}(\delta)\subset{\operatorname{int}}(\tau)$. By Lemmas \ref{le: sum}
and \ref{le: dominate}, $\Delta^\sigma|\tau=
\Delta^\sigma|\omega(\tau)\oplus r(\tau)$ for some
regular cone $r(\tau)$. Since $\delta$ is
$\Gamma$-indecomposable we conclude that $\delta\in
\Delta^\sigma|\omega(\tau)$ and
$\tau=\omega(\tau)\leq \sigma$.
By definition $G^\sigma$ acts on
$\widetilde{X}_{\Delta^\sigma}$ and consequently on
strata from
the stratification
${\rm Sing}^\Gamma(\widetilde{X}_{\Delta^\sigma})$ (see
Lemma \ref{le: singularity type2}).
Let ${\rm sing_0}\in {\rm
Sing}(\widetilde{X}_{\Delta^\sigma})$ denote the stratum
corresponding to indecomposable face $\delta_0\in\Delta^\sigma$.
The images of the $G^\sigma$ action on ${\rm sing_0}$ form a finite invariant
subset of strata
${\rm Sing_0}\subset {\rm Sing}(\widetilde{X}_{\Delta^\sigma})$
. All strata from
${\rm Sing_0}$ correspond to some isomorphic
cones $\delta_i\in \Delta^\sigma$ for $i=0,\ldots,l$.
Let $v\in {\operatorname{int}}(\delta)$ be
a minimal internal vector.
Then by Lemma \ref{le: minimal vectors2} for any
$\widetilde{\phi}\in G^\sigma$ the image
$\widetilde{\phi}_*({\rm val}(v))$
is a valuation on $\widetilde{X}_{\Delta^\tau}$
corresponding to a minimal internal vector of $\delta_i$.
Hence $v$ is $G^\sigma$-semiinvariant and by (1) it is
$G^\sigma$-invariant.
By Lemma \ref{le: minimal0} each minimal
generator is a minimal internal vector of some
indecomposable cone $\delta$, hence by the above it is $G^\sigma$-invariant.
(5) For any $\tau\geq\sigma$ the cone $\delta$ corresponds
to the maximal component
$O_{{\delta}}$ in $\widetilde{X}_{{\Delta^\sigma}}$
which dominates $O_{{\tau}}\subset
\widetilde{X}_{\sigma}$. Then
$G^{\sigma}$ permutes the set of maximal components of the
inverse image of $O_{{\tau}}$
dominating $O_{{\tau}}$ and we repeat the
reasoning from (4).
\qed
\begin{remark}
The above lemma holds if we replace $G^{\sigma}$ with its
connected proalgebraic subgroup.
\end{remark}
As a direct corollary of Lemma \ref{le: minimal vectorss}
is the following
\begin{lemma} \label{le: minimal vectors} Let
$\Delta$ be a
canonical subdivision of an oriented $\Gamma$-semicomplex $\Sigma$.
\begin{enumerate}
\item
All the semistable vectors in $\Sigma$ are stable.
\item For any $\sigma \in \Sigma$ all elements of ${\operatorname{Vert}}(\Delta^\sigma)\setminus
{\operatorname{Vert}}(\overline{\sigma})$ are $\Sigma$-stable.
\item Let $\delta$ be a $\Sigma$-stable face of $\Delta$. Then any minimal
internal vector $v\in {\operatorname{int}}(\delta)$ is $\Sigma$-stable.
\item Let $\delta$ be a
\underline{$\Gamma$-indecomposable} face of $\Delta^\sigma$.
Then $\delta$ is $\Sigma$-stable and all its
minimal internal vectors are stable.
In particular all minimal generators of
any face
$\delta\in\Delta^\sigma$ are $\Sigma$-stable.
\item Let $\delta\in \Delta^\sigma$ be a minimal face such that $
{\operatorname{int}}(\delta)\subset {\operatorname{int}}(\sigma)$, where $\sigma\neq
\{0\}$. Then $\delta$ is $\Sigma$-stable and
${\operatorname{int}}(\delta)$ contains a $\Sigma$-stable
vector. \qed
\end{enumerate}
\end{lemma}
\bigskip
\subsection{Star subdivisions and blow-ups of stable valuations}
\noindent
\begin{proposition} \label{pr: blow-ups}
\begin{enumerate}
\item Let $\Sigma$ be an oriented $\Gamma$-semicomplex and
$v_1,\ldots,v_k$ be $\Sigma$-stable vectors. Then
$\Delta:=\langle v_k\rangle \cdot\ldots\cdot \langle v_1
\rangle\cdot\Sigma$ is a canonical subdivision of $\Sigma$.
\item Let $(X,S)$ be a $\Gamma$-stratified toroidal
variety with an associated oriented $\Gamma$-semicomplex
$\Sigma$. Let $v_1,\ldots,v_k$ be stable vectors
and $\nu_1,\ldots,\nu_k$ be the associated
stable valuations. Then the composite of blow-ups
${\operatorname{bl}}_{\nu_k}\circ\ldots\circ {\operatorname{bl}}_{\nu_1}:
X_k\rightarrow X$ is the toroidal modification of $X$
associated to the canonical subdivision
$\Delta:=\langle v_k\rangle \cdot\ldots\cdot \langle v_1
\rangle\cdot\Sigma$ of $\Sigma$.
\end{enumerate}
\end{proposition}
\noindent{\bf Proof.} Induction on $k$. Let
$\Delta_k:=\langle v_k\rangle \cdot\ldots\cdot \langle v_1
\rangle\cdot\Sigma$ be a canonical subdivision
corresponding to $f:={\operatorname{bl}}_{\nu_k}\circ\ldots\circ {\operatorname{bl}}_{\nu_1}:
X_k\rightarrow X$. Then $v_{k+1}\in{\operatorname{int}}(\sigma)$ defines an
$X$-stable valuation $\nu_{k+1}$ on $X_k$. It follows from \ref{le:
locally monomial} that for any chart $\phi_\sigma: U_\sigma\to X_{\sigma}$ the
blow-up of $f^{-1}(U_\sigma)\subset X_j$ at $\nu_{k+1}=\phi^{f*}({\operatorname{val}}(v_{k+1}))$ corresponds
to the star subdivision of $\Delta^\sigma_{k}$ at $v_{k+1}$. \qed
\section{Correspondence between toroidal morphisms and
canonical subdivisions}
\subsection{Definition of stable support}
By Lemma \ref{le: glueing} we are in a position to glue pieces ${\rm
stab}(\sigma)$ into one topological space.
\begin{definition}\label{de: stable support}
The {\it stable support} of a $\Gamma$-semicomplex $\Sigma$ is the
topological space ${\rm Stab}(\Sigma) :=\bigcup_{\sigma\in
\Sigma} {\rm stab}(\sigma)$.
\end{definition}
Let
$G\subset
{\rm Aut}(\widetilde{X}_\sigma)$ be any abstract
algebraic group.
Let $I_G$ denote the set of ${G}$-invariant toric
valuations on $\widetilde{X}_\sigma$. Set
$${\rm Inv}(G,\sigma):=\{a\cdot v\mid {\operatorname{val}}(v)\in I_G,\,\,
a\in {\bf Q}_{\geq 0}\},\quad {\rm Inv}(\sigma):={\rm Inv}(G^\sigma,\sigma).$$
\begin{lemma}\label{le: stable support} Let $\Sigma$ be a $\Gamma$-semicomplex.
\begin{enumerate}
\item ${\rm stab(\sigma)}\cap {\operatorname{int}}(\sigma)=\bigcap_{\tau\geq \sigma}{\rm
Inv}(\tau)\cap {\operatorname{int}}(\sigma)$.
\item ${\rm stab(\sigma)}=
\bigcup_{\tau\leq \sigma} ({\rm stab}(\tau)\cap {\operatorname{int}}(\tau))$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} Follows from the definitions of stable
valuation and stable support.\qed
\bigskip
\subsection{Convexity of the stable support}
\begin{lemma} \label{le: convex}
\begin{enumerate}
\item For any abstract subgroup
$G\subset {\rm Aut}(\widehat{X}_\sigma)$ the cone
${\rm Inv}(G,\sigma)$ is convex.
\item Let $\Sigma$ be a $\Gamma$-semicomplex. Then for any
$\sigma \in \Sigma$, ${\rm stab}(\sigma)$ is convex.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.}
(1). Let $v_1,v_2\in I_G$ and
${\operatorname{bl}}_{{\rm val}(v_1)}\circ {\operatorname{bl}}_{{\rm
val}(v_2)}: X_{\langle v_1\rangle\cdot \langle v_2
\rangle\cdot{{\sigma}}} \rightarrow
{X}_{\sigma} $ be the toric morphism. The induced morphism
$({\operatorname{bl}}_{{\rm val}(v_1)}\circ {\operatorname{bl}}_{{\rm
val}(v_2)})\widehat{}: X_{\langle v_1\rangle\cdot \langle v_2
\rangle\cdot{{\sigma}}}\times_{X_{\sigma}}
\widehat{X}_{\sigma}\rightarrow
\widehat{X}_{\sigma}$
is $G$-equivariant.
The exceptional divisors $D_1, D_2$ of
${\operatorname{bl}}_{{\rm val}(v_1)}\circ {\operatorname{bl}}_{{\rm
val}(v_2)}$ correspond to $
v_1,v_2 \in \sigma$. The
$D_1$ and $D_2$ intersect along a stratum $O_{\tau}$
corresponding to the cone
$\tau:=\langle v_1, v_2 \rangle$.
Then $\widehat{X}_{\tau}$
is a ${G}$-invariant local scheme of toric variety at the
generic point of the orbit $O_{\tau}$
and $\{D_1,D_2,O_{\tau}\}$ is the
orbit stratification on $(\widehat{X}_{\tau})$
which is also $G$-invariant.
Let $u_1,\ldots,u_k$ denote semiinvariant generators of
$\widehat{{{\mathcal O}}}_{X_{\tau},O_{\tau}} \simeq
K(O_{\tau})[[u_1,\ldots,u_k]]$
Each automorphism from $G$ preserves the orbit stratification and
multiplies the generating monomials $u_i$ by invertible
functions. Therefore it does not change the valuations
${\rm val}(v)$, where $ v\in\tau$,
on $\widehat{{{\mathcal O}}}_{X_{\tau},O_{\tau}} \simeq
K(O_{\tau})[[u_1,\ldots,u_k]]$.
(2) Follows from (1) and from Lemma \ref{le: stable support}
\qed
\bigskip
\subsection{Toroidal embeddings and stable support}
\begin{lemma} \label{le: tembeddings} Let $(X,S)$ be a toroidal
embedding. Then
\begin{enumerate}
\item All integral
vectors in the faces $\sigma$ of $\Sigma$ are stable.
\item
All subdivisions of $\Sigma$ are canonical.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} (1) Each automorphism $g$ from $G^{\sigma}$
preserves divisors, hence
multiplies the generating monomials by invertible
functions. Consequently, it does not change
the valuations ${\rm val}(v)$, where $ v\in\sigma$,
on $\widetilde{X}_{\sigma}$.
(2) Let $\Delta^\sigma$ be a
subdivision of $\sigma$.
For any $\delta\in \Delta^\sigma$
each automorphism $g$ of $\widetilde{X}_{\sigma}$
lifts to an automorphism $g'$ of
$\widetilde{X}_{\delta}=\widetilde{X}_{\sigma}
\times_{X_{\widetilde{\sigma}}}
{X}_{\widetilde{\delta}}$ which also multiplies monomials
by suitable invertible functions. Therefore $g$ lifts
to the scheme $\widetilde{X}_{\Delta^\sigma}$. \qed
\bigskip
\subsection{Minimal vectors and stable support}
\begin{lemma}\label{le: parr}
Let $\sigma$ be a
$\Gamma$-semicone and
$\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ be a $G^{\sigma}$-equivariant
morphism.
\begin{enumerate}
\item Let $\delta\in\Delta^\sigma$ be a simplicial cone
where $\sigma\in\Sigma$. Then all vectors in
${\rm par}(\delta)$ are $\Sigma$-stable.
\item
Let $\delta\in \Delta^\sigma$ be a $G^{\sigma}$-invariant simplicial face.
Then all vectors from
$\overline{{\rm par}(\delta)}\cap {\rm int(\delta)}$ are $G^{\sigma}$-invariant.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) Each $v\in {\rm par}(\delta)$ is a nonnegative
integral combination of
minimal generators of $\delta$. Minimal generators of
$\delta$ are $G^{\sigma}$-invariant by Lemma
\ref{le: minimal vectorss}(4). Their linear combination is
$G^{\sigma}$-invariant by Lemma
\ref{le: convex}.
(2) Let $v\in\overline{{\rm par}(\delta)}\cap {\rm
int(\delta)}$. Write $\delta=\langle
w_1,\ldots,w_k\rangle$ and $v=\sum \alpha_iw_i$, where
$0\leq \alpha_i\leq 1$. If $v$ is a minimal internal vector
then it is $G^{\sigma}$-invariant by Lemma \ref{le: minimal vectorss}(3). If not
then $v=v'+v''$, where $v'=\sum \beta_iw_i$, $0<
\beta_i\leq \alpha_i\leq 1$, is a minimal internal vector and
$v''=\sum (\alpha_i-\beta_i)w_i \in {\rm par}(\delta)$. By
(1) and Lemma \ref{le: convex}, $v'$ is $G^{\sigma}$-invariant.
\qed
A direct corollary from the abo Lemma is the following Lemma.
\begin{lemma}\label{le: par} Let
$\Delta$ be a
canonical subdivision of an oriented $\Gamma$-semicomplex
$\Sigma$.
\begin{enumerate}
\item Let $\delta\in\Delta^\sigma$ be a simplicial cone
where $\sigma\in\Sigma$. Then all vectors in
${\rm par}(\delta)$ are $\Sigma$-stable.
\item
Let $\delta\in \Delta^\sigma$ be a $\Sigma$-stable simplicial cone.
Then all vectors from
$\overline{{\rm par}(\delta)}\cap {\rm int(\delta)}$ are $\Sigma$-stable.
\qed
\end{enumerate}
\end{lemma}
\subsection{Canonical stratification on a toroidal
modification} \label{se: canonical stratification}
\begin{lemma} \label{le: stab-subdivisions}
Let $\Delta$ be a canonical subdivision of an oriented
$\Gamma$-semicomplex $\Sigma$. For any $\sigma\in\Sigma$ set $$
\Delta^\sigma_{\operatorname{stab}}:=\{\delta\in\Delta^\sigma\mid
{\operatorname{int}}(\delta)\cap {\rm stab}(\sigma)\neq\emptyset\}.$$
Then
\begin{enumerate}
\item $\Delta^\sigma_{\operatorname{stab}}\subset\Delta^\sigma$ is an
embedded $\Gamma_\sigma$-semifan in $N_\sigma$.
\item The subset $\Delta^\sigma_{\operatorname{stab}}\subset\Delta^\sigma$
consists of all $\Sigma$-stable faces in $\Delta^\sigma$.
\item For any $\tau\leq\sigma$,
$\Delta^\sigma_{\operatorname{stab}}|\imath^\sigma_\tau(\tau) =
\imath^\sigma_\tau(\Delta^\tau_{\operatorname{stab}})$.
\item For any $\sigma\in\Sigma$, the
stratification $S^{{\operatorname{stab}}}_\sigma$ on
$\widetilde{X}_{\Delta^\sigma}$ determined by the embedded $\Gamma$-semifan
$\Delta^\sigma_{\operatorname{stab}}\subset\Delta^\sigma$ is $G^{\sigma}$-invariant.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) By Lemma \ref{le: convex},
${\operatorname{stab}}(\sigma)$ is convex and there is a
unique maximal face $\omega\preceq\delta$ whose relative
interior ${\operatorname{int}}(\omega)$ intersects the stable support.
Thus $\omega \in \Delta^\sigma_{\operatorname{stab}}$.
By Lemma
\ref{le: minimal vectors}(4), either $\operatorname{sing}^\Gamma(\delta)=\{0\}$
or ${\operatorname{int}}(\operatorname{sing}^\Gamma(\delta))\cap{\operatorname{stab}}(\sigma)\neq \emptyset $.
In both cases $\operatorname{sing}^\Gamma(\delta)\leq \omega$ and consequently
$\delta=\omega\oplus^\Gamma r(\delta)$, for some
regular cone $r(\delta)$.
(2) If $\delta\in\Delta^\sigma$ is $\Sigma$-stable then by Lemma \ref{le:
minimal vectors}(3) its minimal internal
vector is $\Sigma$-stable. Therefore $\delta\in\Delta^\sigma_{\operatorname{stab}}$.
If the relative interior of $\delta\in\Delta^\sigma_{\operatorname{stab}}$
contais a stable vector
$v\in {\operatorname{int}}(\tau)$, where $\tau\leq\sigma$, then
${\operatorname{int}}(\delta)\subset{\operatorname{int}}(\tau)$,
$\delta\in\Delta^\tau_{\operatorname{stab}}$, and for any $\varrho\geq\tau$,
the closure of $O_{\imath^\varrho_\tau(\delta)}\subset
\widetilde{X}_{\Delta^\varrho}$ is exactly the center of ${\operatorname{val}}(v)$,
and therefore is $G^\varrho$-invariant.
(3) (4) follow from (2).
\qed
As a simple corollary of the above we get
\begin{lemma} There is a $\Gamma$-semicomplex $\Delta_{\operatorname{stab}}$
obtained by glueing the semicomplexes $\Delta^{\sigma\,\,{\operatorname{semic}}}_{\operatorname{stab}}$
along $\Delta^{\tau\,\,semic}_{\operatorname{stab}}$, where $\tau\leq\sigma$.\qed
\end{lemma}
\noindent{\bf Proof} For any
$\omega\in\Delta^{\sigma\,\,{\operatorname{semic}}}_{\operatorname{stab}}$ denote by
$\sigma(\omega\leq\sigma$ the semicone in $\Sigma$ for which
${\operatorname{int}}(\omega)\subset{\operatorname{int}}(\sigma(\omega))$. Then
$\omega\in\Delta^{\sigma(\omega)\,\,semic}_{\operatorname{stab}}$.
For $\omega\in \Delta^{\sigma(\omega)\,\,semic}_{\operatorname{stab}}$ and
$\gamma\in \Delta^{\sigma(\gamma)\,\,semic}_{\operatorname{stab}}$,
write $\omega\leq \gamma$ if
$\sigma(\omega)\leq\sigma(\gamma)$ and
$\imath^{\sigma(\gamma)}_{\sigma(\omega)}(\omega)\leq
\gamma$. Then for $\omega\leq \gamma$ we set
$\imath^{\gamma}_{\omega}:=
\imath^{\sigma(\gamma)}_{\sigma(\omega)}$ and
$\Gamma_\omega:= (\Gamma_{\sigma(\omega)})_\omega$.
\qed
\begin{proposition} \label{pr: canonical stratification}
Let $(X,S)$ be an oriented $\Gamma$-stratified
toroidal variety with an associated oriented $\Gamma$-semicomplex
$\Sigma$. Let $Y$ be a toroidal modification of $(X,S)$
corresponding to a canonical subdivision $\Delta$ of $\Sigma$.
Then there is a canonical stratification $R$ on $Y$ with the
following properties:
\begin{enumerate}
\item Let $\phi^f_\sigma: f^{-1}(U_\sigma)\to
{X}_{\Delta^\sigma}$ denote the morphism
induced by a chart $\phi_\sigma:U_\sigma\to
{X}_{\sigma}$. The intersections $r\cap f^{-1}(U_\sigma)$,
$r\in R$ are precisely the inverse images of
the strata associated to the embedded
$\Gamma$-semifan $\Delta^\sigma_{\operatorname{stab}} \subset\Delta^\sigma$.
\item The closures $\overline{r}$ of strata $r\in R$ are centers
of $X$-stable valuations on $Y$.
\item The $\Gamma$-semicomplex
associated to $(Y,R)$ is equal to
$\Sigma_R=\Delta_{\operatorname{stab}}$ and the atlas is given by
$\bigcup_{\sigma\in\Sigma}{\mathcal U}(\Delta^\sigma,\Delta^\sigma_{\operatorname{stab}})$.
\item The stratification $R$ is the finest stratification
on $Y$ satisfying the conditions:
\begin{enumerate}
\item The morphism $f$ maps strata in $R$ onto strata in $S$.
\item
There is an embedded semifan
$\Omega^\sigma\subset\Delta^\sigma$ such that the intersections
$ r\cap f^{-1}(U_\sigma)$, $r\in R$, are inverse images of
strata of $S_{\Omega^\sigma})$ on
$(X_{\Delta^\sigma}$.
\item
For any point $ x$ every $\Gamma_x$-equivariant
automorphism $ \alpha$ of $\widehat{X}_x$
preserving strata and
orientation
can be lifted
to an automorphism $\alpha'$ of $Y\times_X\widehat{X}_x
\to Y$ preserving strata.
\end{enumerate}
\end{enumerate}
\end{proposition}
\noindent{\bf Proof.}
For any face $\omega\in \Delta^\sigma_{\operatorname{stab}}$ we find a
$\Sigma$-stable vector $v_\omega$ in its relative interior. By Lemma
\ref{le: stable valuations} the vector $v_\omega$
corresponds to an $X$-stable valuation
$\nu_\omega$ on $Y$.
We define the closure of a stratum $r\in R$
asociated to $\omega$ to be
$$\overline{r}:=\overline{{\operatorname{strat}}_Y(\omega)}:={\operatorname{Z}}(\nu_\omega).$$
\noindent Then define the stratum $r$ as
$$ r:={\operatorname{strat}}_Y(\omega):=\overline{{\operatorname{strat}}_Y(\omega)}\setminus \bigcup_{\omega'< \omega}
\overline{{\operatorname{strat}}_Y(\omega')}.$$
By Lemma \ref{le: strata} the strata ${\operatorname{strat}}(\omega)$
of the stratification associated to the embedded semifan
$\Delta^\sigma_{\operatorname{stab}} \subset \Delta^\sigma$ satisfy the
condition
$\overline{{\operatorname{strat}}(\omega)}={\operatorname{Z}}({\operatorname{val}}(v_\omega),X_{\Delta^\sigma})$,
${\operatorname{strat}}(\omega):=\overline{{\operatorname{strat}}(\omega)}\setminus \bigcup_{\omega'< \omega}
\overline{{\operatorname{strat}}(\omega')}$. It follows by the above that
the sets $r$ define stratification $R$ satisfying
conditions (1), (2) and (3).
(4) Since the strata of $R$ on $Y$ are defined by centers of
$X$-stable valuations, conditions (a), (b) and (c) are satisfied.
Let $Q$ be a stratification on $Y$ satisfying
conditions (4a), (4b) and (4c).
For the morphism $\phi^f_\sigma$ induced by a chart
$\phi_\sigma$ and for any $x\in
U_\sigma$, let
$\widetilde{\phi}^f_\sigma: Y\times_X\widehat{X}_x\to
\widetilde{X}_{\Delta^\sigma}$
be a $G^\sigma\simeq {\operatorname{Aut}}(\widehat{X}_x,S)^0$-equivariant
isomorphism of formal completions mapping the strata of $Q$
isomorphically onto the strata of
$\widetilde{S}_{\Omega^\sigma}$ on
$\widetilde{X}_{\Delta^\sigma}$ defined by the embedded
semifan $\Omega^\sigma\subset \widetilde{\Delta}^\sigma$
By (4a) the strata of $
\widetilde{S}_{\Omega^\sigma}$
on $\widetilde{X}_{\Delta^\sigma}$ are
$G^{\sigma}$-invariant.
Hence for any two isomorphisms $
\widetilde{\phi}^f_{\sigma,i}: Y\times_X\widehat{X}_x\to
\widetilde{X}_{\Delta^\sigma}$, where $i=1,2$,
induced by
charts $\phi_{\sigma,i}$ the induced automorphism
$\widetilde{\phi}^f_{\sigma,1}(\widetilde{\phi}_{\sigma,2}^{f})^{-1}\in G^\sigma$
of
$\widetilde{X}_{\Delta^\sigma}$ maps
$\widetilde{S}_{\Omega_1^\sigma}$ to $\widetilde{S}_{\Omega_2^\sigma}$.
Since both stratifications are $G^\sigma$-invariant we get
$\widetilde{S}_{\Omega_1^\sigma}= \widetilde{S}_{\Omega_2^\sigma}$,
$\Omega_1^\sigma=\Omega_2^\sigma$.
Hence $\Omega^\sigma$ does not depend upon a chart.
Consequently, if $\tau\leq\sigma$ then
$\Omega^\sigma|\tau=\imath^\sigma_\tau(\Omega^\tau)$.
By condition (4a), the relative interior of a face
$\omega\in \Omega^\sigma$ is contained in the interior of
a face $\tau \leq \sigma$. Moreover the closure $O_{\imath^\varrho_\tau(\omega)}\subset
\widetilde{X}_{\Omega^\varrho}$ is $G^\varrho$-invariant
for any $\varrho\geq\tau$. This
shows that all faces $\omega\in \Omega^\sigma$ are
$\Sigma$-stable. Consequently,
$\Omega^\sigma\subset\Delta_{\operatorname{stab}}^\sigma$ and finally by Lemma \ref{le: strata2}.
the corresponding stratification $S_{\Delta^\sigma}$ is finer
than $S_{\Omega^\sigma}$. \qed
\subsection{Correspondence between toroidal
morphisms and canonical subdivisions}
Proposition \ref{pr: canonical stratification} can be
rephrased as follows:
\begin{theorem} \label{th: correspondence} Let $(X,S)$ be an
oriented stratified toroidal variety
(resp. $\Gamma$-stratified toroidal variety)
with the associated oriented semicomplex (resp. $\Gamma$-semicomplex)
$\Sigma$. There is a $1-1$ correspondence
between the toroidal morphisms of stratified
(resp. $\Gamma$-stratified) toroidal
varieties $f:(Y,R)\to (X,S)$ and
canonical subdivisions $\Delta$
of the oriented semicomplex
(resp. orineted $\Gamma$-semicomplex) $\Sigma$.
Moreover the semicomplex (resp. $\Gamma$-semicomplex)
associated to $(Y,R)$ is given by
$\Sigma_R= \Delta_{{\operatorname{stab}}}$. \qed
\end{theorem}
\begin{remark}
In particular, if $(X,S)$ is a
toroidal embedding (see Lemma \ref{le: tembeddings}) then
$\Sigma$ is a complex and all its subdivisions are
canonical (Lemma \ref{le: tembeddings}). We get a
1-1-correspondence between the subdivisions of the
complex $\Sigma$ and the toroidal morphisms $(Y,R)\to (X,S)$ (see \cite{KKMS}).
\end{remark}
\bigskip
\begin{example}\label{ex: isolated 3} Let $X$ be a variety with isolated
singularity of type $x_1x_2=x_3x_4$ with the
stratification consisting of the singular point and its
complement (as in Example \ref{ex: isolated}). Then the
associated semicomplex $\Sigma$ consists of the cone over a square and
its vertex. As follows from Example \ref{ex: isolated4}
the stable support consists of the ray over the center of this
square. Consequently, by Proposition \ref{pr: simple} there are three canonical nontrivial
subdivisions of $\Sigma$. One is the star subdivision at
the stable ray corresponding to the blow-up of the point. Then
$\Delta_{\operatorname{stab}}$ consists of the ray and its vertex and
corresponds to the toroidal embedding defined by the
exceptional divisor and its complement. The other two are subdivisions defined
by diagonals corresponding to two small resolutions of
singularities. $\Delta_{\operatorname{stab}}$ consists of the cone over the
diagonal and its vertex and corresponds to the smooth
stratified toroidal variety with the stratification defined by the exceptional
curve-the preimage of the singular point and its complement.
\end{example}
The language of stratified toroidal varieties allows a
combinatorial description of the \underline{Hironaka twist}.
\bigskip
\begin{example}(\cite{Hironaka1})\label{ex: Hironaka} Let $X={\bf P}^3$ be a projective
$3$-space containing two curves $l_1$ and $l_2$ intersecting transversally in two
points $p_1$ and $p_2$. These data define a stratified
toroidal variety $(X,S)$. The associated semicomplex
$\Sigma$ consists of two
$3$-dimensional cones $\sigma_1$ and $\sigma_2$
corresponding respectively to the points $p_1$ and $p_2$,
and sharing two $2$-dimensional faces $\tau_1$ and $\tau_2$
corresponding to $l_1$ and $l_2$. The Hironaka` twist $Y$
is obtained by glueing the consecutive blow-ups of
$X\setminus\{p_i\}$ at $l_1$ and $l_2$ taken in two
different orders, along the isomorphic open subsets over
$X\setminus\{p_1\}\setminus\{p_2\}$. Then $Y$ is a
stratified toroidal variety. The preimage of $p_i$ consists
of two irreducible curves $l_{i1}$ and $l_{i2}$
intersecting at $p_{i0}$. The stratification $T$ on $Y$ is determined by
the above four curves, two points and two exceptional
divisors-preimages of curves. Let $v_i$ denote
the sum of the two generators of $\tau_i$. Let $v^i$
denote the sum of the generators of $\sigma_i$.
$\Delta$ is the subdivision
of $\Sigma$ obtained by glueing the consecutive star subdivisions of
$\sigma_i$ at $\langle v_1\rangle $ and $\langle v_2\rangle $ taken in two
different orders, along the star subdivisions of $\tau_i$
at $\langle v_i\rangle$.
It follows from Example \ref{ex: Hironaka2} that
${\operatorname{Stab}}(\Sigma)$ is the union of the cones
$\langle v_1,v_2,v^1\rangle\subset \sigma_1$ and
$\langle v_1,v_2,v^2\rangle\subset \sigma_2$.
Then $\Delta_{{\operatorname{stab}}}$ consists of the relevant
cones in $\Delta$ whose relative interior intersects
${\operatorname{Stab}}(\Sigma)$. These cones correspond to the above
mentioned strata on $Y$.
\end{example}
\section{Canonical subdivisions and stable support}
\subsection{Isomorphisms of local rings and linear
transformations of stable supports}
\begin{lemma} \label{le: linear}
\begin{enumerate}
\item
Let $\sigma$ and $\tau$ be two cones of the same dimension.
Let $\psi:\widehat{X}_{\sigma}\simeq
\widehat{X}_{\tau}$
be an isomorphism preserving the closures of the toric orbits. Then
there is a linear isomorphism
$L_\psi:{\sigma}
\to \tau$ such that
$\psi_*({\operatorname{val}}(v))={\operatorname{val}}(L_\psi(v))$ for any $v\in\sigma$.
\item Let $\sigma$ and $\tau$ be two $\Gamma$-semicones of the same
dimension. Let $\psi:\widetilde{X}_{\sigma}\simeq
\widetilde{X}_{\tau}$ be a $\Gamma$-equivariant isomorphism
preserving strata. Then there is
a linear isomorphism
$L_\psi:{{\operatorname{Inv}}(\sigma)}
\to {{\operatorname{Inv}}(\sigma')}$ such that
$\psi_*({\operatorname{val}}(v))={\operatorname{val}}(L_\psi(v))$ for any $v\in{\operatorname{Inv}}(\sigma)$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) The isomorphism $\psi$ maps the divisors of
the orbit stratification on $\widetilde{X}_{\sigma}$
to the divisors of the orbit stratification on
$\widetilde{X}_{\tau}$. Consequently, the local
semiinvariant parameters at $O_{\sigma}$ are mapped to local
parameters at $O_{\tau}$ which differ
from semiinvariant ones by invertible functions. This defines a
linear isomorphism of cones $L:{\sigma}
\to {\tau}$ mapping faces of
${\sigma}$ corresponding to
strata to suitable faces of ${\tau}$
Let ${\psi}_L: \widehat{X}_\sigma\simeq
\widehat{X}_{\tau}$
denote the induced morphism. Then $\phi:=\psi^{-1}_L\psi$ is an
automorphism of
$\widehat{X}_{\sigma}$ preserving strata.
By Lemma \ref{le: tembeddings}
the automorphism $\phi$ preserves all valuations, hence
$\psi_*({\operatorname{val}}(v))= \psi_{L*}({\operatorname{val}}(v))={\operatorname{val}}(L(v))$.
(2) Let $v_1, v_2\in {\operatorname{Inv}}(\sigma)$. Then
$\psi_*({\operatorname{val}}(v_1))$ and $\psi_*({\operatorname{val}}(v_2))$ are
$G^\tau$-invariant, and correspond to
$v_1',v_2'\in {\operatorname{Inv}}(\tau)$.
Set
$\Delta^\sigma=\langle v_2 \rangle\langle v_1 \rangle\cdot\sigma$
and $\Delta^\tau=\langle v'_2 \rangle\langle v'_1 \rangle\cdot\tau$.
The isomorphism $\psi:\widetilde{X}_{\sigma}\simeq
\widetilde{X}_{\tau}$ lifts to an isomorphism
$\Psi:
\widetilde{X}_{\Delta^\sigma}\simeq
\widetilde{X}_{\Delta^\tau}$. Let $D_1$ and $D_2$ (resp. $D'_1$ and $D'_2$)
denote the exceptional divisors on
$\widetilde{X}_{\Delta^\sigma}$ (resp. $\widetilde{X}_{\Delta^\tau}$)
corresponding to $v_1$ and $v_2$ (resp. $v'_1$ and $v'_2$)
The generic point of intersection $p:=D_1\cap D_2$ is mapped
to the point of intersection $p':=D'_1\cap D'_2$. Set
$\delta_1:=\langle v_1, v_2 \rangle\in \Delta^\sigma$ and
$\delta_2:=\langle v'_1, v'_2 \rangle\in \Delta^\tau$
. Then
$(\widetilde{X}_{\delta_1},\{D_1,D_2,p\})$ is a formal completion of a toroidal
embedding which is mapped to $(\widetilde{X}_{\delta_2},\{D'_1,D'_2,p'\})$.
By (1) $\Psi_*$ determines a linear transformation on
$\langle v_1, v_2 \rangle$ and consequently a linear
transformation of ${\operatorname{Inv}}(\sigma)$ which is a linear isomorphism.
\qed
\begin{definition} Let $\sigma$ be a
$\Gamma$-semicone. A vector $v_\sigma \in {\rm
int}( \widetilde{\sigma})$ is {\it absolutely invariant}
if ${\operatorname{val}}(v_\sigma)$ is
invariant with respect to ${\rm Aut}
(\widetilde{X}_{\sigma})$.
\end{definition}
\begin{lemma} \label{le: invariant} Let $\sigma$ be a
$\Gamma$-semicone. There
exists an absolutely invariant vector $v\in {\rm
int}( \widetilde{\sigma} )$.
\end{lemma}
\noindent{\bf Proof.} We can assume that
$\sigma=\widetilde{\sigma}$ replacing $\sigma$, if
necessary, by the $\Gamma$-semicone
$\widetilde{\sigma}:=\sigma\cup |\widetilde{\sigma}|$.
Let $v\in {\rm
int}( \widetilde{\sigma} )$
be a minimal internal vector which by Lemma \ref{le:
minimal vectors}(3) is a stable vector.
By Lemma \ref{le: minimal vectors2} for any $g\in {\rm Aut}
(\widetilde{X}_{\sigma})$, $g_*({\operatorname{val}}(v))={\operatorname{val}}(v')$, where $v'$,
is also minimal. The set $W$ of minimal vectors is contained in
$\overline{\rm par}(\sigma)$ and therefore it is finite. By
the above, ${\rm Aut}(\widetilde{X}_{\sigma})$ acts on $W$.
Let $v_\sigma$ denote the sum of all the minimal vectors. By
Lemma \ref{le: linear}, ${\rm
Aut}(\widetilde{X}_{\sigma})$ acts on ${\operatorname{Inv}}(\sigma)\supset W$ and therefore it
acts trivially on $v_\sigma$.\qed
\subsection{Initial terms defined by a monomial valuation}
\begin{lemma} \label{le: valuation} Let
$F(x_1,\ldots,x_k)\in K[x_1,\ldots,x_k]$
be a quasihomogeneous polynomial of weight $w(F)$ with
respect to weights
$w(x_1),\ldots,w(x_k)$. Let $\nu$ be a valuation of a ring
$R$ and $u_1,\ldots,u_k\in R$ be such that
$\nu(u_i)\geq w(x_i)$. Then $\nu(F(u_1,\ldots,u_k))\geq
w(F)$. If $F$ belongs to the ideal $ (x_l,\ldots,x_k)$ and $\nu(u_i)> w(x_i)$
for $i=l, \ldots, k$ then $\nu(F(u_1,\ldots,u_k)) >
w(F)$.
\end{lemma}
\noindent{\bf Proof.} The proof can be reduced to the
situation, in which case $F$ is a monomial when it immediately
follows from the definition of valuation. \qed
\begin{definition} Let $\nu$ be a monomial valuation of $K[[x_1,\ldots,x_k]]$
with a basis $x_1,\ldots,x_k$ and weights $a_1,\ldots,
a_k>0$. For $f\in K[[x_1,\ldots,x_k]]$ let
$f=f_0+f_1+\ldots$ be a decomposition of $f$
into an infinite sum of quasihomogeneous polynomials such
that $\nu(f_0)<\nu(f_1)<\ldots$.
Then by the {\it initial term} of $f$ we mean
${\operatorname{in}}_\nu(f)=f_0$. Let $G=(G_1,\ldots,G_k)$ be an endomorphism
of $K[[x_1,\ldots,x_k]$ such that $x_i\mapsto G_i$. Then
$${\operatorname{in}}_\nu(G):=({\operatorname{in}}_\nu(G_1),\ldots,{\operatorname{in}}_\nu(G_k))$$
\noindent is the
endomorphism determined by the initial terms of $G_i$.
\end{definition}
\begin{lemma} \label{le: initial}
Let $G=(G_1,\ldots,G_k)$ be an automorphism of
$K[[x_1,\ldots,x_k]]$ preserving a monomial valuation
$\nu$: $G_*(\nu)=\nu$.
Then ${\operatorname{in}}_\nu(G)$
is an automorphism of $K[[x_1,\ldots,x_k]]$ preserving $\nu$.
Moreover for any $f\in K[[x_1,\ldots,x_k]]$ we have
${\operatorname{in}}_\nu(G(f))={\operatorname{in}}_\nu(G)({\operatorname{in}}_\nu(f))$.
\end{lemma}
\noindent{\bf Proof.} Write
$G_i={\operatorname{in}}_\nu(G_{i})+G_{ih}$, where $\nu(G_{ih})>\nu(G_i)$.
For any $f\in K[[x_1,\ldots,x_k]]$ write
$f={\operatorname{in}}_\nu(f)+f_{h}$, where $\nu(f_{h})>\nu(f)$. Then we have
$G(f)=f(G_1,\ldots,G_n)={\operatorname{in}}_\nu f(G_1,\ldots,G_n)+f_{h}(G_1,\ldots,G_n)$,
where by Lemma \ref{le: valuation},
$\nu(f_{h}(G_1,\ldots,G_n))>\nu(f)$.
${\operatorname{in}}_\nu(f)(x_1+y_1,\ldots,x_k+y_k)$ is a quasihomogeneous
polynomial in $x_1,\ldots,x_k,y_1,\ldots,y_k$ with weights
$\nu(x_1)=\nu(y_1)
,\ldots,\nu(x_k)=\nu(y_k)$. Write
${\operatorname{in}}_\nu f(x_1+y_1,\ldots,x_k+y_k)= {\operatorname{in}}_\nu(f)(x_1,\ldots,x_n)+f_h(x_1,y_1,\ldots,x_n,y_n)$ where
$f_h$ is a quasihomogeneous polynomial in the ideal
$(y_1,\ldots,y_n)\cdot K[[x_1,\ldots,x_k,y_1,\ldots,y_k]]$.
Applying Lemma \ref{le: valuation} to $x_i={\operatorname{in}}_\nu(G_i)$ and ,
$y_i=G_{ih}$ we obtain
\[\begin{array}{rc}
&({\operatorname{in}}_\nu(f)(G_1,\ldots,G_n))=
{\operatorname{in}}_\nu(f)({\operatorname{in}}_\nu(G_1)+G_{1h},\ldots,{\operatorname{in}}_\nu(G_n)+G_{nh}))=\\
&{\operatorname{in}}_\nu(f)({\operatorname{in}}_\nu(G_1),\ldots,{\operatorname{in}}_\nu(G_n)))+
f_h({\operatorname{in}}_\nu(G_1),\ldots,{\operatorname{in}}_\nu(G_n),G_{1h},\ldots,G_{nh})),
\end{array}\]
\noindent where by Lemma \ref{le: valuation},
$\nu(f_h({\operatorname{in}}_\nu(G_1),\ldots,{\operatorname{in}}_\nu(G_n),G_{1h},\ldots,G_{nh}))>\nu(f)$,
which gives $${\operatorname{in}}_\nu(G(f))={\operatorname{in}}_\nu(G)({\operatorname{in}}_\nu(f)).$$
Now let $G^{-1}=(G'_1,\ldots,G'_n)$. Then by the above,
$$ G(x_i)\circ
G^{-1}(x_i)=G'_{i}(G_{1},\ldots,G_{n})=x_i\quad \mbox{and}$$
$${\operatorname{in}}_\nu(G'_{i})({\operatorname{in}}_\nu(G_{1}),\ldots,{\operatorname{in}}_\nu(G_{n}))=x_i.$$
This shows that ${\operatorname{in}}_\nu(G)$ and ${\operatorname{in}}_\nu(G^{-1})$ are automorphisms.
They are
defined by quasihomogeneous polynomials of $\nu$-degrees
$\nu(x_1),\ldots,\nu(x_k)$.
It follows from Lemma \ref{le: valuation} that
$\nu({\operatorname{in}}_\nu(G(f)))\geq \nu(f)$
and $\nu({\operatorname{in}}_\nu(G^{-1}(f)))\geq \nu(f)$ for any $f\in
K[[x_1,\ldots,x_k]]$. Consequently, $\nu({\operatorname{in}}_\nu(G(f)))= \nu(f)$.\qed
\begin{lemma}\label{le: valuation2}
Let $u_1,\ldots,u_n$ be local
semiinvariant parameters on $\widetilde{X}_{\sigma}$. Write
$K[\widetilde{X}_{\sigma}]= K[[x_1,\ldots,x_k]]/I$. Let
$\nu$ be a toric valuation on
$\widetilde{X}_{\sigma}$ invariant with respect to the group of
all automorphisms of $\widetilde{X}_{\sigma}$ centered at
$O_{\widetilde{\sigma}}$.
Then $\nu$ defines a monomial valuation on
$K[[x_1,\ldots,x_k]]$ with basis $x_1,\ldots,x_n$ and
weights $\nu(u_1),\ldots,\nu(u_n)$.
Moreover:
\begin{enumerate}
\item Any automorphism $g$ of
$K[\widetilde{X}_{\sigma}]$ extends to an automorphism $G$
of $K[[x_1,\ldots,x_k]]$ preserving $\nu$ and $I$.
\item The automorphism ${\operatorname{in}}_\nu(G)$ of
$K[[x_1,\ldots,x_k]]$ defines an automorphism
${\operatorname{in}}_\nu(g)$ of $K[\widetilde{X}_{\sigma}]$ preserving $\nu$.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.} Any automorphism $g$ of
$K[\widetilde{X}_{\sigma}]$ sends parameters
$u_1,\ldots,u_n$ to $g_1=g^*(u_1),\ldots,g_n=g^*(u_n)$. Moreover
$\nu(g_i)= \nu(u_i)$ so we may assume that
$g_i=G_i(u_1,\ldots, u_n)$ are given
by some formal power series $G_i\in K[[x_1,\ldots,x_k]]$
of $\nu$-weights $\nu(u_i)$. Let $h=g^{-1}$, and let
$h_1=h^*(u_1),\ldots,h_n=h^*(u_n)$ be given
by some formal power series $H_i\in K[[x_1,\ldots,x_k]]$
of $\nu$-weights $\nu(u_i)$.
Then $\nu(H_i(G_1,\ldots,G_k))\geq \nu(u_i)$. On the other
hand
$H_i(G_1(u_1,\ldots,u_n),\ldots,G_k(u_1,\ldots,u_n))=u_i$.
Hence $(H_i(G_1,\ldots,G_k)-x_i)\in I$. In other words
$H_i(G_1,\ldots,G_k)=x_i+F_i(x_1,\ldots,x_k)$, where
$\nu(F_i)\geq \nu(u_i)$ and $F_i\in I$. Consequently, for
any $F\in K[[x_1,\ldots,x_k]]$, $\nu(F)\geq \nu(F(G_1,\ldots,G_k))$.
Let $m\subset K[[x_1,\ldots,x_k]]$ be the
maximal ideal. By definition $m/(m^2+I)=m/m^2$ and
thus $I\subset m^2$. This gives $F_i\in m^2$.
Then the ring automorphism $\Phi:=H\circ G$ of
$K[[x_1,\ldots,x_k]]$ can be written as a $K$-linear
transformation $\Phi={\operatorname{id}}+F$, where
$F(K[[x_1,\ldots,x_k]])\subset I$ and $F(m^i)\subset m^{i+1}$.
It follows from Lemma \ref{le: valuation} that for any
formal power series $x\in
K[[x_1,\ldots,x_k]]$, $\nu(\Phi(x))\geq\nu(x)$ and hence
$\nu(F(x))\geq \min\{\nu(\phi(x)),\nu(x)\}=\nu(x)$.
The inverse of $\Phi$ can be written as
$\Phi^{-1}={\operatorname{id}}+F'$, where $F'=-F+F^2-F^3+\ldots$. By the above
$F'(K[[x_1,\ldots,x_k]])\subset I$ and
$\nu(F'(x))\geq \nu(x)$, which shows that
$\nu(\Phi^{-1}(x))\geq \nu(x)$ and finally
$\nu(\Phi^{-1}(x))= \nu(x)$.
Hence $G$ and $G^{-1}=\Phi^{-1}\circ H$ are automorphisms
satisfying $\nu(G^{-1}(x)\geq \nu(x)$.
By Lemma \ref{le: valuation} and the above, $\nu(G(x))
\geq \nu(x)$ for any $x\in K[[x_1,\ldots,x_k]]$.
Hence $\nu(G(x))=\nu(x)$.
We have shown that any automorphism $g$ of
$K[\widetilde{X}_{\sigma}]$ is defined by an automorphism
$G$ of
$K[[x_1,\ldots,x_k]]$ preserving the valuation $\nu$ and the ideal $I$.
Then by Lemma \ref{le: initial}, ${\operatorname{in}}_\nu(G)$ is an
automorphism of $K[[x_1,\ldots,x_k]]$ preserving the valuation
$\nu$ and the ideal ${\operatorname{in}}_\nu(I)=I$. Therefore it defines an
automorphism ${\operatorname{in}}_\nu(G)$ of $K[\widetilde{X}_{\sigma}]$. \qed
\subsection{Structure of the group of automorphisms of the
completion of a local ring of a toric variety}
\begin{lemma}
\label{le: group structure}
Let $\sigma$ be a $\Gamma$-semicone.
Let $\nu$ be the valuation on
$\widetilde{X}_{\sigma}$ associated to the absolutely
invariant vector $v_\sigma$. Let $G^\sigma_{\nu}:={\rm Aut}_\nu
(\widetilde{X}_{\sigma})$ denote the group of all
quasihomogeneous automorphisms with respect to
semiinvariant coordinates on $\widetilde{X}_{\sigma}$
with weights detrmined by $\nu$. Then
\begin{enumerate}
\item $G^\sigma_{\nu}$ is an \underline{algebraic group}.
\item There is a
surjective morphism
$\Phi:{\rm Aut}(\widetilde{X}_{\sigma})\to {\rm Aut}_\nu
(\widetilde{X}_{\sigma})$ defined by taking the lowest degree
quasihomogeneous part.
\item The kernel ${\rm ker}_\sigma$ of $\Phi$ is a
connected proalgebraic group.
\item ${\rm Aut}(\widetilde{X}_{\sigma}) =G^\sigma_{\nu}\cdot {\rm ker}_\sigma$.
\item For any $g\in {\rm ker}_\sigma$ there exists a morphism $i_{g}:{\bf A}^1\to {\rm ker}_\sigma$
such that $g= i_{g}(1)$, ${\operatorname{id}}=i_{g}(0)$.
\item For any $g\in {\rm Aut}
(\widetilde{X}_{\sigma})^0$ there exists a morphism
$i_g: W :={\bf A}^1\times (G^{\sigma}_{\nu})^0\to
{\rm Aut}(\widetilde{X}_{\sigma})^0$
such that $\{g,{\operatorname{id}}\}\subset i_g(W)$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.}
Let $u_1,\ldots,u_k$ be semiinvariant parameters on
$\widetilde{X}_{\sigma}$ generating the maximal ideal of
$O_{\widetilde{\sigma}}$. Denote by $m_1,\ldots,m_k\in
{\widetilde{\sigma}}$ the
corresponding characters.
(1) Let ${\rm Aut}_\nu
({{\bf A}^k})$ denote the group of all
quasihomogeneous automorphisms of ${\bf A}^k$. By Lemma \ref{le: valuation2}, ${\rm Aut}_\nu
(\widetilde{X}_{\sigma})$ is isomorphic to the subgroup of ${\rm Aut}_\nu
({{\bf A}^k})$ consisting of the automorphisms preserving
$I$.
The automorphisms of ${\rm Aut}_\nu
({{\bf A}^k})$
are described by sets of $k$ quasihomogeneous polynomials with
weights $\nu(u_1),\ldots,\nu(u_k)$ and having linearly
independent linear terms. Thus ${\rm Aut}_\nu
({{\bf A}^k})$ and ${\rm Aut}_\nu
(\widetilde{X}_{\sigma})$ are algebraic.
(2) This follows from Lemma $\ref{le: valuation}$ and $\ref{le: valuation2}$
(3)(4)(5) The $1$-parameter
subgroup $t\mapsto t^{\nu_\sigma}$ defines the action on
$\widetilde{X}_{\sigma}$ such that
$\psi_t(u_i)=t^{(m_i,v_\sigma)}u_i=t^{\nu(u_i)}u_i=t^{\alpha_i}$,
where $\alpha_i=\nu(u_i)$ is a $\nu$-weight of $u_i$.
The action defines an embedding $K^* < {\rm
Aut}(\widetilde{X}_{\sigma})$.
Write $g\in {\rm ker}_\sigma$ in terms of coordinate functions,
$g=(g_1,\ldots,g_k)$, where $g_i=g^*(u_i)$. Let
$\alpha_{i,0}=\alpha_i$ denote the $\nu$-weight of $g_i$ and let
$g_i=u_i+g_{\alpha_{i,1}}+\ldots+g_{\alpha_{i,j}}+\ldots$
be the decomposition according to $\nu$-weights
$\alpha_{i,0}<\alpha_{i,1}<\ldots<\alpha_{i,j}<\ldots$. Define the morphism
$i_g: K^* \to {\rm ker}_\sigma$ by $i_g(t):={\psi_t}^{-1}\circ g \circ
\psi_t$. Then $(i_g(t))^*(u_i)=({\psi_t}^{-1}\circ g \circ
\psi_t)^*(u_i)=(g \circ
\psi_t)^*t^{-\alpha_i}(u_i)=\psi_t^*(t^{-\alpha_i}g_i)=
t^{-\alpha_i}(t^{\alpha_{i,0}}u_i+t^{\alpha_{i,1}}g_{\alpha_{i,1}}+\ldots+
t^{\alpha_{i,j}}
g_{\alpha_{i,j}}+\ldots)=
=u_i +t^{\alpha_{i,1}-\alpha_{i,0}}
g_{\alpha_{i,1}}+\ldots+
t^{\alpha_{i,j}-\alpha_{i,0}}g_{\alpha_{i,j}}+\ldots$.
The above morphism extends to a morphism
$i_g: {\bf A}^1\to {\rm ker}_\sigma$
''connecting'' ${\operatorname{id}}$ to $g$. This shows that ${\rm ker}_\sigma$ is
connected. Note that $\Psi$ maps the connected component ${\rm Aut}_\nu
(\widetilde{X}_{\sigma})^0$ to the
connected component ${\rm
Aut}(\widetilde{X}_{\sigma})^0$.
(6) Write $g\in {\rm Aut}
(\widetilde{X}_{\sigma},S)^0$ as $g=g_1g_2$, where $g_1\in
{\rm ker}_\sigma$, $g_2\in G^\sigma_{\nu}$. Then for $t\in {\bf A}^1$ and
$h\in G^\sigma_{\nu}$, set
$i_g(t,h):=i_{g_1}(t)\cdot h$. \qed
\begin{lemma}(\cite{Demushkin})\label{le: surjection})
Let $\sigma$ be a $\Gamma$-semicone.
Then ${\rm Aut}(\widetilde{X}_{\sigma})^0\subset
{\rm Aut}(\widetilde{X}_{\sigma})$ is a normal subgroup and
there is a natural surjection
${\rm Aut}(\sigma) \longrightarrow
{\rm Aut}(\widetilde{X}_{\sigma})/{\rm Aut}(\widetilde{X}_{\sigma})^0.$
\end{lemma}
\noindent{\bf Proof.}
Set $G:={\rm Aut}_{\nu}(\widetilde{X}_{\sigma})=
{\rm Aut}_{\nu}({X}_{\widetilde{\sigma}})$
(notation of Lemma \ref{le: group structure}).
It
follows from Lemma \ref{le: group structure} that the
natural inclusion $G\subset {\rm
Aut}(\widetilde{X}_\sigma)$ determines a group isomorphism $G/G^0\simeq
{\rm Aut}(\widetilde{X}_\sigma)/ ({\rm Aut}(\widetilde{X}_\sigma))^0$. Then we
get a surjection $N_G(T)/T\to G/G^0$. Let $H:={\rm
Aut}_T(X_{\sigma})$ be the group of the
$\Gamma$-equivariant automorphisms of $X_\sigma$,
preserving the big torus $T$. It
suffices to show that $N_G(T)=H$ and $H/T\simeq
{\operatorname{Aut}}(\sigma)$. If $g\in N_G(T)$ then $g{
T}g^{-1}=T$. Let $x\in T\subset X_{\widetilde{\sigma}}$. Then $g{
T}g^{-1}x=Tx$ or equivalently ${
T}g^{-1}x=g^{-1}Tx$. Since the latter subset is open,
$g^{-1}x$ is in the big open orbit ${
T}g^{-1}x={
T}x$, in other words, $g^{-1}x\in T$ which shows that
$g\in H$ and $N_G(T)\subset H$.
Let $H^0$ be the connected component of $H$
containing ${\operatorname{id}}$. Then $H_0$ acts trivially on irreducible
components of the complement of $T$,that is, on ${T}$-invariant Weil and Cartier divisors. Hence it
multiplies characters by invertible functions on $X_{\widetilde{\sigma}}$.
Invertible functions on $T$ are monomials. Invertible
monomials on $X_{\widetilde{\sigma}}$ are constants. Thus $H_0$ acts on
characters multiplying them by nonzero constants. This
shows that $H_0=T$. Hence for any $g\in H$, $g{T}g^{-1}=T$, which shows
that $N_G(T)\supset H$.
$H/T$ can be identified with the subgroup of $H$
preserving $1\in T\subset X_{\widetilde{\sigma}}$, that is, the automorphisms
mapping characters to characters. The latter subgroup is
equal to the group ${\rm Aut}(\widetilde{\sigma})$ of
automorphisms of embedded semifan $\sigma\subset
\widetilde{\sigma}$.
The restriction of an automorphism
in ${\rm Aut}(\widetilde{\sigma})$ is an automorphism
of $\sigma$. Any automorphism of $\sigma$ defines an
automorphism of
$\widetilde{\sigma}=\sigma\times{\operatorname{reg}}(\sigma)$. Thus there
are natural group homomorphisms $i: {\rm
Aut}({\sigma})\to {\rm Aut}(\widetilde{\sigma})$ and
$p:{\rm Aut}(\widetilde{\sigma})\to {\rm Aut}({\sigma})$
such that $ p\circ i={\operatorname{id}}_{\sigma}$.
Moreover $p$ corresponds to the restriction of
automorphisms of $X_{\widetilde{\sigma}}$ to the subvariety
$X_{\sigma}\times O_{{\operatorname{reg}}(\sigma)}\subset X_{\widetilde{\sigma}}$.
By Lemma \ref{le: sections}, $p({\rm
Aut}(\widetilde{\sigma})^0={\rm Aut}({\sigma})^0$.
\qed
\subsection{Invariant and semiinvariant valuations}
\begin{lemma} \label{le: semi}
Let $v\in\sigma$ be an integral vector. Then ${\operatorname{val}}(v)$
is an ${\rm
Aut}(\widetilde{X}_{\sigma})^0$-invariant valuation on
$\widetilde{X}_{\sigma}$ iff it is ${\rm
Aut}(\widetilde{X}_{\sigma})$-semiinvariant.
\end{lemma}
\noindent{\bf Proof.} $(\Leftarrow)$. If ${\operatorname{val}}(v)$ is ${\rm
Aut}(\widetilde{X}_{\sigma})$-semiinvariant then it is ${\rm
Aut}(\widetilde{X}_{\sigma})^0$-semiinvariant. Hence by
Lemma \ref{le: minimal vectorss}(1) it is ${\rm
Aut}(\widetilde{X}_{\sigma})^0$-invariant.
$(\Rightarrow)$. If ${\operatorname{val}}(v)$ is ${\rm
Aut}(\widetilde{X}_{\sigma})$-invariant then by Lemma
\ref{le: surjection}, ${\rm
Aut}(\widetilde{X}_{\sigma})$ acts on ${\operatorname{val}}(v)$ as the finite
group ${\rm Aut}(\sigma)\longrightarrow
{\rm Aut}(\widetilde{X}_{\sigma})/ {\rm Aut}(\widetilde{X}_{\sigma})^0 $. Therefore ${\operatorname{val}}(v)$ is ${\rm
Aut}(\widetilde{X}_{\sigma})$-semiinvariant. \qed
\begin{lemma} \label{le: semi2} Let $L$ be any
algebraically closed field containing $K$.
Then ${\operatorname{val}}(v)$ defines ${\rm
Aut}(\widetilde{X}_{\sigma})^0$-invariant valuation on
$\widetilde{X}_{\sigma}$ iff it defines an ${\rm
Aut}_L(\widetilde{X}_{\sigma})^0 $-invariant valuation on
$\widetilde{X}^L_{\sigma}$.
\end{lemma}
\noindent {\bf Proof.} Follows from Lemma \ref{le: base}(3).
\qed
For a $\Gamma$-semicone $\sigma$ denote by
$\widehat{{\operatorname{Inv}}}(\sigma)$ the cone generated by
vectors $v\in |\sigma|$, for which valuations
${\operatorname{val}}(v,\widehat{X}_{\sigma})$ are invariant with
respect to ${\rm
Aut}(\widehat{X}_{\sigma})^0.$
\begin{lemma}\label{le: hat}
For a $\Gamma$-semicone $\sigma$ the two cones are equal
$\widehat{{\operatorname{Inv}}}(\sigma)={{\operatorname{Inv}}}(\sigma)$.
\end{lemma}
\noindent {\bf Proof.} Note that
$K[\widetilde{X}_{\sigma}]=
L[\widehat{X}_{\sigma}]$ for suitable
$L:=K[[x_1,\ldots,x_k]]$. Let $(L)$ denote the algebraic
closere of the quotient
field of $L$.
Each automorphism of $K[[\sigma^\vee]]$ defines the
automorphism of $L[[\sigma^\vee]]$ which is constant on $L$.
Each automorphism in ${\rm^\vee
Aut}(\widehat{X}_\sigma)^0$ determines an automorphism in ${\rm
Aut}(\widetilde{X}_\sigma)^0$, so that there is a
monomorphism of proalgebraic groups $\imath: {\rm
Aut}(\widehat{X}_\sigma)^0 \to {\rm
Aut}(\widetilde{X}_\sigma)^0$. Therefore if
${\operatorname{val}}(v,\widetilde{X}_\sigma)$ is invariant with respect to ${\rm
Aut}(\widetilde{X}_\sigma)^0$ then it is $\imath({\rm
Aut}(\widehat{X}_\sigma)^0)$-invariant.
Since the subring $K[\widehat{X}_{\sigma}]\subset
K[\widetilde{X}_{\sigma}]$ is $\imath({\rm
Aut}(\widehat{X}_\sigma)^0)$-invariant then the
restriction ${\operatorname{val}}(v,\widehat{X}_\sigma)$ of
${\operatorname{val}}(v,\widetilde{X}_\sigma)$ to $\widehat{X}_\sigma$ is ${\rm
Aut}(\widehat{X}_\sigma)^0$-invariant.
Now, if ${\operatorname{val}}(v,\widehat{X}_\sigma)$ is ${\rm
Aut}(\widehat{X}_\sigma)^0$-invariant then by
Lemma \ref{le: semi2}, it defines an ${\rm
Aut}_{(L)}(\widehat{X}^{(L)}_\sigma)^0$-invariant valuation.
By Lemma \ref{le: semi}, ${\operatorname{val}}(v,\widehat{X}^{(L)}_\sigma)$ is ${\rm
Aut}_{(L)}(\widehat{X}^{(L)}_\sigma)$-semiinvariant.
The proalgebraic group ${\rm
Aut}_{L}(\widetilde{X}_\sigma)={\rm
Aut}_{L}(\widehat{X}^L_\sigma)$ is a subgroup (as
an abstract group) of ${\rm
Aut}_{(L)}(\widehat{X}^{(L)}_\sigma)$. Thus the
restriction ${\operatorname{val}}(v,\widetilde{X}_\sigma)$ of
${\operatorname{val}}(v,\widehat{X}^{(L)}_\sigma)$ is ${\rm
Aut}_{L}(\widetilde{X}_\sigma)$-semiinvariant.
By Lemma \ref{le: decomposition} any
automorphism $\phi$ in ${\rm
Aut}(\widetilde{X}_\sigma)$ can be
decomposed as $\phi=\phi_0\phi_1$, where $\phi_0$ preserves
monomials and $\phi_1\in {\rm
Aut}_{L}(\widetilde{X}_\sigma)$.
Therefore ${\operatorname{val}}(v,\widetilde{X}_\sigma)$ is ${\rm
Aut}(\widetilde{X}_\sigma)$-semiinvariant and consequently
by Lemma \ref{le: semi} it is ${\rm
Aut}(\widetilde{X}_\sigma)^0$-invariant. \qed
\begin{lemma}\label{le: stableincl} If $\sigma\leq \tau$ then
\begin{enumerate}
\item ${\operatorname{Inv}}(\sigma)\subset {\operatorname{Inv}}(\tau)$.
\item ${\operatorname{stab}}(\sigma)\subset{\operatorname{stab}}(\tau)\cap \sigma$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.}
(1) Let $v\in {\operatorname{Inv}}(\sigma)$. Any automorphism
$\phi\in {\operatorname{Aut}}(\widetilde{X}_{\tau})$ preserves the
stratum ${\operatorname{strat}}(\sigma)$ and thus defines an automorphism
$\widehat{\phi}$ of the completion
of $Y:=\widetilde{X}_{\tau}$ at ${\operatorname{strat}}(\sigma)$ which by Lemma
\ref{le: 1} is isomorphic to the spectrum of the ring
$K[\widehat{Y}_{{\operatorname{strat}}(\sigma)}]=K_\sigma[\widehat{X}_{\sigma}]$,
where $K_\sigma$ is the residue field of the generic orbit
point $O_\sigma$. By Lemma \ref{le: decomposition} any
automorphism $\widehat{\phi}$ of $K_\sigma[\widehat{X}_{\sigma}]$
can be
written as $\widehat{\phi}=\phi_0\phi_1$, where $\phi_0$ preserves
monomials and $\phi_1\in {\rm
Aut}_{K_\sigma}(\widehat{X}_{\sigma})$. By Lemma \ref{le:
hat}, ${\operatorname{val}}(v,\widehat{X}_{\sigma})$ is ${\rm
Aut}(\widehat{X}_{\sigma})^0$-invariant. Let
$\overline{K_\sigma}$ be the algebraic closure of $K_\sigma$.
By Lemma \ref{le: semi2},
${\operatorname{val}}(v,\widehat{X}^{\overline{K_\sigma}}_{\sigma})$ is ${\rm
Aut}_{K_\sigma}(\widehat{X}^{\overline{K_\sigma}}_{\sigma})^0$-invariant.
By Lemma \ref{le: semi},
${\operatorname{val}}(v,\widehat{X}^{\overline{K_\sigma}}_{\sigma})$ is ${\rm
Aut}_{K_\sigma}(\widehat{X}^{\overline{K_\sigma}}_{\sigma})$-semiinvariant.
Then its restriction
${\operatorname{val}}(v,\widehat{X}^{K_\sigma}_{\sigma})$ is ${\rm
Aut}_{K_\sigma}(\widehat{X}^{K_\sigma}_{\sigma})$-semiinvariant.
Thus $\phi_{1*}({\operatorname{val}}(v,\widehat{X}^{K_\sigma}_{\sigma}))$ can
be one of the finitely many toric valuations on
$\widehat{X}^{K_\sigma}_{\sigma}$ for any $\phi_1\in {\rm
Aut}_{K_\sigma}(\widehat{X}_{\sigma})$. But then
$\widehat{\phi}_{*}({\operatorname{val}}(v,\widehat{Y}_{{\operatorname{strat}}(\sigma)}))=\phi_{1*}({\operatorname{val}}(v,
\widehat{Y}_{{\operatorname{strat}}(\sigma)}))$
can be one of the finitely many toric valuations on
$K[\widehat{Y}_{{\operatorname{strat}}(\sigma)}]$ for all
$\phi\in {\operatorname{Aut}}(\widetilde{X}_{\tau})$. The restrictions of these
valuations to the local ring of $\widetilde{X}_{\tau}$ at
${\operatorname{strat}}(\sigma)$ define finitely many valuations on
$\widetilde{X}_{\tau}$. Hence ${\operatorname{val}}(v,\widetilde{X}_{\tau})$ is semiinvariant and
consequently invariant on $\widetilde{X}_{\tau}$.
(2) Follows from (1) and from Lemma \ref{le: stable support}.
\qed
\subsection{Group of divisor classes of the completion of a local
ring of a toric variety}
\begin{lemma} \label{le: divisor classes} Let $\sigma$ be a
cone in a lattice $N$ and $\Delta$ be a suddivision $\sigma$.
Let $\widehat{X}_\Delta:={X}_\Delta\times_{X_\sigma}\widehat{X}_\sigma$.
The following groups of divisor
classes are isomorphic. (The isomorphisms are determined by the
natural morphisms).
\begin{enumerate}
\item $\operatorname{Cl}(\widehat{X}_\sigma)\simeq \operatorname{Cl}(X_\sigma)
$, ${\operatorname{Pic}}(\widehat{X}_\sigma)\simeq{\operatorname{Pic}}(X_\sigma)
$.
\item $\operatorname{Cl}(\widehat{X}_\Delta)\simeq\operatorname{Cl}({X}_\Delta)$,
${\operatorname{Pic}}(\widehat{X}_\Delta)\simeq{\operatorname{Pic}}({X}_\Delta)$
.
\item For the affine
toric variety $X_\sigma$ set
${\bf A}^1\widehat{\times}\widehat{X}_\sigma:=\lim_{\to} {\bf A}^1\times
X^{(n)}_\sigma=\operatorname{Spec}(K[t][\widehat{X}_\sigma])$. Then \\
$\operatorname{Cl}({\bf A}^1\widehat{\times}\widehat{X}_\sigma)\simeq
\operatorname{Cl}({\bf A}^1{\times} {X}_\sigma)\simeq \operatorname{Cl}({X}_\sigma)$
, ${\operatorname{Pic}}({\bf A}^1\widehat{\times}\widehat{X}_\sigma)\simeq
{\operatorname{Pic}}({\bf A}^1{\times} {X}_\sigma)\simeq {\operatorname{Pic}}({X}_\sigma)$.
\item For any
subdivision $\Delta$ of $\sigma$ set
${\bf A}^1\widehat{\times}\widehat{X}_\Delta:=
({\bf A}^1\widehat{\times}\widehat{X}_\sigma)\times_{X_\sigma}{X}_\Delta$
. Then \\ $\operatorname{Cl}({\bf A}^1\widehat{\times}\widehat{X}_\Delta)\simeq
\operatorname{Cl}({\bf A}^1{\times}\widehat{X}_\Delta)\simeq \operatorname{Cl}({X}_\Delta)$,
${\operatorname{Pic}}({\bf A}^1\widehat{\times}\widehat{X}_\Delta)\simeq
{\operatorname{Pic}}({\bf A}^1{\times}\widehat{X}_\Delta)\simeq {\operatorname{Pic}}({X}_\Delta)$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.} Let $v_1,\ldots,v_k$ denote the
generators of $\sigma\cap N_\sigma$ and let $\tau=\langle e_1,\ldots,e_k
\rangle$ denote the regular
$k$-dimensional cone. Let $\pi:\tau\to \sigma$ be the
projection defined by
$\pi(e_i)=v_i$. Then $\pi$ defines a surjective morphism of
lattices $\pi:N_\tau\to N_\sigma$ whose kernel is a
saturated sublattice $N\subset N_\tau$. Thus the projection
$\pi$
defines the quotient map
$X_\tau \to X_\sigma=X_\tau//T$ for the subtorus $T\subset
T_\tau$ corresponding to the sublattice $N\subset N_\tau$.
This also defines the quotient morphism $\alpha: \widehat{X}_\tau\to \widehat{X}_\sigma$,
where $ K[\widehat{X}_\sigma] = K[\widehat{X}_\tau]^T$.
Denote by $T_\tau\subset X_\tau $, $T_\sigma\subset
X_\sigma $ the relevant tori and set
$\widehat{T}_\tau:=\widehat{X}_\tau\times_{X_\tau}
T_\tau\subset \widehat{X}_\tau$,
$\widehat{T}_\sigma:=\widehat{X}_ \sigma\times_{X_\sigma}
T_\sigma\subset \widehat{X}_\sigma$. Then both schemes are nonsingular and let
$\alpha': \widehat{T}_\tau \to \widehat{T}_\sigma $ be the restriction of $\alpha$. We also have
$ K[\widehat{T}_\sigma] = K[\widehat{T}_\tau]^T$.
The morphism $\alpha$ defines a group homomorphism
$\operatorname{Cl}(\widehat{T}_\sigma)={\operatorname{Pic}}(\widehat{T}_\sigma)
\buildrel \alpha^*\over \longrightarrow
{\operatorname{Pic}}(\widehat{T}_\tau)=\operatorname{Cl}(\widehat{T}_\tau)=0$. The last
group is $0$ since the ring $K[\widehat{X}_\tau]$ as well as
its localization $K[\widehat{T}_\tau]$ are UFD. Let $D$ be
an effective Cartier divisor on $\widehat{T}_\sigma$. Then $\alpha^*(D)$ is $T$-
invariant and principal on
$\widehat{T}_\tau$. This means that the ideal $I$ of $\alpha^*(D)$ is
generated by $f\in K[\widehat{T}_\tau]$. By multiplying
by a suitable monomial we can asume that $f\in
K[\widehat{X}_ \tau]^T$. Note that all $t\cdot f$, $t\in T $
also belong to $I$. Let $M_T$ denote the lattice of
characters of $T$. Let $f_\beta$ denote the component of $f$
with weight $\beta\in M_T$. By considering
$K[\widehat{X}_\tau]/m^k$, where $m\subset K[\widehat{X}_\tau]$ is the maximal ideal,
we see that all $f_\beta+m^k\in I\cdot
K[\widehat{X}_\tau]/m^k$. Moreover $I\cdot
K[\widehat{X}_\tau]/m^k$ is generated by all
$f_\beta+m^k$. This yields $I=(f_\beta)_{\beta\in M_T}$ or
$(1)=(f_\beta/f)$ showing that there is a $\beta_0$ such
that $f_{\beta_0}/f$ is invertible, which means
$I=(f_{\beta_0})$. By multiplying by
monomials we can assume that $f_{\beta_0}$ is $T$-invariant.
Thus $f_{\beta_0}\in
K[\widehat{T}_\sigma]^T=K[\widehat{X}_\tau]$ and generates the
ideal of $D$. But then $D$ is principal on
$\widehat{X}_\tau$ and finally $\operatorname{Cl}(\widehat{T}_\sigma)=0$.
Now any divisor $D$ in $\operatorname{Cl}(\widehat{X}_\sigma)$
(respectively in $\operatorname{Cl}(\widehat{X}_\Delta)$) is linearly
equivalent to the $T$-invariant one
$D':=D-(f_D)$, where $D_{|\widehat{T}_\sigma}=
(f_D)_{|\widehat{T}_\sigma}$.
(3)(4) We repeat the reasoning from (1) and (2) and use the
fact that $K[{\bf A}^1\widehat{\times}\widehat{X}_\tau]=K[t][[x_1,\ldots,x_k]]$ is UFD
(see Bourbaki\cite{Bourbaki}).
\qed
\begin{lemma}\label{le: trivial action}
\begin{enumerate}
\item $G^{\sigma}$ acts trivially on
$\operatorname{Cl}(\widetilde{X}_{\Delta^\sigma})$.
\item Let $\Delta^\sigma$ be a subdivision
of $\sigma$ such that $\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ is $G^{\sigma}$-equivariant. Then
$G^{\sigma}$ acts trivially on $\operatorname{Cl}(\widetilde{X}_{\Delta^\sigma})$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.} We shall use the notation and
results from Lemma \ref{le: group structure}.
(1) The natural morphism
$\widetilde{X}_{\Delta^\sigma}\to {X}_{\Delta^\sigma}$ is
$G^\sigma_{\nu}$-equivariant. By Lemma \ref{le: divisor classes} this
morphism induces a $G^\sigma_{\nu}$-equivariant isomorphism
$\operatorname{Cl}(\widetilde{X}_{\sigma})\to \operatorname{Cl}({X}_{\sigma})$.
By Sumihiro \cite{Sumihiro} the algebraic linear group acts trivially on
$\operatorname{Cl}({X}_{\sigma})$, which yields a trivial action on
$\operatorname{Cl}(\widetilde{X}_{\sigma})$. By Lemma \ref{le: group
structure} it suffices to show that
${\rm ker}_\sigma$ acts trivially on $\operatorname{Cl}(\widetilde{X}_{\sigma})$.
Let $\widetilde{X}_\sigma^{\rm ns}\subset
\widetilde{X}_\sigma$ denote the open subset of
nonsingular points. Its complement is of codimension 2.
We get a $G^{\sigma}$-equivariant isomorphism
${\operatorname{Pic}}(\widetilde{X}_\sigma^{\rm ns})=
\operatorname{Cl}(\widetilde{X}_\sigma^{\rm ns})\simeq
\operatorname{Cl}(\widetilde{X}_\sigma)$. It suffices to prove
that ${\rm ker}_\sigma$ acts trivially on
${\operatorname{Pic}}(\widetilde{X}_\sigma^{\rm ns})$. Fix $g\in G$.
Let $i_{g,{\bf A}^1}:{\bf A}^1\to {\rm ker}_\sigma$ be the morphism from Lemma
\ref{le: group structure}.
Let $\Phi:
G^{\sigma}\widehat{\times}\widetilde{X}_\sigma^{\rm ns}\to
\widetilde{X}_\sigma^{\rm ns} $ be the
action morphism from Lemma \ref{le: group action} and set
$\Phi_{{\bf A}^1}:=i_{g,{\bf A}^1}\circ
\Phi:
{\bf A}^1\widehat{\times}\widetilde{X}_\sigma^{\rm ns}\to
\widetilde{X}_\sigma^{\rm ns} $. Denote by
$p: {\bf A}^1\widehat{\times}\widetilde{X}_\sigma^{\rm ns}\to
\widetilde{X}_\sigma^{\rm ns} $ the standard
projection. By Lemma \ref{le: divisor classes}(4),
$p^*:{\operatorname{Pic}}(\widetilde{X}_\sigma^{\rm ns})\to
{\operatorname{Pic}}({\bf A}^1\widehat{\times}\widetilde{X}_\sigma^{\rm ns})$ is
an isomorphism . Let $D\in {\operatorname{Pic}}(\widetilde{X}_\sigma^{\rm ns})$.
Let $j_g:\widetilde{X}_\sigma^{\rm ns}\to
\{g\}\widehat{\times}\widetilde{X}_\sigma^{\rm ns}
\subset {\bf A}^1\widehat{\times}\widetilde{X}_\sigma^{\rm ns}$
denote the standard embedding. Then $p\circ
j_g={\operatorname{id}}_{|\widetilde{X}_\sigma^{\rm ns}}$. Since
$p^*$ is an isomorphism there is $D'\in
{\operatorname{Pic}}(\widetilde{X}_\sigma^{\rm ns})$ such that
$\Phi_{{\bf A}^1}^*(D)\simeq p^*(D')$.
Therefore
$j_g^*\Phi_{{\bf A}^1}^*(D)=g\cdot D\simeq
j_g^*p^*(D')=D'$. This implies $h\cdot D\simeq D$ for any
$h\in G={\rm ker}_\sigma \cdot{\operatorname{Aut}}_\nu(\widetilde{X}_{\sigma})$
(2) Let $L\subset \widetilde{X}_{\Delta^\sigma}$ denote the
complement of the set where the birational
morphism $\psi:\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\sigma}$ is an isomorphism.
It is a $G^{\sigma}$-invariant subset of
codimension 2. Therefore we get $G^{\sigma}$-equivariant
homomorphisms $\operatorname{Cl}(\widetilde{X}_{\sigma})\simeq
\operatorname{Cl}(\widetilde{X}_{\sigma}\setminus L) \hookrightarrow
\operatorname{Cl}(\widetilde{X}_{\Delta^\sigma})$. Note also that
$\operatorname{Cl}(\widetilde{X}_{\Delta^\sigma})$ is generated by
$\operatorname{Cl}(\widetilde{X}_{\sigma})$ and by the exceptional divisors of $\psi$
which are $G^{\sigma}$-invariant by Lemma \ref{le: divisors}.
Finally, all elements of
$\operatorname{Cl}(\widetilde{X}_{\Delta^\sigma})$ are $G^{\sigma}$-invariant.\qed
\bigskip
\subsection{Simple definition of a canonical subdivision of a
semicomplex}\label{se: simple}
Recall that for any fan $\Sigma$ we denote
by ${\operatorname{Vert}}(\Sigma)$ the set of all $1$-dimensional
rays in $\Sigma$.
\begin{proposition}\label{pr: simple} Let $\Sigma$ be an oriented
$\Gamma$-semicomplex. A subdivision $\Delta$ of $\Sigma$
is canonical if for any $\sigma \in \Sigma$ $${\operatorname{Vert}}(\Delta^\sigma)\setminus
{\operatorname{Vert}}(\overline{\sigma}) \subset{\rm
Stab}(\Sigma).$$
\end{proposition}
\noindent{\bf Proof.} This is an immediate
consequence of Lemma \ref{le: stable support} and the following.
\begin{proposition}\label{pr: ssimple} Let $\sigma$ be a
$\Gamma$-semicone and $\Delta^\sigma$ be a
subdivision of $\sigma$. Then
the following conditions are equivalent:
\begin{enumerate}
\item $\widetilde{X}_{\Delta^\sigma}\to \widetilde{X}_{\sigma}$ is
$G^{\sigma}$-equivariant (where $G^{\sigma}={\operatorname{Aut}}(\widetilde{X}_{\sigma})^0$).
\item ${\operatorname{Vert}}(\Delta^\sigma)\subset
{\operatorname{Vert}}(\overline{\sigma})\cup {\rm Inv}({\sigma})$.
\item $\widehat{X}_{\Delta^\sigma}\to \widehat{X}_{\sigma}$ is
${\operatorname{Aut}}(\widehat{X}_{\sigma})^0$-equivariant.
\end{enumerate}
\end{proposition}
Before the proof of Propositions \ref{pr: ssimple} and
\ref{pr: simple} we need to show a few lemmas below.
\begin{lemma} \label{le: multiplication} Let $\sigma$ be a
$\Gamma$-semicone. If $\Delta_1$ and $\Delta_2$ are
subdivisions of $\sigma$ for which
$\widetilde{X}_{\Delta_i}\to \widetilde{X}_{\sigma}$ is
$G^{\sigma}$-equivariant then for the subdivision
$$\Delta_1\cdot\Delta_2:= \{\sigma_1\cap \sigma_2\mid
\sigma_1\in \Delta_1, \sigma_2\in \Delta_2\}$$\noindent
the morphism
$\widetilde{X}_{\Delta_1\cdot\Delta_2}\to \widetilde{X}_{\sigma}$ is
$G^{\sigma}$-equivariant.
Let $I_1$ and $I_2$ be
$G^{\sigma}$-invariant on $\widetilde{X}_{\sigma}$. Let
$\Delta_1$ and $\Delta_2$ be subdivisions of $\sigma$
corresponding to the normalized blow-ups at $I_1$ and
$I_2$. Then $I_1\cdot I_2$ corresponds to
$\Delta_1\cdot\Delta_2$.
\end{lemma}
\noindent {\bf Proof.} It follows from the universal
property of the fiber product that $\Delta_1\cdot\Delta_2$ corresponds
to the normalization of an irreducible component in
$\widetilde{X}_{\Delta_1}\times_{\widetilde{X}_{\sigma}}
{\widetilde{X}_{\Delta_2}}$.
The second part of the lemma is an
immediate consequence of the relations between ideals
and ${\operatorname{ord}}$-functions: ${\operatorname{ord}}(I_1\cdot
I_2)={\operatorname{ord}}(I_1)+{\operatorname{ord}}(I_2)$ corresponds to $\Delta_1\cdot\Delta_2$(\cite{KKMS}). \qed
\begin{lemma}\label{le: vertices}
Let $\sigma$ be a $\Gamma$-semicone.
Let $\tau\subset |\sigma|$ be a cone
with all rays ${\operatorname{Vert}}(\tau)$ in ${\operatorname{Vert}}(\overline{\sigma})\cup{\rm Inv}(\sigma)
$.
\begin{enumerate}
\item There exists a fan subdivision
$\Delta_\tau$ of $\sigma$ which
contains $\tau$ as its face and for which
$\widetilde{X}_{\Delta^\tau}\to \widetilde{X}_{\tau}$ is
$G^{\sigma}$-equivariant.
\item If all rays of $\tau$ which are not in ${\rm
Inv}(\sigma)$ determine a face $\varrho$ of $|\sigma|$
(and $\tau$) then there exists
a $G^{\sigma}$-invariant ideal $I$ such that the normalized
blow-up of $I$ corresponds to the subdivision $\Delta_\tau$ of
$\sigma$ containing $\tau$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.} Write $\tau=\langle v_1,\ldots,v_l\rangle$,
where $v_1,\ldots,v_k\in {\operatorname{Inv}}(\sigma)$ and
$v_{k+1},\ldots,v_{l}$ are in ${\operatorname{Vert}}(\overline{\sigma})$.
(1) Let $\{\tau_i\mid i\in J_0\}$ denote the set of all
one-codimensional faces of $\tau$.
For any one-codimensional
face $\tau_i$ of $\tau$, where $i\in J_0$,
find an integral functional $H_i$ such that
$H_{i|\tau_i}=0$ and $H_{i|\tau\setminus\tau_i} >0$.
Find functionals $H_j$, $ j\in J_1$ with
common zeros
exactly on ${\rm lin}(\tau)$. Then $\tau=\{x\in
\sigma\mid H_i(x)\geq 0, H_j(x)=0, i\in J_0, j\in J_1\}$.
Let $F_0$ be an integral functional such that for any $i\in
J_0$ and $j\in J_1$, $F_0+H_i$ and $F_0-|H_j|$ are strictly greater than
zero on $\sigma\setminus \{0\}$.
Set $F_0(v_1)=n_1,\ldots,F_0(v_l)=n_l$.
Let $\nu_1,\ldots,\nu_l$ be
valuations corresponding to $v_1,\ldots,v_l$.
Set $$I:=
I_{\nu_1,n_1}\cap\ldots\cap I_{\nu_l,n_l}.$$ Let
$p:\widetilde{X}_{\Delta_0}\to \widetilde{X}_{\sigma}$ be the
composition of the $G^{\sigma}$-equivariant blow-ups ${\operatorname{bl}}_{\nu_k}\circ\ldots\circ {\operatorname{bl}}_{\nu_1}$
corresponding to $G^{\sigma}$-invariant valuations $\nu_1$,\ldots,$\nu_k$.
Then all valuations $\nu_1,\ldots,\nu_l$ correspond to Weil
divisors $D_1,\ldots,D_l$ on $\widetilde{X}_{\Delta_0}.$
Set $D=n_1D_1+\ldots+n_lD_l$. Then $$p_*(I_D)=\{f\in p_*(
{{\mathcal O}}_{\widetilde{X}_{\Delta_\tau}})={{\mathcal O}}_{\widetilde{X}_{\sigma}}\mid
\nu_{i}(f)\geq n_i,
i=1,\ldots,l\}= I.$$ By Lemma \ref{le: trivial action},
$G^{\sigma}$ acts
trivially on $\operatorname{Cl}(\widetilde{X}_{\Delta_0})$; then for any
$g\in G$ we have $g_*(D)=D+(f_g)-(h_g)$, where $f_g, h_g\in
{{\mathcal O}}_{\widetilde{X}_{\sigma}}$. In other words for any
$g\in G$, $$f_g\cdot I_D=h_g\cdot I_{g_*(D)},$$ which implies
$$f_g\cdot I=h_g\cdot {g_*(I)}.$$
This means that the action of $G^{\sigma}$ lifts to the blow-up of
$I$, and to its normalization. The normalized blow-up of $I$
corresponds to a $G^{\sigma}$-equivariant subdivision $\Delta_\tau$ of $\sigma$ into
maximal cones, where the piecewise linear function
${\operatorname{ord}}(I)={\rm min}\{F\in \sigma^\vee\mid F(v_i)\geq n_i\}$ is
linear. We have to show that $\tau\in \Delta_\tau$.
Note that ${\operatorname{ord}}(I)$ is linear on $\tau$ since by
definition ${\operatorname{ord}}(I)_{|\tau}=F_{0|\tau}$.
Let $x\not\in \tau$. Consider first the case $x\in{\rm lin}(\tau)$. There exists
a one-codimensional face $\tau_{i_0}$, $i_0\in J_0$, of
$\tau$ such that $H_{i_0}(x)<0$. Then $F_1:=F_0+H_{i_0} \in \sigma^\vee$
and by definition $F_1\geq
{\operatorname{ord}}(I)$, since $H_{i_0}(v_i)\geq 0$ and $F_1(v_i)=F_0(v_i)+H_{i_0}(v_i)\geq
n_i$. This implies ${\operatorname{ord}}(I)(x)\leq F_1(x)<F_0(x)$. Now let
$x\not\in {\rm lin}(\tau)$. Then there is $H_{j_0}$, $j_0\in
J_1$ such that $H_{j_0}(x)\neq 0$. Set
$F_1:=F_0+H_{j_0}$ if $H_{j_0}(x)<0$ and $F_1:=F_0-H_{j_0}$ otherwise.
The definition of $F_0$ shows that $F_1$ is strictly greater than
zero on $\sigma\setminus \{0\}$. We have $F_1(v_i)=F_0(v_i)+H_{j_0}(v_i)=
n_i$ and consequently ${\operatorname{ord}}(I)\leq F_1$. Again
${\operatorname{ord}}(I)(x)\leq F_1(x)<
F_0(x)$. This implies ${\operatorname{ord}}(I)< F_0$
off ${\rm lin}(\tau)$.
(2) The reasoning is very similar. Let $J_0$ denote the set
of all one-codimensional faces $\tau_i$ of
$\tau$ which contain $\varrho$. For any $i\in J_0$
find an integral functional $H_i$ such that
$H_{i|\tau_i}=0$ and $H_{i|\tau\setminus\tau_i} >0$.
Let $H_j$, $ j\in J_1$, be functionals with
common zeros
exactly on ${\rm lin}(\tau)$. Then $\tau=\{x\in
\sigma\mid H_i(x)\geq 0, H_j(x)=0, i\in J_0, j\in J_1\}$.
Let $F_0$ be an integral functional such that
$F_{0|\tau}=0$, and for any $i\in
I$ and $j\in J$, $F_0+H_i$, $F_0-|H_j|$ are strictly greater than
zero on $\sigma\setminus \varrho$ for $j\in J_1$.
Set $F_0(v_1)=n_1,\ldots,F_0(v_k)=n_k$. Then $F_0(v_{k+1})=F_0(v_l)=0$.
Let $\nu_1,\ldots,\nu_k$ be
valuations corresponding to $v_1,\ldots,v_k$.
Set $$I:=
I_{\nu_1,n_1}\cap\ldots\cap I_{\nu_k,n_k}.$$ By definition $I$
is $G^{\sigma}$-invariant. The normalization of the blow-up of $I$
corresponds to a $G^{\sigma}$-equivariant subdivision
$\Delta_\tau$ of $\sigma$ into
maximal cones, where the piecewise linear function
${\operatorname{ord}}(I)={\rm min}\{F\in \sigma^\vee\mid F(v_i)\geq n_i\}$ is
linear. We have to show that $\tau\in \Delta_\tau$.
Note that ${\operatorname{ord}}(I)$ is linear on $\tau$ since by
definition ${\operatorname{ord}}(I)_{|\tau}=F_{0|\tau}$.
The rest of the proof is the same as in (1).\qed
\begin{lemma} \label{le: Ssigma} Let $\sigma$ be a
$\Gamma$-semicone and $ \tau \subset
|\sigma|$ be a cone all of whose rays are in ${\operatorname{Vert}}({\overline{\sigma}})\cup
{\rm Inv}(\sigma)$. Let $\Delta_\tau$ be a
subdivision of $\sigma$ containing $\tau$ as its face
and for which
$\widetilde{X}_{\Delta_\tau}\to \widetilde{X}_{\sigma}$ is
$G^{\sigma}$-equivariant.
Then
$$S(\tau, \Delta_\tau):=\{\varrho\in \Delta_\tau \mid (\varrho \setminus
\tau) \cap {\rm Inv}(\sigma) = \emptyset \}$$
\noindent is a subfan of $\Delta_\tau$ for which the open subset
$\widetilde{X}_{S(\tau, \Delta_\tau)}={X}_{S(\tau,
\Delta_\tau)}\times_{X_\sigma}\widetilde{X}_\sigma$ of
$\widetilde{X}_{\Delta_\tau}$ is
$G^{\sigma}$-invariant.
\end{lemma}
\noindent {\bf Proof } It suffices to show that
each $G^{\sigma}$-orbit intersecting $\widetilde{X}_{S(\tau, \Delta_\tau)}:=
{X}_{S(\tau, \Delta_\tau)}\times_{{X}_\sigma}\widetilde{X}_{\sigma}$
is contained in this subscheme.
Since $\widetilde{X}_{S(\tau, \Delta_\tau)}$ is an open subset
of $\widetilde{X}_{\Delta_\tau}$
any $G^{\sigma}$-orbit contains the generic point of
$O_\varrho$ for $\varrho\in S(\tau, \Delta_\tau)$. By Lemma
\ref{le: minimal vectorss}(3), ${\operatorname{int}}(\varrho)$ intersects
${\rm Inv}(\sigma)$.
The other toric orbits in the $G^{\sigma}$-orbit
correspond to the cones $\varrho'\in {\rm Star}(\varrho,
\Sigma)$ for which $\varrho'\setminus\varrho =\emptyset$. Thus
$\varrho$ is a face of $\tau$ and all $\varrho'$ belong to $S(\tau, \Delta_\tau)$. \qed
\begin{lemma} \label{le: Ssigma2} Let $\sigma$ be a
$\Gamma$-semicone. Then
$ S(\tau):=S(\tau, \Delta)$ does not depend
upon a subdivision $\Delta$ of $\sigma$ containing $\tau$ and all of whose
rays are in ${\operatorname{Vert}}(\overline{\sigma})\cup
{\rm Inv}(\sigma)$.
\end{lemma}
\noindent {\bf Proof.} Let $\Delta_1$ and $\Delta_2$ be two
subdivisions containing $\tau$ and for
which all rays are in ${\operatorname{Vert}}(\overline{\sigma})\cup
{\rm Inv}(\sigma)$. By Lemmas
\ref{le: multiplication} and \ref{le: vertices} we may find
subdivisions
$\Delta_\varrho$, where
${\varrho\in\Delta_1 \cup \Delta_2}$ and
the common subdivision
$\Delta:=\prod_{\varrho\in\Delta_1 \cup \Delta_2}\Delta_\varrho$
such that $\widetilde{X}_{\Delta}\to
\widetilde{X}_{\sigma}$ is $G^{\sigma}$-equivariant.
Then ${\operatorname{Vert}}(\Delta_1\cdot\Delta_2)\subset
{\operatorname{Vert}}(\Delta)\subset{\operatorname{Vert}}({\sigma})\cup
{\rm Inv}(\sigma)$. Hence we can assume that
$\Delta_1$ is a subdivision of $\Delta_2$ replacing $
\Delta_1$ with $\Delta_1\cdot\Delta_2$ if necessary. Let
$\varrho\in S(\tau, \Delta_2)$. By Lemma \ref{le:
stab-subdivisions}(1) applied to $\Delta_\varrho$ we can
represent $\varrho$ as
$\varrho:=\varrho' \oplus \langle e_1,\ldots,e_k \rangle$, where
$ {\operatorname{int}}( \varrho') \cap {\rm
Inv}(\sigma)\neq \emptyset$, $(\varrho\setminus\varrho')\cap {\rm
Inv}(\sigma)= \emptyset$ and thus $\varrho'\preceq
\tau$. Since all new rays of $\Delta_{1}|\varrho$
are in ${\rm
Inv}(\sigma)$ we find that
$\Delta_{1}|\varrho=\Delta_{1}|\varrho'\oplus \langle e_1,\ldots,e_k \rangle$.
But $\Delta_{1}|\varrho'=\varrho'$ since
$\varrho'\preceq\tau$.
Therefore $\Delta_{1}|\varrho=\varrho$ and
$\varrho\in\Delta_1$, and consequently
$\varrho\in S(\tau, \Delta_1)$.\qed
\begin{lemma} \label{le: Ssimple3} Let $\sigma$ be a $\Gamma$-semicone.
Let $\Delta$ be any subdivision of $\sigma$ with all
''new '' rays belonging to ${\rm Inv}(\sigma)$. Then for
any $\tau\in\Delta$,
$\widetilde{X}_{S(\tau)}$ is $G^{\sigma}$-invariant.
\end{lemma}
\noindent{\bf Proof} By Lemma \ref{le: vertices}, we find a
subidvision $\Delta_\tau$ of $\sigma$. By Lemma \ref{le: Ssigma2},
$\widetilde{X}_{S(\tau)}=\widetilde{X}_{S(\tau,\Delta_\tau)}$.
By Lemma \ref{le: Ssigma}, the latter scheme is $G^{\sigma}$-invariant.
\qed
\bigskip
\noindent {\bf Proof of Proposition \ref{pr: ssimple}.}
$(1)\Rightarrow (2)$ All exceptional divisors determine
$G^{\sigma}$-semiinvariant and therefore $G^{\sigma}$-invariant
valuations. The latter correspond to vectors in
${\operatorname{Vert}}(\Sigma)\subset{\operatorname{Vert}}({\sigma})$.
$(1)\Leftarrow (2)$
Let $\Delta$ be any subdivision of $\sigma$ containing
$\tau$ with all
''new '' rays belonging to ${\rm Inv}(\sigma)$.
Then by Lemmas \ref{le: Ssimple3},
$\widetilde{X}_{\Delta}$ is the
union of open $G^{\sigma}$-invariant neighborhoods
$\widetilde{X}_{S(\tau)}$, where $\tau\in\Delta$, and
consequently, $\widetilde{X}_{\Delta}$ is $G^{\sigma}$-invariant.
$(2)\equiv (3)$ By Lemma \ref{le: hat} vectors defining $G^{\sigma}$-invariant
valuations are the same as those defining \\
${\operatorname{Aut}}(\widehat{X}_{\sigma})^0$-invariant valuations.
Then the proof of the equivalence is the same as for
$(1)\equiv (2)$. We have to
replace $\widetilde{X}_{\sigma}$ with $\widehat{X}_{\sigma}$.
\qed
\begin{lemma}\label{le: ideal} Let $\sigma$ be a
$\Gamma$-semicone and $G^{\sigma}$ be a connected
proalgebraic group acting on $X:=\widetilde{X}_{\sigma}$.
Let $\Delta^\sigma$ be a
subdivision of $\sigma$ such that $G^K$ acts on
$Y:=\widetilde{X}_{\Delta^\sigma}$ and the toric morphism $\psi: Y\to X$
is a proper birational $G^K$-equivariant
morphism. Then there exists a toric ideal $I$ on $X$ such
that the normalization of the blow-up of $I$,
$Z\to X$, factors as
$Z\to Y\to X$.
\end{lemma}
\noindent {\bf Proof.} First subdivide $\Delta^\sigma$ so that
all its faces satisfy the condition of Lemma \ref{le: vertices}(2).
We can make $\Delta^\sigma$ simplicial by applying
additional star subdivisions at all
its $1$-dimensional rays (see \cite{Wlodarczyk1}).
Let $\tau$ be a face of $\Delta^\sigma$ which does not
satisfy the condition of Lemma \ref{le: vertices}(2).
Denote by $\tau'$ a minimal face of $\tau$ with rays
in ${\operatorname{Vert}}(\overline{\sigma})$ and which is not a face of the cone $\sigma$. Then
there is $\sigma'$ which is a
face of $\sigma$ and for which
${\operatorname{int}}(\tau')\subset {\operatorname{int}}(\sigma')$.
Write $\sigma'=\operatorname{sing}^\Gamma(\sigma')\oplus^\Gamma\langle
e_1,\ldots,e_k\rangle$. Then $\tau'\cap \operatorname{sing}^\Gamma(\sigma')$ is a
face of $\tau'$ but it is not a face of $\sigma$.
(Otherwise $\tau'$ was a face of $\sigma$.)
By minimality $\tau'=\tau'\cap \operatorname{sing}^\Gamma(\sigma')\subset
\operatorname{sing}^\Gamma(\sigma')$ and $\sigma'=\operatorname{sing}^\Gamma(\sigma')$ is
indecomposable and since ${\operatorname{int}}(\tau')\subset
{\operatorname{int}}(\sigma')$ we can find by Lemma
\ref{le: minimal vectorss}(5) a vector
$v\in{\operatorname{int}}(\tau)\cap {\operatorname{Inv}}(\sigma)$.
By applying star subdivision at $\langle
v\rangle$ to $\Sigma$ we eliminate the cone $\tau$ (and
some others cones not satisfying the condition of Lemma
\ref{le: vertices}(2)). After
a finite number of steps we arrive at a subdivision $\Delta^\sigma$
with all faces satisfying the condition of Lemma \ref{le: vertices}(2).
Then by Lemma \ref{le: vertices}(2), for any $\tau\in\Delta$, we
construct the ideal $I_\tau$ such that the normalization of
the blow-up of $I_\tau$
corresponds to the subdivision $\Delta^\sigma_\tau$ of $\sigma$
containing $\tau$.
Then by Lemma \ref{le: multiplication} the $G^{\sigma}$-invariant
ideal $I:=\prod_{I_\tau\in
\Sigma}I_\tau$ correponds to the subdivision $\Delta_I:=\prod_{\tau\in
\Sigma}\Delta^\sigma_\tau$ of $\sigma$. \qed
\begin{lemma}\label{le: extendsub}
Let $\sigma$ be a face of a fan $\Sigma$. Let $\Delta_1$ be
a subdivision of $\sigma$ such that ${\operatorname{Vert}}(\Delta_1)={\operatorname{Vert}}(\sigma)$. Then
there is a subdiision $\Delta_2$ of $\Sigma$ such that
$\Delta_2|\sigma=\Delta_1$ and ${\operatorname{Vert}}(\Delta_2)={\operatorname{Vert}}(\Sigma)$.
\end{lemma}
\noindent{\bf Proof.}
Set $\{v_1,\ldots,v_k\}={\operatorname{Vert}}(\Sigma)\setminus
{\operatorname{Vert}}(\sigma)$, $\Delta_0:=\langle v_1 \rangle\cdots
\langle v_k \rangle\cdot\Sigma$. Then $\sigma\in
\Delta_0$. We shall construct
a subdivision $\Delta_2$ for any cone in $\Delta_0$. If
$\tau\in\Delta_0$ does not intersect $\sigma$ then we put
$\Delta_2|\tau=\tau$.
If $\tau\in\Delta_0$ intersects $\sigma$ along a common face
$\sigma_0$ then all rays ${\operatorname{Vert}}(\tau\setminus \sigma_0)$ are
centers of star subdivisions and therefore are
linearly independent of the other rays of $\sigma$
and in particular of $\operatorname{lin}(\sigma_0)$ (After a star
subdivision at $\langle v \rangle$, $v$ becomes linearly
independent of all other rays of cones containing $v$. Hence the
subdivision $\Delta_2|\sigma_0$ extends to a unique
subdivision of $\tau$ with no new rays,
$\Delta_2|\tau:=\{\delta+\sum_{\varrho\in({\operatorname{Vert}}(\tau)\setminus
{\operatorname{Vert}}(\sigma_0))}\varrho \mid \delta\in \Delta|\sigma_0\}$.\qed
\begin{lemma} \label{le: subdivision} Let $\tau\subset\sigma$ be a cone such that
${\operatorname{Vert}}(\tau)\subset{\operatorname{Vert}}(\sigma)$. Then there is a proper
subdivision $\Delta$ of $\sigma$ containing $\tau$ and such
that ${\operatorname{Vert}}(\Delta)={\operatorname{Vert}}(\sigma)$.
\end{lemma}
\noindent{\bf Proof.} By Lemma \ref{le: extendsub}
it suffices to prove the lemma for the situation when
${\operatorname{Vert}}(\sigma)\setminus {\operatorname{Vert}}(\tau)$ consists of one ray.
Then for any $\tau$ we find $\sigma'$ containing $\tau$ and
for which ${\operatorname{card}}({\operatorname{Vert}}(\sigma)\setminus {\operatorname{Vert}}(\tau))=1$. By induction we can
find a subdivision of $\sigma'$ and then by Lemma \ref{le:
extendsub} extend it to a subdivision of $\sigma$.
Let $\tau=\langle v_1,\ldots,v_k\rangle$ and
$\sigma=\langle v_1,\ldots,v_k,v_{k+1}\rangle$.
If $\tau$ is simplicial and $\tau$ is not a face of
$\sigma$ then there is a unique
linear relation between $v_1,\ldots,v_k,v_{k+1}$.
Without loss of generality we can assume that it is
$a_rv_r+a_{r+1}v_{r+1}+\ldots+a_lv_l=a_{l+1}v_{l+1}+\ldots+a_{k+1}v_{k+1}$,
where $r\geq 1$ and all coefficients are positive. Then
there are exactly two
simplicial subdivisions of
$\sigma$ with no
''new'' rays and with maximal cones $\langle
v_1,\ldots,\check{v}_i,\ldots,v_k,v_{k+1}\rangle$, where
$i=r,\ldots l$ and $i=l+1,\ldots,k+1$ (see
\cite{Wlodarczyk1}). The second subdivision
contains $\tau$.
If $\tau=\langle v_1,\ldots,v_k\rangle$ is not simplicial
then for the cone $\tau^\vee$ we
find a simplicial cone $(\tau_0^\vee,M_{\tau_0})$ and a linear epimorphism
$\phi^\vee: (\tau_0^\vee,M_{\tau_0}\to(\tau^\vee,M_\tau)$
mapping rays of $\tau_0^\vee$ to
rays of $\tau_0^\vee$. Moreover let $H^\vee\subset
M_{\tau_0}$ denote the vector
subspace which is the kernel of
$\phi^\vee$. Then $\tau^\vee\simeq(\tau_0^\vee+H^\vee)/H^\vee$.
The dual morphism $\phi: (\tau,N_\tau) \to
(\tau_0,N_{\tau_0})$ is a monomorphism which maps
$\tau$ isomorphically to $\tau_0\cap H$, where $H=\{v\in N_{\tau_0}\mid v_{|H^\vee}=0\}$.
Then also $v_{k+1}\in H\setminus \tau_0$. Note that $v_{k+1}\not\in -\tau_0$
since otherwise $v_{k+1}\in (-\tau_0)\cap H=-\tau$. Thus
we can apply the previous case to
$\sigma_0:=\langle v_{k+1}\rangle + \tau_0\supset \tau_0$ and obtain a
subdivision $\Delta_0$ of $\sigma_0$ containing
$\tau_0$ with no ''new '' rays. Intersecting $\Delta_0$
with $H$ defines a subdivision $\Delta$ of $\sigma$ containing
$\tau$. All rays in $\Sigma$ except $v_{k+1}\in H$ are
obtained by intersecting some faces of $\tau_0$ with $H$,
hence belong to ${\operatorname{Vert}}(\tau)$.\qed
\section{Orientability of stratified toroidal varieties
and resolution of singularities}
\bigskip
\subsection{Orientation group of an affine toric variety}
\begin{definition}\label{de: orientation group}
Let $(X,S)$ be a stratified noetherian scheme over $K$ and
$x\in X$ be a $K$-rational point fixed under the
$\Gamma$-action. By the {\it orientation group} of $(X,S)$ at $x\in
X$ we mean
$\Theta^\Gamma(X,S,x):={\operatorname{Aut}}(\widehat{X}_{X,x})/{\operatorname{Aut}}(\widehat{X}_{X,x})^0$.
\end{definition}
\begin{lemma}\label{le: orientation group}
Let $\sigma$ be a $\Gamma$-semicone. Let $\widehat{{\operatorname{Inv}}}(\sigma)={\operatorname{Inv}}(\sigma)$
be the
set of the vectors $v\in\sigma$ corresponding to
${\operatorname{Aut}}(\widehat{X}_{\sigma})^0$-invariant valuations ${\operatorname{val}}(v)$ on
$\widehat{X}_{\sigma}$ (see Lemmma \ref{le: hat}).
Let
${\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}$ be the group of all elements $g$ of
${\operatorname{Aut}}(\sigma)$ satisfying the following conditions:
\begin{enumerate}
\item $g_{|{\operatorname{Inv}}(\sigma)}={\operatorname{id}}_{|{\operatorname{Inv}}(\sigma)}$.
\item For any subdivision $\Delta$ of $\sigma$ such that
${\operatorname{Vert}}(\Delta)={\operatorname{Vert}}(\overline{\sigma})$, $g$ lifts to an automorphism
of $\Delta$.
\end{enumerate}
Then
${\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}=Aut(\sigma)^0$ and
$\Theta^\Gamma(X_{\sigma},S,O_{\sigma})={\operatorname{Aut}}(\sigma)/{\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}$.
\end{lemma}
\noindent{\bf Proof.} By Lemma \ref{le: surjection} there is
an epimomorphism ${\operatorname{Aut}}(\sigma)\longrightarrow
\Theta^\Gamma(X_{\sigma},S,O_{\sigma})$ with kernel ${\operatorname{Aut}}(\sigma)^0$.
By definition all $g\in
{\operatorname{Aut}}(\sigma)^0$ preserve ${\operatorname{Inv}}(\sigma)$. By
Proposition \ref{pr: simple} the subdivisions $\Delta$ satisfying
condition (2) of the lemma determine
${\operatorname{Aut}}(\widehat{X}_{\sigma})^0$-equivariant
morphisms $\widehat{X}_{\Delta}\to\widehat{X}_{\sigma}$ which
commute with morphisms defined by $g\in {\operatorname{Aut}}(\sigma)^0$. Thus
automorphisms $g\in {\operatorname{Aut}}(\sigma)^0$ lift to automorphisms of $\Delta$.
Therefore ${\operatorname{Aut}}(\sigma)^0\subseteq
{\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}$. We need to prove that ${\operatorname{Aut}}(\sigma)^0\supseteq
{\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}$.
Let $v_1,\ldots,v_k$ denote the generators of $\sigma$. Let
$\sigma_0$ denote the regular cone spanned by the standard
basis $e_1,\ldots,e_k$ of some $k$-dimensional lattice.
Let $\pi:\sigma_0\to\sigma$ be the projection defined by
$\pi(e_i)=v_i$. Each automorphism of $\sigma$ lifts to a unique
automorphism of $\sigma_0$. This defines a group
homomorphism $\pi^*:{\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}\to {\operatorname{Aut}}(\sigma_{0})$.
Let ${\operatorname{Aut}}(\sigma_{0})_{\operatorname{Inv}}$ denote the group of automorphisms of
the cone $\sigma_{0}$ preserving $\pi^{-1}({{\operatorname{Inv}}(\sigma)})$.
\begin{lemma}\label{le: sigma}
${\operatorname{Aut}}(\sigma_{0})_{\operatorname{Inv}}\simeq {\operatorname{Aut}}(\sigma)_{\operatorname{Inv}}$.
\end{lemma}
\noindent{\bf Proof.}
Any automorphism $g\in {\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}$ preserves all vectors from
${\operatorname{Inv}}(\sigma)$ and consequently it preserves the vectors
from $\overline{{\operatorname{Inv}}(\sigma)}:={\rm
lin}({\operatorname{Inv}}(\sigma))\cap\sigma=\operatorname{lin}(\overline{{\operatorname{Inv}}(\sigma)})\cap
\sigma$.
Set
$\overline{{\operatorname{Inv}}(\sigma_0)}:=\pi^{-1}(\overline{{\operatorname{Inv}}(\sigma)})=
\langle w_1,\ldots, w_m \rangle$. Then, by definition
$\overline{{\operatorname{Inv}}(\sigma_0)}=\operatorname{lin}(\overline{{\operatorname{Inv}}(\sigma_0)})\cap
\sigma_0$.
Let $g':=\pi^*(g)$. We shall show that
$g'$ preserves $\overline{{\operatorname{Inv}}(\sigma_0)}$.
Let $w_i\in{\operatorname{int}}(\tau_i)$, where
$\tau_i=\langle e_{ij_1}, \ldots, e_{ij_l}\rangle\preceq
\sigma_0$. Then , by definition, $w_i$ defines a unique ray
of the cone $\tau_i\cap
\overline{{\operatorname{Inv}}(\sigma_0)}=\tau_i\cap
\operatorname{lin}(\overline{{\operatorname{Inv}}(\sigma_0)})$. Then
$\ker(\pi)\subset \operatorname{lin}(\overline{{\operatorname{Inv}}(\sigma_0)})$ and
$\ker(\pi)\cap
\operatorname{lin}(\tau_i)=\ker(\pi)\cap\operatorname{lin}(\overline{{\operatorname{Inv}}(\sigma_0)})\cap
\operatorname{lin}(\tau_i)=\ker(\pi)\cap\operatorname{lin}\{w_i\}=0$. Consequently
$\pi$ maps $\tau_i$ isomorphically onto a cone
$\pi(\tau_i)=\langle v_{ij_1}, \ldots, v_{ij_l}\rangle$
containing a unique ray $\pi(w_i)\in\overline{{\operatorname{Inv}}(\sigma)}$.
By Lemma \ref{le: subdivision} such a cone is a face of some
subdivision $\Delta$ of $\sigma$ satisfying condition (2)
of the lemma. Thus
$\pi(\tau_i)$ is a unique face of $\Delta$ containing
$\pi(w_i)$ in its interior. Therefore it is preserved by
$g$. Consequently, $g'$ preserves
$\tau_i$ and $w_i$, and hence all points from
$\overline{{\operatorname{Inv}}(\sigma_0)}$.
Now, let $g'$ be an automorphism of
$\sigma_0$ preserving $\overline{{\operatorname{Inv}}(\sigma_0)}$. Then
$g'$ defines a permutation $g$ of $v_1,\ldots,v_k$.
We will show that such a
permutation preserves linear relations between
$v_1,\ldots,v_k$.
Let $p:=a_{i_1}v_{i_1}+\ldots +a_{i_r}v_{i_r}=
a_{i_{r+1}}v_{i_{r+1}}+\ldots +a_{is}v_{i_s}$ be a minimal
linear relation, i.e. $s-1=\dim\{v_{i_1}, \ldots, v_{i_s}\}$.
Then both cones $\sigma_1:=\langle v_{i_1}, \ldots,
v_{i_r}\rangle $ and $\sigma_2:=\langle v_{i_{r+1}}, \ldots,
v_{i_\sigma}\rangle $ are simplicial and by Lemma \ref{le: subdivision}
they are faces of two
subdivisions $\Delta_{1}$ and $\Delta_{2}$ of $\sigma$
satisfying the condition 2 of the lemma. Then
$\langle p \rangle \in {\operatorname{Vert}}(\Delta_1\cdot\Delta_2)$ and
by Lemma \ref{le: multiplication} and Proposition
\ref{pr: ssimple}, $p
\in {{\operatorname{Inv}}(\sigma)}$. Consequently
the point $p':=a_{i_1}e_{i_1}+\ldots +a_{i_r}e_{i_r}\in
\overline{{\operatorname{Inv}}(\sigma_0)}$. This implies
$p'=g'(p')=a_{i_1}g'(e_{i_1})+\ldots +a_{i_r}g'(e_{i_r})$ and
$p=a_{i_1}v_{i_1}+\ldots +a_{i_r}v_{i_r}=a_{i_1}g(v_{i_1})+\ldots
+a_{i_r}g(v_{i_r})$. Analogously $p=a_{i_{r+1}}g(v_{i_{r+1}})+
\ldots +a_{is}g(v_{i_s})$. Finally, $a_{i_1}g(v_{i_1})+\ldots
+a_{i_r}g(v_{i_r})=a_{i_{r+1}}g(v_{i_{r+1}})+
\ldots +a_{is}g(v_{i_s})$ and $g$ defines a linear
automorphism preserving vectors from
$\overline{{\operatorname{Inv}}(\sigma)}$.
We need to show that $g$ satisfies condition (2).
Let $\Delta$ be a
subdivision satisfying (2). If $\tau$ is any face
of $\Delta$ whose relative interior intersects
${{\operatorname{Inv}}(\sigma)}\subset \overline{{\operatorname{Inv}}(\sigma)}$
then it lifts to a face $\tau'$ of $\sigma_0$ whose relative interior
intersect $\overline{{\operatorname{Inv}}(\sigma_0)}$. Thus $\sigma'$ is
preserved by $g'$ and $\sigma$ is
preserved by $g$. Therefore all
faces of $\Delta$ whose relative interior intersect
${{\operatorname{Inv}}(\sigma)}$ are preserved. If the relative interior of $\tau\in
\Delta$ does not intersect ${{\operatorname{Inv}}(\sigma)}$ then $\tau\in
S(\sigma')\subset \Delta$, where $\sigma'\in \Delta$, is the
maximal face of $\tau$ whose relative interior intersects
${{\operatorname{Inv}}(\sigma)}$ (see Lemma \ref{le: Ssigma}). By the above,
$\sigma'$ is preserved by $g$. Thus by
\ref{le: Ssigma2} $g(\tau)\in
S(\sigma',g(\Delta))=S(\sigma',\Delta)=S(\sigma') $. Consequently, all
faces of $\Delta$ are mapped to faces of $\Delta$ and
finally $g$ defines
an automorphism of $\Delta$. Lemma \ref{le: sigma} is
proven. \qed
Write $X_{\sigma}=X_{\sigma_0}//T_0$ for the torus
$T_0$ corresponding to the kernel of $\pi$. Also
$\widehat{X}_{\sigma}=\widehat{X}_{\sigma_0}//T_0$ and
$K[\widehat{X}_{\sigma}]=K[\widehat{X}_{\sigma_0}]^{T_0}$.
Let ${{\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})}_{{\operatorname{Inv}}}$ denote the
group of all $T_0$-{equivariant} automorphisms $g$
in ${\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})$ which preserve invariant
valuations, i.e $g_*({\operatorname{val}}(v))={\operatorname{val}}(v)$
for
any $ v\in\overline{{\operatorname{Inv}}(\sigma_0)}$. By definition
each automorphism $g\in
{{\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})}_{{\operatorname{Inv}}}$ defines an
automorphism of $K[\widehat{X}_{\sigma_0}]^{T_0}$. This
gives the homomorphism
$p: {{\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})}_{{\operatorname{Inv}}}\longrightarrow
{\operatorname{Aut}}(\widehat{X}_{\sigma})$.
It now suffices to prove the following Lemma:
\begin{lemma}\label{le: connect}
${\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})_{{\operatorname{Inv}}}$ is
connected.
\end{lemma}
Consequently, ${\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}\subset
p({\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})_{{\operatorname{Inv}}})\subset
{\operatorname{Aut}}(\widehat{X}_{\sigma})^0$ and
${\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}\subset{\operatorname{Aut}}(\sigma)^0$. \qed
\noindent {\bf Proof of Lemma \ref{le: connect}}.
The same reasoning as in Lemma \ref{le: smooth}. Let
$x_1,\ldots, x_k$ be the standard coordinates on $X_{\sigma_0}={\bf A}^k$.
Any automorphism $g\in
{\operatorname{Aut}}(\widehat{X}_{\sigma_{0}})_{{\operatorname{Inv}}}$ is given by
$\Gamma$-semiinvariant functions
$g^*(x_1),\ldots,g^*(x_k)$ with
the corresponding $\Gamma$-weights and such that
${\operatorname{val}}(v)(g^*(x_i))={\operatorname{val}}(v)((x_i))$, where
$v\in \overline{{\operatorname{Inv}}}({\sigma_{0}})\cap N_{\sigma_0}$. There is a
birational map $\alpha: {\bf A}^1\to G$ defined by
$$\alpha(z):=(x_1,\ldots,x_k)\mapsto
((1-z)g^*(x_1)+zx_1,\ldots,(1-z)g^*(x_k)+zx_k).$$
$\alpha(z)$
defines a $\Gamma$-equivarinat automorphism of
$\widehat{X}_{\sigma_{0}}$ for all $z$ in the open subset $U$ of ${\bf A}^1$,
where the linear parts of coordinates of $\alpha(z)$ are
linearly independent. Note that for any integral vector
$v\in \overline{{\operatorname{Inv}}}({\sigma_{0}})$ and $z\in U$,
${\operatorname{val}}(v)((1-z)g^*(x_i)+zx_i)\geq {\operatorname{val}}(v)(x_i)$. By Lemma
\ref{le: monomial2}, $\alpha(z)$ preserves ${\operatorname{val}}(v)$.
\qed
\begin{example}\label{ex: isolated2} Let $X_\sigma\subset
{\bf A}^4$ be a toric variety described by $x_1x_2=x_3x_4$. Then
$\sigma=\langle v_1,v_2,v_3,v_4 \rangle\subset N^{\bf Q}_\sigma$ is a cone
over a
square, where $N^{\bf Q}_\sigma:=\{(x_1,x_2,x_3,x_4)\mid
x_1+x_2-x_3-x_4\}\subset {\bf Q}^4$, $v_1=(1,0,0,1)$, $v_2=(0,1,1,0)$,
$v_3=(1,0,1,0)$ $v_4=(0,1,0,1)$.
Then ${\operatorname{Aut}}(\sigma)={\bf D}_8$ consists of all
isometries
of $\sigma$. ${\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}$ consists of all isometries preserving
diagonals: the reflections with respect to the diagonals and
rotation through $\pi$. Consequently
$$\Theta(X_\sigma,S,O_\sigma)={\operatorname{Aut}}(\sigma)/{\operatorname{Aut}}(\sigma)_{{\operatorname{Inv}}}
\simeq {\bf Z}_2.$$
\end{example}
\bigskip
\subsection{Stratified toroidal varieties are orientible
}\label{se: existence}
\begin{lemma}\label{le: orientation0}
Let
$\phi_i: X \to X_\sigma$ for $i=1,2$ be
$\Gamma$-smooth morphisms of a $\Gamma$-stratified toroidal
scheme $(X,S)$ to a
$\Gamma$-stratified toric variety $(X_\sigma,S_\sigma)$
such that strata of $S$ are precisely the inverse images of
strata of $S_\sigma$.
Denote by $\Delta_1,\ldots,\Delta_l$ all
subdivisions of $\sigma$ for which
${\operatorname{Vert}}(\Delta_i)={\operatorname{Vert}}(\overline{\sigma})$. Let $v_1,\ldots,v_r$ be stable
vectors such that $\operatorname{lin}(v_1,\ldots,v_r)\supseteq
{\operatorname{Inv}}(\sigma)$. Denote by $\Delta_{l+1},\ldots,\Delta_{l+r}$
the star subdivisions of
$\sigma$ at
$\langle v_1 \rangle,\ldots, \langle v_r \rangle$.
The following conditions are equivalent:
\begin{enumerate}
\item $\phi_1$ and $\phi_2$ determine the same orientation at
a $K$-rational point $x\in s=\phi_i^{-1}({\operatorname{strat}})\sigma))$.
\item
There is an open
neighborhood $U$ of $x$ such that
$U_{j1}:=U\times_{X_\sigma}X_{\Delta_j}$, defined via
$\phi_1$, and $U_{j2}:=U\times_{X_\sigma}X_{\Delta_j}$, defined via
$\phi_2$, are isomorphic over $U$ for any $j=1,\ldots,{l+r}$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.}
$(\Rightarrow)$ Suppose $\phi_1$ and $\phi_2$ determine the same
orientation at $x$. By Proposition \ref{pr: simple} the
morphism $\widetilde{X}_{\Delta_j}\to \widetilde{X}_\sigma$
is $G^\sigma$-equivariant. Therefore by Lemma \ref{le: isomorphisms2} for
any subdivision $\Delta_j$ of $\sigma$ there is an open
neighborhood $U_j$ of $x$ such that
$U'_{j1}:=U_j\times_{X_\sigma}X_{\Delta_j}$, defined via
$\phi_1$, and $U'_{j2}:=U_j\times_{X_\sigma}X_{\Delta_j}$, defined via
$\phi_2$, are isomorphic over
$U_j$. It suffices to put $U=\bigcap U_j$.
$(\Leftarrow)$
Suppose $\phi_1$ and $\phi_2$ do not determine the same
orientation at $x$.
Then by Lemma \ref{le: orientation group}
we find an automorphism $\phi$ of $\sigma$ such
that for the corresponding automorphism
$\overline{\phi}\in{\operatorname{Aut}}(X_\sigma)$, $\phi_1$ and
$\overline{\phi}\phi_2$ determine the same
orientation at $x$. By the above there is an open $U\ni x$ for which
$U_{j,\phi_1}$ and $U_{j,\phi\phi_2}$ are isomorphic.
But $\overline{\phi}$ does not preserve orientation so by
\ref{le: orientation group}, $\phi$ does not satisfy one of
the conditions (1)-(2)
of the lemma and does not
lift to an automorphism of some
$\Delta_j\neq \phi(\Delta_j)$. Consequently,
$U_{j,\phi\phi_2}\simeq U_{j,\phi_1}\simeq U\times_{X_\sigma}X_{\Delta_j}$ is
not isomorphic to $U_{j,\phi_2}\simeq U\times_{X_\sigma}X_{\phi(\Delta_j)}$ for any open neighborhood
$U$ of $x$. \qed
\begin{lemma}\label{le: orientation1}
Let $(X_\Delta,S_\Omega)$ be a $\Gamma$-stratified toric
variety corresponding
to an embedded semifan $\Omega\subset\Delta$.
Let
$\phi_i: X \to X_\Delta$ for $i=1,2$ be
$\Gamma$-smooth morphisms of a $\Gamma$-stratified toroidal
scheme $(X,S)$ to a $\Gamma$-stratified toric
variety $(X_\Delta,S_\Omega)$ such that the strata of $S$
are precisely the inverse images of strata of $S_\Omega$.
For a cone $\omega\in\Omega$ denote by
${\omega}(\Delta)$ the subset $\{\tau\in \Delta\mid
\omega(\tau)=\omega\}$ of $\Delta$.
Let $\Delta_1,\ldots,\Delta_l$ be all
subdivisions of $\omega$
satisfying condition (2) of Lemma
\ref{le: orientation group}. Let $\Delta_{l+1},\ldots,\Delta_{l+r}$
denote the star subdivisions of
$\omega$ at stable vectors
$\langle v_1 \rangle,\ldots, \langle v_r \rangle$ for which
$\operatorname{lin}(v_1,\ldots,v_r)\supseteq
{\operatorname{Inv}}(\sigma)$.
Denote by ${\Delta_j}(\Delta)$ the subdivisions of
${\omega}(\Delta)$ induced by ${\Delta_j}$, i.e. for
any $\delta=\omega(\delta)\oplus r(\delta)\in
{\omega}(\Delta)$,
${\Delta_j}(\Delta)|\delta=\Delta_j\oplus r(\delta)$.
Let $x\in \phi_i^{-1}({\operatorname{strat}}(\omega))$
be a $K$-rational point.
The following conditions are equivalent:
\begin{enumerate}
\item $\phi^*_1({\mathcal U}(\Delta,\Omega))$ and
$\phi^*_2({\mathcal U}(\Delta,\Omega))$ are compatible at $x$.
\item
There is an open
neighborhood $U$ of $x$ such that
$U_{j1}:=U\times_{X_{{\omega}(\Delta)}}X_{{\Delta_j}(\Delta)}$, defined via
$\phi_1$, and $U_{j2}:U\times_{X_{{\omega}(\Delta)}}X_{{\Delta_j}(\Delta)}
$, defined via
$\phi_2$,
are isomorphic over $U$ for any $j=1,\ldots,{l+r}$
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.} Let $\phi_i(x)\in\delta_i$, where
$\delta_i\in\Delta$ and $i=1,2$. Denote by $U_i$ the
inverse image $\phi^{-1}_i(X_{\delta_i})$. Since
$x\in \phi_i^{-1}({\operatorname{strat}}(\sigma))$, we have $\omega{(\delta_i)}=\omega$.
The atlases $\phi^*_i({\mathcal U}^{\operatorname{\rm can}}(X_\Delta,S_\Delta))$
are compatible at $x$ if the morphisms
$\pi_\sigma^{\sigma_i}\phi_i: U_i \to X_\omega$ determine the same
orientation at $x$. This is equivalent by Lemma \ref{le:
orientation0} to the fact
that $U\times_{X_{\sigma}}X_{\Delta_j}$, defined via
$\pi^\sigma_{\sigma_i}\phi_{i}$, are isomorphic for some
$U$. The latter schemes are isomorphic to
$U\times_{X_{{\sigma}_i}}X_{{\Delta_j}(\Delta)|{\delta_i}}=
U\times_{X_{{\omega}(\Delta)}}X_{{\Delta_j}(\Delta)}$.\qed
\begin{lemma}\label{le: orientation2}
Let
$\phi_i: (X,S) \to (X_\Delta,S_{\Omega})$ for $i=1,2$ be
$\Gamma$-smooth morphisms of a $\Gamma$-stratified toroidal scheme to a
$\Gamma$-stratified toric variety $(X_\Delta,S_{\Omega})$
such that the strata of $S$ are precisely the inverse images
of strata of $S_\Omega$.
Assume that a stratum $s\in S$ is a variety over $K$ and
$\phi^*_1({\mathcal U}(\Delta,\Omega))$ and
$\phi^*_2({\mathcal U}(\Delta,\Omega))$ are compatible at some
point $x\in s$. Then they are compatible along $s$.
\end{lemma}
\noindent {\bf Proof.} By Lemma \ref{le:
orientation1}, $\phi^*_1({\mathcal U}(\Delta,\Omega))$ and
$\phi^*_2({\mathcal U}(\Delta,\Omega))$ are compatible at any
$x\in s$ iff they are compatible at all points of $s\cap
U_x$, where $U_x$
is an open neighborhood of $x$. Since $s$ is a variety the
sets $s\cap
U_x$ and $s\cap U_y$ intersect for any $x,y\in s$.\qed
\begin{lemma} \label{le: openorient}
Let
$\phi_1, \phi_2: X \to X_{\sigma}$ be two
$\Gamma$-smooth morphisms of a $\Gamma$-stratified toroidal
variety $(X,S)$ into the $\Gamma$-stratified toric variety $X_{\sigma}$, such that the strata of $S$
are precisely the inverse images of strata of $S_\sigma$.
Assume that $\phi_1, \phi_2$
determine the same orientation at a point $x\in \phi_i^{-1}({\operatorname{strat}}(\sigma))$. Then $\phi^*_i({\mathcal U}^{\operatorname{\rm can}}(X_{\sigma},S_{\sigma}))$ and
$\phi^*_2({\mathcal U}^{\operatorname{\rm can}}(X_{\sigma},S_{\sigma}))$ are compatible on $X$.
\end{lemma}
\noindent{\bf Proof.} Let
$\Delta_{1},\ldots,\Delta_{l},\ldots,\Delta_{l+r}$
be subdivisions of $\tau\leq \sigma$ as in
Lemma \ref{le: orientation0}. By Lemma \ref{le: extendsub},
the subdivisions $\Delta_{1},\ldots,\Delta_{l}$ can be
extended to subdivisions ${\Delta}_{1}(\sigma),\ldots,{\Delta}_{l}(\sigma)$
of $\sigma$ for which ${\operatorname{Vert}}(\Delta_i)={\operatorname{Vert}}(\overline{\sigma})$.
By Lemma
\ref{le: stableincl} ${\operatorname{Inv}}(\tau)\subset {\operatorname{Inv}}(\sigma)$ and
we can extend the star subdivisions
$\Delta_{l+1},\ldots,\Delta_{l+r}$ to the star
subdivisions $\Delta_{l+1}(\sigma),\ldots,\Delta_{l+r}(\sigma)$ of
$\sigma$ at $v_1,\ldots,v_r\in
{\operatorname{Inv}}(\tau)\subset {\operatorname{Inv}}(\sigma)$.
By Lemma \ref{le: isomorphisms2} there is an open
neighborhood $U$ of $x$
for which $U_{j,\phi_i}=U\times_{X_\sigma}X_{\Delta_j}$
definded for $\phi_1$ and $\phi_2$ are pairwise isomorphic for any $j$.
Let ${\tau}(\sigma)=\{\varrho\preceq \sigma\mid \omega(\varrho)=\tau\}$.
Then there is an open subset
$U'=\phi_1^{-1}(X_{{\tau}(\sigma)})\cap
\phi_2^{-1}(X_{{\tau}(\sigma)})\cap U $ intersecting
$\phi_i^{-1}({\operatorname{strat}}(\tau))$ such that
$U'\times_{X_\sigma}X_{\Delta_j}=
U'\times_{X_{{\tau(\sigma)}}} X_{{\Delta_j(\sigma)}|{{\tau(\sigma)}}}$
defined by the restrictions of $\phi_{i}$ are isomorphic.
This implies by Lemma \ref{le: orientation1} that
$\phi^*_1({\mathcal U}^{\operatorname{\rm can}}(X_{\sigma},S_{\sigma})$ and
$\phi^*_2({\mathcal U}^{\operatorname{\rm can}}(X_{\sigma},S_{\sigma}))$ are compatible
at $x\in {\operatorname{strat}}(\tau) \cap U'$ and
consequently by Lemma \ref{le: orientation2}
they are compatible at all the points of $\phi_i^{-1}({\operatorname{strat}}(\tau))$.\qed
\begin{lemma} \label{le: orientation4} The $\Gamma$-semicomplex
$\Sigma$
associated to an oriented
$\Gamma$-stratified toroidal variety $(X,S)$
is oriented.
\end{lemma}
\noindent{\bf Proof.}
For any $\sigma\leq \tau$, the inclusion morphism
$\imath^\tau_{\sigma}: \sigma\to \tau$
identifies $\sigma$ with a face of $\tau$. Let
$\overline{\pi}^\sigma_{\tau}: N_\tau\to N_\sigma$ denote a
projection such that $\overline{\pi}^\sigma_{\tau}\circ \imath^\tau_{\sigma}=
{\operatorname{id}}_{|\sigma}$. Let ${\pi}^\sigma_{\tau}: X_\sigma\to
X_\tau $ denote the
induced toric morphism. Consider faces $\sigma\leq \tau
\leq \varrho$ of the semicomplex $\Sigma$. Let
$\phi_{\sigma}: U_\sigma \to X_\sigma, \phi_\tau: U_\tau
\to X_\tau, \phi_\varrho: U_\varrho \to X_\varrho$ denote
any charts on $(X,S)$ corresponding to the cones $\sigma,
\tau$ and $\varrho$. Since $\tau\leq \varrho$, the open subset
$U_\varrho$ intersects $\tau$. For a generic point $x_\tau$
of ${\operatorname{strat}}(\tau)\cap U_\varrho $, the morphism
$\phi_\varrho$ sends $x_\tau$ to the toric orbit $O_\tau\subset X_\varrho$.
Since $(X,S)$ is oriented we find that
$\pi^\tau_{\varrho}\phi_{\varrho}: U^\tau_\varrho \to X_\tau$
and $\phi_{\tau}: U_\tau \to X_\tau$
determine the same orientation at $x_\tau$. Analogously
$\pi^\sigma_{\varrho}\phi_{\varrho}$, $\phi_{\sigma}$ and
$\pi^\sigma_{\tau}\phi_{\tau}$ determine the same orientation
for a generic point $x_\sigma$ of ${\operatorname{strat}}(\sigma)$.
Applying Lemma \ref{le: openorient} to $\pi^\tau_{\varrho}\phi_{\varrho}$ and $\phi_{\tau}$ we see that
$\pi^\sigma_{\tau}\pi^\tau_{\varrho}\phi_{\varrho}$ and
$\pi^\sigma_{\tau}\phi_{\tau}$ determine the same orientation for
a generic point of ${\operatorname{strat}}(\sigma)$. Finally
$\pi^\sigma_\tau\pi^\tau_\varrho\phi_{\varrho}$ and
$\pi^\sigma_\varrho\phi_{\varrho}$ determine the same orientation and
thus so do $\pi^\sigma_\tau\pi^\tau_\varrho$ and
$\pi^\sigma_\varrho$.
Write $\imath^\varrho_\tau \imath^\tau_\sigma=\imath^\varrho_\sigma
\alpha_{\sigma}$,
where $\alpha_{\sigma}\in {\rm Aut}(\sigma)$.
Then $\alpha_{\sigma}{\pi}^\sigma_\tau{\pi}^\tau_\varrho
\imath^\varrho_\sigma=
\alpha_{\sigma}{\pi}^\sigma_\tau{\pi}^\tau_\varrho
\imath^\varrho_\tau \imath^\tau_\sigma\alpha_{\sigma}^{-1}=
{\operatorname{id}}_{|\sigma}$.
By Lemma \ref{le: sections} the toric morphisms induced by
the projections
$\alpha_{\sigma}{\pi}^\sigma_\tau{\pi}^\tau_\varrho$ and
${\pi}^\sigma_\varrho$ determine the same orientation. Thus the
morphisms induced by
$\alpha_{\sigma}{\pi}^\sigma_\tau{\pi}^\tau_\varrho$ and
${\pi}^\sigma_\tau{\pi}^\tau_\varrho$ determine the same
orientation, which gives $\alpha_{\sigma}\in {\rm Aut}(\sigma)^0$.
\qed
\bigskip
\begin{lemma} \label{le: existence}
On any $\Gamma$-stratified toroidal
variety $(X,S)$ there exists an atlas ${{\mathcal U}}$ such that
$(X,S,{{\mathcal U}})$ is oriented.
\end{lemma}
\noindent{\bf Proof.} We shall improve the charts from
${{\mathcal U}}$ and inclusion maps in the associated $\Gamma$-semicomplex
$\Sigma$.
For any $\sigma \in \Sigma$ fix a chart $\phi_{\sigma,a_\sigma}:
U_{\sigma,a_\sigma}\to X_{\sigma}$.
By Lemma \ref{le: surjection} for any chart $\phi_{\sigma,a}:U_{\sigma,a}\to X_{\sigma}$
we can find $\psi_{\sigma,a}$ of $X_{\sigma})$ induced by
an automorphism of $\sigma$
such
that $\phi_{\sigma,a_\sigma}$ and
$$\overline{\phi}_{\sigma,a}:=\psi_{\sigma,a}\circ\phi_{\sigma,a}$$
determine the same orientation for some points of ${\operatorname{strat}}(\sigma)\cap U_{\sigma,a_\sigma}\cap
U_{\sigma,a}$.
By Lemma \ref{le: orientation2} the above morphisms determine
the same orientation for all points from ${\operatorname{strat}}(\sigma)\cap U_{\sigma,a_\sigma}\cap
U_{\sigma,a}$.
Let $\sigma\leq \tau$. Denote by $\sigma'\preceq\tau$ the
face cone in $N_\tau$ such that $\underline{\sigma'}=\sigma$.
By Lemma \ref{le: surjection} find
$\alpha_{\sigma}\in {\rm Aut}(\sigma)$ such that
for the inclusion map
$\overline{\imath^\tau_{\sigma'}}:={\imath^\tau_{\sigma}}
\circ\alpha_\sigma$
any two charts $\phi_{\sigma,a_\sigma}: U_{\sigma,a_\sigma}\to
X_{\sigma}$,
$\pi_{\sigma'}^{\sigma}\phi^{\sigma'}_{\tau,a_{\tau}}:
U^{\sigma'}_{\tau,a_{\tau}}\to
X_{\sigma}$ define the same orientation at some point of
$x\in {\operatorname{strat}}(\sigma)\cap
U_{\sigma,a_\sigma}\cap U^{\sigma'}_{\tau,a_{\tau}}
$.
By Lemma \ref{le: orientation2} they determine the same
orientation at all points of ${\operatorname{strat}}(\sigma)\cap
U_{\sigma,a_\sigma}\cap U^{\sigma'}_{\tau,a_{\tau}}
$. We need to verify that for any two charts $\phi_{\sigma,a}$ and
$\phi_{\tau, b}$, where $\sigma\leq\tau$, and any
$\tau'\preceq \tau$ such that $\omega(\tau')=\sigma$ the morphisms
$\phi_{\sigma,a}$ and
$\phi^{\tau'}_{\tau,b}$ determine the same orientation
for all $x\in {\operatorname{strat}}(\sigma)\cap U_{\sigma,a}\cap U^{\tau'}_{\tau,b}$.
By Lemma \ref{le: orientation2} it suffices to show that the
above morphisms determine the same orientation for some $x\in
{\operatorname{strat}}(\sigma)\cap U_{\sigma,a}\cap
U_{\sigma,a_\sigma} \cap U^{\tau'}_{\tau,b}\cap U^{\sigma}_{\tau,a_\tau}$.
By the above $\phi_{\sigma,a}$ and $\phi_{\sigma,a_\sigma}$
determine the same orientation at all $x\in
{\operatorname{strat}}(\sigma)\cap U_{\sigma,a}\cap
U_{\sigma,a_\sigma}$. Also $\phi_{\tau,b}$ and $\phi_{\tau,a_\tau}$
determine the same orientation at all $x\in {\operatorname{strat}}(\tau)\cap U_{\tau,b}\cap
U_{\tau,a_\tau}$. By Lemma \ref{le: openorient},
$\phi_{\tau,b}^*({\mathcal U}^{\operatorname{\rm can}}(X_\tau,S_\tau))$ and
$\phi_{\tau,a_\tau}^*({\mathcal U}^{\operatorname{\rm can}}(X_\tau,S_\tau))$ are
compatible on $U_{\tau,b}\cap
U_{\tau,a_\tau}$. Hence $\phi^{\tau'}_{\tau,b}$ and
$\phi^{\sigma'}_{\tau,a_\sigma}$ determine the same
orientation at points $x\in
{\operatorname{strat}}(\sigma) \cap U^{\tau'}_{\tau,b}\cap U^{\sigma'}_{\tau,a_\tau}$.
Finally $\phi_{\sigma,a}$ and
$\phi^{\tau'}_{\tau,b}$ determine the same orientation for
all $x\in {\operatorname{strat}}(\sigma)\cap U_{\sigma,a}\cap U^{\tau'}_{\tau,b}$.
\qed
\bigskip
\subsection{Resolution of singularities
of toroidal varieties in arbitrary characteristic}\label{se: resolution}
We call a
$\Gamma$-semicomplex $\Sigma$ {\it regular} if all cones
$|\sigma|$, where $\sigma\in\Sigma$ are regular.
\begin{proposition} \label{pr: Danilov} For any oriented
$\Gamma$-semicomplex $\Sigma$ there exists a sequence of
stable vectors
$v_1,\ldots,v_m\in {\rm Stab}(\Sigma) $ such that the subdivision
$\Delta:=\langle v_k\rangle \cdot\ldots\cdot \langle v_1
\rangle\cdot\Sigma$ of $\Sigma$ is a regular
$\Gamma$-semicomplex.
\end{proposition}
\noindent{\bf Proof.}
Let $\sigma_1,\ldots,\sigma_k$ denote all the faces of $\Sigma$.
We shall construct by induction on $i$ the canonical
subdivision $\Delta_i$ of $\Sigma$ such that the
fan ${\Delta_i^{\sigma_j}}$ is regular for all
$j\leq i$. Suppose $\Delta_i$ is
already constructed.
By \cite{KKMS} or \cite{Danilov1}
there exist centers $v^i_1,\ldots,v^i_{k_i}$ for the fan
$\Delta^{\sigma_{i+1}}_i$ in $N_{\sigma_{i+1}}$ such that
$\langle v^i_{k_i} \rangle\cdot\ldots\cdot\langle v^i_1
\rangle\cdot\Delta^{\sigma_{i+1}}_i$ is regular.
The subsequent centers $v^i_j$ of star subdivisions coul be chosen to be minimal internal vectors in indecomposable faces of
$\langle v^i_{j-1}\rangle \cdot\ldots\cdot \langle v^i_1
\rangle\cdot\Delta^{\sigma_{i+1}}_i$.
Therefore by Lemma \ref{le: minimal vectors} all centers
in the desingularization process are $\Sigma$-stable. Hence by
Proposition \ref{pr: blow-ups}, $\Delta_{i+1}:
=\langle v^i_k \rangle\cdot\ldots\cdot\langle v^i_1
\rangle\cdot\Delta_{i}$ is canonical. By construction
$\Delta^{\sigma_{i+1}}_{i+1}$ is regular. Note that
$\Delta^{\sigma_{j}}_{i+1}=\Delta^{\sigma_j}_{i}$ for
$j<i+1$ since during the desingularization regular cones
remain unaffected. \qed
\begin{theorem} \label{th: resolution}
For any $\Gamma$-stratified toroidal variety $(X,S)$
(respectively a $\Gamma$-toroidal variety $X$) there
exists a sequence of $\Gamma$-equivariant blow-ups at locally monomial valuations
${\operatorname{bl}}_{\nu_k}\circ\ldots\circ {\operatorname{bl}}_{\nu_1}(X)$ which is a
resolution of singularities.
\end{theorem}
\noindent{\bf Proof.} In the case of a nonstratified
toroidal variety consider $X$ with stratification ${\rm
Sing}^\Gamma(X)$. By Lemma \ref{le: existence} we can assume that
the $\Gamma$-stratified toroidal variety is oriented.
By Lemma \ref{le: orientation2} the associated semicomplex
is oriented.
By
Proposition \ref{pr: Danilov} and Proposition \ref{pr:
blow-ups}, ${\operatorname{bl}}_{\nu_k}\circ\ldots\circ {\operatorname{bl}}_{\nu_1}(X)$ is smooth. \qed
\section{Orientation of toroidal modifications}
\subsection{Lifting group actions}
Let $U$ be an affine variety. Let $\widehat{X}_x$ be a
completion of a variety $X$ at its closed point $x$. Let
$\Theta: Y\to X$ be a proper morphism and
$\widehat{Y}_x:=\widehat{X}_x\times_XY$.
Set $$U\widehat{\times}\widehat{X}_x:=\operatorname{Spec}(\lim_{\leftarrow}
K[U]\otimes {\mathcal O}_x^n/m_{X,x}),$$
$$U\widehat{\times}\widehat{Y}_x:={(U\widehat{\times}\widehat{X}_x)}\times_XY.
$$
\begin{lemma}\label{le: normal0}
If $U$, $X$ are normal varieties then the scheme
$U\widehat{\times}\widehat{X}_x$ is normal. If $U$, $Y$ are
normal varieties then the scheme
$U\widehat{\times}\widehat{Y}_x$ is normal.
\end{lemma}
\noindent{\bf Proof.}
(1) $\pi_X: U\widehat{\times}\widehat{X}_x \to
U\times \widehat{X}_x$ is \'etale at $(u,x)$ for $u\in U$
since it determines an
isomorphism of the completions of the local rings at $(u,x)$. $U\times
\widehat{X}_x$ is normal therefore $U\widehat{\times}\widehat{X}_x$ is
normal at $(u,x)\in U\times\{x\}$. But all closed points of
$U\widehat{\times}\widehat{X}_x$ are in $
U\times\{x\}\subset U\widehat{\times}\widehat{X}_x$. The morphism
$\pi$ is \'etale in a neighborhood of
$U\widehat{\times}\widehat{X}_x$. Such a neighborhood is
equal to $U\widehat{\times}\widehat{X}_x$ since its
complement if nonempty would contain some closed points.
(2) $\pi_Y: U\widehat{\times}\widehat{Y}_x \to
U\times \widehat{Y}_x$ is \'etale at $(u,y)$ for $u\in U$
and $y\in \theta^{-1}(x)$ since $\pi_Y$ is a pull-back of
$\pi_X$. The rest of the reasoning is the same. \qed
\begin{definition}
\label{de: lifting}
Let a proalgebraic group $G$ act on $\widehat{X}_x$. Let
$\Phi:G\widehat{\times}\widehat{X}_x\to\widehat{X}_x$ be
the action morphism and
$\Psi:G\widehat{\times}\widehat{X}_x\to
G\widehat{\times}\widehat{X}_x$
be the action automorphism (see Lemma \ref{le: group action}).
Let $\theta: U\to G$
be a morphism from a normal algebraic variety $U$ to $G$.
Then by the {\it action morphism with
respect to $U$} or simply {\it action morphism} we mean the
induced morphism
$\Phi_{U,\widehat{X}_x}:U\widehat{\times}\widehat{X}_x\to\widehat{X}_x$.
By the {\it action automorphism with
respect to $U$} or simply {\it action automorphism} we mean the
induced morphism
$\Psi_{U,\widehat{X}_x}:U\widehat{\times}\widehat{X}_x\to
U\widehat{\times}\widehat{X}_x$.
Let $\alpha: Y\to\widehat{X}_x$ be a $G^K$-equivariant
birational morphism.
By the {\it lifting} of the action morphism (resp. automorphism)
we mean the morphism
$\Phi_{U,Y}:U\widehat{\times}Y\to Y$ (resp. the automomophism
$\Psi_{U,Y}:U\widehat{\times}Y\to U\widehat{\times}Y$)
commuting with the induced morphism ${\operatorname{id}}\widehat{\times}\alpha:U\widehat{\times}Y\to
U\widehat{\times} \widehat{X}_x$ and with $\alpha$.
\end{definition}
\begin{lemma}\label{le: blow-up} Let $G$ be a
proalgebraic group acting on $\widehat{X}_x$. Let $U\to G$
be any morphism from a normal algebraic variety $U$.
Let $I$ be
a $G$-invariant ideal on $\widehat{X}_x$. Let $\phi: Y\to
\widehat{X}_x$ be the normalization of the blow-up at
$I$. Then there is a lifting of the action automorphism
(and action morphism)
$\Psi_{U,\widehat{X}_x}: U\widehat{\times} \widehat{X}_x \to
U\widehat{\times} \widehat{X}_x$ to $\Psi_{U,Y}: U\widehat{\times}
Y \to U \widehat{\times} Y$.
\end{lemma}
\noindent {\bf Proof.} Let $\Phi:=\Phi_{U,\widehat{X}_x}: U\widehat{\times}
\widehat{X}_x \to \widehat{X}_x$ denote the action morphism.
Set $\Psi:=\Psi_{U,\widehat{X}_x}$. By Lemma \ref{le: group
ideal} we know
that $J=:p^*(I)=\Phi^*(I)=\Psi^*p^*(I)=\Psi^*(J)$ is
preserved by the action automorphism.
Therefore $J$ lifts to the blow-up of
$U\widehat{\times} \widehat{X}_x$ at $J$ and to its
normalization.
Since $U\widehat{\times} \widehat{X}_x\to
\widehat{X}_x$ is \'etale, the normalization of the blow-up of
$U\widehat{\times} \widehat{X}_x$ at $J$ is isomorphic to
$U\widehat{\times} Y$.
\qed
\begin{proposition} \label{pr: liftings} Let $\sigma$ be a cone of
maximal dimension in $N_\sigma$. Let $G$ be a connected
proalgebraic group acting on $X:=\widehat{X}_\sigma$.
Let $g:U\to G$ be a morphism from a normal algebraic
variety $U$. Let $G^K$ act on
$Y:=\widehat{X}_\Sigma$ such that $\psi: Y\to X$ is a
$G^K$-equivariant proper birational
morphism. Then there is a lifting of the action automorphism
$\Psi_{U,X}: U\widehat{\times} X \to U\widehat{\times} X$ to
$\Psi_{U,Y}: U\widehat{\times} Y \to U\widehat{\times} Y$.
\end{proposition}
\noindent {\bf Proof.} Let $\Psi_{U,Y}: U\times Y {{-}\to} U\times Y$ be
a birational map which is a lifting of $\Phi_{U,Y}$. By
Lemmas \ref{le: ideal} and \ref{le: blow-up} we can find a
factorization $Z\to Y\to X$ giving the diagram of proper morphisms
\[\begin{array}{rcccccc}
&U\widehat{\times} Z&&\buildrel \Psi_{U,Z}\over\longrightarrow &&
U\widehat{\times} Z&\\
&&\searrow&&\swarrow&\\
&\downarrow&&W&&\downarrow&\\
&&\swarrow f_1&&\searrow f_2&&\\
&U\widehat{\times} Y&&\buildrel \Psi_{U,Y} \over {{-}\to} &&
U\widehat{\times} Y&\\
&\downarrow\phi_{U\widehat{\times} Y}&&&&
\downarrow\phi_{U\widehat{\times} Y}&\\
&U\widehat{\times} X&&\buildrel \Psi_{U,X} \over \longrightarrow&&
U\widehat{\times} X&
\end{array}\]
\noindent where $W$ is the irreducible component of the
fiber product of $U\widehat{\times}
Y\times_{U\widehat{\times} X} U\widehat{\times} Y$ which
dominates $U\widehat{\times} Y$. (The fiber product is
given by the morphisms $\phi_{U\widehat{\times} Y}$ and
$\Psi_{U,X}\phi_{U\widehat{\times} Y}.)$
Let $u\in U$ be a closed point. Take a pull-back of the above
diagram via the closed embedding:
$\{u\}\widehat{\times} X\to U\widehat{\times} X$
\[\begin{array}{rcccccc}
&\{u\}\widehat{\times} Z&&\buildrel \Psi_{u,Z}\over\longrightarrow &&
\{u\}\widehat{\times} Z&\\
&&\searrow&&\swarrow&\\
&\downarrow&&W_u&&\downarrow&\\
&&\swarrow &&\searrow &&\\
&\{u\}\widehat{\times} Y&&\buildrel \Psi_{u,Y} \over {\longrightarrow} &&
\{u\}\widehat{\times} Y&\\
&\downarrow&&&&\downarrow&\\
&\{u\}\widehat{\times} X&&\buildrel \Psi_{u,X} \over \longrightarrow&&
\{u\}\widehat{\times} X&
\end{array}\]
The birational map $\Psi_{u,Y}$ is the isomorphism defined
by the
action of $g(u)\in G$ on $Y$. $W_u$ consists of some
components of the
fiber product $\{u\}\widehat{\times}
Y\times_{\{u\}\widehat{\times} X} \{u\}\widehat{\times} Y$
. In particular it contains the irreducible component dominating
$\{u\}\widehat{\times} Y$ which is isomorphic to $Y$. On
the other hand $\{u\}\widehat{\times} Z\to W_u$ is a proper
surjective morphism. Consequently, $W_u$ is irreducible and
isomorphic to $Y$. By Lemma \ref{le: normal0},
$U\widehat{\times}Y$ is normal. Thus $f_i:W\to
U\widehat{\times}Y$ is a proper birational
morphism onto a normal scheme which is bijective
on the set of closed ($K$-rational) points. By the Zariski theorem
$f_i$ is an isomorphism and
$\Psi_{U,Y}=f_1^{-1}f_2$ is an action automorphism. \qed
\bigskip
\subsection{Orientation of toroidal modifications}
\begin{proposition} \label{pr: canonical orientation} Let $(X,S)$ be an
oriented $\Gamma$-stratified toroidal variety
with the associated $\Gamma$-semicomplex $\Sigma$. Let $f:(Y,R)\to (X,S)$
be the toroidal morphism associated to a
canonical subdivision $\Delta$ of $\Sigma$.
Then
\begin{enumerate}
\item The stratified toroidal variety $(Y,R)$ is \underline{oriented}
with charts from
$\bigcup_{\sigma\in\Sigma}\phi_\sigma^{f*}({\mathcal U}(\Delta^\sigma,\Delta^\sigma_{\operatorname{stab}}))$
induced by the charts $\phi_\sigma: U\to X_\sigma$ on $X$
and the associated oriented semicomplex $\Sigma_R=\Delta_{\operatorname{stab}}$.
\item
For any point $x$ in a stratum $s\in S$ every
$\Gamma_s$-equivariant automorphism
$\alpha$ of $\widehat{X}_x$ preserving strata and orientation
lifts
to a $\Gamma_s$-equivariant automorphism
$\alpha'$ of $Y\times_X\widehat{X}_x$ such that the atlases
\\
${\alpha'}^*\hat{\phi}_\sigma^{f*}
({\mathcal U}(\Delta^\sigma,\Delta^\sigma_{\operatorname{stab}}))$ and $\hat{\phi}_\sigma^{f*}
({\mathcal U}(\Delta^\sigma,\Delta^\sigma_{\operatorname{stab}}))$ are
compatible along $f^{-1}(x)\subset Y\times_X\widehat{X}_x$.
Here $\hat{\phi}_\sigma^{f}: Y\times_X\widehat{X}_x\to
X_{\Delta^\sigma}$ denotes the morphism induced by ${\phi}_\sigma^{f}$.
\item The stable support of the oriented $\Gamma$-semicomplex
$\Sigma_R=\Delta_{\operatorname{stab}}$ is contained in the stable support of $\Sigma$.
\end{enumerate}
\end{proposition}
\noindent{\bf Proof.}$(\Leftarrow)$
Let $\phi_{\sigma,1}:U_{\sigma,1}\to
X_{\sigma}$ and $\phi_{\tau,2}: U_{\tau,2}\to
X_{\tau}$ be two charts on $X$, where $\sigma\leq
\tau$. For (1) we have to show that
$\phi_{\sigma,1}^{f*}({\mathcal U}(\Delta^\sigma, \Delta^\sigma_{\operatorname{stab}}))$
and $\phi_{\tau,2}^{f*}({\mathcal U}(\Delta^{\tau}, \Delta^\tau_{\operatorname{stab}}))$
are compatible on $f^{-1}(U_{\sigma,1}\cap U_{\tau,2})$ along
$f^{-1}(x)$ for any point $x\in
{\operatorname{strat}}(\sigma)\cap(U_{\sigma,1} \cap U_{\tau,2})$.
Let $\tau'\preceq \tau$ denote a face of $\tau$ such that
${\omega}(\tau')=\sigma$ and $\phi_{\tau,2}(x)\in O_{\tau'}$.
Then $\tau'=\sigma\times r(\tau')$, where
$r(\tau')$ denotes a regular cone and
$\tau'\setminus\sigma$ is disjoint from ${\operatorname{Stab}}(\Sigma)$.
By Lemma \ref{le: sum},
$\Delta^\tau|\tau'=\Delta^\tau|\sigma
\oplus r(\tau')=\Delta^\sigma\times r(\tau')$. Also,
$\Delta^\tau_{\operatorname{stab}}|\tau'=\Delta^\sigma_{\operatorname{stab}}\times\{0\}$.
Denote by $\phi^{\tau'}_{\tau,2}: U^{\tau'}_{\tau,2}\to
X_{\tau'}$, the
restriction of $\phi_{\tau,2}$ and by
$\phi^{\tau'f}_{\tau,2}$ its lifting.
Then
$\phi^{\tau'f*}_{\tau,2}({\mathcal U}(\Delta^\tau|\tau', \Delta_{\operatorname{stab}}^\tau|\tau'))
$ is the restriction of $\phi^{f*}_{\tau,2}
({\mathcal U}(\Delta^\tau, \Delta_{\operatorname{stab}}^\tau))
$ to the open set $f^{-1}(U^{\tau'}_{\tau,2})$ containing
$f^{-1}(x)$.
Set $U_{\sigma,2}:=U^{\tau'}_{\tau,2}$,
$\phi_{\sigma,2}:={\pi}^\sigma_{\tau'}\phi^{\tau'}_{\tau,2}:
U_{\sigma,2}\to
X_{\sigma}$, where ${\pi}^{\sigma}_{\tau'}: X_\tau'
\to
X_\sigma$ is defined by any projection which is identical on
$\sigma\preceq \tau'$. Let
$\phi^f_{\sigma,2}:={\pi}^{\sigma f}_{\tau'}\phi^{\tau'
f}_{\tau,2}: f^{-1}(U_{\sigma,2})\to
X_{\sigma}$ be the lifting of $\phi_{\sigma,2}$. Then
$$\phi^{\tau'f*}_{\tau,2}({\mathcal U}(\Delta^\tau|\tau',\Delta_{\operatorname{stab}}^\tau|\tau'))=
\phi^{\tau'f*}_{\tau,2}{\mathcal U}(\Delta^\sigma\times
r(\tau'), \Delta_{\operatorname{stab}}^\sigma\times
\{0\})$$ \noindent is compatible along $f^{-1}(x)$ with
$$\phi^{\tau'f*}_{\tau,2}
{\pi}^{\sigma f*}_{\tau'}({\mathcal U}(\Delta^\sigma, \Delta_{\operatorname{stab}}^\sigma))=
\phi^{f*}_{\sigma,2}({\mathcal U}(\Delta^\sigma, \Delta_{\operatorname{stab}}^\sigma)).$$
It suffices to show that
$\phi^{f*}_{\sigma,1}({\mathcal U}(\Delta^\sigma,
\Delta_{\operatorname{stab}}^\sigma))$ and
$\phi^{f*}_{\sigma,2}({\mathcal U}(\Delta^\sigma,
\Delta_{\operatorname{stab}}^\sigma))$ are compatible along $f^{-1}(x)$.
Let $U\subset U_{\sigma,1}\cap U_{\sigma,2}$, be an open
subset for which there exist \'etale
extensions $\widetilde{\phi}_{i}: U \to
X_{\widetilde{\sigma}}$ of $\phi_{\sigma,i}$, $i=1,2$.
Assume that
$\widetilde{\phi}_{i}(x)=O_{\widetilde{\sigma}}=
O_{{\sigma}\times
{\operatorname{reg}}(\sigma)}$, where ${\operatorname{reg}}(\sigma)$ is a regular cone of
dimension $k=\dim({\operatorname{strat}}(\sigma))$.
Let $\widetilde{\phi}_{fi}$ be an \'etale extension
of ${\phi}_{fi}$ which is a lifting of $\widetilde{\phi}_{i}$.
Consider the fiber squares
\[\begin{array}{rccccccccccccc}
&\widetilde{\phi}_i:&U&\to&
X_{\widetilde{\sigma}}&=X_{\sigma\times
{\operatorname{reg}}(\sigma)}&&&&
\widehat{\phi}_i:&\widehat{X}_x&\to& \widetilde{X}_{\sigma}&\\
&&\uparrow&&\uparrow f_{\widetilde{\Delta}}
&&&&&&\uparrow&&\uparrow \widetilde{f}_{{\Delta}}&\\
&\widetilde{\phi}_{fi}:&f^{-1}(U)&\to&
X_{\widetilde{\Delta}^\sigma}&=X_{\Delta^\sigma\times
{\operatorname{reg}}(\sigma)}&&&&\widehat{\phi}_{fi}:
&Y\times_X \widehat{X}_x&\to & \widetilde{X}_{\Delta^\sigma}&
\end{array}\]
The morphism $\widetilde{f}_{{\Delta}}$
is a pull-back of the morphism $f_{\widetilde{\Delta}}$. Therefore
$\widetilde{f}_{{\Delta}}^{-1}(O_{\widetilde{\sigma}})$ is
a $K$-subscheme of $ \widetilde{X}_{\Delta^\sigma}$.
The automorphism
$\widehat{\phi}:=\widehat{\phi}_{2}\widehat{\phi}^{-1}_{1}$
of $\widetilde{X}_{\sigma}\simeq \widehat{X}_x$
preserves strata and orientation.
For the proof of conditions (1) and (2) we need to show
that any such automorphism $\widehat{\phi}$ preserving
strata and orientation
lifts to the automorphism
$\widehat{\phi}^f=\widehat{\phi}^f_{2}(\widehat{\phi}_{1}^{f})^{-1}$
of $\widetilde{X}_{\Delta^\sigma}\simeq Y\times_X \widehat{X}_x$
such that
$
\widehat{\phi}^{f*}({\mathcal U}(\Delta^\sigma,
\Delta^\sigma_{\operatorname{stab}})$ and ${\mathcal U}(\Delta^\sigma, \Delta^\sigma_{\operatorname{stab}})$ are compatible along
$\widetilde{f}_{{\Delta}}^{-1}(O_{\widetilde{\sigma}_\sigma})$.
By Lemma \ref{le: group structure} there is
an action morphism
$\Phi: W \widehat{\times}\widetilde{X}_{\sigma} \to
\widetilde{X}_{\sigma}$ such that
\begin{enumerate}
\item $\Phi_{e}:=\Phi_{|e\widehat{\times}\widetilde{X}_{\sigma}}=
{\operatorname{id}}_{X_{\sigma}}$.
\item There exists $g\in W$ such that
$\Phi_{g}:=\Phi_{|g\widehat{\times}\widetilde{X}_{\sigma}}
=\widehat{\phi}$.
By Proposition \ref{pr: liftings}, $\Phi$ can be lifted to ${\Phi}^f
: W\widehat\times
\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\Delta^\sigma}$ such that
\item
${\Phi}^f_{e}={\operatorname{id}}_{\widetilde{X}_{\Delta^\sigma}}$.
\item
${\Phi}^f_{g}=\widehat{\phi}^f$.
\end{enumerate}
Let ${\Pi}: W\widehat\times
\widetilde{X}_{\sigma}\to
\widetilde{X}_{\sigma}$ and ${\Pi}^f: W\widehat\times
\widetilde{X}_{\Delta^\sigma}\to
\widetilde{X}_{\Delta^\sigma}$ denote the natural projections.
Note that
${\Pi}^{-1}(O_{\widetilde{\sigma}})={\Phi}^{-1}(O_{\widetilde{\sigma}})=
W\widehat{\times}{O}_{\widetilde{\sigma}}\simeq
W{\times}{O}_{\widetilde{\sigma}}\simeq W$.
The morphisms $\Pi^f$ and $\Phi^f$ are pull-backs of
$\Pi$ and $\Phi$ induced by $\widetilde{f}_\Delta$. Thus
$({\Pi}^{f})^{
-1}(f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}}))=
({\Phi}^{f})^{ -1}(f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})=
W\widehat{\times}f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})
\simeq
W{\times}f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})$
is a $K$-subscheme of $W\widehat{\times}\widetilde{X}_{\Delta^\sigma}$.
For the sake of our considerations we shall enrich
the stratification $R$ by adding to strata
$r\in R$ their intersections with $f^{-1}(x)$.
Set $\overline{R}=\{r\setminus f^{-1}(x)\mid r\in R\}
\}\cup\{r \cap f^{-1}(x)\mid r\in R \}$.
Strata of $R$ on $f^{-1}(U)$ correspond to
the embedded semifan $\Delta^\sigma_{\operatorname{stab}}\subset
\Delta^\sigma\times {\operatorname{reg}}(\sigma)$. Strata of $\overline{R}$
correspond to the embedded semifan $\Omega\subset \Delta^\sigma\times
{\operatorname{reg}}(\sigma)$, where $\Omega:= \Delta^\sigma_{\operatorname{stab}} \cup
\{\omega\times{\operatorname{reg}}(\sigma)\mid \omega\in
\Delta^\sigma_{\operatorname{stab}},
{\operatorname{int}}(\omega)\subset{\operatorname{int}}(\sigma)\}$.
For any stratum ${\operatorname{strat}}_Y(\omega)\in R$, $\omega\in\Delta^\sigma_{\operatorname{stab}}$,
dominating the stratum
${\operatorname{strat}}_X(\sigma)$ let ${\operatorname{strat}}(\omega)$ denote the
corresponding stratum on $\widetilde{X}_{\Delta^\sigma}$.
Let $\widetilde{\omega}:=
\omega\times{\operatorname{reg}}(\sigma)\in\Omega$. Then
${\operatorname{strat}}(\omega)\cap
f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})=
{\operatorname{strat}}(\widetilde{\omega})\subset
f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})
\subset \widetilde{X}_{\Delta^\sigma}$ is a
$G^{\sigma}$-invariant stratum corresponding to
$\widetilde{\omega}\in\Omega$
. Since ${\operatorname{strat}}(\widetilde{\omega})$ is an irreducible locally closed subset of
$f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})$ it is an
algebraic variety.
The natural isomorphism $(\Phi^{f})^{-1}({\operatorname{strat}}(\widetilde{\omega}))=
(\Pi^{f})^{-1}({\operatorname{strat}}(\widetilde{\omega}))= W\widehat{\times}
{\operatorname{strat}}(\widetilde{\omega})\longrightarrow
W{\times}{\operatorname{strat}}(\widetilde{\omega}) $ is a pull-back of the
isomorphism
$W\widehat{\times}{O}_{\widetilde{\sigma}}\longrightarrow
W{\times}{O}_{\widetilde{\sigma}}$ induced by $
f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})\to
O_{\widetilde{\sigma}}$.
Thus $W\widehat{\times}{\operatorname{strat}}(\widetilde{\omega})=
W{\times}{\operatorname{strat}}(\widetilde{\omega})$ is an algebraic variety.
By definition we have $\Psi^f_{e}= \Pi^f_{e}=
{\operatorname{id}}_{|{\operatorname{strat}}(\widetilde{\omega})}$. Hence by Lemma \ref{le: sections2}
applied to the natural projection of
$W\widehat{\times}{\operatorname{strat}}(\widetilde{\omega})$
on $W$
the collections of charts $\Psi^{f*}
({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},S_\Omega))$
and $\Pi^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma}, S_\Omega))$ are compatible along
$\{e\}\widehat{\times}{\operatorname{strat}}(\widetilde{\omega})$.
Hence by Lemma \ref{le: orientation2},
$\Psi^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma}, S_\Omega))$
and $\Pi^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma}, S_\Omega))$ are compatible along
$\{g\}\widehat{\times}{\operatorname{strat}}(\widetilde{\omega})$. Again by
Lemma \ref{le: sections2},
$\Psi_{g}^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma}, S_\Omega))=
\widehat{\psi}^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_\Omega))$
and
$\Pi_{g}^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_\Omega))=
{\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_\Omega)$ are compatible
along ${\operatorname{strat}}(\widetilde{\omega})$. The charts
$\phi_{\omega'}$, $\omega'\geq\omega$, of
${\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_{\Delta^\sigma_{\operatorname{stab}}})$
are obtained by composing the charts
$\phi_{\widetilde{\omega'}}$ of
${\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_\Omega)$ with the natural projection
$X_{\widetilde{\omega'}}\to X_{\omega'}$. Thus the atlases
${\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_{\Delta^\sigma_{\operatorname{stab}}})$ and
$\widehat{\psi}^{f*}({\mathcal U}^{\operatorname{\rm can}}(\widetilde{X}_{\Delta^\sigma},
S_{\Delta^\sigma_{\operatorname{stab}}})$
are compatible
along ${\operatorname{strat}}({\omega})\cap
f_{\widetilde{\Delta}}^{-1}(O_{\widetilde{\sigma}})={\operatorname{strat}}(\widetilde{\omega})$.
(3) By Lemma \ref{le: orientation4}, $\Sigma_R=\Delta_{\operatorname{stab}}$ is an
oriented $\Gamma$-semicomplex and the notion of stability
makes sense for $\Sigma_R$. Let $v\in {\operatorname{int}}(\omega)$,
where $\omega\in\Delta^\sigma_{\operatorname{stab}}$, be a
$\Sigma_R$-stable vector.
Assume that ${\operatorname{int}}(\omega)\subset {\operatorname{int}}(\sigma)$. We need
to show that for any $\tau\geq\sigma$,
any automorphism $\alpha$ of ${\operatorname{Aut}}(\widetilde{X}_\tau)^0$ preserves
${\operatorname{val}}(v,\widetilde{X}_\tau)$. Find the cone $\omega'\in
{\Delta^\tau}$ such that $\omega'\geq\omega$ and
${\operatorname{int}}(\omega')\subset{\operatorname{int}}(\tau)$. By the convexity of
${\operatorname{stab}}(\tau)$, ${\operatorname{int}}(\omega')$ intersects ${\operatorname{stab}}(\tau)$
and $\omega'\in {\Delta^\tau}_{\operatorname{stab}}$.
Denote by
$\pi:\widetilde{X}_{\Delta^\tau}\to {X}_{\Delta^\tau}$ the
standard projection.
By (2) $\alpha$ lifts to an
automorphism $\alpha^f$ of $\widetilde{X}_{\Delta^\tau}$
such that $\alpha^{f*}\pi^*({\mathcal U}(\Delta^\tau,
\Delta^\tau_{\operatorname{stab}})$ and $\pi^*({\mathcal U}(\Delta^\tau,
\Delta^\tau_{\operatorname{stab}})$ are compatible along
$f_\Delta^{-1}(O_{\widetilde{\tau}})$.
Let $p:=O_{\omega'\times{\operatorname{reg}}(\tau)}=O_{\omega'}\cap f_\Delta^{-1}
(O_{\widetilde{\tau}}) \subset
\widetilde{X}_{\Delta^\tau}$. Set
$Y:=\widetilde{X}_{\Delta^\tau}$. By Lemma \ref{le: 1},
$\widehat{Y}_p=\widehat{X}_{\omega'\times reg(\tau)}=
\widetilde{X}_{\omega'}$. By assumption
${\operatorname{val}}(v,\widehat{Y}_p)$ is invariant with respect to any
$\Gamma_{\omega'}=\Gamma_\tau$-equivariant automorphisms of
$\widehat{Y}_p=\widetilde{X}_{\omega'}$ preserving
orientation and strata.
The automorphism $\alpha^{f}$ preserves $p$ and induces an
automorphism $\widehat{\alpha}_p^{f}$ of $\widehat{Y}_p$.
Denote by $\widehat{\pi}: \widehat{Y}_p\to X_{\omega'}$ the
morphism induced by $\pi$.
Since the atlases $\alpha^{f*}_p\pi^*({\mathcal U}(\Delta^\tau,
\Delta^\tau_{\operatorname{stab}})$ and $\pi^*({\mathcal U}(\Delta^\tau,
\Delta^\tau_{\operatorname{stab}}))$ are compatible the morphisms
$\widehat{\pi}\widehat{\alpha}_p^{f}:\widehat{Y}_p\to
X_{\omega'}$ and $\widehat{\pi}$ determine the same
orientation at $p$. Thus $\widehat{\alpha}_p^{f}$ preserves
orientation at $p$ and $\widehat{\alpha}_{p*}^{f}({\operatorname{val}}(v,Y_p))={\operatorname{val}}(v,Y_p)$.
Then also
${\alpha}_{*}^{f}({\operatorname{val}}(v,\operatorname{Spec}({\mathcal O}_{Y,p}))={\operatorname{val}}(v, \operatorname{Spec}({\mathcal O}_{Y,p})$.
Consequently,
$\alpha_{*}^{f}({\operatorname{val}}(v,\widetilde{X}_{\Delta^\tau})=
{\operatorname{val}}(v,\widetilde{X}_{\Delta^\tau})$
and $\alpha_{*}({\operatorname{val}}(v,\widetilde{X}_{\tau})=
{\operatorname{val}}(v,\widetilde{X}_{\Delta^\tau})$.
\qed
\section{The $\pi$-desingularization lemma of Morelli}
\subsection{Local projections of $\Gamma$-semicomplexes}
\begin{definition} Let $\Sigma$ be a simplicial
$\Gamma$-semicomplex and $\Delta$ be its canonical subdivision.
Let $\pi_{\sigma}: \sigma\rightarrow
\sigma^\Gamma$ be the projection defined by the quotient map $X_{\sigma}\to
X_{\sigma}/\Gamma_\sigma=X_{\sigma^\Gamma}$, for any $\sigma\in
\Sigma$. For any
$\delta\in\Delta^\sigma$ set
$\Gamma_{\delta}=(\Gamma_{\sigma})_{\delta}=
\{g\in\Gamma_\sigma\mid gx=x, x\in O_\delta\}$.
By $\pi_{\delta}: \delta\rightarrow
\delta^\Gamma$ denote the projection defined by the quotient map $X_{\delta}\to
X_{\delta}/\Gamma_\delta=X_{\delta^\Gamma}$.
We say that $\Sigma$ is {\it strictly
$\pi$-convex} if for any $\sigma \in \Sigma$,
$\pi_{\sigma}(\sigma)= \sigma^\Gamma$ is strictly
convex (contains no line). We call a semicomplex $\Sigma$
{\it simplicial} if all cones $|\sigma|$, $\sigma\in\Sigma$, are simplicial.
\end{definition}
\begin{lemma} \label{le: projections2}
\begin{enumerate}
\item Let $\Sigma$ be a
strictly convex $\Gamma$-semicomplex.
Then for any $\tau\leq \sigma$, there exists an inclusion
$\imath^{\sigma\Gamma}_\tau : \tau^\Gamma \hookrightarrow
\sigma^\Gamma $ and the commutative diagram of inclusions:
\[\begin{array}{rccc}
\imath^{\sigma}_\tau:&\tau &\hookrightarrow&\sigma\\
&\pi_{\tau}\downarrow&&\pi_{\sigma}\downarrow\\
\imath^{\sigma\Gamma}_\tau:&\tau^\Gamma&\hookrightarrow&
\sigma^\Gamma
\end{array}\]
\item Let $\Delta$ be a canonical subdivision of $\Sigma$.
Then for any $\delta\in\Delta^\sigma_{\operatorname{stab}}$
we have the commutative diagram of inclusions
\[\begin{array}{rccc}
&\delta &\subset&\sigma\\
&\pi_{\delta}\downarrow&&\pi_{\sigma}\downarrow\\
&\delta^\Gamma&\hookrightarrow&
\sigma^\Gamma
\end{array}\]
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) Consider the commutative diagram
\[\begin{array}{rccccc}
&
X_{(\tau,N_\sigma)}&\simeq& X_{\tau}\times T&&\\
&\downarrow &&\downarrow&&\\
&X_{(\tau,N_\sigma)}//\Gamma_\tau&
\buildrel i\over\simeq&X_\tau//\Gamma_\tau
\times T&&\\
&\downarrow &&\downarrow&&\\
&X_{\pi_{\sigma}(\tau,N_{\sigma})}=
X_{(\tau,N_{\sigma})}/\Gamma_\sigma&\buildrel j\over\simeq
&X_{(\tau,N_\sigma)}//\Gamma_\sigma
\times T/\Gamma_{\sigma}&\simeq& X_{(\tau^\Gamma,N_{\sigma}^\Gamma)}\\
\end{array}\]
\noindent where $T$ is the relevant torus.
Note that $\Gamma_\tau$ acts trivially on
${\operatorname{strat}}(\tau)\subset X_\sigma$, hence in particular on the torus $T$ in
$X_{\tau}\times T$. This gives the isomorphism
$i$. Since $\Gamma_\sigma/\Gamma_\tau$ acts freely on
$X_{(\tau,N_\sigma)}//\Gamma_\tau=X_{\tau}//\Gamma_\sigma
\times T$ we get the
isomorphism $j$. Consequently,
$(\tau^\Gamma,N_{\sigma}^\Gamma)\simeq\pi_{\sigma}(\tau,N_{\sigma})
\subset \sigma^\Gamma$.
(2) Repeat the reasoning from (1) with appropriate inclusions.
\qed
\bigskip
\subsection{Dependent and independent cones}
In further considerations we shall assume $\Gamma= K^*$.
\begin{lemma} Let $\Sigma$ be a simplicial strictly
convex $K^*$-semicomplex.
Let ${\operatorname{Dep}}(\Sigma):=\{\sigma \in \Sigma\mid
\Gamma_\sigma=K^*\}$ and ${\operatorname{Ind}}(\Sigma):=\{\sigma \in \Sigma\mid
\Gamma_\sigma\neq K^*\}$. Then
\begin{enumerate}
\item For $\sigma \in {\operatorname{Ind}}(\Sigma)$, $\pi_{\sigma}$ is an immersion into the lattice
$N^\Gamma_\sigma$ of the same dimension.
\item For $\sigma \in {\operatorname{Dep}}(\Sigma)$, $\pi_{\sigma}$ is a
submersion onto the lattice $N^\Gamma_\sigma$ such that
$\dim(N_\sigma)= \dim(N^\Gamma_\sigma)+1$.
Moreover there is a vector $v_\sigma\in N_\sigma$ determined by a
$K^*$-action which spans the kernel of $\pi_{\sigma}$.
\end{enumerate}
\end{lemma}
\noindent {\bf Proof.} If $\sigma \in {\operatorname{Ind}}(\Sigma)$ then
$\Gamma_\sigma$ is finite and there is an immersion
$M_\sigma^\Gamma\to M_\sigma$, of lattices of the same dimension, where
$M^\Gamma$ is a lattice of $\Gamma$-invariant characters.
The dual morphism $N_\sigma\to N^\Gamma_\sigma$ is also an immersion of
lattices of the same dimension.
If $\sigma \in {\operatorname{Dep}}(\Sigma)$ then $\Gamma_\sigma=K^*$ is a
subtorus corresponding to the sublattice $N:=({\bf Q}\cdot
v_\sigma)\cap N_\sigma$ of $N_\sigma$. The projection
$\pi_\sigma$ is defined by the quotient map $N_\sigma\to
N_\sigma/N\simeq N_\sigma^\Gamma$.
\qed
\begin{lemma} Let $\Sigma$ be a simplicial strictly
convex $K^*$-semicomplex. Then for any faces
$\tau\leq\sigma$ in ${\operatorname{Dep}}(\Sigma)$,
$\imath^\sigma_\tau(v_\tau)=v_\sigma$.
\end{lemma}
\noindent{\bf Proof.} The induced toric morphism $X_\tau\to X_\sigma$
is $K^*$-equivariant. \qed
\begin{definition} (see Morelli \cite{Morelli1})
Let $\Delta$ be a canonical subdivision of a
$K^*$-semicomplex $\Sigma$. Let $\sigma\in \Sigma$.
\begin{enumerate}
\item A cone $\delta \in \Delta^\sigma$ is called {\it independent}
if the restriction of $\pi_\sigma$ to $\delta$ is a lattice
immersion.
Otherwise $\delta \in \Delta^\sigma$ is called {\it
dependent}.
\item
A minimal dependent face of $\Delta^\sigma$ is called a
{\it circuit}.
\item We call an independent face $\tau$
{\it up-definite} (respectively
{\it down-definite}) with
respect to a dependent face $\delta$ if
there exists a nonzero functional $F$ on
$\delta\subset N_\sigma^{\bf Q}$ such that
$F(v_\sigma)> 0$ (respectively $F(v_\sigma)< 0$), and $\tau=\{v \in
\sigma\mid F(v)=0\}$.
\end{enumerate}
\end{definition}
\begin{lemma}
Each dependent cone
$\delta=\langle v_1,\ldots,v_k\rangle \in \Delta^\sigma$ defines a
unique linear dependence relation
$$r_1\pi_{\sigma}(v_1)+\ldots+r_k\pi_{\sigma}(v_k)=0
\eqno{*}.$$ \noindent This
relation is determined up to proportionality.
\end{lemma}
\noindent {\bf Proof.} $\pi_\sigma(\delta)$ is $k-1$-
dimensional cone spanned by $k$ rays.\qed
\begin{definition}
\begin{enumerate}
\item The relation (*) is {\it positively normalized} if
the rays
$\langle v_i \rangle$
for which $r_i>0$ form an up-definite face
\item The rays for which $r_i>0$ (in a positively normalized
relation) are called {\it positive}, the rays for which $r_i<0$ are called
{\it negative}, the rays for which $r_i=0$ are called {\it
null} rays. The face $\tau$ of a cone $\delta$
is called {\it codefinite} if it contains only positive or only
negative rays.
\item For any cone $\delta$ we denote by $\delta_{-}$ (respectively
$\delta_{+}$) the fan
consisting of all faces of $\delta$ which are
down-definite (respectively up-definite). By $\delta^-$ (respectively
$\delta^+$) we denote the face of $\delta$
spanned by all negative rays and null rays
(respectively by all positive rays and null rays).
\end{enumerate}
\end{definition}
\begin{lemma} \label{le: weights} The relation (*) is
positively normalized iff
there exists $\alpha>0$, such that
$$r_1v_1+\ldots+r_kv_k=\alpha v_\delta$$
\end{lemma}
\noindent{\bf Proof.} Let $F$ be a nonnegative
functional on $\delta$ which is $0$ exactly on $\langle
v_i\mid r_i>0 \rangle$. Then $\langle
v_i\mid r_i>0 \rangle$ is up-definite iff $F(v_\delta)>0$.
The latter is equivalent to
$\alpha>0$. \qed
\begin{lemma}
Let $\tau=\langle v_1,\ldots,\check{v}_i,\ldots,v_k \rangle$ be a face
of $\delta$ of codimension $1$. Then $\tau$ is up-definite
iff $r_i>0$.
\end{lemma}
\noindent{\bf Proof.} Let $F$ be a nonnegative
functional on $\delta$ which is $0$ exactly on $\tau$.
Then by Lemma \ref{le: weights}, $r_i>0$ iff
$F(v_\delta)>0$. The latter means that
$\tau$ is up-definite.\qed
\begin{lemma} If $\tau\preceq\delta$ is up-definite
(respectively down-definite) then any face
$\tau'\preceq\tau$ is up-definite (respectively
down-definite) with respect to $\delta$. \qed
\end{lemma}
\noindent{\bf Proof.} Let $F$ be a nonnegative
functional on $\delta$ which is $0$ exactly on $\tau$ and
such that $F(v_\delta)>0$. Let $F'$ be a nonnegative
functional on $\tau$ which is $0$ exactly on $\tau'$. Then
$nF+F'$, where $n>>0$, is a nonnegative
functional on $\delta$ which is $0$ exactly on $\tau'$ and
$(nF+F')(v_\delta)>0$.\qed
\begin{lemma} \label{le: projection} The set $\delta_+$ (respectively $\delta_-$)
is a subfan of $\overline{\delta}$ with maximal faces of
the form $\langle v_1,\ldots,\check{v}_i,\ldots,v_k
\rangle$, where $r_i>0$ (respectively $r_i<0$). In
particular, each boundary face is up-definite or down-definite.
The projection $\pi_\delta$ maps $\delta_+$ (respectively
$\delta_-$) onto the
subdivision $\pi_\delta(\delta_+)$ (respectively
$\pi_\delta(\delta_-)$ of $\pi_\delta(\delta)$. Moreover
the restriction of $\pi$ to any boundary face is a linear isomorphism.
\end{lemma}
\noindent{\bf Proof.} Let $v$ be a vector in
$|\delta^\Gamma|$. Then either the line $\pi^{-1}(v)=\{v\}+\operatorname{lin}(v_\delta)$
intersects the relative interiors of exactly two boundary
faces at one point each and $\pi^{-1}(v)\cap \delta$ is an interval or
$\pi^{-1}(v)$ intersects the relative interior of one
boundary face at a point
and $\pi^{-1}(v)\cap \delta$ is the point. In the first case
one of the faces is up-definite and one is down-definite.
In the second case let $\tau$ be the boundary face
containing $p=\pi^{-1}(v)\cap \delta$. Let $F$ be a
nonnegative
functional on $\delta^\Gamma$ which is zero exactly on
$\tau$. Let $F'$ be any functional on $N^{\bf Q}_\delta$
which equals $0$ on $\tau$ and
such that $F'(v_\delta)\geq 0$ . Then $(nF\circ\pi)\pm F'$, where
$n>>0$, is nonnegative on $\delta$ and equals $0$ exactly
on $\tau$. Moreover $(nF\circ\pi+F')(v_\delta)>0$ and
$(nF\circ\pi-F')(v_\delta)<0$. Hence $\tau$ is up-definite
and down-definite.\qed
\begin{lemma} \label{le: weights2} Let
$\delta\in\Delta^\sigma$ be a dependent cone. Set
$\delta^+:=\langle v_i\mid r_i> 0 \rangle$.
The
ideal $I_{\overline{O}_{\delta^+}}$ of the closure of
${O}_{\delta^+}$ is generated by the set of all characters
with positive $K^*$-weights.
\end{lemma}
\noindent{\bf Proof.} Let $x^m$, $m\in \delta^\vee\cap
M_\delta$ be the character with the positive weight. Then
$(m,v_\delta)> O$. By Lemma \ref{le: weights}, there is
$r_i> 0$ in a positively normalized relation such that $(m,v_i)>0$.
Then $x^m$ is zero on the divisor corresponding to $v_i$
and on ${O}_{\delta^+}$. On the other hand for any $r_i>0$ we can find a
functional $m_i$ such that $(m_i,v_i)>0$,
$(m_i,v_\delta)>O$, $(m_i,v_j)=0$ , where $j\neq i$. Then
$\overline{O}_{\delta^+}$ is the set of
the common zeros of $x^{m_i}$, $r_i> 0$. \qed
\bigskip
We associate with a cone $\delta \in
\Delta^\sigma$ and an integral vector $ v \in
\delta^\Gamma$ a vector ${\rm Mid}(v,\delta)\in \delta$ (\cite{Morelli1}.
If $\delta$
is independent face, then ${\rm Mid}(v,\delta)$ is defined to
be the primitive
vector of the ray ${\bf Q}_{\geq
0}\cdot\pi_{\sigma}^{-1}(v) \subset \delta$.
If $\delta$ is dependent let $\pi_{\delta_{-}}=\pi_{\sigma|\delta_-}$ and
$\pi_{\delta_{+}}= \pi_{\sigma|\delta_+}$ be the
restrictions of $\pi_\sigma$ to $\delta_{-}$ and
$\delta_{+}$. Then
${\rm
Mid}(v,\delta)$ is the primitive vector of the ray spanned by
$\pi_{\delta_{-}}^{-1}(v)+\pi_{\delta_{+}}^{-1}(v)
\in \delta.$
\begin{remark} By Lemma \ref{le: projections2} the local
projections $\pi_{\sigma}:N_\sigma\to N^\Gamma_\sigma$
commute with subdivisions and face restrictions therefore
the above notions are
coherent.
\end{remark}
\bigskip
\subsection{Stable vectors on simplicial
$K^*$-semicomplexes}\label{se: pi-stable}
In the section $\Sigma$ denotes an oriented
strictly convex $K^*$-semicomplex.
\begin{lemma} \label{le: tild1} Let $\sigma$ be a semicone
in $\Sigma$. Then
$G_\sigma$ acts on
$$\widetilde{X}_{\sigma^\Gamma}:=\widetilde{X}_\sigma/\Gamma_\sigma$$
\noindent and the morphism
$\widetilde{X}_{\sigma}\to
\widetilde{X}_{\sigma^\Gamma}$
is $G_\sigma$-equivariant. Moreover
$\widetilde{X}_{\sigma^\Gamma}=\widehat{X}_{\sigma^\Gamma\times{\operatorname{reg}}(\sigma)}$.
\end{lemma}
\noindent{\bf Proof.}
Follows from the fact that the action of $G^\sigma$
commutes with $\Gamma_\sigma$.\qed
\begin{lemma} \label{le: tild2} Let $\sigma$ be a
dependent cone in $\Sigma$.
Then \begin{enumerate}
\item
$\widetilde{X}_{\sigma_{-}}:= {X}_{\sigma_{-}}
\times_{{X}_{\sigma}}\widetilde{X}_{\sigma} \subset
\widetilde{X}_{\sigma}$ is $G_\sigma$-invariant.
\item
$\widetilde{X}_{{\pi_{\sigma_-}}({\sigma_{-}})}:=
{X}_{{\pi_{\sigma_-}}({\sigma_{-}})}
\times_{{X}_{{\pi(\sigma}})}
\widetilde{X}_{\pi({\sigma})}\to
\widetilde{X}_{{\pi}({\sigma})}$ is proper $G_\sigma$-equivariant.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof}
(1) By Lemma \ref{le: projection}, $\sigma_-=
\overline{\sigma}\setminus\operatorname{Star}(\sigma^+,\overline{\sigma})$.
By
Lemma \ref{le: weights}, the ideal of
$\overline{O}_{{\sigma}^+}=\widetilde{X}_{\sigma}\setminus
\widetilde{X}_{\sigma_{-}}$
is
generated by all
semi-invariant functions with positive weights (see also
\cite{Wlodarczyk2}, Example 2).
(2) Follows from Proposition \ref{pr: simple} since
${\operatorname{Vert}}(\pi_{\sigma_-}({\sigma_{-}}))={\operatorname{Vert}}(\pi({\sigma}))$ \qed
\begin{lemma} Let $\sigma$ be a cone of the maximal
dimension. Let $F$ denote the functional on
$\tau:=\sigma^{\vee}$ defined by some integral vector
$v\in N_\sigma$ which is not in $\sigma$. For
any integral $k$
set $\tau_k:=\{v\in
\tau\mid F(v)=k\}$. Then there are a finite number of vectors
$w_{k1},\ldots,w_{kl_k}\in \tau_k$ such that $$\tau_k=
\bigcup_{i=1,\ldots,l_k} (w_{ki}+\tau_0).$$
\end{lemma}
\noindent {\bf Proof.} We can replace $\tau$ by a regular
cone $\tau_0$ by considering the epimorphism
$p:\tau_0\to \tau$ mapping generators to generators. Then
$F$ defines a functional on $\tau_0$. Let
$\tau_{0k} :=\{v\in \tau_0\mid F(v)=k\}$. Let
$x_1,\ldots,x_m$ define the standard coordinates on $\tau\simeq
{\bf Z}^m_{\geq 0}$.
Without loss of generality we can write $F=n_1x_1+\ldots+
n_lx_l- n_{l+1}x_{l+1}-\ldots-n_{r}x_{r}$, where $l<r\leq m$ and
all $n_i>0$. Then $\tau_{0k}=(\tau_{0k}\cap ({\bf Z}^r_{\geq 0}\times \{0\}))+
(\{0\}\times {\bf Z}^{m-r})$, where $
\{0\}\times {\bf Z}^{m-r}\subset \tau_{00}$.
By a {k-\it minimal} vector we shall mean a vector
$v\in \tau_{0k}\cap ({\bf Z}^r_{\geq 0}\times \{0\})$ such that
there is no $w\in \tau_{00}\cap ({\bf Z}^r_{\geq 0}\times
\{0\})$ for which $v-w\in\tau_{0k}$.
It suffices to show that the number of $k$-minimal vectors
is finite.
Suppose
$x_{i_0}(v)\geq (n_1+\ldots +n_l)(n_{l+1}+\ldots+n_r)+|k|$ for some
$i_0 \leq l$ and let $x_{j_0}:=\max\{x_i(v)\mid l+1\leq i\leq
r\}$. Then $x_{j_0}\geq
\frac{n_1x_1+\ldots+n_lx_l}{n_1+\ldots+n_l}=
\frac{n_{l+1}x_{l+1}+\ldots+n_{r}x_{r}+k}{n_{l+1}+\ldots+n_r}\geq
(n_1+\ldots +n_l)(n_{l+1}+\ldots+n_r)/(n_{l+1}+\ldots+n_r)=n_1+\ldots +n_l$
Then $w:=(0,\ldots,n_{j_0},\ldots,0,\ldots,n_{i_0}\ldots 0)\in
\tau_{00}$, where the
$i_0$-th coordinate is $n_{j_0}$ and the $j_0$-th coordinate is $n_{i_0}$.
Thus $v-w\in\tau_{0k}$ and $v$ is not $k$-minimal.
Consequently, all minimal $k$-vectors satisfy
$\max\{x_i(v)\mid 1\leq i\leq r\}<
(n_1+\ldots +n_l)(n_{l+1}+\ldots +n_r)+|k|$. \qed
As a corollary we get
\begin{lemma} \label{le: decomposition2} Let $K[\widetilde{X}_{\sigma}]=\bigoplus
K[\widetilde{X}_{\sigma}]^k$ be the decomposition
according to weights with respect to the $\Gamma_\sigma=K^*$-action.
Then $K[\widetilde{X}_{\sigma}]^k$ is a finitely
generated $K[\widetilde{X}_{\sigma}]^0$-module.
\qed
\end{lemma}
\begin{lemma} \label{le: tild3} Let $\sigma$ be a
dependent cone in $\Sigma$.
Denote by ${j}_{\pi_{\sigma_-}}:
{X}_{\sigma_{-}}\to {X}_{\sigma_{-}}/\Gamma_\sigma$ and \\
$
\widetilde{j}_{\pi_{\sigma_-}}:
\widetilde{X}_{\sigma_{-}}\to \widetilde {X}_{\sigma_{-}}/\Gamma_{\sigma}$
the quotient morphisms.
Then \begin{enumerate}
\item $\widetilde{X}_{\sigma_{-}}/\Gamma_\sigma\simeq
\widetilde{X}_{\pi_{\sigma_-}(\sigma_{-})}$.
\item $\tilde{j}_{\pi_{\sigma_-}}: \widetilde{X}_{\sigma_{-}}\to
\widetilde{X}_{\pi_{\sigma_-}({\sigma_{-}})}$ is $G_\sigma$-equivariant.
\item For any $v\in \pi(\sigma_{-})$, ${j}_{\pi}^*
({\operatorname{val}}(v,{X}_{\pi_{\sigma_-}({\sigma_{-}})} ))
={\operatorname{val}}(\pi_{\sigma_-}^{-1}(v),{X}_{\sigma_{-}})$.
\item If
${\operatorname{val}}(v,\widetilde{X}_{\pi_{\sigma_-}({\sigma_{-}})} )$
is $G^\sigma$ invariant then
${\tilde{j}_{\pi_{\sigma_-}}}^*
({\operatorname{val}}(v,\widetilde{X}_{\pi_{\sigma_-}({\sigma_{-}})} ))
={\operatorname{val}}(\pi_{\sigma_-}^{-1}(v),\widetilde{X}_{\sigma_{-}})$
is $G^\sigma$-invariant.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof.}
(1) For $\tau\in\sigma_{-}$ set
$\widetilde{X}_{\tau}:={X}_{(\tau,N_\sigma)}\times_{{X}_{\sigma}}
\widetilde{X}_{\sigma}$,\,\,\,\\
$\widetilde{X}_{\pi_{\sigma_-}({\tau},N_\sigma)}
:={X}_{\pi_{\sigma_-}({\tau})}
\times_{{X}_{\pi({\sigma)}}}
\widetilde{X}_{{\pi({\sigma})}}$.
Then $K[\widetilde{X}_{\tau}]=
K[{X}_{(\tau,N_{\sigma})}]
\otimes_{K[\widetilde{X}_{\sigma}]}K[\widetilde{X}_{\sigma}]$.
The elements of this ring are finite sums
$\sum\chi_if_i$, where $f_i\in
K[\widetilde{X}_{\sigma}]$ and $\chi_i\in \tau^\vee$
is a character.Set
$R_1:=K[\widetilde{X}_{\tau}/\Gamma_\sigma]=
K[\widetilde{X}_{\tau}^{\Gamma_\sigma}]=
(K[{X}_{\tau}]
\otimes_{K[{X}_{\sigma}]}
K[\widetilde{X}_{\sigma}])^{\Gamma_\sigma}$,
$R_2:=K[\widetilde{X}_{\pi_{\sigma_-}({\tau})}]=
K[{X}_{\pi_{\sigma_-}({\tau})}]\otimes
_{K[{X}_{\pi({\sigma)}}]}
K[\widetilde{X}_{\pi({\sigma})}]\subset R_1$.
$R_2$
consists of elements
$\sum\chi_if_i$, where $\chi_i\in\tau^\vee$ and
$f_i\in K[\widetilde{X}_{\sigma}]$ have weight $0$.
$R_1$
consists of elements
$\sum\chi_if_i$, where $\chi_i\in\tau^\vee$ and
$f_i\in K[\widetilde{X}_{\sigma}]$ have
opposite weights. By Lemma \ref{le: decomposition2} there
is a finite
number of characters $\chi_{ij}\in
\sigma^{\vee}$ such that
$f_i=\sum\chi_{ij}f_{ij}$, where $f_{ij}$ have weight zero.
Finally,
$\sum\chi_if_i=\sum_i\chi_i(\sum_j\chi_{ij}f_{ij})=
\sum_i\sum_j(\chi_i\chi_{ij})f_{ij}\in R_2$.
(2) Follows from (1).
(3) By Lemma \ref{le: projection} the projection $\pi:=\pi_{\sigma_{-}}$ maps
cones of $\sigma_{-}$ isomorphically onto cones of
$\pi(\sigma_-)$.
By \cite{KKMS}, Ch. I Th. 9, the sheaf of
ideals ${\mathcal I}:={\mathcal I}_{{\operatorname{val}}(\pi(v)),d}$ on
$X_{\pi_{\sigma_{-}}(\sigma_-)}$ corresponds to the strictly
convex piecevise linear function ${\operatorname{ord}}({\mathcal I})$. By definition
and Lemma \ref{le: center}, ${\operatorname{ord}}({\mathcal I})$ equals $0$ on
all cones not containing $v$.
Let $\tau\in \sigma_-$ be a cone containing $v$.
By the proof of Lemma \ref{le: blow-up valuation}, if $d$ is
sufficiently divisible then for any face
$\delta$ of $\tau$ such that $\delta$ does not contain $v$,
there is an integral vector $m_{\pi(\delta),\pi(\tau)}\in \pi(\tau)^\vee$ for which$ (m_{\pi(\delta),\pi(\tau)},\pi(v))=d$ , $m_{\pi(\delta),\pi(\tau)}$ is
$0$ on $\pi(\delta)$ and
${\operatorname{ord}}({\mathcal I})$ equals $m_{\pi(\delta),\pi(\tau)}$ on $\pi(\delta+\langle
v\rangle)$.
In particular ${\mathcal I}$ is generated on each cone $X_{\pi(\tau)}$ by
$m_{\pi(\delta),\pi(\tau)}$, where of $\delta$ is a face of
$\tau$ that does not contain
$v$. Let $m_{\delta}=m_{\pi(\delta)}\circ\pi\in \tau^\vee$ be the
induced functional on $M_\tau$.
Thus the ideal $\pi^*({\mathcal I})$ is generated on each cone $X_{\tau}$, where
$\tau\in\sigma_-$ contains $v$, by
$m_{\delta,\tau}$, where $\delta$ doesn not contain $v$.
Then $(m_{\delta,\tau},v)=d$ , $m_{\delta}$ is
$0$ on $\delta$ and
${\operatorname{ord}}(\pi^*({\mathcal I}))$ equals $m_{\delta,\tau}$ on $\delta+\langle
v\rangle$. This gives by the proof of Lemma \ref{le: blow-up valuation},
$\pi^*({\mathcal I}_{{\operatorname{val}}(\pi(v)),d})={\mathcal I}_{{\operatorname{val}}(v),d}$.
(4) Follows from (2) and (3). \qed
\begin{lemma} \label{le: inclu}
Let $\tau\leq\sigma$ be faces of $\Sigma$. Then either the inclusion
$\imath^{\sigma \Gamma}_\tau$ maps $\tau^\Gamma$ to a face of
$\sigma^\Gamma$, or $\tau$ is independent, $\sigma$ is dependent, and
$\tau^\Gamma$ is mapped isomorphically onto the face of $\pi(\sigma_{-})$ (or
$\pi(\sigma_{+})$).
\end{lemma}
\noindent{\bf Proof.} If both faces $\tau$ and $\sigma$ are independent then
it follows that $\pi_\sigma$ and $\pi_\tau$ are linear isomorphisms. Then
$\imath^{\sigma \Gamma}_\tau(\tau^\Gamma)$ is a face of $\sigma^\Gamma$.
If $\tau$ and $\sigma$ are both dependent then the kernel
of the projection
$\pi_\sigma$ is contained in $\operatorname{lin}(\tau)$. Then $\pi(\tau)$
is a face of $\pi(\sigma)$.
If $\tau$ is an independent face of a dependent cone $\sigma$ then $\tau$ is a
face of $\sigma_-$ or $\sigma_+$. Both fans consist of independent faces of
$\sigma$ and project onto the subdivisions $\pi(\sigma_{-})$ and
$\pi(\sigma_{+})$ of $\pi(\sigma)$.
\qed
\begin{definition}
A vector $v\in {\rm
int}({\sigma^\Gamma})$, where $\sigma\in \Sigma$,
is $\Sigma$-{\it stable} if for any $\tau\geq
\sigma$ the corresponding
valuation ${\rm val}(\imath^{\tau \Gamma}_{\sigma}(v))$ on
$\widetilde{X}_{\tau^\Gamma}$
is $G^\tau$-invariant. A vector $v\in \sigma^\Gamma$ is $\Sigma$-{\it
stable} if there is a $\Sigma$-stable vector $v_0\in {\rm
int}({\sigma^\Gamma_{0}})$, where $\sigma_0\leq \sigma$, for which
$v=\imath^{\sigma\Gamma}_{\sigma_0}(v_0)$.
\end{definition}
\begin{proposition} \label{pr: G-stab} Let $\sigma$ be a
semicone in $\Sigma$.
A vector $v\in{\operatorname{int}}(\sigma)$ is
$\Sigma$-stable if $\pi_\sigma(v)\in \sigma^\Gamma$
is $\Sigma$-stable.
\end{proposition}
\noindent{\bf Proof.}
A stable vector $v\in
{\rm int}(\sigma)$ determines a $G^\tau$-invariant
valuation on any $\widetilde{X}_{\tau}$ for $\tau\geq
\sigma$. Consequently, it
determines a $G^\tau$-invariant
valuation on any $\widetilde{X}_{\tau}/\Gamma_{\tau}$ and
finally $\pi_{\sigma}(v)$ is $\Sigma$-stable.
Now
let $v\in {\rm int}(\sigma)$ be an integral vector
such that $\pi_{\sigma}(v)\in
{\operatorname{int}}(\sigma^\Gamma)$ is stable. Then
${\operatorname{val}}{(\pi_{\sigma}(v))}$ is $G^\tau$-invariant on
$\widetilde{X}_{\tau}/\Gamma_{\tau}$.
We have to prove that $v$ is stable, or equivalently, that for any $\tau\geq \sigma$,
${\rm val}(v)$ is $G^\tau$-invariant on
$\widetilde{X}_{\tau}$. Consider two cases:
(1) $\tau\in {\operatorname{Ind}}(\Sigma)$.
The morphism $\tilde{j}_{\pi_{\tau}}: \widetilde{X}_{\tau}\to
\widetilde{X}_{\tau}/\Gamma_\tau=\widetilde{X}_{\tau^\Gamma}$
is $G^\tau$-equivariant.
The valuation
$\tilde{j}_{\pi_{\tau}*}({\operatorname{val}}{(v)})={\operatorname{val}}{(\pi_{\tau}(v))}$ is
$G^\tau$-invariant on
$\widetilde{X}_{\tau}/\Gamma_\tau$. Thus for
any $g\in G^\tau$, $g_*\tilde{j}_{\pi_\tau *}({\operatorname{val}}{(v)})=
\tilde{j}_{{\pi_{\tau}}*}g_*
({\operatorname{val}}{(v)})=\tilde{j}_{{\pi_{\sigma}}*}({\operatorname{val}}{(v)})$.
This means that ${\operatorname{val}}(v)$ and $g_*({\operatorname{val}}(v))$ define the same
functional on the lattice of characters $M^\Gamma_\tau$ and
consequently $M_\tau$ (since $M^{\bf
Q}_\tau=(M^\Gamma_\tau)^{\bf Q}$ ). This shows by Lemma \ref{le: monomial} that
$g_*({\operatorname{val}}(v))\geq {\operatorname{val}}(v)$ and finally by Lemma \ref{le:
monomial2} we conclude $g_*({\operatorname{val}}(v)) = {\operatorname{val}}(v)$.
(2) $\tau\in {\operatorname{Dep}}(\Sigma)$. By Lemma \ref{le: tild2}(2),
$\widetilde{ i_{/K^*}}:\widetilde X_{\pi(\tau_{-})} \to
\widetilde{X}_{\pi(\tau)}$ is a
$G^\tau$-equivariant proper birational morphism.
Then the valuation ${\rm val}(v,\widetilde{X}_{\pi(\tau_-)})=
(\widetilde{i_{/K^*}}_*^{-1}({\rm
val}(v,\widetilde{X}_{\pi(\tau)}))$ is $G^\tau$-invariant.
By Lemma \ref{le: tild3}(4),
$\widetilde{j}_{\pi_{\tau_{-}}}^*({\rm
val}(\pi(v),\widetilde{X}_{\pi(\tau_-)}))= {\rm
val}(\pi_{\tau_{-}}^{-1}\pi(v),\widetilde{X}_{\tau_-})$ is
$G^\tau$-equivariant.
We get $\pi_{\tau_{-}}^{-1}(\pi(v))\in {\rm Inv}(\tau)$.
Analogously $\pi_{\tau_{+}}^{-1}(\pi(v))\in
{\rm Inv}(\tau)$. By Lemma \ref{le: convex} this gives
$v\in \langle\pi_{\tau_{-}}^{-1}(\pi(v)),
\pi_{\tau_{+}}^{-1}(\pi(v))\rangle\subset
{\rm Inv}(\tau)$.
\qed
\begin{proposition} \label{pr: G-stab 2} Let $\sigma \in
{\operatorname{Ind}}(\Sigma)$ be an
independent face of $\Sigma$. Then
\begin{enumerate}
\item All vectors in $ \overline{\rm
par}(\sigma^\Gamma)\cap {\operatorname{int}}(\sigma^\Gamma)$
are $\Sigma$-stable.
\item All vectors in ${\rm
par}(\sigma^\Gamma)$ are $\Sigma$-stable.
\end{enumerate}
\end{proposition}
\noindent {\bf Proof}
(1) Let $v\in
\overline{\rm par}(\sigma^\Gamma)\cap {\rm
int}(\sigma^\Gamma)$. We have to prove that it defines a
$G^\tau$-invariant valuation on any
$\widetilde{X}_{\pi_\tau(\tau)}$ for $\tau\geq \sigma$.
By Lemma \ref{le: inclu} either $\sigma^\Gamma$ is a face of $\tau^\Gamma$ or
$\tau\in {\operatorname{Dep}}(\Sigma)$ and $\sigma^\Gamma$ is a face of $\pi_\tau(\tau_-)$ (or
$\pi_\tau(\tau_+)$).
In the first case $\sigma^\Gamma$ is a is a
$G^\tau$-invariant face of
$\tau^\Gamma$. By Lemma
\ref{le: parr}(2), ${\operatorname{val}}(v)$ defines a $G^\tau$-invariant
valuation on
$\widetilde{X}_{\tau^\Gamma}$. In the second case $\sigma$ is
a $G^\tau$-invariant face of $\tau_-$ (or $\tau_+$). Consequently,
the closure of the
orbit $O_{\sigma^\Gamma}$ in $\widetilde{X}_{\pi_\tau(\tau_-)}$
is $G_\tau$-invariant.
Let $\widetilde{i_{/K^*}}:
\widetilde{X}_{\pi_\tau(\tau_{-})} \longrightarrow
\widetilde{X}_{\pi_\tau(\tau)}$ be
the $G_\tau$-equivariant morphism defined by the
subdivision $\pi_\tau(\tau_{-})$ of
$\pi_\tau(\tau)$. Then by Lemma
\ref{le: parr}(2), $\overline{\rm
par}(\sigma^\Gamma)\cap {\operatorname{int}}(\sigma^\Gamma)$ is contained
${\operatorname{Inv}}(\pi_\tau(\tau))$.
(2) Let $v\in
{\rm par}(\pi_\sigma(\sigma))$. Then $v\in \overline{\rm par}
(\pi_\sigma(\tau))\cap {\rm
int}(\tau)$ for some $\Gamma$-indecomposable
face $\tau$ of $\sigma$. We apply (1).
\begin{lemma} \label{le: centers2}
Let
$\sigma=\langle v_1,\ldots,v_k\rangle$ be a
circuit in $\Sigma$ with the unique relation
$\sum\alpha_iw_i=0,$
where all $\alpha_i\neq 0$ and $w_i={\operatorname{prim}}(\pi_\sigma(v_i))$.
Then
${\rm Ctr}_-(\sigma):=\sum_{\alpha_i<0}w_i$ and
${\rm Ctr}_+(\sigma):=\sum_{\alpha_i>0}w_i$
are $\Sigma$-stable.
\end{lemma}
\noindent{\bf Proof.} We have to show that
for any $\tau\geq \sigma$ the valuation ${\operatorname{val}}({\rm
Ctr}_-(\sigma),\widetilde{X}_{\tau^\Gamma})$ is
$G_\tau$-invariant on $\widetilde{X}_{\tau^\Gamma}$.
By Lemmas \ref{le: weights2} and \ref{le: tild2}(1),
the face $\sigma^-=\langle
v_i\rangle_{\alpha_i>0}\in \sigma_{-}\subset \tau_{-}$
corresponds to the $G^\tau$-invariant closed subscheme of
$\widetilde{X}_{\tau_-}\subset \widetilde{X}_{\tau}$.
Therefore the orbit closure
scheme $\overline{O_{\pi_\sigma(\sigma^-)}}\subset
\widetilde{X}_{\pi_\tau(\tau_-)}$ is $G^\tau$-invariant.
By Lemma \ref{le: parr}(2), applied to the subdivision
${\pi_\tau(\tau_-)}$ of ${\pi_\tau(\tau)}$,
$v:={\rm Ctr}_-(\sigma)$ defines a $G^\tau$-invariant valuation on $
\widetilde{X}_{\pi_\tau(\tau)}$.
\qed
\begin{lemma} \label{le: centers} Let $\Delta$ be a
canonical subdivision of $\Sigma$.
\begin{enumerate}
\item If
$v\in {\rm par}(\pi(\delta))$, where $\delta \in
\Delta^\sigma$, is an independent cone, then
${\rm Mid}(v,\delta)$ is $\Sigma$-stable.
\item Let
$\delta\in\Delta^\sigma$ be a
circuit.
Then
${\rm Mid}({\rm Ctr}_{+}(\delta),\delta)$,
${\rm Mid}({\rm Ctr}_{-}(\delta),\delta)$
are $\Sigma$-stable.
\end{enumerate}
\end{lemma}
\noindent{\bf Proof}
(1) There is a $\Gamma$-indecomposable cone $\delta'\preceq\delta$
such that $v\in {\rm par}(\pi(\delta'))$. Then $\delta'\in \Delta^\sigma_{\operatorname{stab}}$.
We apply Lemma \ref{le: centers2} and Proposition \ref{pr: G-stab} to $\delta'\in
\Sigma'=\Delta_{\operatorname{stab}}$ to deduce that
${\rm Mid}(v,\delta)={\rm Mid}(v,\delta')$ is
$\Sigma'$-stable. By Proposition \ref{pr: canonical orientation}(3), it is
$\Sigma$-stable.
(2) The circuit $\delta\in \Delta^\sigma$ is
$\Gamma$-indecomposable. Thus $\delta\in \Delta_{\operatorname{stab}}$.
We apply Lemma \ref{le: centers2} and Proposition \ref{pr: G-stab} to $\delta\in
\Sigma'=\Delta_{\operatorname{stab}}$ to infer that ${\rm Mid}({\rm Ctr}_{+}(\delta),\delta)$,
${\rm Mid}({\rm Ctr}_{-}(\delta),\delta)$
are $\Sigma'$-stable. Consequently, by Proposition \ref{pr:
canonical orientation}(3) they are
$\Sigma$-stable. \qed
\subsection{The $\pi$-desingularization Lemma of Morelli}
\begin{definition} An independent cone $\delta \in
\Delta^\sigma$ is $\pi$-{\it nonsingular} if
$\pi_{\sigma}(\delta)$ is regular. A subdivision $\Delta$ is
$\pi$-{\it nonsingular} if all independent cones in $\Delta^\sigma$,
where $\sigma\in\Sigma$, are $\pi$-{\it nonsingular}.
\end{definition}
\begin{definition}(Morelli \cite{Morelli1}, \cite{Morelli2},
\cite{Abramovich-Matsuki-Rashid})\label{de: Morelli}
Let $\pi:N^{{\bf Q}+}:=N^{\bf Q}\oplus {\bf Q}\to N^{\bf Q}$ be the
natural projection and $v=(\{0\}\times 1)\in N^{\bf Q}$.
A fan $\Sigma$ in $N^{{\bf Q}+}$ is
called a {\it polyhedral cobordism} or simply a cobordism if
\begin{enumerate}
\item For any cone $\sigma\in\Sigma$ the image $\pi(\sigma)$ is
strictly convex (contains no line).
\item The sets of cones
$$\Sigma_+: =\{\sigma \in\Sigma\mid \mbox{there exists
} \quad \epsilon>0, \mbox{such that } \sigma+\epsilon\cdot v\not\in
|\Sigma| \}$$ \noindent and
$$\Sigma_-=\{\sigma \in\Sigma\mid \quad\mbox{there exists} \quad
\epsilon>0, \mbox{such that}\sigma+\epsilon\cdot
v\not\in |\Sigma|\}$$ are
subfans of $\Sigma$ and
$\pi(\Sigma_-):=\{\pi(\tau)\mid\tau\in \Sigma_-\}$ and
$\pi(\Sigma_+):=\{\pi(\tau)\mid\tau\in \Sigma_+\}$
are fans in $N^{\bf Q}$.
\end{enumerate}
\end{definition}
\begin{lemma}(Morelli \cite{Morelli1}, \cite{Morelli2},
\cite{Abramovich-Matsuki-Rashid}) \label{le: Morelli}
Let $\Sigma$ be a simplicial cobordism in $N^{{\bf Q}+}$. Then there exists a
simplicial cobordism $\Delta$ obtained from $\Sigma$ by a sequence of star
subdivisions such that $\Delta$ is $\pi$-nonsingular.
Moreover, the sequence can be taken so that any independent
and already $\pi$-nonsingular face of $\Sigma$ remains
unaffected during the process. Moreover
all centers of star subdivisions are of
the form (1), (2) from Lemma \ref{le: centers}.$\qed$
\end{lemma}
A simple corollary of the above lemma is
\noindent \begin{lemma} \label{le: pi-lemma}
Let $\Sigma$
be a simplicial strictly convex $K^*$-semicomplex. Then there exists a
canonical subdivision $\Delta$ of
$\Sigma$ which is a sequence of star
subdivisions such that $\Delta$ is $\pi$-nonsingular.
Moreover, the sequence can be taken so that any independent
and already $\pi$-nonsingular face of $\Sigma$ remains
unaffected during the process. Moreover all centers are of
the form (1), (2) from Lemma \ref{le: centers} and
therefore are $\Sigma$-stable.
\end{lemma}
\begin{lemma} \label{le: normal} (\cite{Wlodarczyk1},Lemma
10, \cite{Morelli1}) Let $w_1,\ldots,w_{k+1}$ be integral vectors in
${\bf Z}^{k}\subset {\bf Q}^{k}$
which are not contained in a proper vector subspace of ${\bf Q}^{k}$.
Then $$\sum_{i=1}^n(-1)^i\det(w_1,\ldots,\check{w_i},\ldots,w_{k+1})\cdot w_i=0 $$ is the
unique (up to proportionality) linear relation between $w_1,\ldots,w_{k+1}$.
\end{lemma}
\noindent{\bf Proof.}
Let $v:=
\sum_{i=1}^n (-1)^i\det(w_1,\ldots,\check{w_i},\ldots,w_{k+1})\cdot w_i$.
Then for any $i<j$,
\[\begin{array}{rc}
&\det(w_1,\ldots,\check{w_i},\ldots,\check{w_j},\ldots,w_{k+1},v)=
\det(w_1,\ldots,\check{w_i},\ldots,w_{k+1})\cdot
\det(w_1,\ldots,\check{w_i},\ldots,\check{w_j},\ldots,w_{k+1},w_i)+
\\
&\det(w_1,\ldots,\check{w_j},\ldots,w_{k+1})\cdot
\det(w_1,\ldots,\check{w_i},\ldots,\check{w_j},\ldots,w_{k+1},w_j)=\\
&(-1)^i(-1)^{k-i}\det(w_1,\ldots,\check{w_j},\ldots,w_{k+1})
\cdot\det(w_1,\ldots,\check{w_i},\ldots,w_{k+1})+\\
&(-1)^j(-1)^{k-j+1}\det(w_1,\ldots,\check{w_i},\ldots,w_{k+1})
\cdot\det(w_1,\ldots,\check{w_j},\ldots,w_{k+1})=0
\end{array}\]
Therefore $v\in \bigcap_{i,j}
\operatorname{lin}\{w_1,\ldots,\check{w_i},\ldots,\check{w_j},\ldots,w_{k+1}\}=\{0\}$.\qed
\noindent \begin{lemma} \label{le: centers3} If $\Sigma$ is a simplicial
cobordism which is $\pi$-nonsingular and $\sigma\in \Sigma$
is a circuit then \\${\rm Mid}({\rm Ctr}_{+}(\sigma),\sigma)=
{\rm Mid}({\rm Ctr}_{-}(\sigma),\sigma)$ and
${\rm Mid}({\rm Ctr}_{+}(\sigma),\sigma)\cdot\Sigma$ is $\pi$-nonsingular.
\end{lemma}
\noindent{\bf Proof.}
Let $\sigma=\langle v_1,\ldots,v_k\rangle$ be a
curcuit and $w_i:={\operatorname{prim}}(\pi(\langle v_i \rangle))$..
\\Then by Lemma \ref{le: normal}, in the relation the unique relation
$\sum\alpha_iw_i=0$ all
$\alpha_i=\det[w_1,\ldots,\check{w}_i,\ldots,
w_k]=\pm 1$
and projections of ''new independent faces''are regular
since they are obtained by
regular star subdivisions applied to projections of "old"
independent faces. \qed
\noindent{\bf Proof of Lemma \ref{le: pi-lemma}.}
First note that
the local projections $\pi_{\sigma}: N_\sigma\to
N^\Gamma_\sigma$ glue together
and commute with subdivisions by Lemma \ref{le: projections2} and therefore the notion of
$\pi$-nonsingularity is coherent.
Let $\sigma_1,\ldots,\sigma_k$ denote all the faces of $\Sigma$.
We shall construct by induction on $i$ a canonical
$\Gamma$-subdivision $\Delta_i$ of $\Sigma$ such that the
fan ${\Delta_i^{\sigma_j}}$ is $\pi$-nonsingular for all
$j\leq i$. Suppose $\Delta_i$ is
already constructed. By Lemma \ref{le: Morelli}
there exist centers $v_1,\ldots,v_k$ of the appropriate type for the cobordism
$\Delta^{\sigma_{i+1}}_i$ in $N_{\sigma_{i+1}}$ such that
$\langle v_k \rangle\cdot\ldots\cdot\langle v_1
\rangle\cdot\Delta^{\sigma_{i+1}}_i$ is
$\pi$-nonsingular. It follows from Lemma
\ref{le: centers} that all centers
in the $\pi$-desingularization process are $\Sigma$-stable, hence by
Proposition \ref{pr: blow-ups}, $\Delta_{i+1}:
=\langle v_k \rangle\cdot\ldots\cdot\langle v_1
\rangle\cdot\Delta_{i}$ is canonical. By construction
$\Delta^{\sigma_{i+1}}_{i+1}$ is $\pi$-nonsingular. Note that
$\Delta^{\sigma_{j}}_{i+1}$ for $j<i+1$ is obtained by star
subdivisions of the $\pi$-nonsingular
cobordisms $\Delta^{\sigma_j}_{i}$ at
centers of type ${\operatorname{Mid}}({\operatorname{Ctr}}_\pm(\delta),\delta)$ only and hence by Lemma \ref{le:
centers3} it remains $\pi$-nonsingular. \qed
\bigskip
For completeness we give a simple proof of Lemma
\ref{le: Morelli}. The original
proof of Morelli used different centers of star subdivisions but finally it
was modified by the author
(\cite{Morelli1},\cite{Morelli2}).
The complete published proof of the lemma is given in
\cite{Abramovich-Matsuki-Rashid} (\cite{Morelli2}).
The present proof is based upon the algorithm developed
in \cite{Wlodarczyk1}. All algorithms in their final versions use the same centers of subdivisions and are
closely related.
\noindent{\bf Proof of Lemma \ref{le: Morelli}.}
\begin{definition} Let $\delta=\langle v_1,\ldots,v_k\rangle$ be a
dependent cone and $w_i:={\operatorname{prim}}(\pi(\langle v_i \rangle))$.
Then we shall call the relation $\sum_{i=1}^k r_iw_i=0$ {\it
normal} if it is positively normalized and
$|r_i|=|\det(w_1,\ldots,\check{w_i},\ldots,w_{k})|$ for $i=1,\ldots,k$.
An independent cone $\sigma$ is called an {\it $n$-cone}
if $|\det(\sigma)|=n$. A dependent cone $\sigma$
is called an {\it $n$-cone} if one of its independent faces
is an $n$-cone and the others are $m$-cones, where $m\leq n$.
\end{definition}
\begin{lemma} \label{le: normal2} Let $\Sigma$ be a
simplicial coborism and $\delta=\langle v_1,\ldots,v_k\rangle$ be a
maximal dependent cone in $\Sigma$. Set
$w_i:={\operatorname{prim}}(\pi(\langle v_i \rangle))$. Let
$$\sum_i r_iw_i=0 \eqno(0),$$ \noindent where
$|r_i|=|\det(w_1,\ldots,\check{w_i},\ldots,w_{k})|$,
be its normal relation (up to sign).
\noindent (1) Let $v={\operatorname{Mid}}({\operatorname{Ctr}}_+(\delta),\delta)\in{\operatorname{int}}
\langle v_i\mid r_i\neq 0\rangle$. Let $m_w\geq 1$ be the integer
such that the vector
$w=\frac{1}{m_w}(w_1+\ldots+w_k)$ is primitive. Then the maximal
dependent cones in $\langle v \rangle\cdot\delta$ are of the
form $\delta_{i}=\langle
v_1,\ldots,\check{v}_{i},\ldots,v_k,v\rangle$.
\noindent 1a. Let $r_{i_0}>0$. Then for the maximal dependent cone
$\delta_{i_0}=\langle
v_1,\ldots,\check{v_{i_0}},\ldots,v_k,v\rangle$ in
$\langle v \rangle\cdot\delta$, the normal relation
is given (up to sign) by $$\sum_{r_i> 0, i\neq i_0}\frac{r_i-r_{i_0}}{m_w}w_i+
\sum_{r_i< 0} \frac{r_i}{m_w}w_i+r_{i_0}w=0. \eqno(1a)$$
\noindent 1b. Let $r_{i_0}<0$. Then for the maximal dependent cone
$\delta_{i_0}=\langle
v_1,\ldots,\check{v}_{i_0},\ldots,v_k,v\rangle$ in $\langle
v \rangle\cdot\delta$, the normal relation
is given (up to sign) by $$\sum_{r_i< 0}-\frac{r_{i_0}}{m_w}w_i+
r_{i_0}w=0. \eqno(1b)$$
(2) Let $\sigma=\langle v_i\mid i\in I\rangle$ be a
codefinite face of
$\delta$. For simplicity assume that $r_i\geq 0$ for $i\in I$.
Let
$w=\sum_{i\in I}\alpha_iw_i\in
{\operatorname{par}}(\pi(\sigma))\cap{\operatorname{int}}(\pi(\sigma))$ and $v={\operatorname{Mid}}(w,\sigma)\in{\operatorname{int}}
(\sigma)$. Then the maximal dependent cones in
$\langle v \rangle\cdot\delta$
are of the form $\delta_{i_0}=
\langle v_1,\ldots,\check{v}_{i_0},\ldots,v_k,v\rangle$,
where $i_0\in I$.
2a. Let $i_0\in I$ and $r_{i_0}>0$. Then for the maximal
dependent cone $\delta_{i_0}=
\langle v_1,\ldots,\check{v}_{i_0},\ldots,v_k,v\rangle$ in
$\langle v \rangle\cdot\delta$,
the normal relation
is given (up to sign) by
$$\sum_{i\in I\setminus \{i_0\},r_i>0}
(\alpha_{i_0}r_{i}-
\alpha_{i}r_{i_0})w_{i}+\sum_{i\not\in I, r_i>0}
\alpha_{i_0}r_{i}w_i+\sum_{
i\in I,r_i=0} -\alpha_{i}r_{i_0}w_{i}+
\sum_{r_i<0}\alpha_{i_0} r_iw_{i}+r_{i_0}w=0. \eqno(2a)$$
2b. Let $i_0\in I$ and $r_{i_0}=0$. For the maximal dependent cone
$\delta_{i_0}=\langle
v_1,\ldots,\check{v}_{i_0},\ldots,v_k,v\rangle$,
the normal relation
is given (up to sign) by
$$\sum_{i\in I\setminus\{i_0\}} \,\,\alpha_{i_0}r_iw_i+0w_{i_0}=0. \eqno(2b)$$
\end{lemma}
{\bf Proof.}
It is straightforward to see that the above equalities hold. We only
need to show that the relations considered are normal. For that
it suffices to show that one of the coefficients is equal
(up to sign) to the corresponding determinant
Comparing the coefficients of $w$ in the above
relations with the normal
relations from Lemma \ref{le: normal} we get
1a. $\det(w_1,\ldots,\check{w}_{i_0},\ldots,w_k)=
r_{i_0}$.
1b. The coefficient of $w$ is equal to
$\det(w_1,\ldots,\check{w}_{i_0},\ldots,w_k))=r_{i_0}$,
2a. The coefficient of $w$ is equal to
$\det(w_1,\ldots,\check{w}_{i_0},\ldots,w_k)=r_{i_0}$.
2b. The coefficient of $w_i$, where $r_i>0$, is equal to \\
$\det(w_1,\ldots,
\check{w}_{i_0},\ldots,\check{w_{i}},\cdots,w_k,w)=\alpha_{i_0}
\det(w_1,\ldots,
\check{w}_{i_0},\ldots,\check{w_{i}},\cdots,w_k,w_{i_0})=\\
(-1)^{k-i_0}\alpha_{i_0}
\det(w_1,\ldots,\check{w}_{i},\cdots,w_k,w)=
(-1)^{k-i_0} \alpha_{i_0}r_i$.
\qed
Let $\delta=\langle v_1,\ldots,v_k\rangle$ be a dependent cone. Let
$\sum\,\, r_iw_i=0$ be its normal relation.
We shall consider the following kinds of dependent cones.
Type $I$: the cones which are {\it pointing down}: ${\operatorname{Card}}\{r_i\mid
r_i<0\}=1$.
and the cones {\it pointing up}: ${\operatorname{Card}}\{r_i\mid r_i>0\}=1$.
Type $I(n,n)$: the cones of type $I$ for which
$\max\{|r_i|\mid r_i>0\}=\max\{|r_i|\mid r_i<0\}=n$.
Type $I(n,*)$: the cones pointing down for which
$\max\{|r_i|\mid r_i>0\}<n$ and $\max\{|r_i|\mid r_i<0\}=n$
and the cones pointing up for which
$\max\{|r_i|\mid r_i<0\}<n$ and $\max\{|r_i|\mid r_i>0\}=n$.
Type $I(*,n)$: the cones pointing down for which
$\max\{|r_i|\mid r_i<0\}<n$ and $\max\{|r_i|\mid r_i>0\}=n$,
and the cones pointing up for which
$\max\{|r_i|\mid r_i>0\}<n$ and $\max\{|r_i|\mid r_i<0\}=n$.
\bigskip
Type $II$: the cones which are neither pointing down nor
pointing up: $ {\operatorname{Card}}\{r_i\mid
r_i>0\}>1$ and $ {\operatorname{Card}}\{r_i\mid
r_i<0\}>1$.
Type $II(n,n)$: the cones of type $II$ for which
$\max\{|r_i|\mid r_i>0\}=\max\{|r_i|\mid r_i<0\}=n$.
Type $II(n,*)$: the cones of type $II$ for which
$\max\{|r_i|\mid r_i\neq 0\}=n$ but
$\max\{|r_i|\mid r_i>0\}<n$ or $\max\{|r_i|\mid r_i<0\}<n$.
\bigskip
Type $III$: the cones which are pointing down and pointing up
at the same time:\\ ${\operatorname{Card}}\{r_i\mid
r_i>0\}={\operatorname{Card}}\{r_i\mid
r_i<0\}=1$.
Type $III(n,n)$: the cones of type $III$ for which
$\max\{|r_i|\mid r_i>0\}=\max\{|r_i|\mid r_i<0\}=n$.
\bigskip
Types $I(*,*)$, $II(*,*)$, $III(*,*)$ the cones of type $I$, $II$,
$III$, respectively, for which $\max\{|r_i|\mid r_i\neq 0\}<n$.
\bigskip
Denote
by $n$ the maximum of the determinants of the projections
of independent cones in the simplicial cobordism $\Sigma$.
The $\pi$-desingularization algorithm consists of
eliminating all dependent $n$-cones according to the above classification.
All normal relations below are considered up to sign.
\bigskip
{\bf Step 1}. {\it Eliminating all cones of type $II(n,n)$}
Let $\delta$ be a maximal dependent cone of type $II(n,n)$
in $\Sigma$ with normal
relation (0) (see Lemma \ref{le: normal2}). By Lemma \ref{le:
normal2} after the star
subdivision at ${\operatorname{Mid}}({\operatorname{Ctr}}_+,\delta)$ all new dependent
$\delta'=\delta_{i_0}$ cones have normal relations of type
(1a) if $r_{i_0}> 0$ or (1b) if $r_{i_0}> 0$.
If $r_{i_0}> 0$ in the relation (1a) the absolute values of
the coefficients satisfy $\frac{1}{m_w}|r_i-r_{i_0}|<n$, the absolute values of
the other coefficients can be $n$ only for some $r_i<0$
and $r_{i_0}$. If $r_{i_0}<n$ then there is no
positive coefficient in (1a) which is equal to $n$.
Hence $\delta'$ can be of type
$I(n,*)$, $I(*,n)$, $I(*,*)$, $II(n,*)$ or $II(*,n)$. If $r_{i_0}=n$ then there is only
one positive coefficient (of $w$) in (1a). The other
coefficients are
$\frac{r_i-r_{i_0}}{m_w}=\frac{r_i-n}{m_w} \leq 0$ and $r_i<0$. Thus
$\delta'$ is of the form $I(n,*)$ or $I(n,n)$.
If $\delta'=\delta_{i_0}$ for $r_{i_0}<0$
then the normal relation is of the form (1b). If
$|r_{i_0}|=n$ then $\delta'$
is of the form $I(n,n)$. If $|r_{i_0}|<n$ then $\delta'$
is of the form $I(*,*)$.
By taking the star subdivision we eliminate dependent cones
in $\operatorname{Star}(\delta_+,\Sigma)$, with the normal relation
proportional to (0). All the dependent $n$-cones
we create are either of type $I$ or $II(n,*)$.
\bigskip
{\bf Step 2}. {\it Eliminating all cones of type $I(n,n)$}.
Let $\delta$ be a maximal dependent cone of type $I(n,n)$.
Without loss of generality we may assume that $\delta$ is
pointing down (one negative ray). Then $\delta^+$ is of
codimension $1$ and
$\det(\pi(\delta^+))=n$.
Find
$v_+\in {\operatorname{par}}(\pi(\delta^+))$. Then $v_+\in {\operatorname{int}}(\pi(\sigma_v))$, where
$\sigma_v$ is a face of $\delta^+$. Thus $\sigma_v$ is
independent.
{\bf Step 2a}. {\it Making $\sigma_v$ codefinite with respect to all
dependent cones in $\operatorname{Star}(\sigma_v,\Sigma)$}.
By applying certain star subdivisions we eliminate all
cones for which $\sigma_v$ is not codefinite.
Let $\delta'\in\operatorname{Star}(\sigma_v,\Sigma)$ be any dependent cone
for which $\sigma_v$ is not codefinite. Note that $\delta'$
is not of type $III$ since in this case the circuit consists
of one positive and one negative ray and the noncodefinite face $\sigma_v$
containing positive and negative rays would contain a
circuit, which is impossible for an independent cone.
Therefore $\delta'$ is either of type $I$ or $II$.
Suppose first that $\delta'=\langle
v_1,\ldots,v_{k}\rangle$ is of type $I(n,n)$
or
$I(*,n)$. Without loss of generality we may assume that
there is only one coefficient $r_{i_1}>0$ (equal to $n$)
in the normal relation
(0) from Lemma \ref{le: normal2}(1).
Since $\sigma_v$ is not codefinite
with respect to $\delta'$, we have $v_{i_1}\in \sigma_v$. Note
that ${\operatorname{Ctr}}_+(\delta')={\operatorname{prim}}(\pi(\langle v_{i_1}\rangle))$ and
for $v:={\operatorname{Mid}}({\operatorname{Ctr}}_+(\delta'),\delta')$ we have
${\operatorname{prim}}(\pi(\langle v_{i_1}\rangle))={\operatorname{prim}}(\pi(\langle
v_{i_1}\rangle))$. Let $w_i:={\operatorname{prim}}(\pi(\langle
v_{i_1}\rangle))$ and $w:={\operatorname{prim}}(\pi(\langle v\rangle))$.
Then $w=w_{i_1}$.
After the star subdivision of $\delta'$ at
$\langle v\rangle$ by Lemma \ref{le: normal2}(1a),
we create a dependent cone $\delta'_{i_1}$ in $\langle
v\rangle\cdot\delta'$
with the
normal relation obtained from the relation (0) by replacing $w$ with
$w_{i_1}$ . Since $\sigma_v$ contains only negative rays of
$\delta'_{i_1}$ it is codefinite. We also create
dependent cones $\delta'_{i_0}$, where
$i_0\neq i_1$, of type $III$
($III(n,n)$ or $III(*,*)$) with the normal relation
$r_{i_0}w-r_{i_0}w_{i_1}=0$ (case 1b). The face $\sigma_v$ is
codefinite with respect to $\delta'_{i_0}$.
Thus after blow-up we create one cone of the type $I(n,n)$
and cones of type $III(n,n)$ or $III(*,*)$.
Since at the same
time we eliminate the cone $\delta'$, the number of maximal cones of type $I(n,n)$
remains the same. After the star subdivision
$\sigma_v$ is
codefinite with respect to all new dependent cones.
Suppose $\delta'$ is of type $I(n,*)$, $I(*,*)$ $II(n,*)$
or $II(*,*)$. Without loss of generality
we may
assume that in
the normal relation (0) for $\delta'$ (see Lemma \ref{le: normal2}(1)):
$\max\{r_i\mid r_i>0\}\geq \max\{|r_i|\mid r_i<0\}$.
In particular $\max\{|r_i|\mid r_i<0\}< n$.
We shall assign to such a cone $\delta'$ for which
$\sigma_v$ is not codefinite, the invariant
${\operatorname{inv}}(\delta'):=(\max\{r_i\mid r_i>0\},\max\{|r_i|\mid
r_i>0\}$), ordered lexicogragraphically.
By Lemma \ref{le: normal2}(1)), after the star subdivision at
${\operatorname{Mid}}({\operatorname{Ctr}}_+(\delta'),\delta')$ all new dependent cones
$\delta''=\delta'_{i_0}$ have normal relations of type
(1a) if $r_{i_0}> 0$ or (1b) if $r_{i_0}< 0$.
If $r_{i_0}=\max\{r_i\mid r_i>0\}$ then in the relation
(1a) there is only one positive coefficient of $w$ which is
equal to $r_{i_0}$. All other coefficients:
$\frac{r_i-r_{i_0}}{m_w}$ and $r_i<0$ are nonpositive with
the absolute values $<n$. Thus $\delta'_{i_0}$ is of the
type $I(n,*)$ or $I(*,*)$. The face $\sigma_v$ is
codefinite with respect to $\delta'_{i_0}$.
If $0<r_{i_0}<\max\{r_i\mid r_i>0\}$ then
the absolute values of
the coefficients $\frac{r_i-r_{i_0}}{m_w}$, $r_{i_0}$ and
$r_i\leq 0$, are $<n$. The cone $\delta'_{i_0}$ is of type
$I(*,*)$, $II(*,*)$, $III(*,*)$ and
${\operatorname{inv}}(\delta'_{i_0})<{\operatorname{inv}}(\delta')$.
If $r_{i_0}< 0$ then the absolute values of
all coefficients in the relation
(1b) are equal to $|r_{i_0}|<n$ or $|\frac{r_{i_0}}{m_w}|<n$ and
$\delta'_{i_0}$ is a dependent cone of type $I(*,*)$ for which
$\sigma_v$ is not codefinite.
We repeat the procedure for all cones $\delta''$ for which
$\sigma_v$ is not codefinite.
The procedure
terminates since for all new dependent cones for which $\sigma_v$ is
not codefinite the invarinat ${\operatorname{inv}}(\delta')$ drops.
{\bf Step 2b}. {\it Eliminating $\delta$}.
Apply the star subdivision at $\langle v\rangle$ to the
resulting cobordism. We eliminate all cones in
$\operatorname{Star}(\sigma_v)$. In particular $\delta\in\operatorname{Star}(\sigma_v)$ will be eliminated.
By Step 2(a) $\sigma_v$
is codefinite with respect to all dependent cones in
$\operatorname{Star}(\sigma_v)$. All such cones $\delta'$ can be of the
type $I(n,n)$, $I(*,n)$ $I(n,*)$, $I(*,*)$, $II(n,*)$, $II(*,*)$, $III(n,n)$, $III(*,*)$.
By Lemma \ref{le: normal2}(2) all new
dependent cones after the star subdivision
have normal relations of the form (2a) and (2b).
Since all the coefficients in normal relations of the form
(2b) are $<n$, the corresponding dependent cones are of
the form $I(*,*)$, $II(*,*)$, $III(*,*)$.
In the normal relation of the form (2a) for $\delta'_{i_0}$, the absolute
values
of the coefficients of $w_i$, for $i\neq i_1$, are $<n$. The
coefficient $r_{i_0}$ of $w$ is the only one which can be
equal to $n$. Therefore all new dependent cones we create
are of the type $I(n,*)$, $I(*,*)$, $II(n,*)$, $II(*,*)$, $III(*,*)$.
\bigskip
{\bf Step 3}. {\it Eliminating all cones of type $I(*,n)$ and $II(n,*)$}.
Let $\delta=\langle v_1,\ldots,v_{k}\rangle$ be a cone of
the type $I(*,n)$ or $II(n,*)$.
We may assume that the positive coefficients
$r_{i}$ in the normal relation are
$\leq n$ and the absolute values
of all negative coefficients are $< n$. Then $\delta^+$
contains at least two positive rays.
By Lemma \ref{le: normal2}(1) after the star subdivision at
${\operatorname{Mid}}({\operatorname{Ctr}}_+(\delta),\delta)$ we create new dependent
$n$-cones $\delta_{i_0}$ only for $r_{i_0}=n$ in case
(1a). Then by the same reasoning as in Step1,
$\delta_{i_0}$ is of type $I(n,*)$.
\bigskip
{\bf Step 4}. {\it Eliminating all cones of type $III(n,n)$}.
Let $\delta$ be a cone of type $III(n,n)$.
Then $\pi(\delta^+)=\pi(\delta^-)$.
Find
$w\in {\operatorname{par}}(\pi(\delta^+))$ and let $v:={\operatorname{Mid}} (w, \delta^+)$.
Then $v$ is in the relative interior of
a face $\sigma_v$ of $\delta^+$.
{\bf Step 4a}. {\it Making $\sigma_v$ codefinite with respect to all
dependent cones in $\operatorname{Star}(\sigma_v,\Sigma)$}.
Let $\delta'\in\operatorname{Star}(\sigma_v,\Sigma)$ be any dependent cone
for which $\sigma_v$ is not codefinite. As in Step 2,
$\delta'$ can be only of type $I$ or $II$. Then $\delta'$ is
of type $I(n,*)$, $I(*,*)$ or $II(*,*)$.
Suppose $\delta'$ is of type $I(n,*)$.
Without loss of generality we may assume that in the normal
relation for $\delta'$
there is only one positive coefficient $r_{i_1}$ equal to $n$ and
$|r_{i}|<n$ for all negative coefficients $r_{i}$.
Since $\sigma_v$ is not codefinite
with respect to $\delta'$, we have $v_{i_1}\in \sigma_v$.
Note that ${\operatorname{Ctr}}_+(\delta')=\langle v_{i_1}\rangle$ and
for $v:={\operatorname{Mid}}({\operatorname{Ctr}}_+(\delta'),\delta')$ we have $w=w_{i_1}$.
After the star subdivision at
$\langle v'\rangle$ by Lemma \ref{le: normal2}(1)
we create one dependent cone $\delta'_{i_1}$
with the
normal relation obtained from the normal relation for
$\delta'$ by replacing $v$ with
$v_{i_1}$, and some dependent cones $\delta'_{i_0}$, for
$i_0\neq i_1$, which are of type
$III(*,*)$ with the normal relation
$r_{i_0}w=r_{i_0}w_{i_1}$.
After the star subdivision $\sigma_v$ is not
codefinite with respect to any new dependent cones.
Suppose $\delta'$ is of type $I(*,*)$
or $II(*,*)$. Without loss of generality
we may
assume that in
the normal relation (0) for $\delta'$ (see Lemma \ref{le: normal2}(1)):
$\max\{r_i\mid r_i>0\}\geq \max\{|r_i|\mid r_i<0\}$.
In particular $\max\{|r_i|\mid r_i<0\}< n$.
We shall assign to such a cone $\delta'$ for which
$\sigma_v$ is not codefinite, the invariant
${\operatorname{inv}}(\delta'):=(\max\{r_i\mid r_i>0\},\max\{|r_i|\mid
r_i>0\}$), ordered lexicogragraphically.
By Lemma \ref{le: normal2}(1)), after the star subdivision at
${\operatorname{Mid}}({\operatorname{Ctr}}_+(\delta'),\delta')$ all new dependent cones
$\delta''=\delta'_{i_0}$ have normal relations of type
(1a) if $r_{i_0}> 0$ or (1b) if $r_{i_0}< 0$.
If $r_{i_0}=\max\{r_i\mid r_i>0\}$ then in the relation
(1a) there is only one positive coefficient of $w$ which is
equal to $r_{i_0}<n$. All other coefficients:
$\frac{r_i-r_{i_0}}{m_w}$ and $r_i<0$ are nonpositive with
the absolute values $<n$. Thus $\delta'_{i_0}$ is of the
type $I(*,*)$. The face $\sigma_v$ is
codefinite with respect to $\delta'_{i_0}$.
If $0<r_{i_0}<\max\{r_i\mid r_i>0\}$ then
the absolute values of
the coefficients $\frac{r_i-r_{i_0}}{m_w}$, $r_{i_0}$ and
$r_i\leq 0$, are $<n$. The cone $\delta'_{i_0}$ is of type
$I(*,*)$, $II(*,*)$, $III(*,*)$ and
${\operatorname{inv}}(\delta'_{i_0})<{\operatorname{inv}}(\delta')$.
If $r_{i_0}< 0$ then the absolute values of
all coefficients in the relation
(1b) are equal to $|r_{i_0}|<n$ or $|\frac{r_{i_0}}{m_w}|<n$ and
$\delta'_{i_0}$ is a dependent cone of type $I(*,*)$ for which
$\sigma_v$ is not codefinite.
We repeat the procedure for all cones $\delta''$ for which
$\sigma_v$ is not codefinite.
The procedure
terminates since for all new dependent cones for which $\sigma_v$ is
not codefinite the invarinat ${\operatorname{inv}}(\delta')$ drops.
{\bf Step 4b}. {\it Eliminating $\delta$}.
Apply the star subdivision at $\langle v\rangle$ to the
resulting cobordism. We eliminate all cones in
$\operatorname{Star}(\sigma_v)$. In particular $\delta\in\operatorname{Star}(\sigma_v)$ will be eliminated.
By Step 4(a) $\sigma_v$
is codefinite with respect to all dependent cones in
$\operatorname{Star}(\sigma_v)$. All such cones $\delta'$ can be of the
type $I(n,*)$, $I(*,*)$, $II(*,*)$, $III(n,n)$, $III(*,*)$.
By Lemma \ref{le: normal2}(2) all new
dependent cones after the star subdivision
have normal relations of the form (2a) and (2b).
Since all the coefficients in normal relations of the form
(2b) are $<n$, the corresponding dependent cones we create are of
the form $I(*,*)$, $II(*,*)$, $III(*,*)$.
If $\delta'$ is of type $I(*,*)$, $II(*,*)$, $III(*,*)$ then
all the coefficients in normal relations of the form
(2a) are $<n$ and the corresponding dependent cones we
create are of
the form $I(*,*)$, $II(*,*)$, $III(*,*)$.
If $\delta'$ is of type $I(n,*)$ then we can assume that
$\sigma_v$ is a face of ${\delta'}^+$. If ${\delta'}$ is
pointing down then all positive
coefficients $r_i$ in the normal relation for $\delta'$ are
$<n$. In the normal relation (2a) for $\delta'_{i_0}$,
where $r_{i_0}>0$, all coefficients are $<n$. If ${\delta'}$ is
pointing up then there is only one positive
coefficient $r_{i_1}$. In the normal relation (2a) for $\delta'_{i_1}$,
the first two sums disappear and there is only one
positive coefficient of $w$ equal to $r_{i_0}\leq n$. By
applying the star subdivision at $\langle v \rangle$ to
$\delta'$ we create one maximal dependnent cone of type
$I(n,*)$. At the same time we eliminate the cone $\delta'$ of type $I(n,*)$.
If $\delta'$ is of type $III(n,n)$ (for instance if
$\delta'=\delta$) then there is one positive coefficient $r_{i_1}=n$
and one negative coefficient $r_{i_2}=-n$ in the normal
relation for $\delta'$.
After the star subdivision we create only one maximal
dependent cone $\delta'_{i_0}$ for
$i_0=i_1$ having the normal relation of the form (2a).
In the relation
(2a) the first two sums disappear and there is only one
positive coefficient of $w$ equal to $r_{i_0}\leq n$. The
absolute values of
the other negative coefficients are $< n$.
Thus we create one maximal dependent
cone of type $I(n,*)$.
\bigskip
{\bf Step 5}. {\it Eliminating all dependent cones of type $I(n,*)$
and all n-cones which are not faces of dependent cones}.
Let $\delta$ be a cone of type $I(n,*)$ or
an $n$-cone which is not a face of a dependent cone.
In the first case without loss
of generality we may assume that
in the normal relation there is one negative
coefficient equal to $-n$ and all positive coefficients are
$<n$ ($\delta$ is pointing down).
Then $\pi(\delta^+)=\pi(\delta)$.
Find $w\in {\operatorname{par}}(\pi(\delta^+))$ and let $v={\operatorname{Mid}}
(w,(\delta_+))$.
Then $v\in {\operatorname{int}} (\sigma_v)$, where
$\sigma_v$ is a face of $\delta_+$.
{\bf Step 5a}. {\it Making $\sigma_v$ codefinite with respect to all
dependent cones in $\operatorname{Star}(\sigma_v,\Sigma)$}.
We repeat word by word the procedure in Step 4a.
{\bf Step 5b}. {\it Eliminating $\delta$}. Apply the star subdivision at $\langle v\rangle$ to the
resulting cobordism. We eliminate all cones in
$\operatorname{Star}(\sigma_v)$.
By Step 5(a) $\sigma_v$
is codefinite with respect to all dependent cones in
$\operatorname{Star}(\sigma_v)$. All such cones $\delta'$ can be of the
type $I(n,*)$, $I(*,*)$, $II(*,*)$, $III(*,*)$. We repeat
word by word the procedure in Step 5a excluding the last case $III(n,n)$.
By applying the star subdivision at $\langle v\rangle$ we
decrease the number of maximal cones of type $I(n,*)$ (or
independent $n$-cones). All new dependent
cones we create are of type $I(n,*)$, $I(*,*)$, $II(*,*)$.
\qed
\section{Birational cobordisms}
\subsection{Definition of a birational cobordism}
\noindent \begin{definition} (\cite{Wlodarczyk2}):
Let $X_1 $ and $X_2$ be two birationally equivalent
normal varieties.
By a {\it birational cobordism} or simply a {\it cobordism}
$B:=B(X_1,X_2)$ between them we understand a
normal variety $B$ with an algebraic action of $K^*$ such that the
sets
\[\begin{array}{rccc}
&B_-:=\{x \in B\mid \, \lim_{{\bf t}\to 0} \, {\bf t}x \,\, \mbox{does
not exist}\}& \, \mbox{and}& \\
&B_+:=\{x \in B\mid \, \lim_{{\bf t}\to\infty} \, {\bf t}x \,\,
\mbox{does not exist}\} &&
\end{array}\]
\noindent are nonempty and open and there exist
geometric quotients
$B_-/ K^*$ and $B_+\// K^*$ such that $B_+/ K^*\simeq X_1$ and $B_-\//
K^*\simeq X_2$ and the birational map $X_1\mathrel{{-}\,{\rightarrow}}
X_2$ is given by the above
isomorphisms and the open embeddings
$B_+ \cap B_-\// K^*$ into
$B_+\//K^*$ and $B_-\//K^*$ respectively.
\end{definition}
\bigskip
\subsection{Collapsibility}
Let $X$ be a variety with an action of $K^*$. Let
$F\subset X$ be a set consisting of fixed points. Then we define
$$F^+(X)=F^+=\{x\in X\mid
\, \lim_{{\bf t}\to 0} {\bf t}x \in F\},\, F^-(X)=F^-=\{x\in X\mid
\, \lim_{{\bf t}\to \infty} {\bf t}x \in F\}.$$
\begin{definition} (\cite{Wlodarczyk2}).\label{de: order} Let $X$ be a
cobordism or any variety with a $K^*$-action.
\begin{enumerate}
\item We say that a connected component $F$ of the fixed point
set is an {\it immediate predecessor} of a component
$F'$ if there exists a nonfixed point $x$ such that
$\lim_{{\bf t}\to 0} {\bf t}x\in F$ and
$\lim_{{\bf t}\to
\infty} {\bf t}x\in F'$.
\item
We say that $F$ {\it precedes} $F'$ and
write $F<F'$ if there exists a sequence of connected fixed
point set components $F_0=F ,F_1,\ldots,F_l=F'$ such that
$F_{i-1}$ is an immediate predecessor of $F_i$
(see \cite{BB-S} , Def. 1. 1).
\item
We call a cobordism (or a variety with $K^*$-action) {\it collapsible} (see
also Morelli \cite{Morelli1}) if the relation $<$ on its set of
connected components of the fixed point set
is an order. (Here an
order is just required to be transitive.)
\end{enumerate}
\end{definition}
\begin{lemma}(\cite{Wlodarczyk2}). A projective cobordism is
collapsible. \qed
\end{lemma}
\begin{lemma}\label{le: collapsible} Let $B$ be a collapsible
variety and $I$ be a $K^*$-invariant sheaf of ideals.
Then the blow-up $B':={\operatorname{bl}}_I(B)$ is also collapsible.
\end{lemma}
\noindent {\bf Proof.} Set $\pi:B'\to B$.
For any connected fixed point component $F'$ on $B'$
by abuse of notation denote by $\pi(F')$ the fixed point
component containing the image $\pi(F')$. Then for any
$F'_0$ and $F'_1$ on $B'$ the relation $F'_0<F'_1$ implies
that either $\pi(F'_0)<\pi(F'_1)$ or $\pi(F'_0)=\pi(F'_1)$.
Consider the latter case. Let $x\in F'$ and $u_1,\ldots,u_k$
be semiinvariant generators. Then at least one of the
functions $\pi^*(u_{i})$, say $\pi^*(u_{i_0})$,
generates $\pi^{-1}(I)\cdot{{\mathcal O}}_{B'}$ at $x$. We let ${\rm
ord}(F)$ denote the weight of
$\pi^*(u_{i_0})$ with respect to the $K^*$-action. This definition does not depend on the
choice of the semiinvariant generator. For any two generators
$\pi^*(u_{i_0})$ and $\pi^*(u_{i_1})$ their quotient is a
semiinvariant function which is invertible at the fixed point
$x\in B'$, which implies that it is invariant function. This
definition is locally constant, which means that it does not
depend on the choice of $x\in X$. Moreover the functions
$\pi^{*}(u_1),\ldots,\pi^*(u_k)$ are sections generating
${\mathcal O}_{\pi^{-1}(U)}(-D)$, where $D$ is the exceptional
divisor of $\pi$. This gives a $K^*$-equivariant morphism
$\phi:\pi^{-1}(U)\to {\bf P}^k$
into a projective
space with semiinvariant
coordinates $x_1,\ldots x_k$ having the same weights as $u_1,\ldots,u_k$, say
$a_1,\ldots,a_k$. We can assume that $a_1\leq\ldots\leq a_k$. Then
the function ${\rm ord}(F)=\min\{a_i \mid x_i(\phi(F))\neq 0\}$ coincides with the one
induced on ${\bf P}^k$, and $F < F'$ implies
${\rm ord}(F)<{\rm ord}(F')$. Define $F\prec F'$ if
either $\pi(F)<\pi(F')$, or $\pi(F)=\pi(F')$ and
${\rm ord}(F)\leq{\rm ord}(F')$. Since $F<F'$ implies
$F\prec F'$ it follows that $\prec$ is an order on the fixed
point components on $B'$. \qed
\subsection{Decomposition of a birational cobordism}
\begin{definition} (\cite{Wlodarczyk2}). Let $B$ be a collapsible
cobordism and $F_0$ be a minimal
component. By an {\it elementary collapse with respect to $F_0$}
we mean the cobordism $B^{F_0}:=B\setminus F_0^-$ (see
section \ref{se: pi-stable}).
By an {\it elementary cobordism with respect to
$F_0$} we mean the cobordism $B_{F_0}:=B\setminus \bigcup_
{F\neq F_0} F^+$.
\end{definition}
\begin{proposition} (\cite{Wlodarczyk2}). Let $F_0$ be a minimal component
of the fixed point
set in a collapsible cobordism $B$. Then the
elementary collapse $B^{F_0}$ with respect to $F_0$ is again a
collapsible cobordism, in particular it satisfies:
\begin{enumerate}
\item $B^{F_0}_+=B_+$ is an open subset of $B$,
\item $F_0^-$ is a closed subset of $B$ and equivalently $B^{F_0}$
is an
open subset of $B$.
\item $B^{F_0}_-$ is an open subset of $B^{F_0}$ and
$B^{F_0}_-=B^{F_0}\setminus \bigcup_{F\neq F_0} F^+$.
\item The elementary cobordism $B_{F_0}$ is an open subset
of $B$ such that
$$\displaylines{ (B_{F_0})_-=B_{F_0} \setminus F_0^+=B_-\cr
(B_{F_0})_+=B_{F_0} \setminus F_0^- =B^{F_0}_-.\cr} $$
\item There exist good and respectively geometric quotients $B_{F_0}//{K^*}$
and $B^{F_0}_-/{K^*}$ and moreover the natural embeddings $ i_- : B_-\subset
B_{F_0}$ and $i_+ : B^{F_0}_-\subset B_{F_0}$ induce proper morphisms
$ i_{-/K^*} : B^{F_0}_-/{K^*}\rightarrow B_{F_0}//{K^*}$
and $ i_{+/K^*} : B_-/{K^*}\rightarrow B_{F_0}//{K^*}$. \qed
\end{enumerate}
\end{proposition}
We can extend the order $<$ from Definition \ref{de: order}
to a total order.
As a corollary from the above we obtain:
\begin{proposition} \label{pr: factorization} Let $F_1<\ldots<F_k$ denote
the connected fixed point components of a cobordism $B$.
Set $B_i:=B\setminus (\bigcup_{F_j<F_i}F_j^-\cup \bigcup_{F_j>F_i}F_j^+)$.
Then \begin{enumerate}
\item
$(B_1)_-=B_-$, $(B_k)_+=B_+$.
\item $(B_{i+1})_-=(B_i)_+$.
\item There is a factorization
$$B_-/K^*=(B_1)_-/K^*{{-}\to} (B_1)_+/K^*= (B_2)_-/K^*{{-}\to}
\ldots (B_{k-1})_+/K^*= (B_k)_-/K^* {{-}\to}
(B_k)_+/K^*=B_+/K^*.$$ \qed
\end{enumerate}
\end{proposition}
\subsection{Existence of a smooth birational cobordism}\label{se: construction}
\begin{proposition} (see also \cite{Wlodarczyk2})\label{pr: construction} Let $\phi: Y\to X$ be a birational projective
morphism between smooth complete varieties over an algebraically
closed field of characteristic $0$. Let $U\subset X, Y$ be
an open subset , where $\phi$ is an isomorphism.
\begin{enumerate}
\item There exists a smooth
collapsible cobordism $B=B(X,Y)$ between $X$ and $Y$ such that
$B\supset U\times K^*$
\item If $D_X:=X\setminus U$ and $D_Y:=Y\setminus U$ are divisors
with normal crossings then there exists
a cobordism $\tilde{B}$ between $\tilde{X}$ and
$\tilde{Y}$ such that
\begin{itemize}
\item $\tilde{X}$ and $\tilde{Y}$ are obtained from $X$ and $Y$ by a sequence of
blow-ups at centers which have normal crossings with
components of the total transforms of $D_X$ and $D_Y$ respectively.
\item $U\times K^* \subset \tilde{B} $ and
$\tilde{B}\setminus (U\times K^*)$ is a
divisor with normal crossings.
\end{itemize}
\end{enumerate}
\end{proposition}
\noindent {\bf Proof.} (1) If $X$ and $Y$ are projective the
construction of $B$ is given in Proposition 2 of \cite{Wlodarczyk2}; in
general we use the construction of D.Abramovich (Remark after
Proposition 2 in \cite{Wlodarczyk2}). $B$ is an open subset of $\overline{B}$
where $\overline{B}$ is
obtained from $X\times {\bf P}^1$ by a sequence of blow-ups at
ideals, hence by a single blow-up of an ideal. By Lemma
\ref{le: collapsible}, $\overline{B}$ is collapsible. Consequently
$B$ is a collapsible cobordism.
(2) The sets $B_-$ and $B_+$ are locally
trivial $K^*$-bundles with projections $\pi_-:B_-\to B_-/K^*\simeq
X$ and $\pi_+:B_+\to B_+/K^*\simeq
Y$. Let $Z:=B\setminus (U\times K^*)$. Then $Z\cap B_-$ and
$Z\cap B_+$ are normal crossing divisors and $\pi_-(Z\cap
B_-)=D_X$ and $\pi_+(Z\cap
B_+)=D_Y$.
Let $f:\tilde{B}\to B$ be a canonical principalization of
$I_Z$ (see Hironaka \cite{Hironaka1},
Villamayor \cite{Villamayor} and Bierstone-Milman \cite{Bierstone-Milman}). Let $f_+:f^{-1}(B_+)\to B_+$
(resp.$f_-:f^{-1}(B_-)\to B_-$) be
its restriction. By functoriality
$f_+$ (resp. $f_-$) is a canonical
principalization of $B_+$ (resp. $B_-$) which commutes
with a canonical principalization $\tilde{Y}$ of $I_{D_Y}$ on $Y$
(resp. $\tilde{X}$ of $I_{D_X}$ on $D_X$)
In particular all centers are
$K^*$-invariant of the form $\pi_+^{-1}(C)$ (resp.
$\pi_-^{-1}(C)$), where $C$ have
normal crossings with components of the total transform of $D_Y$ (resp. $D_X$).
\qed
\section{Weak factorization theorem}
\subsection{Construction of a $\pi$-regular cobordism}
\begin{definition} \begin{enumerate}
\item We say that a cobordism $B$ is
{\it $\pi$-regular} if for any $x$ in $B$ which is not a fixed point of
the $K^*$-action in $B$ there exists an
affine invariant
neighborhood $U$ without fixed points such that $U/K^*$ is
smooth.
\item We call a cobordism $B$ {\it $K^*$-stratified} if it
is a $K^*$-stratified toroidal variety.
\end{enumerate}
\end{definition}
\begin{proposition} \label{pr: pi-lemma2} Let $(B,S)$ be a smooth stratified
cobordism
between smooth varieties $X$ and $X'$. There
exists a $K^*$-toroidal modification $\tilde{B}$ of $B$ obtained
as a sequence of blow-ups of stable valuations such that
$\tilde{B}$ is a $\pi$-regular
$K^*$-stratified cobordism between $X$ and $X'$. Moreover if $U\subset X$
is an open affine invariant fixed point free subset such
that $U/K^*$ is smooth then all centers of blow-ups are
disjoint from $U$.
\end{proposition}
\noindent {\bf Proof.} By Lemma \ref{le: asemicomplex} we can associate with $(B,S)$ a
$K^*$-semicomplex $\Sigma$. Since for any fixed point component $F$
in $B$ the sets $F^+$ and $F^-$ are not dense it follows
that $\Sigma_S$ is strictly convex. By Lemma \ref{le: pi-lemma}
there is a $K^*$-canonical subdivision $\Delta$ of
$\Sigma$ which is $\pi$-nonsingular and does not affect
any $\pi$-nonsingular cones in $\Sigma$. By
Theorem \ref{th: correspondence}, $\Delta$ corresponds
to a $K^*$-stratified cobordism between $X$ and $Y$. \qed
\subsection{Factorization determined by a $\pi$-regular cobordism}
\begin{proposition} \label{pr: regular factorization}
Let $(B,R)$ be a $\pi$-regular collapsible
$K^*$-stratified cobordism between smooth stratified toroidal varieties
$(X,S_X)=(B_-/K^*,(R\cap B_-)/K^*)$, $(Y,S_Y)=(B_+/K^*,(R\cap B_+)/K^*)$.
Let $B_i$ be elementary cobordisms as in Proposition
\ref{pr: factorization}.
Then there exists a sequence of smooth stratified toroidal varieties
$(X_i,S_i):=((B_i)_-/K^*, (R\cap (B_i)_-)/K^*)$
and a factorization
\[\begin{array}{rcccccccccccccccc}
&&Y_0&&&&Y_1&&&&&&&Y_{n-1}&&&\\
&g_0\swarrow&& \searrow f_0&& g_1\swarrow&& \searrow
f_1&&&&&g_{n-1}\swarrow&& \searrow f_{n-1}&&\\
X=X_0&& {{-}{\to}}&& X_1&&
{{-}{\to}}&&X_2&&\ldots& X_{n-1}&&
{{-}{\to}}& X_n&=Y&,
\end{array}\]
\noindent where each $X_i$ is a smooth variety and $f_i:Y_i\to X_{i+1}$
and $g_i:Y_i\to X_i$ for $i=0,\ldots,n-1$ are blow-ups at
smooth centers which have normal crossings with the closures of
strata in $S_{i}$ (resp. $S_{i+1}$).
\end{proposition}
\noindent{\bf Proof.} Follows from the following Lemma
\ref{le: pi-lemma2} and Proposition \ref{pr: factorization}.
\qed
\begin{lemma} \label{le: pi-lemma2} Let $(B,R)$ be a
$\pi$-regular elementary
$K^*$-stratified cobordism. Then $X=B_-/K^*$ and $Y=B_+/K^*$ are smooth
stratified toroidal varieties with stratifications $S_X$ and
$S_Y$ induced
by $R\cap B_+$ and $R\cap B_-$ and there exists a smooth
variety $Z$, obtained by a blow-up
at a smooth center $C_X$ (resp. $C_Y$),
having only normal crossings with closures of strata in
$S_X$ (resp. $S_X$).
\end{lemma}
\noindent{\bf Proof.}
For any $X$ let $\widetilde{X}$ denote its normalization.
The fixed
point component is equal to the closure
$\overline{{\operatorname{strat}}_B(\sigma)}$ of the maximal stratum
, corresponding to a circuit
$\sigma$.
Consider the
commutative diagram
\[\begin{array}{rccccc}
&B_+/K^* &&&&B_-/K^*\\
&&\searrow&& \swarrow&\\
&&&B//K^*&&\\
\end{array}\]
This diagram can be completed to
\[\begin{array}{rccccccc}
&&&Z:=({B_+/K^*
\times_{B//K^*} B_-/K^*})\widetilde{}&&\\
&&&\psi\downarrow&&&&\\
&&&B_+/K^* \times_{B//K^*} B_-/K^*&&\\
&&\swarrow&& \searrow&&&\\
&B_+/K^* &&&&B_-/K^*&& (*)\\
&&\searrow&& \swarrow&&&\\
&&&B//K^*&&&&\\
\end{array}\]
\noindent where the morphism $\psi$ is given by the normalization.
For any $x\in \overline{{\operatorname{strat}}(\sigma)}$ find $\tau\geq\sigma$ such that
$x\in{\operatorname{strat}}(\tau)$.
Let $\phi_{\tau}: U_{\tau}\to X_{\tau}$ be the
relevant chart.
The above diagram defines locally a diagram
\[\begin{array}{rccccccc}
&&&((U_{\tau})_+/K^*\times_{U_{\tau}/K^*}
(U_{\tau})-/K^*)\widetilde{} &&&&\\
&&&\downarrow&&&&\\
&&&(U_{\tau})_+/K^*\times_{U_{\tau}/K^*}{(U_{\tau})_-/K^*}&&&&\\
&&\swarrow&& \searrow&&&\\
&(U_{\tau})_+/K^*&&&&(U_{\tau})_-/K^*&&\\
&&\searrow&& \swarrow&&& \\
&&&U_{\tau}//K^*&&&&\\
\end{array}\]
This diagram is a pull-back via a smooth morphism
$\phi_{\tau/K^*}:U_{\tau}//K^*\to X_{\tau}/K^*$
of the diagram of toric varieties
\[\begin{array}{rccccc}
&&&X_{\Sigma} :=((X_{\tau})_-/K^*
\times_{X_{\tau}/K^*}(X_{\tau})_+/K^*)\widetilde{}
&&\\
&&&\downarrow&&\\
&&&(X_{\tau})_-/K^*\times_{X_{\tau}/K^*} (X_{\tau})_+/K^*&&\\
&&\swarrow&& \searrow&\\
&X_{\Sigma_1}:=(X_{\tau})_-/K^*&&&&X_{\Sigma_2}:=(X_{\tau})_+/K^*\\
&&\searrow&& \swarrow&\\
&&&X_{\tau^\Gamma}=X_{\tau}//K^*&&\\
\end{array}\]
\noindent It follows from the universal property of the fiber product that
$X_\Sigma$ is a
normal toric variety whose fan consists of the cones $
\{\tau_1\cap\tau_2 \mid \tau_1\in \Sigma_1, \tau_2\in \Sigma_2\}.$
The cone $\tau$ is $\pi$-nonsingular and by Lemma \ref{le: normal} we can write the
only dependence relation of
$\tau^\Gamma=\langle v_1,\ldots,v_k,w_1,\ldots,w_m,q_1,\ldots,q_l \rangle$, where
$\langle v_1,\ldots,v_k,w_1,\ldots,w_m \rangle$ is a curcuit,
as follows:
$$e:=v_1+\ldots+v_k=w_1+\ldots+w_m.$$
Finally, $\Sigma=\langle e \rangle\cdot\Sigma_1$ and
$\Sigma=\langle e\rangle \cdot \Sigma_2$
are regular star subdivisions.
We have shown that the morphisms $Z\to X$ and
$Z\to Y$ defined by the diagram (*) are
blow-ups at smooth centers. These centers have normal
crossings with closures of strata since locally the centers and
the closures of strata are determined by the closures of orbits on a
smooth toric variety.
\qed
\subsection{Nagata's factorization}
\begin{lemma}(Nagata)\label{le: Nagata}
Let $X\supset U \subset Y$ be two complete varieties over
an algebraically closed field, containing an open subset $U$.
Then there exists a variety $Z$ which is simultaneously a
blow-up of a coherent sheaf of ideals ${\mathcal I}_X$ on $X$ and
a blow-up of a coherent sheaf of ideals ${\mathcal I}_Y$ on $Y$.
Moreover the supports of ${\mathcal I}_X$ and ${\mathcal I}_Y$ are disjoint from $U$
\end{lemma}
The above lemma is a refinement of the
original lemma of Nagata
\begin{definition}(Nagata \cite{Nagata}) \label{de: Na}
Let $X_i\supset U $, for $i=1,\ldots,n$ be complete varieties over
an algebraically closed field, containing an open subset $U$.
By the {\it join} $X_1*\ldots *X_k$ we mean
the closure of the diagonal $\Delta(U) \subset
X_1\times\ldots\times X_k$.
\end{definition}
\begin{lemma}(Nagata \cite{Nagata}) \label{le: Na}
Let $X\supset U \subset Y$ be two complete varieties over
an algebraically closed field, containing an open subset $U$.
Then there exist coherent sheaves of ideals ${\mathcal I}_1,\ldots,{\mathcal I}_k$ on
$X$ with supports disjoint from $U$ and blow-ups
$X_1={\operatorname{bl}}_{{\mathcal I}_1}X,\ldots,X_k={\operatorname{bl}}_{{\mathcal I}_k}X$
such that the join
$X_1*\ldots *X_k$
dominates $Y$.\qed
\end{lemma}
\noindent {\bf Proof of Lemma \ref{le: Nagata}}
Let $X_i$ be as in the assertion of Lemma \ref{le: Na}.
Note that $X_1\times\ldots\times X_k$ is projective over
$X\times\ldots \times X$. This implies that $X_1*\ldots *X_k$ is
projective over $\Delta(X)\simeq X$. Thus there exists a
sheaf of ideals ${\mathcal J}$ on $X$ such that $X_1*\ldots *X_k\to X$ is
the blow-up at ${\mathcal J}$.
By Nagata and the above find a blow-up $X'$ of $X$ at ${\mathcal J}$
such that $X'$ dominates $Y$. Denote the blow-up by $\phi:X'\to Y$.
Analogously find a blow-up $Z$ of $Y$ at ${\mathcal I}_Y$
such that $Z$ dominates $X'$.
Thus $Z\to X'$ is a blow-up at $\phi^{-1}({\mathcal I}_Y)$ and $Z\to
X$ is a composite of blow-ups and consequently a single blow-up
at some sheaf of ideals ${\mathcal I}_X$. \qed
\subsection{Proof of the Weak Factorization Theorem}\label{se: factorization}
\bigskip
\noindent {\bf Step 1} By Nagata's Lemma \ref{le: Nagata}
we find a $Z'$ obtained from $X$ and $Y$ respectively
by blow-ups at
centers disjoint from $U$. If $X$ and $Y$ are projective we
can take $Z'$ to be the graph of $\phi: X{-\to} Y$.
Let $Z''$ be a
resolution of singularities of $Z'$ and $Z$ be a canonical
principalization of the ideal of the set $Z''\setminus U$
(this is needed for part (2) of the theorem only). Then
$Z\setminus U$
is a divisor with normal crossings.
It suffices to prove the theorem for the projective
morphisms $Z\to X$ (or $Z\to Y$).
\noindent {\bf Step 2}
By Proposition \ref{pr: construction} there exists
a smooth collapsible cobordism $B=B(Z,X)$. If $D_X:=X\setminus U$ and $D_Z:=Z\setminus U$
are divisors with normal crossings then there exists a
cobordism $\tilde{B}$ between $\tilde{X}$ and $\tilde{Y}$
as in the proposition such that $\tilde{B}\setminus
(U\times K^*)$ is a normal crossings divisor. For
simplicity denote $\tilde{X}, \tilde{Z},\tilde{B}$ by $X
,Z, B$ respectively.
\noindent {\bf Step 3}
By Lemma
\ref{le: existence2} there exists a $K^*$-invariant stratification $S$
on $B$ such that $(B,S)$ is a $K^*$-stratified
toroidal variety. If $B\setminus
(U\times K^*)$ is a normal crossings divisor
then the
components of this divisor are the closures of some strata
in $S$.
\noindent {\bf Step 4}
By Proposition
\ref{pr: pi-lemma2} we can find a $\pi$-regular collapsible $K^*$-stratified
cobordism $(\overline{B},\overline{R})$ between $X$ and $Z$
by applying a sequence of blow-ups at stable valuations to $(B,S)$
and
using the combinatorial algorithm. The subsets
$B_+=\overline{B}_+$ and $B_-=\overline{B}_-$ are unaffected.
\noindent {\bf Step 5} By Proposition \ref{pr: regular
factorization} the $\pi$-regular
cobordism $(\overline{B},\overline{R})$ determines a
factorization of the map $Z\to X$ into a sequence of blow-ups and blow-downs.
All centers have only normal crossings with closures of the
induced strata on intermediate varieties.
\qed
\section{Comparison with another proof. Torification
and stable support}\label{se: comparison}
\subsection{Comparison with another proof}
In \cite{AKMW} another proof of the Weak
Factorization Theorem is given , where the theorem is stated
an proved in a more general version. In particular the
assumption of the base field to be algebraically closed was
removed. The theorem was also formulated and proved for
bimeromorphic maps of complex compact analytic manifolds.
The proof in \cite{AKMW} uses the idea of torific ideals due to
Abramovich and De Jong.
The idea of the proof is
to consider an ideal on a smooth cobordism whose blow-up
determines a structure of toroidal embedding compatible
with the $K^*$-action. Such an ideal, called {\it ''torific''},
and the the blow-up procedure , called {\it
torification} were introduced by D.Abramovich and
J.de Jong. The torific ideal is defined in an invariant
neighborhood $U$ of
the fixed point component of the smooth variety with $K^*$-action.
Let $u_1,\ldots,u_k$ denote the semiinvariant parameters with weights
$a_1,\ldots,a_k$. We define the torific ideal to be
$I=I_{a_1}\cdot\ldots\cdot I_{a_k}$, where $I_a$ denotes the
ideal generated by all semiinvariant functions of weight $a$.
The blowing-up of the torific ideal
induces a structure of toroidal embedding, locally on each elementary piece
of cobordism. Torification allows one to pass to the category of toroidal
embeddings, where the factorization problem has already been
solved using combinatorial algorithms. However the torific
ideal can be constructed only locally on elementary
cobordisms. Hence after the torification is done on
each elementary cobordism some glueing procedure should be
applied. Using canonical principalizations of torific ideals
and canonical resolution of singularities of intermediate varieties
gives a factorization on an elementary cobordism between
canonical resolution of singularities.
Patching these factorizations together one
gets a decomposition of the given birational map.
\subsection{Torification and stable support}
The notion of torification can be understood and generalized
on the ground of the theory of stratified toroidal varieties.
\noindent
\begin{definition} Let $(X,S)$ be an oriented stratified toroidal
variety (with or without group action). By a {\it stable
ideal} ${\mathcal I}$ we mean a coherent sheaf of ideals ${\mathcal I}$ satisfying the following
conditions :
\begin{enumerate}
\item The support of ${\mathcal I}$ is a union of strata in $S$.
\item For any points $x_1$ and $x_2$ in a stratum $s$ and any
isomorphism $\phi:\widehat{X}_{x_1}\to \widehat{X}_{x_2}$
preserving strata and orientation (respectively equivariant
with respect to the $\Gamma_\sigma$-action),
$\phi^*({\mathcal I})\cdot
\widehat{{\mathcal O}}_{x_2}={\mathcal I}\cdot\widehat{{\mathcal O}}_{x_1}$.
\end{enumerate}
\end{definition}
\begin{proposition} Let ${\mathcal I}$ be a stable ideal on an oriented stratified toroidal
variety $(X,S)$.
Then for any $\sigma\in \Sigma$ there is a $G^\sigma$-invariant toric
ideal $I_\sigma$ on ${X}_{\sigma}$ such that
\begin{enumerate}
\item The induced ideal
on $\widetilde{X}_{\sigma}$ is $G^\sigma$-invariant.
\item For any
chart $\phi_\sigma:U\to X_{\sigma}$, ${\mathcal I}_U=\phi_\sigma^*(I_\sigma)$.
\item For any $\tau\leq \sigma$, $I_\tau\cdot
{\mathcal O}_{x,X_{\sigma}}$ is the restriction of $I_{\sigma}$
to the open subset $X_{(\tau, N_\sigma)}\subset X_{\sigma}$.
\end{enumerate}
\end{proposition}
One can rephrase the above proposition in combinatorial
language, using the correspondence between toric ideals ${\mathcal I}$
and ${\operatorname{ord}}({\mathcal I})$-functions in \cite{KKMS} and Lemma \ref{le: vertices}.
\begin{proposition} Let $(X,S)$ be an oriented stratified toroidal
variety with an associated oriented semicomplex $\Sigma$.
The stable complete coherent sheaves of ideals ${\mathcal I}$ on $(X,S)$ are in $1-1$
correspondence with collections of functions ${\operatorname{ord}}(I_\sigma)$,
$\sigma\in \Sigma$,
satisfying:
\begin{enumerate}
\item
${\operatorname{ord}}(I_\sigma):\sigma\to {\mathbb{Q}}$ is a convex piecewise
linear function ${\operatorname{ord}}(I_\sigma)({N}_\sigma)\to {\bf Z}$.
\item Denote by $\Delta(I_\sigma)$ the subdivision
of $\sigma$ into the maximal cones where ${\operatorname{ord}}(I_\sigma)$ is
linear. Then ${\operatorname{Vert}}(\Delta(I_\sigma))\subset{\operatorname{stab}}(\sigma)\cup
{\operatorname{Vert}}(\overline{\sigma})$, and ${\operatorname{ord}}(I_\sigma)(v)=0$ for any $v\in
{\operatorname{Vert}}(\Delta(I_\sigma))\setminus {\operatorname{stab}}(\sigma)$.
\item For any $\tau\leq \sigma$, the restriction of ${\operatorname{ord}}(I_{\sigma})$ to
$\tau$ is equal to ${\operatorname{ord}}(I_{\tau})$. \qed
\end{enumerate}
\end{proposition}
It follows that the functions ${\operatorname{ord}}(I_\sigma)$ patch together and
give a function ${\operatorname{ord}}({\mathcal I})$ on the totality of vectors in
faces of $\Sigma$.
\begin{lemma}
Denote by $\Delta({\mathcal I})$ the subdivision of $\Sigma$ obtained
by glueing the subdivisions $\Delta(I_\sigma)$. Then
$\Delta({\mathcal I})$ is a canonical subdivision of $\Sigma$
correspondinding to the blow-up of $(X,S)$ at ${\mathcal I}$. \qed
\end{lemma}
\begin{definition} Let $(X,S)$ be an oriented stratified toroidal
variety (with or without group action). Then an ideal
${\mathcal I}$ is called {\it torific} if it is stable and the
normalization of its blow-up is a
canonical toroidal morphism $(Y,R)\to (X,S)$ such that $(Y,R)$ is a
toroidal embedding. Such a blow-up is called a {\it
torification}.
\end{definition}
\begin{lemma}
Let $(X,S)$ be an oriented stratified toroidal variety with
an
associated semicomplex $\Sigma$. Then a stable ideal ${\mathcal I}$ is
torific iff any $\sigma \in \Delta({\mathcal I})$ whose
relative interior intersects the stable support ${\operatorname{Stab}}(\Sigma)$,
is contained in ${\operatorname{Stab}}(\Sigma)$.
\end{lemma}
\noindent{\bf Proof.}
If ${\operatorname{int}}(\sigma)\cap {\operatorname{Stab}}(\Sigma) \neq\emptyset$ then
$\sigma\in \Delta({\mathcal I})_{{\operatorname{stab}}}$. By Theorem
\ref{th: correspondence} and Lemma \ref{le: associated
semicomplex}, $\Delta({\mathcal I})_{{\operatorname{stab}}}$ is a complex corresponding
to the toroidal embedding. Hence all vectors from $\sigma$
are stable. \qed
Torification is not always possible.
However we can perform torification locally in the following sense.
\begin{proposition}\label{pr: torification} Let $(X,S)$ be an
oriented stratified toroidal variety with an associated
oriented semicomplex $\Sigma$.
Then for any $\sigma\in \Sigma$ there is a $G^\sigma$-invariant
ideal $$I_\sigma:=(G^\sigma\cdot u_1)\cdot\ldots\cdot (G^\sigma\cdot
u_k),$$ where $u_1,\ldots,u_k$ are toric paramaters on $\widetilde{X}_{\sigma}$
such that $\Delta(I_\sigma)$ contains the face ${\operatorname{Inv}}(\sigma)$.
In particular $${\operatorname{Inv}}(\sigma)=
{\rm conv}(({\operatorname{Vert}}(\Delta(I_\sigma))\setminus
{\operatorname{Vert}}(\sigma)) \cup {\operatorname{Vert}}^{\operatorname{semic}}(\sigma)),$$
\noindent where ${\operatorname{Vert}}^{\operatorname{semic}}(\sigma)$ denotes the set of
$1$-dimensional faces in the semicone $\sigma$.
\end{proposition}
\noindent{\bf Proof.} Denote by $v_\nu$
the vector corresponding to a toric valuation $\nu$.
Note that a toric valuation $\nu$ is $G^\sigma$-invariant iff
$\nu(g_*(u_i))=\nu(u_i))$ for any $i=1,\ldots,k$ and $g\in
G^\sigma$. It follows from Lemma \ref{le: monomial2} that the latter
statement is equivalent to the following one:
$\nu(g_*(u_i))\geq\nu(u_i)$
for any $g\in G^\sigma$. This can be reformulated as follows: $\nu$
is $G^\sigma$-invariant iff
${\operatorname{ord}}{(I_\sigma)}(v_\nu)= \nu(u_1\cdot\ldots\cdot u_k)$.
Finally, ${\operatorname{Inv}}(\sigma)$
is a set where ${\operatorname{ord}}{(I_\sigma)}$ is linear and equal to
the functional of $u_1\cdot\ldots\cdot u_k$. Hence ${\operatorname{Inv}}(\sigma)$ occurs as a cone in
$\Delta(I_\sigma)$. All vertices of this cone belong to
${\operatorname{Inv}}(\sigma)$
and hence they have to be in the form as in the
proposition. \qed
\begin{remark}The above proposition generalizes the construction of
torific ideal due to Abramovich and de Jong (\cite{Abramovich-de-Jong}).
\end{remark}
\subsection{Computing the stable support}
Proposition \ref{pr: torification} provides a method of
computing the stable support of semicomplexes.
\begin{example}\label{ex: isolated4} Let $(X,S)$ be a toroidal variety with an isolated
singularity $x_1x_2=x_3x_4$ as in Example \ref{ex:
isolated}. The associated semicomplex $\Sigma$
consists of the cone over a square and the apex of this cone. Then $G^\sigma\cdot
x_i=(x_1,x_2,x_3,x_4)$ is the ideal $m_p$ of the singular point $p$. Hence the
torific ideal $I_\sigma$, where $s=\{p\}$, is equal to $m_p^4$.
$\Delta(I_\sigma)$ is the star
subdivision at the ray $\varrho$ over the centre of the square. By
Proposition \ref{pr: torification}, $${\operatorname{Stab}}(\Sigma)=
{\operatorname{Inv}}(\sigma) =\varrho.$$
\end{example}
\begin{example}\label{ex: Hironaka2} Let $(X,S)$ be a
smooth stratified toroidal variety of dimension $3$
with the stratification determined by two curves $l_1$ and
$l_2$ meeting transversally at some point $p$. The
associated semicomplex $\Sigma$ consists of the regular
$3$-dimensional cone $\sigma=\langle w_1,w_2,w_3 \rangle $ and its two $2$-dimensional
faces $\tau_1=\langle w_1,w_3 \rangle $ and $\tau_2=
\langle w_2,w_3 \rangle $. Set $v_1:=w_1+w_3$,
$v_2:=w_2+w_3$, $v:=w_1+w_2+w_3$.
Write $l_1: x_1=x_3=0$, $l_2: x_2=x_3=0$. Then $(G^\sigma\cdot
x_1)=I_{l_1}=(x_1,x_3)$, $(G^\sigma\cdot
x_2)=(x_2,x_3)$,
$(G^\sigma\cdot x_3)=I_{l_1\cup l_2}=(x_3,x_1x_2)$. Then
$\Delta(I_\sigma)$ is determined by two star subdivisions:
$\langle v_1 \rangle\cdot\sigma$, $\langle v_2 \rangle\cdot\sigma$,
and the subdivision $\Sigma_0$ of $\sigma$ determined by
${\operatorname{ord}}((x_3,x_1x_2)):=\min(w^*_3, w_1^*+w_2^*)$. The latter
consists of two cones separated by the $2$-dimensional
face $\langle v_1, v_2 \rangle$.
Then ${\operatorname{Inv}}(\sigma)=\langle v_1, v_2, v \rangle$.
Also ${\operatorname{Inv}}(\tau_i)= \langle v_i \rangle$. Finally,
$${\operatorname{Stab}}(\Sigma)={\operatorname{Inv}}(\sigma)=\langle v_1, v_2, v \rangle.$$
\end{example}
The above considerations can be generalized to any regular
semicomplex. We shall skip the details of the proof.
\begin{proposition}
Let $\Sigma$ be a regular semicomplex. For any cone
$\sigma=\langle v_1,\ldots, v_{k_\sigma} \rangle$ set \\ $v_\sigma:=
v_1+\ldots+v_{k_\sigma}$.
Then $${\operatorname{stab}}(\sigma)=\bigoplus_{\tau\leq \sigma} {\bf Q}\cdot v_{\tau}.$$
\qed
\end{proposition}
This formula can be extended to any oriented simplicial semicomplex.
\begin{conjecture}
Let $\Sigma$ be an oriented simplicial semicomplex. For any
$\sigma\in \Sigma$ let $V_\sigma$ denote the set of all the minimal
internal vectors of $\sigma$ and $W_\sigma:=\bigcup_{\tau\leq \sigma}
V_{\tau}$. Then
$${\operatorname{stab}}(\sigma)= \bigoplus_{v\in W_\sigma} {\bf Q}\cdot v.$$
\end{conjecture}
|
\section*{Abstract}
A multi-cloud model is presented which explains the soft X-ray excess
in NGC 4051 and, consistently, the
optical line spectrum and the SED of the continuum.
The clouds are heated and ionized by the photoionizing
flux from the active center and by shocks.
Diffuse radiation, partly absorbed throughout the clouds,
nicely fits the bump in the soft X-ray domain,
while bremsstrahlung radiation from the gaseous clouds
contribute to the fit of the continuum SED.
Debris of high density fragmented clouds are necessary to explain
the absorption oxygen throats observed at 0.87 and 0.74 keV.
The debris are heated by shocks of about 200-300 $\rm km\, s^{-1}$ .
Low velocity ($\leq$ 100 $\rm km\, s^{-1}$ ) - density (100 $\rm cm^{-3}$) clouds contribute to the
line and
continuum spectra, as well as high velocity (1000 $\rm km\, s^{-1}$ ) - density (8000 $\rm cm^{-3}$)
clouds
which are revealed by the FWHM of the line profiles.
The SED in the IR is explained by reradiation of
dust, however, the dust-to-gas ratio is not particularly high ($\leq
3\times 10^{-15}$).
Radio emission is well fitted by synchrotron radiation created
at the shock front by Fermi mechanism.
\newpage
\section{Introduction}
NGC 4051 is a SAB galaxy at z=0.0023. It is classified as a Seyfert 1 galaxy,
and
is characterized by unusually narrow permitted lines, only
moderately wider than the forbidden lines (Osterbrock 1977).
Simultaneous observations by ROSAT-IUE and GINGA
of a sample of 8 Seyfert 1 galaxies (Walter et al 1994),
including NGC 4051, show that the
UV to X-ray spectral energy distribution (SED) can be
decomposed into two major distinct components:
a nonthermal hard X-ray continuum and a broad emission excess (bump)
spanning from UV to soft X-rays.
All models (power-law, thin disk, bremsstrahlung, black body)
are able to reproduce the soft X-ray
spectra in the Walter et al. sample, except the power-law model for NGC 4051.
The evidence that the power-law model is not a good representation
of the X-ray spectrum of NGC 4051,
when the absorbing column density is fixed to the galactic value, contrasts with
the long established situation that
the power-law is the best $simple$ fit to X-spectra of active galactic nuclei (AGN).
This indicates that the spectrum is
more complicated, and, for example, could be affected by intrinsic absorption.
The addition of a soft component to the X-ray spectrum,
which accounts for an excess in the 0.1-2 keV band,
is also raised by Fiore et al. (1992). From the study of
GINGA data they found a spectral variability
consistent with a constant underlying power-law slope modified by partial
covering
or by a 'warm absorber'.
On the other hand, the soft excess frequently found in EXOSAT spectra
were formerly
interpreted as thermal emission from the innermost regions of a viscous
heated accretion disk (Arnaud et al. 1985).
As suggested by Pounds et al. (1994), both
the soft excess and the blue bump emission may
arise, instead, from reprocessing the hard X-rays in dense cold cloudlets
surviving close to the
central source. The ionizing photons absorbed in optically thick material
will be reemitted at the black-body equilibrium temperature ($10^5-10^6$ K)
as long as the density of the absorbing gas is sufficiently high.
Maximum temperatures of the bump component are evaluated to about 5$\times
10^5$ K (Walter et al. 1994).
Variability on small scale of NGC 4051 is used to reveal the characteristics
of the emitting clouds and of the velocity field.
The observed spectral variability on scale of hours can be
explained in terms of a change in ionization parameter plus an emerging soft
excess
(Pounds et al. 1994).
The assumption of a typical variability time scale of one hour leads to a matter
density
of $\geq 5 \times 10^7$ $\rm cm^{-3}$ and a thickness of the photoionized gas of $\leq
1.4 \times
10^{14}$ cm. Considering that the radius of the source
of the optical and UV continuum is larger than a few $10^{14}$ cm,
Walter et al. (1994) claim that it can be covered in
less than 3 years if the velocity of the absorbing clouds is larger than
200 $\rm km\, s^{-1}$ .
From the spectral observations in the optical range (De Robertis \&
Osterbrock 1984)
it appears that there may not be a simple dichotomy between the
broad line region (BLR) and the narrow line region (NLR) in NGC 4051.
Instead, the continuum of line widths suggests
that the emitting regions may be
inhomogeneusly filled with clouds or filaments showing densities in a very
large range, so that the division
into a BLR and a NLR is an extreme simplification.
The asymmetry of the narrow-line profiles is consistent with radial outflow
or expansion of the gas.
The presence of strong blue wings (Veilleux 1991) favors models
with radial motion and a source of obscuration.
NGC 4051 is peculiar also because its UV spectrum is very steep and probably
affected by intrinsic reddening.
The presence of dust in the nucleus of NGC 4051 has been suggested by several
groups (Walter et al 1994).
Veilleux (1991) claims that dust is probably present and
is the source of line asymmetry and that the differences in profiles
of H$\beta$ ~and H$\alpha$ are due to reddening and/or optical depth
effects. Balmer fluxes are known to vary over periods shorter than
one year.
The basic model of a warm absorber (Pounds et al. 1994) consists of
a gaseous region close to the BLR photoionized by the central radiation,
originating absorption features in the central
X-ray radiation. This radiation is represented by a power-law characterized
by a photon index
and normalized by the intensity at 1 keV (photons $\rm cm^{-2}s^{-1}
keV^{-1}$).
The procedure is first to fit the power-law plus a cold absorbing column.
Then, add one or more components seeking the best fit.
The warm absorber model introduces two free parameters: the column density, N,
and
the ionization parameter, U.
Principal absorption edges are identified with ionized carbon, nitrogen, and
oxygen.
The effect of a warm absorber can be interpreted also as a set of emission
features.
Komossa \& Fink (1997) have recently modeled the
absorbed spectrum of NGC 4051 from ROSAT observations
in terms of warm absorption. Their excellent model can
explain most of the features observed in the X-ray spectrum.
However, some discrepancies with a general scenario still appear, e.g., reduced
solar
abundances are in contrast with the absence of dust, a single component warm
absorber
is in contrast with the plurality of cloud conditions found by
De Robertis \& Osterbrock (1984).
In this work we consider NGC 4051 in the light of a multi-cloud model,
as was found appropriate also for other galaxies (e.g. the Circinus galaxy
model by
Contini, Prieto \& Viegas 1998).
The clouds move radially outward the galaxy.
A composite model which consistently accounts for the effects of the radiation
from the active center and of the shocks on the emitting clouds is adopted.
Particularly, we suggest that the soft X-ray excess can be
reproduced by reprocessed radiation (diffuse radiation) emitted from the
hot slabs of gas after being partially
absorbed by optically thick regions throughout
the clouds.
The consistency of the model is checked up by fitting both the SED of the
continuum
in a large frequency range and the emission lines.
In \S 2 the general model is described. In \S 3 the soft X-ray excess
is modeled. The results
of model calculations are compared with the observed continuum SED and
the line spectrum in \S 4 and \S 5, respectively. Final remarks
follow in \S 6.
\section{The model}
Clouds in the NLR with different physical conditions and with
radial outward motion are assumed by the model.
A shock forms on the outer edge of the clouds while the power-law
radiation from the active center (a.c.) reaches the inner edge.
The flux from the central source and the shock are the primary sources
of ionization and heating of the gas.
The cloud may be depicted as consisting of a large number of parallel slabs
in which the conditions within any given increment are essentially uniform.
The gas is ionized and originates a diffuse radiation field through
recombination and lines. The intensity of the diffuse radiation depends on the
source
function which cannot be determined unless the ionization equilibrium is already
known (Williams 1967).
In the source function are therefore implicit the effects of collisional
ionization.
The role of diffuse secondary radiation
in shocked clouds is illustrated by Viegas, Contini \& Contini (1998).
Diffuse radiation which emerges from the heated slabs
is partially absorbed throughout the cloud.
We suggest that the soft X-ray excess may be produced by the reprocessed
radiation, e.g. diffuse radiation, emitted from a group of clouds in
different physical conditions.
The SUMA code is adopted (Viegas \& Contini 1994 and references therein).
A composite ionizing spectrum with spectral index $\alpha_{UV}$ = 1.4 and
$\alpha_X$ = 0.4 is assumed for all models as for previous modeling (see
Contini, Prieto, \& Viegas 1998).
The other input parameters, i.e., the shock velocity, $\rm V_{s}$, the preshock
density, $\rm n_{0}$, the radiation flux
intensity at 1 Ryd, $\rm F_{\nu}$ (in number of photon $\rm cm^{-2} \, s^{-1} \, eV^{-1}$),
and the dust-to-gas ratio by number, d/g, are chosen from the observational
evidence and are adjusted by the modeling.
Cosmic abundances are assumed (Allen 1973) and the preshock magnetic field
$\rm B_{0}$ = $10^{-4}$ gauss.
\section{The soft X-ray excess}
The observational data are
taken from Komossa \& Fink (1997, Fig. 5).
We assume that many clouds, at different physical conditions, contribute to the
'warm absorption'.
A large range of densities is considered. Actually, a continuum distribution
of densities could be present.
Velocities are in agreement with the observed emission line FWHM (De Robertis \&
Osterbrock 1974, Veilleux 1991).
Some clouds are shock dominated, other are radiation dominated with radiation
intensities in the range usually found for the NLR of Seyfert galaxies.
The input parameters of the models used to fit the X-ray excess are listed
in Table 1.
The best fit of the X-ray excess by model calculations
is shown in Figure 1.
Models 12, 13, and 14 are not included in the figure
and will be discussed further on.
Each model represents one type of cloud with a characteristic
geometrical width, D, which is also given in Table 1. Shock dominated (SD)
models are calculated adopting $\rm F_{\nu}$ = 0. Notice that model 8 is a SD
model corresponding to the radiation dominated (RD) model 9. For each model,
the diffuse radiation emitted by each cloud is calculated.
The thick solid line in Fig. 1 represent the weighted summed spectrum from the
models.
In the last column of Table 1 the weights, W, adopted
for single-cloud models in the average sum are shown.
The weights
reflect the dilution factor $\rm (r/d)^2$ (r is the distance
from the clouds to the center and d is the distance to earth)
and the covering factor.
It can be noticed that an acceptable fit to the observations
is obtained by the present sample of models, considering that the
observed data are contaminated by residuals and errors.
X-ray data at energies higher than 1.3 KeV can be well fitted by the flat
power-law radiation from the central source, which is directly reaching the
observer.
Lower energy photons are generally absorbed by the cloud. This can be easily
seen in the curve representing model 5, which corresponds to high density
clouds.
The spectrum given by model 7 shows a trend similar to the observed one,
however lacks the throats of absorption in the critical edges.
The two primary flux models have the lowest weights.
High density clouds are invoked to fit the deep throat at about 0.85 KeV,
which is due to the O VIII edge.
The geometrical thickness of the clouds are small, particularly for dense
clouds, indicating that fragmentation is rather strong. This is
consistent with a regime of turbulence in the presence of shocks.
The weights of the high density models
are very high, particularly for model 3.
In the corresponding clouds the cooling rate is
high, due to the high density; moreover, the hot gas emitting region
is very small. Consequently, the flux emitted by each cloud is weak
and many clouds are necessary to fit the data.
This is consistent with the small D.
Model 6 with a larger D
has also a high weight. In this case most of the gas
inside the cloud is cold and neutral because
the intensity of the central radiation is relatively low.
Modeling implies the choice of a composite model which is seldom unique.
The validity of the present composite model for the warm absorber
will be checked in the next sections by the consistent fit to
the observed continuum SED and to the line spectrum.
\section{The continuum}
References of the observed continuum below $10^{16}$ Hz are given in Table
2.
Data in the X-ray range are from Komossa \& Fink (1997).
The SED of the observed continuum is plotted in Fig. 2.
Optical observations integrated over the whole galaxy are not included.
For energies less than 13.6 eV, the continuum calculated by the models,
which is essentially the sum of bremsstrahlung
radiation emitted from the gas within the clouds, roughly fit the data.
The weights adopted to sum up the models are the same as listed in Table 1.
Reradiation by dust in the IR depends
strongly on the shock velocity. The observed IR maximum constrains the
dust-to-gas ratio (d/g value), while the
frequency corresponding to the maximum depends on the dust temperature.
The grains are heated by radiation and by collisions. Dust and gas mutual
heating
and cooling determine the temperature of dust which follows the temperature of
the gas (Viegas \& Contini 1994).
For all the models dust-to-gas ratios in the range
1-3$\times 10^{-15}$ are adopted.
The SED of the continua corresponding to single models 1,2,3,4,6,8,9,10, and
11
(dotted lines) and their weighted sum (solid lines) are shown in Fig.2. The
three components originating in the clouds (synchrotron emission due to Fermi
mechanism, dust emission, and free-free emission) are shown separetely.
It can be noticed
that most of the models are below the lower edge of the figure because their
weight is very low. The weighted sum corresponds to the SED of model 3 the
weight of which largely prevails.
Depending on the models, the bremsstrahlung component peaks at a different
frequency. So, the weighted average shows two peaks, one at $\sim 10^{14}$ Hz
and another at 3 $\times 10^{16}$ Hz.
Actually, absorption by ISM peaks at 3 $\times 10^{16}$ Hz (Zombeck 1990).
In the radio range ($< 10^{10}$ Hz), the
free-free emission is higher than the observational data, which are nicely
fitted
by synchrotron emission due to Fermi mechanism at the shock front. As happens for
Circinus,
the bremmstrahlung emission at such low frequencies is probably absorbed.
In fact, if
we assume an average temperature for the clouds of about 10$^4$ K, the optical
depth for free-free absorption is greater than unity for $\nu \leq 10^{11}$,
increasing at lower frequencies.
Notice, however, that the observed optical continuum is not well fitted yet.
Moreover, reradiation by dust calculated by the models peaks at $10^{13}$ Hz,
while the data peaks at $\sim 3 \times 10^{12}$ Hz.
Therefore, the ensemble of clouds which explain the X-ray data
is not complete, and models representing other clouds at different physical
conditions must be included in the multi-cloud model.
A final choice of the best fitting models will be possible after
discussing the line spectrum. In fact, modeling the line and continuum
spectra simultaneously implies cross checking of one another
until a fine tuning of the models is obtained.
\section{The optical - near-UV line spectrum}
The observed line spectrum is taken from Malkan (1986, Table 1).
A typing error crept in the published data has been corrected
(H$\beta$ = 31. and not 3.1, M. Malkan, private communication).
The data are reddening corrected adopting E(B-V) = 0.32
which represent the obscuration inside the clouds (Malkan 1986).
This is higher than the
intrinsic reddening, E(B-V)=0.08 and galactic reddening E(B-V)=0.02.
Notice that Walter et al. (1994) obtain E(B-V)=0.05-0.13.
The calculated line intensities relative to H$\beta$
are compared to the observations in Table 3.
Radiation dominated models provide relatively high HeII/H$\beta$ ~(e.g. model 9),
while shock dominated models provide higher [OII]/H$\beta$ ~and
[OIII]4363/H$\beta$ ~(e.g. model 8).
The results presented in Table 3 indicate that both radiation-dominated
and shock-dominated clouds should be taken into account for the final
multi-cloud model.
Model AV0 corresponds to the weighted average of the single-cloud
models accounting for the X-ray data.
This average model gives line ratios practically identical to model 3,
which, in fact, largely prevails. However, the fit to the observed
line ratios is not good enough.
Therefore, models 12, 13, and 14, which are negligible in the fit of the
soft X-ray excess, are invoked to improve the fit of the line ratios
observed in the optical-near UV range.
The most noticeable features
in the spectrum of NGC 4051 are that the ratio of the high ionization line
widths
to the low ionization line widths is considerably smaller than for other objects
in the sample of De Robertis \& Osterbrock (1984) and that blue wings
up to -800 $\rm km\, s^{-1}$ are present in all the forbidden line profiles (Veilleux
1991).
Veilleux (1991) also noticed four 'shoulders' (at -40, -110, -180,
and -350 $\rm km\, s^{-1}$ ) in the observed line profiles. The line intensity ratios
observed by Veilleux are not included in Table 3 which
shows the line ratios to H$\beta$. In fact, the narrow and broad component of H$\beta$
~could not be deblended.
The 'shoulders' indicate the velocities of emitting
clouds which are represented by models 14, 13, 3, in addition
to those represented by models 4 to 10.
Model 12 represents the high velocity gas. The preshock density is high enough
to cause the rapid cooling of the gas downstream. So, low ionization line
ratios relative to
H$\beta$ ~are particularly high (Table 3), and bremsstrahlung emission
in the X-ray domain is low (Fig. 3).
Dust reemission is completely annihilated by the sputtering of the grains
in the immediate postshock region.
Model 13 is characterized by a low $\rm V_{s}$ ~(100 $\rm km\, s^{-1}$ ), a low $\rm n_{0}$ (100 $\rm cm^{-3}$), and a
low primary flux ($\rm F_{\nu}$ = 5 $10^{10}$ units).
These conditions generally represent the clouds either in the outer NLR or
in Liners (Contini 1997).
This model shows a too high [OIII] 4959/[OII] 3727 line ratio ,
while model 14, which is a SD model characterized by a low $\rm V_{s}$ (50 $\rm km\, s^{-1}$ ),
provides a very strong [OII]/H$\beta$ .
The averaged spectrum (AV1) is given in the last column of Table 3 and shows an
acceptable fit to the observations.
The relative contributions of models 3,12,13 and 14 to single line fluxes
are shown in Table 4.
As the models are distinguished particularly by the shock velocities, the results
refer to the line profile features observed by Veilleux (1991).
Model 3 is chosen to represent the contribution of the clouds generating the
X-ray excess.
Interestingly, the shock velocities adopted to fit the spectra
confirm the observed prevailing FWHM of about 200-300 $\rm km\, s^{-1}$ in the
observed line profiles
of H$\beta$, [OIII] 4363, and [OI]. Models calculated with $\rm V_{s}$ = 100 $\rm km\, s^{-1}$
provide 98-99 \% of the [OIII] 4959+5007 and HeII 4686 lines, 94 \% of
He II 3200, and 85-86 \% of the [NeV] 3426 and [NeIII] 3869 lines, respectively.
Models calculated with $\rm V_{s}$ = 50 $\rm km\, s^{-1}$ contribute essentially to
the [OII], [NII], and [SII] 6717,6730 lines. The high velocity model (12)
contributes to all the lines, except [OIII], [NeIII], [NeV], and HeII 4686.
Model results are not in full agreement with
observations for all the lines, but they roughly show how complex is the
structure of the NLR which
extends from the edge of the BLR to the outskirts of the galaxy.
Models 13 and 14 are chosen also to improve the fit of the continuum SED.
In fact,
the mutual heating of dust and gas provides a dust temperature low enough to
settle the peak in the IR at about 3 $10^{12}$ Hz and the trend
of model AV1 in the optical range improves the fit which was obtained by model
AV0 (Fig. 3). Clouds corresponding to models 12, 13, and 14
contribute mostly to emission lines but not to the X-ray excess.
\section{Final Remarks}
Komossa \& Fink find that the X-ray spectrum consists of a power-law modified by
absorption edges and an additional soft excess during the high-state in source
flux.
Their results indicate a column density of ionized material of log N = 22.7
and a ionization parameter of log U = 0.4. The underlying power-law is in its
steepest
observed state with photon index $\Gamma_X$ = -2.3.
They assume that the absorber is one-component with a
gas temperature of 3 $\times 10^5$ K, metal abundances up to 0.2 $\times$
solar,
electron density $\rm n_{e}$ $\leq 3 \times 10^7$ $\rm cm^{-3}$, a thickness D $\geq 2 \times
10^{15}$
cm, at a distance r $\geq 3 \times 10^{16}$ cm from the central power
source ,
and no dust. Moreover, they claim that no emission line component can be fully
identified with the warm absorber. Notice that in photoionization models,
the high temperature is associated
to a low chemical abundance. However, it is well known
that the galaxies often show an abundance gradient, indicating higher abundances
in the central regions. Thus unless the depletion is due to dust, it seems
unreal to assume such low abundances in the central region of NGC 4051.
In the previous sections we have selected the models
which consistently fit the observed continuum in all the
frequency ranges and the line ratios.
Our results show that the so-called warm absorber is composed by many
clouds in different physical conditions. The column density
within each cloud contributing to the soft X-ray excess
does not exceed 5$\times 10^{20}$ $\rm cm^{-2}$,
considering a postshock compression of $\sim$ 10 for low velocity clouds
(200-300 $\rm km\, s^{-1}$ ).
Comparing with the results obtained by Komossa \& Fink this
indicates that hundreds of clouds form the warm absorber.
The preshock densities span from the values which fit the NLR
to values approaching those of the BLR, i.e. between 400 and $10^7$ $\rm cm^{-3}$,
in agreement with the predictions of De Robertis \& Osterbrock (1984) that a
large
range of densities characterize the emitting clouds in NGC 4051.
Obviously, the clouds responsible of the edge absorption are the densest,
in agreement with the density indicated by Komossa \& Fink.
Moreover, they predict no dust, while
the modeling of the continuum in the present work shows that
dust is present inside the clouds to explain the IR emission.
However, the dust-to-gas ratio is rather low, even lower by a factor $\geq 3$
than found for Liners by Viegas \& Contini (1994).
The central radiation flux reaching the clouds
ranges from $10^{11}$ to $10^{13}$ photons per cm$^{-2}$
s$^{-1}$ eV$^{-1}$ at the Lyman limit which are "normal"
values in the NLR of AGN (Viegas \& Contini 1994). However,
the high density clouds which explain the soft X-ray excess at $\sim$ 1 keV
are all shock dominated. This is an interesting result which shows that the
gas is heated by the shock. Temperatures of about 6$\times 10^{5}$ K correspond
in fact to shocks of $\sim$200 $\rm km\, s^{-1}$ . These temperatures are in agreement
with the temperatures
predicted by previous models (Komossa \& Fink 1997 and references therein)
which were explained by very strong
radiation from the active center photoionizing and heating a gas
characterized by low metal abundances, in order to reach such high
temperatures.
Our model shows that shock dominated clouds are present in large number,
indicating
that the central source radiation is screened, probably by the BLR clouds,
and that the filling factor is high.
Because the high temperature is due to shock, the fit to the
observations is obtained with cosmic abundances.
Finally, in the present warm absorber model we assume that some very dense
clouds
are characterized by relatively low velocities (Table 1, models 1, 2, and 3).
In the nuclear region of the Circinus galaxy some clouds characterized by
velocities of 250 $\rm km\, s^{-1}$ and preshock densities of 5000 $\rm cm^{-3}$ were invoked in
order to fit the emission spectra (Contini et al. 1998).
Generally, in AGN, higher densities correspond to higher velocities
as for the clouds corresponding to model 12 ($\rm n_{0}$ = 8000 $\rm cm^{-3}$, $\rm V_{s}$ = 1000 $\rm km\, s^{-1}$ ).
Notice that the high density low velocity clumps are characterized by a very
small geometrical thickness in NGC 4051, therefore, they could be identified
with
the debris of high density-velocity clouds from the BLR edge which
have been fragmented by cloud collision in a turbulent regime. Fragmentation
is generally accompanied by a considerable loss of kinetic energy.
If model 3 represents these debries, their distance from the
active center can be calculated from
$\rm F(H\beta)_{obs}$ $\rm d^2 $ = $ \rm F(H\beta)_{calc}$ $\rm r^2$,
where $ \rm H\beta_{obs}$ ~is the absolute flux of H$\beta$ ~observed at earth
(Malkan 1994),
d is the
distance of the galaxy from earth (d=14 Mpc), $ \rm H\beta_{calc}$ ~is the
absolute
flux of H$\beta$ ~calculated at the gaseous clump, and r is the distance of the
clumps from the
active center. Adopting $ \rm H\beta_{obs}$ = 31.$\times 10^{-14}$ $\rm erg\, cm^{-2}\, s^{-1}$
and $ \rm H\beta_{calc}$ = 27. $\rm erg\, cm^{-2}\, s^{-1}$ (see Table 3), r results 1.5 pc. The high
density
components of the warm absorber are thus located between the NLR and the BLR.
In conclusion, a multi-cloud model can explain the soft X-ray excess
in NGC 4051, the optical emission line spectrum and the SED of the continuum.
Indeed, modeling implies the choice of some conditions
which should actually prevail. So single-cloud models
must be considered as prototypes.
The clouds are heated and ionized by the photoionizing
flux from the active center and by the shocks. Due to the high temperature
gas, diffuse radiation, partly absorbed throughout the clouds,
is used to explain the bump in the soft X-ray domain,
while free-free emission from lower temperature gas
and dust reradiation from the ensemble
of the clouds mainly fit the far IR to optical continuum.
The fit of the continuum in the IR shows that the
dust-to-gas ratio is not particularly high ($\leq 3\times 10^{-15}$).
Radio emission is well fitted by synchrotron radiation created
at the shock front by Fermi mechanism.
Debris of high density fragmented clouds are necessary to explain
the absorption oxygen throats observed at 0.87 and 0.74 keV.
The debris are heated by shocks of about 200-300 $\rm km\, s^{-1}$ .
Low velocity ($\leq$ 100 $\rm km\, s^{-1}$ ) - density (100 $\rm cm^{-3}$) clouds
and high velocity (1000 $\rm km\, s^{-1}$ ) - density (8000 $\rm cm^{-3}$) clouds,
which are revealed from the FWHM of the line profiles,
contribute to the line and continuum spectra.
\bigskip
\noindent
{\it Acknowledgements}.
This paper was partially supported by the Brazilian funding agencies
PRONEX/Finep, CNPq, and FAPESP.
\newpage
{\bf References}
\bigskip
\vsize=26 true cm
\hsize=14 true cm
\baselineskip=18 pt
\def\ref {\par \noindent \parshape=6 0cm 12.5cm
0.5cm 12.5cm 0.5cm 12.5cm 0.5cm 12.5cm 0.5cm 12.5cm 0.5cm 12.5cm}
\ref Allen, C.W. 1973 in "Astrophysical Quantities" (Athlon)
\ref Arnaud, K.A. et al. 1985, MNRAS, 217, 105
\ref Balzano,V.A., \& Weedman,D.W. 1981, ApJ, 243,756
\ref Contini, M. 1997, A\&A, 323, 71
\ref Contini, M., Prieto, M.A., \& Viegas, S.M. 1998, ApJ, 505, 621
\ref De Robertis, M.M. \& Osterbrock, D.E. 1984, ApJ, 286, 171
\ref De Vaucouleurs,G., De Vaulcouleurs, A., Corwin, G.H.G. et al.
1991, Third Reference Catalogue of Bright Galaxies, Version 3.9
\ref De Vaucouleurs, A., Longo, G., 1988, Catalogue of Visual and Infrared
Photometry of Galaxies from 0.5 $\mu$m to 10 $\mu$m (1961-1985)
\ref Fiore, F. et al. 1992, A\&A, 262, 37
\ref Gregory,P.C. \& Condon,J.J. 1991 ApJS, 75,1011
\ref Ficarra,A., Grueff,G., \& Tomasetti,G. 1985, A\&AS, 59,255
\ref Komossa, S. \& Fink, H. 1997, A\&A, 322, 719
\ref Lebofsky,M.J. \& Rieke,G.H. 1979, ApJ, 229,111
\ref McAlary,C.W., McLaren,R.A., \& Crabtree,D.R. 1979, ApJ, 234,471
\ref Malkan, M. 1986, ApJ, 310, 679
\ref Moshir, M. et al. 1990, Infrared Astronomical Satellite Catalogs,
The Faint Source Catalog, Version 2.0
\ref Osterbrock, D.E. 1977, ApJ, 215, 733
\ref Penston,M.V., Penston,M.J., Selmes, R. A., Becklin, E. E.,
\& Neugebauer, G. 1974, MNRAS, 169,357
\ref Pounds, K.A., Nandra, K., Fink, H.H., \& Makino, F. 1994, MNRAS, 267, 193
\ref Rieke,G.H. 1978, ApJ, 226,550
\ref Rieke,G.H. \& Low, F.J. 1972, ApJL, 176,L95
\ref Stein,W.A., \& Weedman,D.W. 1976, ApJ,205,44
\ref Soifer,B.T., Boehmer, L., Neugebauer, G., \&Sanders, D. B.
1989, AJ, 98,766
\ref Veilleux, S. 1991, ApJ, 369, 331
\ref Viegas, S.M. \& Contini, M. 1994, ApJ, 428, 113
\ref Viegas, S.M., Contini, M., \& Contini, T. 1998, A\&A submitted
\ref Walter, R. et al. 1994, A\&A, 285, 119
\ref Williams, R.E. 1967 ApJ, 147, 556
\ref Wisniewski,W.Z. \& Kleinmann,D.E. 1968, AJ,73,866
\ref Zombeck, M.V. 1990 in "Handbook of Space Astronomy and Astrophysics"
Cambridge University Press, p. 199
\newpage
{\bf Figure Captions}
\bigskip
Fig. 1
The X-ray spectrum of NGC 4051. Filled squares indicate the data
Single model results are indicated by numbers which refer
to Table 1. The thick solid line represents the weighted sum
which best fits the data.
\bigskip
Fig. 2
The fit of the SED of the continuum by the ensemble of models
which form the warm absorber. Dotted lines represent
the single models and solid lines the weighted sum.
Filled squares refer to observations in the X-ray (see Fig. 1).
Open squares indicate the observation data at lower
frequencies.
\bigskip
Fig. 3
Same as Fig. 2 including models 12 (dotted line), model 13
(short-dashed lines), and model 14 (long-dashed lines).
The dot-dashed line shows that free-free emission is absorbed
in the radio range.
The thick solid lines refer to model AV1 and the thin solid lines
to AV0.
\newpage
\begin{table}
\centerline{Table 1}
\centerline{The input parameters of the models}
\begin{center}
\small{
\begin{tabular}{rrr rlr }\\ \hline \\
\ model & $\rm n_{0}$ & $\rm V_{s}$ & D &log($\rm F_{\nu}$) & log(W) \\
\ &($\rm cm^{-3}$)& ($\rm km\, s^{-1}$ ) & (cm)& - &- \\
\hline \\
\ 1&1.(6)&300 & $<$4.5(13)&sd &-6.8 \\
\ 2&5.(5)&200 & 2.(12)& sd&-4.8 \\
\ 3&4.(5)&200 & 3.(13)&sd& +0.1 \\
\ 4&800& 400 & $<$2.(16)& sd&-10.4 \\
\ 5& 800 & 420 & 5.(16)& 12 & -13.2 \\
\ 6& 800 & 300 & 2.(17)& 11 & -6.9 \\
\ 7& 600 & 300 & 1.(16) &11 & -13.0 \\
\ 8&600 & 300 & $<$1.(16)& sd &-9.4 \\
\ 9&600 & 300 & 1.(16)& 11 & -10.2 \\
\ 10&500 &400 &3.(16)&13&-10.7 \\
\ 11&400 & 200 & 7.(16)& 12& -8.4 \\
\ 12 &8000 & 1000 & 3.(17) & sd&-4.9 \\
\ 13 &100 & 100 & 3.(16) &10.3 & +4.0 \\
\ 14 &100 & 50 & $<$3(16)& sd& +3.8 \\
\hline \\
\end{tabular}}
\end{center}
\end{table}
\begin{table}
\centerline{Table 2}
\centerline{The data of the continuum}
\tiny{
\centerline{
\begin{tabular}{l ll } \\ \hline \\
Wavelength &Frequency (Hz) & Reference \\ \hline \\
4400 A & 6.81 $10^{14}$ & 1 \\
5530 A & 5.42 $10^{14}$ & 1, 2, 3 \\
7000 A & 4.28 $10^{14}$ & 2, 3 \\
9000 A & 3.33 $10^{14}$ & 2 \\
1.23 $\mu$m & 2.44 $10^{14}$ & 4, 5 \\
1.25 $\mu$m & 2.40 $10^{14}$ & 6 \\
1.26 $\mu$m & 2.38 $10^{14}$ & 2 \\
1.60 $\mu$m & 1.84 $10^{14}$ & 6, 7 \\
1.66 $\mu$m & 1.81 $10^{14}$ & 4, 5 \\
2.20 $\mu$m & 1.36 $10^{14}$ & 7 \\
2.22 $\mu$m & 1.35 $10^{14}$ & 2, 3, 4, 5, 6 \\
3.45 $\mu$m & 8.69 $10^{13}$ & 6 \\
3.50 $\mu$m & 8.57 $10^{13}$ & 8, 9 \\
3.54 $\mu$m & 8.47 $10^{13}$ & 3 \\
3.65 $\mu$m & 8.21 $10^{13}$ & 4 \\
4.65 $\mu$m & 6.45 $10^{13}$ & 4 \\
5.00 $\mu$m & 6.00 $10^{13}$ & 9 \\
10.5 $\mu$m & 2.86 $10^{13}$ & 9 \\
10.6 $\mu$m & 2.83 $10^{13}$ & 6, 10 \\
12.0 $\mu$m & 2.50 $10^{13}$ & 11, 12 \\
21.0 $\mu$m & 1.43 $10^{13}$ & 10 \\
25.0 $\mu$m & 1.20 $10^{13}$ & 11 \\
60.0 $\mu$m & 5.00 $10^{12}$ & 11, 12 \\
100.0 $\mu$m & 3.00 $10^{12}$ & 11, 12 \\
6.18 cm&4.85 $10^9$ & 13 \\
73.5 cm& 4.08 $ 10^8$ & 14 \\ \hline \\
\end{tabular}}}
1: De Vaucouleurs,G. et al. 1991 ; 2: Wisniewski,W.Z. \& Kleinmann,D.E. 1968;
3: De Vaucouleurs,G. et al. 1988; 4: McAlary,C.W., McLaren,R.A., \&
Crabtree,D.R. 1979;
5: Balzano,V.A., \& Weedman,D.W. 1981; 6: Rieke,G.H. 1978;
7: Penston,M.V. et al. 1974; 8: Stein,W.A., \& Weedman,D.W.
1976; 9: Rieke,G.H. \& Low, F.J. 1972; 10: Lebofsky,M.J. \& Rieke,G.H.
1979; 11: Moshir, M. et al. 1990; 12: Soifer,B.T. et al. 1989;
13: Gregory,P.C. \& Condon,J.J. 1991; 14: Ficarra,A., Grueff,G.,
\& Tomasetti,G. 1985.
\end{table}
\oddsidemargin 0.1cm
\evensidemargin 0.1cm
\vsize=24 true cm
\hsize=17 true cm
\begin{table}
\centerline{Table 3}
\centerline{The line spectra}
\tiny{
\begin{tabular}{lll llll l l l l llllll}\\ \hline \\
\ & $\rm obs^1$ & mod & mod& mod & mod &mod &mod&mod&mod&mod& mod& mod & mod &
mod & mod \\
\ & &1&2&3& 4 &6&8&9 &10& 11& AV0 &12& 13 & 14 & AV1 \\
\hline \\
\ [OIII] 4959 & 0.45&1.5(-4)&0.019&0.013&0.18&0.14&3.5&7.7&0.06&3.6
&0.013&2(-4)&2.5 & 0.075&0.76 \\
\ HeII 4686 & 0.2&1.7(-3)&2.6(-3)&1.5(-3)&0.027&3.3(-4)&0.04&0.68&0.89
&0.85&1.5(-3)&6(-6)&0.9&0.0&0.27 \\
\ $\rm H_{\gamma}$&0.40&0.46&0.45&0.45&0.45&0.45&0.43&0.47&0.47&0.47&
0.45&0.45&0.45&0.45&0.46 \\
\ [OIII] 4363 &$\uparrow$&3(-3)&0.15&0.08&0.06&1(-4)&0.44&0.35&0.01&0.27&0.08&
4(-4) & 0.14 & 0.014 & 0.09 \\
\ [SII] 4072+ & -&1.&0.09&0.06&0.47&0.05&0.48&1.2&0.0&1(-5)& 0.06 & 1.2& 2(-5)
& 0.87 & 0.16 \\
\ [NeIII] 3869 & 0.25&0.025&0.085&0.05&0.36&7.3(-3)&1.32&4.4&0.014&0.8&0.05 &
3(-3)& 0.74 & 0.05 & 0.26 \\
\ [OII] 3727+& 0.42 &4(-4)&1(-3)&7(-4)&0.6&0.046&3.7&0.04&0.0&0.023& 7(-4) &
0.16&0.11 & 120. & 0.5 \\
\ [NeV] 3426+ &0.59&0.01&0.3&0.17&0.34&9(-5)&2.65&0.05&9.3&6.17&0.17 & 1(-6) &
2.24 & 0.0 & 0.79 \\
\ HeII 3204 &0.1 &8(-4)&2(-3)&1.1(-3)&0.014&1.2(-4)&0.034&0.28&0.4&0.37 &
1.1(-3) & 0.06 & 0.37 & 0.0 & 0.12 \\
\ $\rm H\beta^2$ & - &1.6(3)&18.&27.&0.063&81.7&5(-3)&0.45&0.62&0.67 &-&4(5)
&1.2(-3) & 2.5(-5) & - \\
\hline \\
\end{tabular}}
$^1$ corrected ($\rm E_{(B-V)}$=0.32)
$^2$ in $\rm erg\, cm^{-2}\, s^{-1}$
\end{table}
\begin{table}
\centerline{Table 4}
\centerline{The \% of the different components in single line fluxes}
\begin{center}
\begin{tabular}{lll ll}\\ \hline \\
\ & mod & mod & mod& mod \\
\ & 3&12 & 13 & 14 \\ \hline \\
\ [SII] 6730 & 7.2 & 6.7 & 2.0 & 84. \\
\ [SII] 6717 & 2.64 & 2.4 & 1.7 & 93. \\
\ [NII] 6583 & 2.55 & 37.4 & 12.8 & 47.2 \\
\ [OI] 6300+& 92.2 & 6.4 & 0.0 & 1.37 \\
\ [OIII] 4959 & 1.15 & 2.6(-3) & 98.8 & 0.04 \\
\ HeII 4686 & 0.37 & 7.4(-3) & 99.6 & 0.0 \\
\ [OIII] 4363 &56. & 0.04 & 43.7 & 0.057 \\
\ [SII] 4072+ & 24.7 & 73.2 & 3.6(-3) & 2.1 \\
\ [NeIII] 3869 & 13.2 & 0.12 & 86.6 & 0.078 \\
\ [OII] 3727+& 0.089 & 3.0 & 6.0 & 90.7 \\
\ [NeV] 3426+ &14.6 & 0.0 & 85.4 & 0.0 \\
\ HeII 3204 &0.62 & 5.0 & 94.2 & 0.0 \\
\ $\rm H\beta$ & 62.6 & 9.27 & 27.8 & 0.37 \\
\hline \\
\end{tabular}
\end{center}
\end{table}
\newpage
\topmargin 0.01cm
\oddsidemargin 0.01cm
\evensidemargin 0.01cm
\begin{figure}
\begin{center}
\centerline{Fig. 1}
\mbox{\psfig{file=f1.eps,clip=,height=12.cm,width=12.cm}}
\end{center}
\end{figure}
\begin{figure}
\begin{center}
\centerline{Fig. 2}
\mbox{\psfig{file=f2.eps,clip=,height=14.6cm,width=14.6cm}}
\end{center}
\end{figure}
\begin{figure}
\begin{center}
\centerline{Fig. 3}
\mbox{\psfig{file=f3.eps,clip=,height=14.6cm,width=14.6cm}}
\end{center}
\end{figure}
\end{document}
|
\section{INTRODUCTION}
This is the second part of our eight presentations in which we consider
applications of methods from wavelet analysis to nonlinear accelerator
physics problems. This is a continuation of our results from [1]-[8],
which is based on our approach to investigation
of nonlinear problems -- general, with additional structures (Hamiltonian,
symplectic or quasicomplex), chaotic, quasiclassical, quantum, which are
considered in the framework of local (nonlinear) Fourier analysis, or wavelet
analysis. Wavelet analysis is a relatively novel set of mathematical
methods, which gives us a possibility to work with well-localized bases in
functional spaces and with the general type of operators (differential,
integral, pseudodifferential) in such bases.
In this part we consider orbital motion in transverse plane for a single
particle in a circular magnetic lattice in case when we take into account
multipolar expansion up to an arbitrary finite number. We reduce initial
dynamical problem to the finite number (equal to the number of n-poles) of standard
algebraical problem and represent all dynamical variables as expansion in the
base of periodical wavelet functions. Our consideration is based on
generalization of variational wavelet approach from part 1. After introducing
our starting points related to multiresolution in
section 3, we consider methods which allow us to construct wavelet representation for
solution in periodic case in section 4.
\section{Particle in the Multipolar Field}
The magnetic vector potential of a magnet with $2n$ poles in Cartesian
coordinates is
\begin{equation}
A=\sum_n K_nf_n(x,y),
\end{equation}
where $ f_n$ is a homogeneous function of $x$ and $y$ of order $n$.
The real and imaginary parts of binomial expansion of
\begin{equation}
f_n(x,y)=(x+iy)^n
\end{equation}
correspond to regular and skew multipoles. The cases $n=2$ to $n=5$
correspond to low-order multipoles: quadrupole, sextupole, octupole, decapole.
Then we have in particular case the following equations of motion for
single particle in a circular magnetic lattice in the transverse plane
$(x,y)$ ([9] for designation):
\begin{eqnarray}
&&\frac{\mathrm{d}^2x}{\mathrm{d} s^2}+ \left(\frac{1}{\rho(s)^2}-k_1(s)\right)x=\nonumber\\
&&{\cal R}e\left[\sum_{n\geq 2}\frac{k_n(s)+ij_n(s)}{n!}\cdot(x+iy)^n\right],\\
&&\frac{\mathrm{d}^2y}{\mathrm{d} s^2}+k_1(s)y=\nonumber\\
&&-{\cal J}m\left[\sum_{n\geq}
\frac{k_n(s)+ij_n(s)}{n!}\cdot(x+iy)^n\right]\nonumber
\end{eqnarray}
and the corresponding Hamiltonian:
\begin{eqnarray}\label{eq:ham}
&&H(x,p_x,y,p_y,s)=\frac{p_x^2+p_y^2}{2}+\nonumber\\
&&\left(\frac{1}{\rho(s)^2}-k_1(s)\right)
\cdot\frac{x^2}{2}+k_1(s)\frac{y^2}{2}\\
&&-{\cal R}e\left[\sum_{n\geq 2}
\frac{k_n(s)+ij_n(s)}{(n+1)!}\cdot(x+iy)^{(n+1)}\right]\nonumber
\end{eqnarray}
Then we may take into account arbitrary but finite number in expansion
of RHS of Hamiltonian (\ref{eq:ham}) and
from our point of view the corresponding Hamiltonian equations of motions are
not more than nonlinear ordinary differential equations with polynomial
nonlinearities and variable coefficients.
\section{Wavelet Framework}
Our constructions are based on multiresolution approach. Because affine
group of translation and dilations is inside the approach, this
method resembles the action of a microscope. We have contribution to
final result from each scale of resolution from the whole
infinite scale of spaces. More exactly, the closed subspace
$V_j (j\in {\bf Z})$ corresponds to level j of resolution, or to scale j.
We consider a r-regular multiresolution analysis (MRA) of $L^2 ({\bf R}^n)$
(of course, we may consider any different functional space)
which is a sequence of increasing closed subspaces $V_j$:
\begin{equation}
...V_{-2}\subset V_{-1}\subset V_0\subset V_{1}\subset V_{2}\subset ...
\end{equation}
satisfying the following properties:
\begin{eqnarray}
&\displaystyle\bigcap_{j\in{\bf Z}}V_j=0,\quad
\overline{\displaystyle\bigcup_{j\in{\bf Z}}}V_j=L^2({\bf R}^n),\nonumber\\
&f(x)\in V_j <=> f(2x)\in V_{j+1}, \nonumber\\
&f(x)\in V_0 <=> f(x-k)\in V_0, \ \forall k\in {\bf Z}^n.
\end{eqnarray}
There exists a function $\varphi\in V_0$ such that
\{$\varphi_{0,k}(x)=$
$\varphi(x-k)$, $k\in{\bf Z}^n$\} forms a Riesz basis for $V_0$.
The function $\varphi$ is regular and localized:
$\varphi$ is $C^{r-1}$,\ $\varphi^{(r-1)}$ is almost
everywhere differentiable and for almost every $x\in {\bf R}^n$, for
every integer $\alpha\leq r$ and for all integer p there exists
constant $C_p$ such that
\begin{equation}
\mid\partial^\alpha \varphi(x)\mid \leq C_p(1+|x|)^{-p}
\end{equation}
Let
$\varphi(x)$ be
a scaling function, $\psi(x)$ is a wavelet function and
$\varphi_i(x)=\varphi(x-i)$. Scaling relations that define
$\varphi,\psi$ are
\begin{eqnarray}
\varphi(x)&=&\sum\limits^{N-1}_{k=0}a_k\varphi(2x-k)=
\sum\limits^{N-1}_{k=0}a_k\varphi_k(2x),\\
\psi(x)&=&\sum\limits^{N-2}_{k=-1}(-1)^k a_{k+1}\varphi(2x+k).
\end{eqnarray}
Let indices $\ell, j$
represent translation and scaling, respectively and
\begin{equation}
\varphi_{jl}(x)=2^{j/2}\varphi(2^j x-\ell)
\end{equation}
then the set $\{\varphi_{j,k}\}, {k\in {\bf Z}^n}$ forms a Riesz basis for $V_j$.
The wavelet function $\psi $ is used to encode the details between
two successive levels of approximation.
Let $W_j$ be the orthonormal complement of $V_j$ with respect to $V_{j+1}$:
\begin{equation}
V_{j+1}=V_j\bigoplus W_j.
\end{equation}
Then just as $V_j$ is spanned by dilation and translations of the scaling function,
so are $W_j$ spanned by translations and dilation of the mother wavelet
$\psi_{jk}(x)$, where
\begin{equation}
\psi_{jk}(x)=2^{j/2}\psi(2^j x-k).
\end{equation}
All expansions which we used are based on the following properties:
\begin{eqnarray}
&&\{\psi_{jk}\}, \quad j,k\in {\bf Z}\quad
\mbox{is a Hilbertian basis of } L^2({\bf R})\nonumber\\
&&\{\varphi_{jk}\}_{j\geq 0, k\in {\bf Z}} \quad\mbox{is an orthonormal
basis for} L^2({\bf R}),\nonumber\\
&& L^2({\bf R})=\overline{V_0\displaystyle\bigoplus^\infty_{j=0} W_j},\\
&& \mbox{or}\qquad
\{\varphi_{0,k},\psi_{j,k}\}_{j\geq 0,k\in {\bf Z}}\nonumber \\
&&\mbox{is
an orthonormal basis for}
L^2({\bf R}).\nonumber
\end{eqnarray}
Fig.1 and Fig.2 give the representation of some function and
corresponding MRA on each level of resolution.
\begin{figure}[ht]
\centering
\epsfig{file=tha135a.eps, width=82.5mm, bb=0 200 599 590, clip}
\caption{Analyzed function.}
\end{figure}
\begin{figure}[ht]
\centering
\epsfig{file=tha135b.eps, width=82.5mm, bb=0 200 599 590, clip}
\caption{MRA representation.}
\end{figure}
\section{VARIATIONAL WAVELET APPROACH\\ FOR PERIODIC TRAJECTORIES}
We start with extension of our approach from part 1 to the case
of periodic trajectories. The equations of motion corresponding
to Hamiltonian (4) may also be formulated as a particular case of
the general system of
ordinary differential equations
$
{dx_i}/{dt}=f_i(x_j,t)$, $ (i,j=1,...,n)$, $0\leq t\leq 1$,
where $f_i$ are not more
than polynomial functions of dynamical variables $x_j$
and have arbitrary dependence of time but with periodic boundary conditions.
According to our variational approach from part 1 we have the
solution in the following form
\begin{eqnarray}
x_i(t)=x_i(0)+\sum_k\lambda_i^k\varphi_k(t),\qquad x_i(0)=x_i(1),
\end{eqnarray}
where $\lambda_i^k$ are again the roots of reduced algebraical
systems of equations
with the same degree of nonlinearity and $\varphi_k(t)$
corresponds to useful type of wavelet bases (frames).
It should be noted that coefficients of reduced algebraical system
are the solutions of additional linear problem and
also
depend on particular type of wavelet construction and type of bases.
This linear problem is our second reduced algebraical problem. We need to find
in general situation objects
\begin{eqnarray}
\Lambda^{d_1 d_2 ...d_n}_{\ell_1 \ell_2 ...\ell_n}=
\int\limits_{-\infty}^{\infty}\prod\varphi^{d_i}_{\ell_i}(x)\mathrm{d} x,
\end{eqnarray}
but now in the case of periodic boundary conditions.
Now we consider the procedure of their
calculations in case of periodic boundary conditions
in the base of periodic wavelet functions on
the interval [0,1] and corresponding expansion (14) inside our
variational approach. Periodization procedure
gives us
\begin{eqnarray}
\hat\varphi_{j,k}(x)&\equiv&\sum_{\ell\in Z}\varphi_{j,k}(x-\ell)\\
\hat\psi_{j,k}(x)&=&\sum_{\ell\in Z}\psi_{j,k}(x-\ell)\nonumber
\end{eqnarray}
So, $\hat\varphi, \hat\psi$ are periodic functions on the interval
[0,1]. Because $\varphi_{j,k}=\varphi_{j,k'}$ if $k=k'\mathrm{mod}(2^j)$, we
may consider only $0\leq k\leq 2^j$ and as consequence our
multiresolution has the form
$\displaystyle\bigcup_{j\geq 0} \hat V_j=L^2[0,1]$ with
$\hat V_j= \mathrm{span} \{\hat\varphi_{j,k}\}^{2j-1}_{k=0}$ [10].
Integration by parts and periodicity gives useful relations between
objects (15) in particular quadratic case $(d=d_1+d_2)$:
$$
\Lambda^{d_1,d_2}_{k_1,k_2}=(-1)^{d_1}\Lambda^{0,d_2+d_1}_{k_1,k_2},\
\Lambda^{0,d}_{k_1,k_2}=\Lambda^{0,d}_{0,k_2-k_1}\equiv
\Lambda^d_{k_2-k_1}
$$
So, any 2-tuple can be represent by $\Lambda^d_k$.
Then our second additional linear problem is reduced to the eigenvalue
problem for
$\{\Lambda^d_k\}_{0\leq k\le 2^j}$ by creating a system of $2^j$
homogeneous relations in $\Lambda^d_k$ and inhomogeneous equations.
So, if we have dilation equation in the form
$\varphi(x)=\sqrt{2}\sum_{k\in Z}h_k\varphi(2x-k)$,
then we have the following homogeneous relations
\begin{equation}
\Lambda^d_k=2^d\sum_{m=0}^{N-1}\sum_{\ell=0}^{N-1}h_m h_\ell
\Lambda^d_{\ell+2k-m},
\end{equation}
or in such form
$A\lambda^d=2^d\lambda^d$, where $\lambda^d=\{\Lambda^d_k\}_
{0\leq k\le 2^j}$.
Inhomogeneous equations are:
\begin{equation}
\sum_{\ell}M_\ell^d\Lambda^d_\ell=d!2^{-j/2},
\end{equation}
where objects
$M_\ell^d(|\ell|\leq N-2)$ can be computed by recursive procedure
\begin{eqnarray}
&&M_\ell^d=2^{-j(2d+1)/2}\tilde{M_\ell^d}, \\
&&\tilde{M_\ell^k}=
<x^k,\varphi_{0,\ell}>=\sum_{j=0}^k {k\choose j} n^{k-j}M_0^j,\quad
\tilde{M_0^\ell}=1.\nonumber
\end{eqnarray}
So, we reduced our last problem to standard
linear algebraical problem. Then we use the same methods as in part 1.
As a result we obtained
for closed trajectories of orbital dynamics described by Hamiltonian (4)
the explicit time solution (14) in the base of periodized wavelets (16).
We are very grateful to M.~Cornacchia (SLAC), W.~Herrmannsfeldt (SLAC),
Mrs. J.~Kono (LBL) and
M.~Laraneta (UCLA) for their permanent encouragement
|
\section{Introduction}
Precise tracking of the Pioneer 10/11, Galileo and Ulysses spacecraft
\cite{Anderson98}
have shown an anomalous constant acceleration for Pioneer 10 with a magnitude
$\sim -8.5\times 10^{-10}\, \mbox{m.s}^{-2}$.
Additional analysis by the same team \cite{Turyshev99} provide a new
value for the acceleration
$(-7.5\pm 0.2)\times 10^{-10}\, \mbox{m.s}^{-2}$
(where the uncertainty is estimated from points in their Fig 1)
and also reveal that there is an additional
annual periodic component with a amplitude of
$\sim 2\times 10^{-10}\, \mbox{m.s}^{-2}$ directed towards the sun.
The main method for monitoring the spacecraft is to measure the frequency
shift of the signal returned by an active transponder.
Any variation in this frequency shift that is not actually due to motion can be
confused with
a Doppler shift and would be attributed to anomalous velocities
and accelerations.
This paper argues that there are is an additional frequency shift in the
spacecraft signal
due to a gravitational interaction with the intervening material.
Because the frequency shift is proportional to the distance to the spacecraft it
can easily mimic an acceleration.
\section{The explanation for the constant acceleration}
In previous papers \cite{Crawford79,Crawford87A,Crawford91}
it was argued that photons have a
gravitational interaction. This claim is based on the premise that in curved
space a bundle of geodesics is focused (the "focusing theorem",
\cite{Misner73})
and as a consequence photons are also focused.
This leads to an interaction in which low energy photons are emitted and the primary photon losses energy. The effect can be observed as a frequency shift in a signal that is a function of distance traveled and the density of the local medium.
Although the cosmological consequences of such an interaction are profound
\cite{Crawford99}, it also leads to predictions which can be tested locally,
including the prediction \cite{Crawford91} that a
frequency shift should be seen in the signals from spacecraft.
For a signal passing through a medium with matter density $\rho$ the rate of
change of frequency, $f$, with distance is \cite{Crawford87A,Crawford91}
\begin{equation}
\label{e1}
\frac{df}{dx} = -\left(\frac{8\pi G\rho}{c^{2}}\right)^{1/2}\!\! f.
\end{equation}
Note that although point masses may distort and deviate the geodesic
bundle they do not focus it and so that there is no frequency shift
predicted for signals passing near stars or planets.
Since the effect is very small we can write it in effective velocity units as
\begin{equation}
\label{e2}
\Delta v = -\sqrt{8\pi G\rho}\,\Delta x .
\end{equation}
Differentiating gives an apparent acceleration of $a = -\sqrt{8\pi G\rho}\,V$
where $V$ is the velocity of the spacecraft (or earth) and $\rho$ is the density at
the current positions.
It is not an average density over the path length.
Using the observed anomalous acceleration of
$ -7.5\times 10^{-10}\, \mbox{m.s}^{-2}$ ,
and a Pioneer 10 velocity of $12.3 \mbox{km.s}^{-1}$, the
required density for the two-way path is
$5.5\times10^{-19}\,\mbox{kg.m}^{-3}$.
The only constituent of the interplanetary medium that approaches this
density is dust.
One estimate \cite{Sergeant80} of the interplanetary dust
density at 1 AU is
$1.3\times10^{-19}\,\mbox{kg.m}^{-3}$ and more recently Gr\"{u}n \cite{Grun99}
suggests a value of
$10^{-19}\,\mbox{kg.m}^{-3}$
which is consistent with his earlier estimate of
$9.6\times 10^{-20}\,\mbox{kg.m}^{-3}$ \cite{Grun85}.
Although the authors do not give uncertainties it is clear that the densities could
be in error by a factor of two or more.
The main difficulties are the paucity of information and that the observations do
not span the complete range of grain sizes.
Taking a density of
$10^{-19}\,\mbox{kg.m}^{-3}$
the computed (anomalous) acceleration is
$-3.4\times10^{-10}\,\mbox{ m.s}^{-2}$,
smaller by a factor of two than the observed anomalous acceleration.
However the density is required at the distance of Pioneer 10 in 1998 of 72 AU
in the plane of the ecliptic (ecliptic latitude of Pioneer 10 is 3$^\circ$).
The meteroid experiment on-board Pioneer 10 measures the flux of grains
with masses larger than $10^{-10}\,$g.
the results show that after it left the influence
of Jupiter the flux \cite{Landgraf99} was essentially constant
(in fact there may be a slight rise) out to a distance of 18 AU.
It is thought that most of the grains are being continuously produced
in the Kuiper belt. As their orbits evolve inwards due to
Poynting-Robertson drag and planetary perturbations they
achieve a roughly constant spatial density.
Given the large uncertainties in both the observed density at 1 AU (due to the
limitations of the detectors), and the extrapolation of the density to 72 AU,
the conclusion is that interplanetary dust
could provide the required density to explain the "anomalous acceleration"
by a frequency shift due to the gravitational interaction.
\section{The explanation for the annual acceleration}
Figure 1B in \cite{Turyshev99} shows a time varying acceleration
that has a period
of one year and an amplitude that both fluctuates and decreases with time.
(It may not be a valid decrease but be due to the solar cycle.)
Their figure shows 50-day averages after the best-fit constant anomalous
acceleration has been removed.
For the years 1987 to 1993 where the curve is well defined the maxima occur at
$0.94\pm 0.03\,$yr and the minima at $0.45\pm 0.03\,$yr.
The amplitude changes from $\sim 2.5\times 10^{-10} \,\mbox{m.s}^{-2}$ in 1988 to
$\sim 1.5\times 10^{-10} \,\mbox{m.s}^{-2}$ in 1992.
In principle the gravitational interaction can explain this acceleration
but now the relevant velocity is not that of Pioneer 10 but the
orbital velocity of the earth.
Taking the earth's velocity as $30\,\mbox{km.s}^{-1}$ and a dust density of
$10^{-19}\,\mbox{kg.m}^{-3}$
the predicted annual acceleration in 1989 has an amplitude of
$7.6\times10^{-10}\,\mbox{ m.s}^{-2}$.
Although this acceleration is a factor of three too large a more
significant objection is that the predicted phase disagrees
with the observations.
With this model the maximum accelerations should occur
when the earth has a maximum velocity relative to Pioneer 10,
namely when it has maximum elongation as seen from the spacecraft.
Since in 1989 Pioneer 10 had an ecliptic longitude of $\sim 72^\circ$
these should occur at 0.17\,yr and 0.68\,yr.
The discrepancy in phase of $97^\circ \pm 11^\circ$ means that the gravitational
interaction does not directly explain the annual variation.
However since the gravitational interaction was not included in in the
complex calculations used to
compute the trajectory it is feasible that the effect has been
compensated for by small adjustments to
other parameters and all that is left is a distorted residual.
If mistakenly interpreted as a Doppler shift the annual component of
the gravitational interaction
is equivalent to an additional velocity of the earth (as seen by Pioneer 10) of
$3.8\,\mbox{mm.s}^{-1}$.
For a circular orbit of the earth this is equivalent to a shift in the
longitude of Pioneer 10 of 0.026 arcseconds.
Thus if there is a gravitational interaction it could be masked by a small error
in longitude.
In practice the position of Pioneer 10 must be consistent with celestial
mechanics and many other observations and it is unlikely that there would
be complete compensation.
The final analysis requires the inclusion of the gravitational interaction into
the orbit calculations.
\section{Conclusion}
It has been argued that the gravitational interaction with a
interplanetary dust density of
$10^{-19}\,\mbox{kg.m}^{-3}$
predicts an anomalous acceleration of Pioneer 10 at 72 AU of
$-3.4\times10^{-10}\,\mbox{ m.s}^{-2}$
to be compared with the observed value of
$(-7.5\pm 0.2)\times 10^{-10}\, \mbox{m.s}^{-2}$.
The largest uncertainty is in the estimate of the interplanetary dust density.
Since the annual period in the gravitational interaction is easily masked by
small shift in the longitude of Pioneer 10 its
effects are unlikely to be observed.
However the predicted magnitude is in the right range
and the observed annual acceleration could be the residuals after
a partial compensation.
\section{Acknowledgments}
This work is supported by the Science Foundation for Physics within the
University of Sydney,
and use has made of NASA's Astrophysics Data System Abstract Service.
|
\section{Introduction}
In connection with the studies of
the copper oxide high-$T_{\rm c}$ superconductors with
CuO$_{2}$ planes,
the electronic states in two-dimensional systems
has been intensively studied.
Especially, the possibility of
various ordered states has been discussed.
One characteristic feature is the proximity of superconductivity
and antiferromagnetism.
In our previous work (hereafter referred to as I),\cite{ore2and3}
we have shown in the mean field approximation
that the coexistent state with
$d$-wave superconductivity (dSC),
commensurate spin-density-wave (SDW)
and $\pi$-triplet pair can be stabilized near half filling
by {\em repulsive} backward scattering
('Umklapp' and 'exchange') processes
between electrons around the saddle points ($\pi$,0) and (0,$\pi$).
As we shall show later,
this model with such a particular type of interaction
have similar features to those in a square lattice model with
on-site {\em repulsion} $U>0$
and nearest-neighbor {\em attraction} $V<0$, i.e.,
an {\em extended Hubbard model}.\cite{EHM2D}
Therefore, it is interesting to examine in more detail
the possibility of the above coexistent state
by use of this model.
At the same time, the extended Hubbard model with
both on-site and nearest-neighbor {\em repulsion} ($U>0$ and $V>0$),
is also of physical interest.
In the 2D extended Hubbard model for $U>0$,
it has been shown based on the mean field approximation that
extended-$s$-,
$p$- and $d$-wave superconductivity
can arise depending on the electron density $n$ for $V<0$,
\cite{EHM2D,dag,sdd,onlyV}
and commensurate charge- and spin-density wave (CDW and SDW)
can appear at half filling $n=1$ for $V>0$.\cite{EHM2D,dag}
However,
the property of the ground state for finite carrier doping and
the relationship among various order parameters, especially between
dSC and SDW, have not been understood yet,
even in the mean field approximation.
From these points of view, we will study possible
ordered states,
especially possible coexistence of different orders,
in the 2D extended Hubbard model on a square lattice
near half filling for $U>0$ {\em and} $V\neq 0$,
with emphasis on electrons around the saddle points
($\pi$,0) and (0,$\pi$).
In \S\ref{ehh}, the extended Hubbard model is introduced
and its relationship to our previous model used in I is referred to.
In \S\ref{mfa}, the phase diagram at absolute zero, $T=0$,
is determined in the mean field approximation.
In \S\ref{rg},
the effects of fluctuation on the mean field ordered states
are examined based on the renormalization method
applicable only for the special case that
the saddle points ($\pi$,0) and (0,$\pi$) lie just on the Fermi surface.
\section{Extended Hubbard Hamiltonian}\label{ehh}
The extended Hubbard Hamiltonian, $H=H_{0}+H_{U}+H_{V}$,
is written as follows:
\begin{subequations}
\begin{eqnarray}
H_{0}&=& \sum_{p\sigma}\xi_{p}c^{\dagger}_{p\sigma}c_{p\sigma},\\
\label{2DSL}
H_{U}&=&U\sum_{i}n_{i\uparrow}n_{i\downarrow}
=\frac{U}{N}\sum_{q}n_{q\uparrow}n_{-q\downarrow},\\
H_{V}&=&\frac{V}{2}\sum_{i\hat{\rho}}
n_{i}n_{i+\hat{\rho}}=\frac{1}{N}\sum_{q}V_{q}n_{q}n_{-q},
\end{eqnarray}
where
$\sigma$ is the spin index taking a value of $+1$ ($-1$) for
$\uparrow$ ($\downarrow$) spin, and
the opposite spin to $\sigma$ is denoted by $\overline{\sigma}\equiv -\sigma$.
$N$ is the total number of lattice sites,
$\xi_{p}=\epsilon_{p}-\mu$ is the one-particle energy dispersion
relative to the chemical potential $\mu$,
including nearest-neighbor- ($t$) and next-nearest-neighbor- ($t'$) hopping
integrals,
\begin{equation}
\epsilon_{p}=-2t(\cos p_{x}+\cos p_{y})-4t'\cos p_{x}\cos p_{y},
\end{equation}
$n_{q}=\sum_{\sigma}n_{q\sigma}=\sum_{k\sigma}
c^{\dagger}_{k\sigma}c_{k+q\sigma}$,
$\hat{\rho}=\pm\hat{x},\pm\hat{y}$ is the unit lattice vector and
\begin{equation}
V_{q}=V(\cos q_{x}+\cos q_{y}).
\end{equation}
\end{subequations}
The energy dispersion $\epsilon_{p}$ has two independent
saddle points, $(\pi,0)$ and $(0,\pi)$.
In this paper, we fix $t'/t=-1/5$ with $t>0$,
in which case the Fermi surface in the absence of interaction
approaches $(\pi,0)$ and $(0,\pi)$
as the {\em hole} doping rate, $\delta\equiv 1-n$, is increased
from half filling, $\delta=0$.
Here, we examine the relationship between the 'g-\' ology' model used in I
and the present extended Hubbard model.\cite{oreD}
In I,
we have treated the backward scattering with large momentum transfer
between electrons around $(\pi,0)$ and $(0,\pi)$,
i.e., 'Umklapp' ($g_{3\perp}$) and 'exchange' ($g_{1\perp}$) processes,
and considered three types of the scattering processes, i.e.,
(1) Cooper-pair, (2) density-wave and (3) $\pi$-pair channels.\cite{ore2and3}
(Here we denote the Hamiltonian for these channels
as $H_{1}$, $H_{2}$ and $H_{3}$.)
Especially for the repulsive case, $g_{3\perp}>0$ and $g_{1\perp}>0$,
we have shown that
the coexistent state with dSC, SDW and $\pi$-triplet pair
can be stabilized near half filling at low temperature.
However, the above effective interaction is too simplified in that
(a) forward scattering processes are not included and
(b) the $k$-dependence of interaction is ignored.
Therefore, by transforming $H_{i}$ ($i=1,2,3$),
into real-space representation,
we will obtain a well-defined model on a square lattice.
If we keep on-site and nearest-neighbor
density-density Coulomb interaction, we obtain
\begin{subequations}
\begin{eqnarray}
H_{1}&\sim&\frac{g_{3\perp}}{2N}\biggr\{
\sum_{i}n_{i\uparrow}n_{i\downarrow}-\alpha \sum_{<ij>}\sum_{\sigma}
n_{i\sigma}n_{j\overline{\sigma}}\biggr\},\\
H_{2}&\sim&\frac{g_{\perp}}{2N}\biggr\{
\sum_{i}n_{i\uparrow}n_{i\downarrow}
-\sum_{<ij>}\sum_{\sigma}n_{i\sigma}n_{j\overline{\sigma}}\biggr\},\\
H_{3}&\sim&\frac{g_{1\perp}}{2N}\biggr\{
\sum_{i}n_{i\uparrow}n_{i\downarrow}-\alpha \sum_{<ij>}\sum_{\sigma}
n_{i\sigma}n_{j\overline{\sigma}}\biggr\},
\end{eqnarray}
\label{tolattice}
\end{subequations}
where $g_{\perp}\equiv g_{3\perp}+g_{1\perp}$,
$<ij>$ stands for a bond connecting site $i$ and its nearest-neighbor
site $j$
and
$\alpha=16/\pi^{4}\sim 0.164$.
Each $H_{i}$ is seen to describe on-site {\em repulsion} and nearest-neighbor
{\em attraction} for $g_{3\perp},g_{1\perp}>0$.
Therefore,
also in the extended Hubbard model for $U>0$ and $V<0$,
the coexistent state with dSC, SDW and $\pi$-triplet pair
is expected to be stabilized.
\section{Mean Field Analysis}\label{mfa}
First, we determine the phase diagram at absolute zero, $T=0$,
near half filling in the mean field approximation.
We fix the electron density to $n=0.9$.
With this choice of parameters, the Fermi surface of noninteracting
electrons is of the YBCO- or BSCCO-type and lies close to
the saddle points ($\pi$,0) and (0,$\pi$),
as shown in Fig.~\ref{fspart2}.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=4cm
\epsfbox{fstpm02n09.eps}
\end{center}
\caption{The Fermi surface in the absence of interaction
for $t'/t=-1/5$ and $n=0.9$.}
\label{fspart2}
\end{figure}
\subsection{Nearest-Neighbor Attraction $V<0$}\label{vnega}
We start with the case $V<0$.
As we saw in \S \ref{ehh},
the ordered states
with dSC, SDW(= antiferromagnetism, AF) and
$\pi$-triplet pair is expected.
We consider the following order parameters,
\begin{subequations}
\begin{eqnarray}
<c_{i\sigma}c_{i+\hat{\rho},\overline{\sigma}}>&\equiv& \sigma s_{\hat{\rho}}
+q_{\hat{\rho}}\cos(QR_{i}),\\
<n_{i\sigma}>&\equiv &\frac{n}{2}+\sigma m \cos(QR_{i}),
\end{eqnarray}
\end{subequations}
where
$Q=(\pi,\pi)$,
$s_{-\hat{\rho}}=s_{\hat{\rho}}$ and $q_{-\hat{\rho}}=q_{\hat{\rho}}$.
$s_{\hat{\rho}}$ ($q_{\hat{\rho}}$) stands for
a spin-singlet (triplet) pair of two electrons with a total momentum $0$ ($Q$)
and total spin $S=0$ ($S=1$ and $S_{z}=0$), i.e.,
Cooper-pair ($\pi$-triplet pair),
and $m$ for the local staggered spin moment.
We take
$s_{\hat{x}}=-s_{\hat{y}}=s_{0}=real$ and
$q_{\hat{x}}=-q_{\hat{y}}=q_{0}=real$,
i.e., consider $d_{x^{2}-y^{2}}$wave pairing,
which is favored near half filling.\cite{EHM2D}
While $s_{0}$ describes dSC,
$q_{0}$ describes an electron-pair of $p_{x}-p_{y}$-wave symmetry.
\cite{DZ,SCZhang}
This can be easily seen by writing the operator of $\pi$-triplet pair
in $k$-space,
\begin{subequations}
\begin{eqnarray}
\hat{O}_{\pi}&\equiv&\frac{1}{\sqrt{2}}
\sum_{p\sigma}w_{p}c_{-p+Q\overline{\sigma}}c_{p\sigma},\\
&=&\frac{1}{\sqrt{2}}
\sum_{p\sigma}w_{\frac{Q}{2}+p}
c_{\frac{Q}{2}-p\overline{\sigma}}c_{\frac{Q}{2}+p\sigma},
\end{eqnarray}
\end{subequations}
where $w_{p}\propto\cos p_{x}-\cos p_{y}$
is the $d_{x^{2}-y^{2}}$-wave factor and
$<\hat{O}_{\pi}>\propto q_{0}$.
The orbital function
as a function of the {\em relative} momentum $p$ of two electrons,
$w_{\frac{Q}{2}+p}$, is proportional to $\sin p_{x}-\sin p_{y}$.
The order parameter in the mean filed Hamiltonian are given as follows
in terms of $s_{0}$, $q_{0}$ and $m$,
\begin{subequations}
\begin{eqnarray}
\Delta_{dSC}=-2|V|s_{0},&&
\Delta_{\pi}=-2|V|q_{0},\\
\Delta_{SDW}&=&Um,
\end{eqnarray}
\end{subequations}
where $\Delta_{dSC}$ and $\Delta_{\pi}$ include only $V$ because
$s_{\hat{\rho}}$ and $q_{\hat{\rho}}$ are defined on a bond, and
$\Delta_{SDW}$ does not include $V$
because $<n_{i}>=\sum_{\sigma}<n_{i\sigma}>$ is independent of $m$.
The pure $\pi$-triplet pairing state with $\Delta_{\pi}\ne 0$ and
$\Delta_{dSC}=\Delta_{SDW}=0$ is always energetically unfavorable
compared with the pure dSC state with
$\Delta_{dSC}\ne 0$ and $\Delta_{SDW}=\Delta_{\pi}=0$.
However, since the coexistence of dSC and SDW ($\Delta_{dSC}\neq 0$
{\em and} $\Delta_{SDW}\neq 0$)
generally results in nonzero $t_{0}$ ({\em and} nonzero $\Delta_{\pi}$ here),
the self-consistency of mean field calculation requires the consideration
of $\pi$-triplet pair into account from the outset.
\cite{ore2and3,chmAAcoe}
The important fact
that the coexistence of
spin-singlet Cooper-pair and SDW always leads to
nonzero spin-triplet pair amplitude with finite total momentum
had been recognized by Psaltakis
{\em et al.} in a slightly different context.\cite{crete}
The close relationship among the order parameters of dSC, SDW and
$\pi$-triplet pair is discussed in Appendix \ref{sym}.
The mean field phase diagram in the plane of $U$ and $|V|$ is shown in
Fig.~\ref{pdvnega}.
While the dSC state is stabilized for small $U/t$,
the coexistent state with dSC, SDW and $\pi$-triplet pair is possible
for large $U/t$,
and the phase boundary between these two states
is shown by solid line.
Although the pure $\pi$-triplet pairing state cannot be stabilized,
$\pi$-triplet pair can condensate as a result of the coexistence of
dSC and SDW.
Since there is {\em attractive} interaction for spin-triplet channel
in the present model,
the coexistent region of dSC and SDW is widened by
the inclusion of $\pi$-triplet pair.
We note that the Fermi surface remains in the SDW state near half filling.
As we saw in I, when SDW appears first as the temperature is lowered,
the coexistent state can be stabilized at lower temperature,
especially at $T=0$, near half filling.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=5cm
\epsfbox{pdvnega.eps}
\end{center}
\caption{The mean field phase diagram at $T=0$
for $U>0$, $V<0$, $n=0.9$ and $t'/t=-1/5$.
Solid line stands for the boundary between
the coexistent state and dSC state in the mean field approximation.
Dotted and broken lines stand for the boundary
in the right side of which
the phase separation occurs in the random phase approximation
for $V'=0$ and $V'=|V|/2$, respectively,
where $V'$ is the next-nearest-neighbor density-density interaction.
$N$ stands for the normal state.}
\label{pdvnega}
\end{figure}
Generally, in the presence of finite-range attractive
density-density interaction,
the system can be hampered by the phase separation (PS).
In order to examine the PS transition,
we calculate the charge compressibility, $\kappa$,
given by static and uniform charge susceptibility,
in the ground state, i.e., dSC or coexistent state.
This phase boundary is determined from $\kappa^{-1}=0$.
Here we use the random phase approximation (RPA),
and take the RPA diagram, shown in Fig.~\ref{RPAdiagram1}, into account.
The explicit form of $\kappa$ in the RPA is shown in Appendix \ref{RPA}.
The PS transition line is shown in Fig.~\ref{pdvnega} by dotted line.
It is seen that the coexistent state is severely suppressed by PS
but survives.
This PS is expected to be suppressed
if we take the long-range Coulomb repulsion into account.
Here, for simplicity,
we consider
the {\em next}-nearest-neighbor density-density repulsion $V'>0$,
\begin{subequations}
\begin{eqnarray}
H_{V'}&=&\frac{V'}{2}\sum_{i\hat{l}}n_{i}n_{i+\hat{l}}
=\frac{1}{N}\sum_{q}V'_{q}n_{q}n_{-q},\\
V'_{q}&=&2V'\cos q_{x}\cos q_{y},
\end{eqnarray}
\end{subequations}
where $\hat{l}=\pm(\hat{x}+\hat{y}),\pm(\hat{x}-\hat{y})$,
and incorporate $V'$ into the RPA calculation, i.e.,
we replace $V_{q}$ with $V_{q}+V'_{q}$.
We note that this replacement,
which does not alter the mean field equations,
can bring about the nontrivial effect on $\kappa$ in the coexistent state,
from eq. (\ref{kappaRPA}).
For $V'=|V|/2$, the PS transition line is shifted,
as shown in Fig.~\ref{pdvnega},
from dotted line to broken line.
It is seen that the long-range Coulomb interaction
does lead to the suppression of the PS,
which is less prominent in the coexistent state than in the dSC state.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=1.2cm
\epsfbox{RPAdiagram1.eps}
\end{center}
\caption{The RPA diagram for charge susceptibility.}
\label{RPAdiagram1}
\end{figure}
The calculation of $\kappa$ in the RPA had been carried over by
Micnas {\em et al.},\cite{EHM2D} {\em only} in the normal state.
Dagotto {\em et al.}\cite{dag} have shown
based on quantum Monte Carlo (QMC) simulation that
(1) PS drastically reduces the size of the mean field dSC region
and (2) the enhancement of $d_{x^{2}-y^{2}}$ pairing correlation itself
is not found.
This QMC result of (2) is different from the present mean field calculation.
We note that
this QMC calculation is limited to the case of half filling $n=1$ and
relatively high temperature $T=t/6$.
The coexistence of dSC and SDW near half filling has been found
also in the $t$-$J$ model in the slave-boson mean field
approximation\cite{inaba}
and by use of variational Monte Carlo calculation (VMC),\cite{Chen,giar,Himeda}
and in the repulsive Hubbard model ($V=0$) by use of VMC.\cite{giar,yamaji}
In these studies, however, $\pi$-triplet pair has not been taken into account.
The effect of $\pi$-triplet pair on the coexistence of dSC and SDW
has been recently examined by Arrachea {\em et al.}\cite{chmAAcoe}
based on a generalized Hubbard model.
In the repulsive Hubbard model,
the nearest-neighbor hopping term is modified as the correlated one,
\begin{eqnarray}
H_{ch}&=&-\sum_{<ij>\sigma}
\left\{c^{\dagger}_{i\sigma}c_{j\sigma}
+c^{\dagger}_{j\sigma}c_{i\sigma}\right\}\nonumber\\
&\times&
\left\{
t_{AA}(1-n_{i\overline{\sigma}})(1-n_{j\overline{\sigma}})
+t_{BB}n_{i\overline{\sigma}}n_{j\overline{\sigma}}\right.\nonumber\\
&&+t_{AB}[n_{i\overline{\sigma}}(1-n_{j\overline{\sigma}})+(1-n_{i\overline{\sigma}})n_{j\overline{\sigma}}]
\left.\right\},
\end{eqnarray}
where three hopping integrals, $t_{AA}$, $t_{BB}$ and $t_{AB}$,
incorporate many-body effects into one-particle hopping processes
phenomenologically.
$t_{AA}$ and $t_{BB}$ do not change the number of doubly occupied sites,
and $t_{AB}$ does, as shown in Fig.~\ref{chopping}.
It is to be noted that $H_{ch}$ can be rewritten as
\begin{subequations}
\begin{eqnarray}
H_{ch}&=&\sum_{<ij>\sigma}
\left\{c^{\dagger}_{i\sigma}c_{j\sigma}
+c^{\dagger}_{i\sigma}c_{j\sigma}\right\}\nonumber\\
&\times&
\left\{-t+
t_{2}(n_{i\overline{\sigma}}+n_{j\overline{\sigma}})
+t_{3}n_{i\overline{\sigma}}n_{j\overline{\sigma}}\right\},
\end{eqnarray}
where
\begin{equation}
t\equiv t_{AA},\:\:t_{2}\equiv t_{AA}-t_{AB},\:\:
t_{3}\equiv 2t_{AB}-t_{AA}-t_{BB}.
\end{equation}
\end{subequations}
The $t_{2}$ term can be also deduced
from the bare Coulomb interaction\cite{BCR1D,BCRHS,PH1D}
or by including the effects of phonon
in the antiadiabatic approximation $M\rightarrow 0$
(where $M$ is the phonon mass),\cite{BCAAL}
and the $t_{3}$ term describes the three-body interaction.
Arrachea {\em et al.} have shown
in the mean field approximation for
$t_{AB}>t_{AA}=t_{BB}$ (i.e., $t_{2}<0$ and $t_{3}=-2t_{2}>0$)
and $t'=0$ that
the coexistence of dSC and SDW is possible but prevented
by $\pi$-triplet pair
(and ruled out for large $U$),
due to {\em repulsive} spin-triplet pairing interaction.
Hence, the effect of $\pi$-triplet pair on the coexistence of
dSC and SDW is different from that in the present extended Hubbard model.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=4.5cm
\epsfbox{CHM.eps}
\end{center}
\caption{The correlated hopping processes.}
\label{chopping}
\end{figure}
\subsection{Nearest-Neighbor Repulsion $V>0$}\label{vposi}
Next we treat the case $V>0$.
In this case,
not only charge- and spin-density-wave states (CDW and SDW),
but also orbital antiferromagnetic (OAF) and spin nematic (SN) states,
in which the local staggered currents of charge and spin
circulate, respectively,\cite{fluxphase,fsinstability,OAF,SN}
are expected.\cite{CG}
Here we include ferromagnetism (FM) for the reason
as we shall describe later.
The order parameters are
\begin{subequations}
\begin{equation}
<n_{i\sigma}>\equiv \frac{n}{2}+\sigma f+
(p+\sigma m) \cos(QR_{i}),\label{dw}
\end{equation}
\begin{equation}
<c^{\dagger}_{i\sigma}c_{i+\hat{\rho},\sigma}>\equiv
(g_{\hat{\rho}}+\sigma l_{\hat{\rho}})\cos(QR_{i}),\label{OAFSN2D}
\end{equation}
\end{subequations}
where $f$, $p$ and $m$ are real,
$g_{-\hat{\rho}}=-g^{\ast}_{\hat{\rho}}$ and
$l_{-\hat{\rho}}=-l^{\ast}_{\hat{\rho}}$.
$f$, $p$ and $m$ describes FM, CDW and SDW, respectively.
We take $g_{\hat{\rho}}$
and $l_{\hat{\rho}}$ to be of $d_{x^{2}-y^{2}}$wave symmetry,
which is favored near half filling.\cite{CG}
In this case, $g_{\hat{\rho}}$
and $l_{\hat{\rho}}$ become pure imaginary,
$g_{\pm\hat{x}}=-g_{\pm\hat{y}}={\rm i}g$ and
$l_{\pm\hat{x}}=-l_{\pm\hat{y}}={\rm i}l$,
where $g$ and $l$ are real.
The states with $g\ne 0$ and $l\ne 0$ are called as OAF and SN ones,
\cite{fluxphase,fsinstability,OAF,SN}
respectively,
in which there exist the staggered
local currents of charge and spin
on a bond ($i,i+\hat{\rho}$),
\begin{subequations}
\begin{eqnarray}
<j^{c}_{i,i\pm\hat{x}}>=-<j^{c}_{i,i\pm\hat{y}}>&\propto &g\cos(QR_{i}),\\
<j^{s}_{i,i\pm\hat{x}}>=-<j^{s}_{i,i\pm\hat{y}}>&\propto &l\cos(QR_{i}),
\end{eqnarray}
\end{subequations}
where
\begin{equation}
j^{\nu}_{i,i+\hat{\rho}}\equiv {\rm i}\sum_{\sigma}v_{\nu}
(c^{\dagger}_{i\sigma}c_{i+\hat{\rho},\sigma}-
c^{\dagger}_{i+\hat{\rho},\sigma}c_{i\sigma}),
\end{equation}
and $v_{c}=1$ and $v_{s}=\sigma$.
In the OAF (SN) state, the local staggered current of charge (spin)
circulates around the plaquettes,
as schematically shown in Fig.~\ref{oafsn},
and the bond-ordered wave (BOW) does not exist, i.e.,
$<c^{\dagger}_{i\sigma}c_{i+\hat{\rho},\sigma}+
c^{\dagger}_{i+\hat{\rho},\sigma}c_{i\sigma}>\equiv 0$.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=2.5cm
\epsfbox{OAFandSN.eps}
\end{center}
\caption{The OAF (SN) state in which the local staggered
charge (spin) current circulates around the plaquettes.}
\label{oafsn}
\end{figure}
The order parameters in the mean field Hamiltonian
are given as follows in terms of $f$, $p$, $m$, $g$ and $l$,
\begin{subequations}
\begin{eqnarray}
\Delta_{FM}&=&Uf,\\
\Delta_{CDW}=(8V-U)p,&&\Delta_{SDW}=Um,\\
\Delta_{OAF}=2Vg,&&\Delta_{SN}=2Vl,
\end{eqnarray}
\end{subequations}
where $\Delta_{OAF}$ and $\Delta_{SN}$ include only $V$
because $g_{\hat{\rho}}$ and $l_{\hat{\rho}}$ are defined on a bond,
and $\Delta_{FM}$ does not include $V$
because $<n_{i}>=\sum_{\sigma}<n_{i\sigma}>$ is independent of $f$.
The order parameters of CDW, SDW, OAF and SN are closely related to
each other, which is shown in Appendix \ref{sym}.
The mean field phase diagram in the plane of $U$ and $V$ is shown
in Fig.~\ref{pdvposi}.
We have found that a coexistent solution
with nonzero $\Delta_{CDW}$, $\Delta_{SDW}$
{\em and} $\Delta_{FM}$
can be stabilized for $n\neq 1$.
This state is {\em ferrimagnetic}, as shown in Fig.~\ref{ferri}.
We note that the coexistence of CDW and SDW
($\Delta_{CDW}\neq 0$ {\em and} $\Delta_{SDW}\neq 0$)
generally results
in nonzero $f$ ({\em and} nonzero $\Delta_{FM}$ here),
which had been indicated by Dzyaloshinski\u\i\cite{scvHs}
based on a qualitative symmetry analysis.
This is the reason why we take FM into account from the outset.
With the present choice of parameters,
pure FM state with only $\Delta_{FM}\neq 0$ cannot be stabilized, but
FM can arise as a result of the coexistence of CDW and SDW.
In the present case, the coexistent region of CDW and SDW
is widened by the inclusion of FM.
We note that the Fermi surface remains in the CDW or SDW state
near half filling.
It is to be noted that neither OAF nor SN can been stabilized solely
in the mean field approximation,
{\em independent} of $t'$, $U$, $V$, $T$ and $n$.
This conclusion is contrary to that of
Chattopadhyay {\em et al.}\cite{CG}
that pure OAF or SN state has lower ground-state energy
than pure CDW or SDW state for the half-filled case
by introducing finite $t'$.
Moreover, a state where
local-current (OAF or SN) and density-wave (CDW or SDW) coexist
cannot be also stabilized.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=5cm
\epsfbox{pdvposi.eps}
\end{center}
\caption{The mean field phase diagram at $T=0$ for $U>0$, $V>0$, $n=0.9$
and $t'/t=-1/5$.}
\label{pdvposi}
\end{figure}
\begin{figure}
\begin{center}
\leavevmode\epsfysize=3cm
\epsfbox{ferri.eps}
\end{center}
\caption{The ferrimagnetic coexistent state.
Each lattice site is shaded according to electron density.
The length of each arrow is proportional to
the magnitude of local spin moment.
Lattice sites with larger local spin moment have higher electron density.}
\label{ferri}
\end{figure}
For $U/t=4.0$,
the $V$ dependences of $\Delta_{CDW}$, $\Delta_{SDW}$,
$\Delta_{FM}$ and
the difference between the energy of the
pure state (CDW for $U<4V$ or SDW for $U>4V$), $E_{p}$, and that of
the coexistent state (CDW+SDW+FM), $E_{c}$,
are shown in Fig.~\ref{cdwsdw}.
In the coexistent state with CDW, SDW and FM,
$|\Delta_{SDW}|$ is larger than $|\Delta_{CDW}|$
for $U>4V$ and vice versa for $U<4V$, and
the first-order phase transition occurs at $U=4V$.
For fixed $U$, $|\Delta_{FM}|$ rapidly saturates as a function of $V$.
It is seen that the energy gain in the coexistent state is very small.
\begin{fullfigure}
\begin{center}
\leavevmode\epsfysize=5cm
\epsfbox{orderenergy.eps}
\end{center}
\caption{The $V$ dependences of (a) $\Delta_{CDW}$, $\Delta_{SDW}$
$\Delta_{FM}$ and
(b)the energy difference $E_{p}-E_{c}$ (scaled by $t$)
between the pure state (CDW or SDW) and the coexistent state (CDW+SDW+FM)
for $U/t=4.0$, $n=0.9$ and $t'/t=-1/5$.}
\label{cdwsdw}
\end{fullfigure}
The energy dispersion in the ferrimagnetic coexistent state is
shown in Fig.~\ref{cdwsdwdisp}.
There are four energy bands, and
the Fermi surface remains as in the CDW or SDW state.
However, the lower band of electrons with majority spin
(up spin for $\Delta_{FM}>0$) is fully occupied and
the Fermi level crosses only the lower band of electrons
with minority spin (down spin for $\Delta_{FM}>0$).
Therefore, this coexistent state is {\em half metallic}.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=5cm
\epsfbox{energydispCOE.eps}
\end{center}
\caption{The energy dispersion relative to the Fermi level
in the coexistent state with CDW, SDW and FM
for $U/t=4.0$, $V/t=1.5$, $n=0.9$ and $t'/t=-1/5$.
Full (dotted) lines stand for that of electrons with up (down) spin,
respectively. The lower band of electrons with up spin is fully occupied.}
\label{cdwsdwdisp}
\end{figure}
The coexistence of CDW and SDW with {\em same} wave vectors
has also been found in a 1D modified Hubbard model for a quarter-filled band
in the mean field approximation.\cite{1D2kf}
In this coexistent state,
the wave vector of charge and spin density, $q$,
and the phase difference between CDW and SDW, $\Delta\theta$,
are equal to $2k_{F}\equiv \pi/2$ and $\pi/2$, respectively,
and the magnitude of local spin moment is equal at each site,
as shown in Fig.~\ref{kogata}.
On the other hand, in our coexistent state,
$q=Q\equiv (\pi,\pi)$ and $\Delta\theta =0$,
and the magnitude of local spin moment is different
at each sublattice, as shown in Fig.~\ref{ferri}.
This {\em ferrimagnetic} coexistent state is
the 2D version of that found in the 3D Hubbard model ($V=0$)
which had been denoted as the special ferrimagnetic (S.F.) state. \cite{penn}
In the present 2D case, this coexistent state for $V=0$
can be stabilized for $12\mbox{\raisebox{-0.6ex}{$\stackrel{<}{\sim}$}} U/t\mbox{\raisebox{-0.6ex}{$\stackrel{<}{\sim}$}} 14$
(not shown in Fig.~\ref{pdvposi}).
\begin{figure}
\begin{center}
\leavevmode\epsfysize=0.7cm
\epsfbox{2kFCDWSDW1D.eps}
\end{center}
\caption{The coexistent state with $2k_{F}$ CDW and $2k_{F}$ SDW
found in a quarter-filled 1D modified Hubbard model.
\protect\cite{1D2kf}
Each lattice site is shaded according to electron density.
The length of each arrow, proportional to
the magnitude of local spin moment,
is equal at each site.}
\label{kogata}
\end{figure}
\section{Renormalization Group Analysis}\label{rg}
In the last section,
we examined possible ordered states for $U>0$ and $V\ne 0$
in the mean field approximation.
In this section,
we examine the effects of fluctuation on these ordered states
which are not taken into account in the mean field calculation.
As a theoretical treatment beyond the mean field level,
we adopt the renormalization group (RG) method for the saddle points
which has been applied to the Hubbard model ($V=0$),
\cite{Schulz,Lederer,scvHs,vHsRG,ttu,ttu2,FurukawaRice}
and determine the most dominant correlation in the normal state.
We also discuss the possibility of the coexistent states
beyond the mean field approximation.
\subsection{Saddle Point Singularity}
We consider the {\em special} case
where the Fermi level in the absence of interaction lies
{\em just} on the saddle points $Q_{A}\equiv (\pi,0)$
and $Q_{B}\equiv (0,\pi)$, i.e., $\mu=4t'$ ($n\sim 0.83$),
and focus on electrons at these two saddle points
{\em on the Fermi surface}, just as two Fermi points in 1D electron systems.
The Fermi surface is shown in Fig.~\ref{juston}.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=4cm
\epsfbox{fstpm02n083.eps}
\end{center}
\caption{The Fermi surface in the absence of interaction
for $t'/t=-1/5$ and $\mu=4t'$, in which case $n\sim 0.83$.}
\label{juston}
\end{figure}
First, we examine the behavior of the following
particle-particle (K) and particle-hole (P) correlation functions,
\begin{subequations}
\begin{eqnarray}
K_{\alpha\alpha'}&=&\lim_{q\rightarrow 0}\int_{k,\epsilon}
G_{\alpha}(k,\epsilon)G_{\alpha'}(-k+q,-\epsilon),\label{ppcf}\\
P_{\alpha\alpha'}&=&\lim_{q\rightarrow 0}\int_{k,\epsilon}
G_{\alpha}(k,\epsilon)G_{\alpha'}(k+q,\epsilon)\label{phcf},
\end{eqnarray}
\end{subequations}
for $\alpha,\alpha'=A,B$, where
\begin{equation}
G_{\alpha}(k,\epsilon)\equiv\frac{1}{{\rm i}\epsilon
-\xi_{Q_{\alpha}+k}},
\end{equation}
is the one-particle Green function in the absence of interaction
for electrons near the saddle point $Q_{\alpha}$ and
\begin{equation}
\int_{k,\epsilon}\equiv \int_{|k|<k_{c}}
\frac{{\rm d}^{2}k}{(2\pi)^{2}}\int\detp,
\end{equation}
where the cutoff around the saddle points, $k_{c}$, is introduced.
We note that
\begin{subequations}
\begin{equation}
K_{1}\equiv K_{AA}=K_{BB},\:\:K_{2}\equiv K_{AB}=K_{BA},\label{K1K2}
\end{equation}
stand for Cooper- and $\pi$-pair correlation, respectively, and
\begin{equation}
P_{1}\equiv P_{AA}=P_{BB},\:\:P_{2}\equiv P_{AB}=P_{BA},\label{P1P2}
\end{equation}
\end{subequations}
for uniform and staggered density-density correlation, respectively.
For $\mu=4t'$, these correlation functions are logarithmically divergent,
\begin{subequations}
\begin{equation}
K_{1}\sim \frac{c}{8\pi^{2}t}\log^{2}\frac{E_{c}}{\omega}, \:\:
P_{1}\sim -\frac{c}{4\pi^{2}t}\log\frac{E_{c}}{\omega},
\end{equation}
\begin{equation}
K_{2}\sim\left\{
\begin{array}{cc}
\frac{c''}{4\pi^{2}t}\log\frac{E_{c}}{\omega}
& \mbox{for } \omega \ll rE_{c},\\
-P_{1} & \mbox{for } \omega \gg rE_{c},
\end{array}\right.
\end{equation}
\begin{equation}
P_{2}\sim
\left\{
\begin{array}{cc}
-\frac{c'}{4\pi^{2}t}\log\frac{E_{c}}{\omega}
& \mbox{for } \omega \ll rE_{c},\\
-K_{1} & \mbox{for } \omega \gg rE_{c},
\end{array}
\right.
\end{equation}
\label{bubbles2}
\end{subequations}
where $E_{c}>0$ and $\omega>0$ ($\omega\ll E_{c}$)
are the ultraviolet and infrared energy cutoff, respectively,
\begin{subequations}
\begin{eqnarray}
c&\equiv &\frac{1}{\sqrt{1-4r^{2}}},\\
c'&\equiv&\log\frac{1+\sqrt{1-4r^{2}}}{2r},\\
c''&\equiv&\frac{1}{2r}\arctan(\frac{2r}{\sqrt{1-4r^{2}}}),
\end{eqnarray}
\label{coefc}
\end{subequations}
and $r\equiv |t'|/t$.
$c$, $c'$ and $c''$ as a function of $r$ are shown in Fig.~\ref{coef}.
Especially for small $r$,
\begin{equation}
c,c''\sim 1,\:\: c'\sim -\log r.\label{smallt'}
\end{equation}
For $r>r_{c}\sim 0.276$, $c>c'$ and $P_{1}$ is more divergent than
$P_{2}$ for $\omega\rightarrow 0$.
We note $c''<$max$\{c,c'\}$, i.e., $\pi$-pair susceptibility $K_{2}$
is always less divergent than particle-hole susceptibility.
For $t'/t=-0.2$, $c''<c<c'$ and they are comparable in magnitude.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=4.5cm
\epsfbox{coef.eps}
\end{center}
\caption{$c$, $c'$ and $c''$ as a function of $r$.}
\label{coef}
\end{figure}
\subsection{Renormalization Group Method for the Saddle Points}\label{RGtech}
In the last subsection,
we saw that the saddle points on the Fermi surface lead to
logarithmic divergence of particle-particle and particle-hole
correlation functions.
This implies that
the fluctuation effect becomes strong.
In the RG approach,
we assume that the single renormalization group variable
$x\equiv\log\frac{E_{c}}{\omega}$ determine the behavior of the system.
The increase of $x$ represents renormalization towards lower energy scale.
For simplicity, we neglect
(1) the deformation of the Fermi surface
by interaction and
(2) $k$-dependence of interaction for small $|k|<k_{c}$,
i.e., we consider only eight coupling constants,
$g_{is}$ ($i=1,2,3,4$ and $s=\perp,\parallel$).
This interaction of the g-\' ology type is shown in Fig.~\ref{gology}.
$g_{1}$ and $g_{3}$
($g_{2}$ and $g_{4}$)
stand for the backward (forward) scattering processes with large (small)
momentum transfer, respectively.
Especially, $g_{1}$ and $g_{3}$
describe 'exchange' and 'Umklapp' processes.
In I, only $g_{1\perp}$ and $g_{3\perp}$ were treated
and taken to be momentum-independent
all over the magnetic Brillouin zone.\cite{ore2and3}
\begin{figure}
\begin{center}
\leavevmode\epsfysize=4cm
\epsfbox{gology2D.eps}
\end{center}
\caption{The scattering processes.
Solid and dashed lines stand for electrons
near $Q_{A}=(\pi,0)$ and $Q_{B}=(0,\pi)$, respectively.}
\label{gology}
\end{figure}
We start with
the renormalization of the couplings in the one-loop approximation.
One-loop diagrams are shown in Fig.~\ref{G1loop}.
The scaling equations are
\begin{subequations}
\begin{eqnarray}
\dot{g_{1\perp}}&=&
-2g_{1\perp}g_{2\perp}\dot{K_{2}}-2g_{1\perp}g_{4\perp}\dot{P_{1}}
\nonumber\\&&
+2g_{1\perp}(g_{1\parallel}-g_{2\parallel})\dot{P_{2}},\\
\dot{g_{1\parallel}}&=&
-2g_{1\parallel}g_{2\parallel}\dot{K_{2}}-2g_{1\parallel}g_{4\parallel}
\dot{P_{1}}\nonumber\\&&
+[2g_{1\parallel}(g_{1\parallel}-g_{2\parallel})
+(g_{1\perp}^{2}-g_{1\parallel}^{2})\nonumber\\&&
\:\:\:\:\:\:+(g_{3\perp}^{2}-g_{3\parallel}^{2})]\dot{P_{2}},\\
\dot{g_{2\perp}}&=&
-(g_{1\perp}^{2}+g_{2\perp}^{2})\dot{K_{2}}-2g_{4\perp}(g_{1\parallel}-
g_{2\parallel})\dot{P_{1}}
\nonumber\\&&
-(g_{2\perp}^{2}+g_{3\perp}^{2})\dot{P_{2}},\\
\dot{g_{2\parallel}}&=&
-(g_{1\parallel}^{2}+g_{2\parallel}^{2})\dot{K_{2}}
-2(g_{4\parallel}g_{1\parallel}-g_{4\perp}g_{2\perp})\dot{P_{1}}
\nonumber\\&&
-(g_{2\parallel}^{2}+g_{3\parallel}^{2})\dot{P_{2}},\\
\dot{g_{3\perp}}&=&-2g_{3\perp}g_{4\perp}\dot{K_{1}}
-2g_{3\perp}(g_{2\perp}+g_{2\parallel}-g_{1\parallel})\dot{P_{2}},
\label{g3perp}\\
\dot{g_{3\parallel}}&=&-2g_{3\parallel}g_{4\parallel}\dot{K_{1}}
-2(2g_{3\parallel}g_{2\parallel}-g_{3\perp}g_{1\perp})\dot{P_{2}},\\
\dot{g_{4\perp}}&=&-(g_{3\perp}^{2}+g_{4\perp}^{2})\dot{K_{1}}
\nonumber\\&&-[g_{1\perp}^{2}+g_{4\perp}^{2}+
2g_{2\perp}(g_{1\parallel}-g_{2\parallel})]\dot{P_{1}},\\
\dot{g_{4\parallel}}&=&-(g_{3\parallel}^{2}+g_{4\parallel}^{2})\dot{K_{1}}
\nonumber\\&&-[g_{1\parallel}^{2}+(2g_{4\parallel}^{2}-g_{4\perp}^{2})
+2g_{2\parallel}(g_{1\parallel}-g_{2\parallel})
\nonumber\\&&\:\:\:\:\:\:
-(g_{2\perp}^{2}-g_{2\parallel}^{2})]\dot{P_{1}},
\end{eqnarray}
\label{intflow}
\end{subequations}
which are to be solved with the initial conditions
$g_{is}(x=x_{i})=g^{0}_{is}$ ( $\dot{}$ $\equiv $d/d$x$),
where $g^{0}_{is}$ are the bare coupling constants.
For example, $g_{3\perp}$ has the following form
to one loop order,
\begin{equation}
g_{3\perp}=-2g^{0}_{3\perp}g^{0}_{4\perp}K_{1}
-2g^{0}_{3\perp}(g^{0}_{2\perp}+g^{0}_{2\parallel}-g^{0}_{1\parallel})
P_{2}.
\end{equation}
By differentiating this equation by $x$ and
replace $g^{0}_{is}$ by $g_{is}$, i.e.,
the bare coupling constants by the renormalized ones,
we obtain the scaling equation eq. (\ref{g3perp}).\\
If we take $g^{0}_{i\perp}=g^{0}_{i\parallel}$ as the initial conditions,
the relation $g_{i\perp}=g_{i\parallel}$ holds all through the flow.
Therefore, the above scaling equations are simplified
as follows,\cite{FurukawaRice}
\begin{subequations}
\begin{eqnarray}
\dot{g_{1}}&=&-2g_{1}g_{2}\dot{K_{2}}-2g_{1}g_{4}\dot{P_{1}}
-2g_{1}(g_{2}-g_{1})\dot{P_{2}},\\
\dot{g_{2}}&=&-(g_{1}^{2}+g_{2}^{2})\dot{K_{2}}
+2g_{4}(g_{2}-g_{1})\dot{P_{1}}\nonumber\\&&
-(g_{2}^{2}+g_{3}^{2})\dot{P_{2}},\\
\dot{g_{3}}&=&-2g_{3}g_{4}\dot{K_{1}}-2g_{3}(2g_{2}-g_{1})\dot{P_{2}},\\
\dot{g_{4}}&=&-(g_{3}^{2}+g_{4}^{2})\dot{K_{1}}\nonumber\\&&
-[g_{1}^{2}+g_{4}^{2}+
-2g_{2}(g_{2}-g_{1})]\dot{P_{1}}.
\end{eqnarray}
\label{intflow2}
\end{subequations}
where $g_{i}\equiv g_{i\perp}=g_{i\parallel}$.
The divergence of $g_{is}(x)$ at a {\em finite} $x$
indicates the existence of the strong coupling fixed point,
i.e., signals the development of an ordered state,
at {\em finite} energy scale or {\em finite} temperature.
(Strictly speaking, this {\em finite} onset temperature
is an artifact of the present approximation in the 2D systems,
and should be interpreted as a crossover temperature, or a
critical temperature when finite three-dimensionality is assumed.)
The properties of this strong coupling fixed point
can be obtained qualitatively from various response functions.
The response functions in the one-loop approximation
are obtained from one-loop diagrams shown in Fig.~\ref{R1loop}.
The response function,
\begin{equation}
R_{\nu}=\int_{0}^{\beta}{\rm d}\tau {\rm e}^{{\rm i}0\tau}\cdot
\frac{1}{N}<T_{\tau}\hat{O}^{\dagger}_{\nu}(\tau)\hat{O}_{\nu}>,
\end{equation}
where $\nu$ stands for the kind of correlation ($\nu=$dSC, SDW, $\cdots$),
has the following form to one-loop order,
\begin{equation}
R_{\nu}=R_{\nu}^{0}+\frac{1}{4}g^{0}_{\nu}(R_{\nu}^{0})^{2},
\end{equation}
where $g^{0}_{\nu}$ is the coupling constant (linear combination of
$g^{0}_{is}$) and $R_{\nu}^{0}$ is the simple bubble.
If we differentiate this equation by $x$ and
replace $g_{\nu}^{0}$, $R_{\nu}^{0}$ by the renormalized ones,
$g_{\nu}$ and $R_{\nu}$,
we obtain
\begin{equation}
\dot{R_{\nu}}=\dot{R_{\nu}^{0}}
\left\{1+\frac{1}{2}g_{\nu}R_{\nu}\right\}.\label{Rnflow}
\end{equation}
This equation is to be solved
with the initial condition that $R_{\nu}(x=x_{i})\sim 0$.
Since $\dot{R_{\nu}}^{0}$ is positive,
$R_{\nu}$ can be divergent for $g_{\nu}>0$ and are suppressed to zero for
$g_{\nu}<0$.
In this paper,
we consider the response functions shown in Fig~\ref{gn},
where
$\hat{O}_{sSC}$, $\hat{O}_{\eta}$ and $\hat{O}_{PS}$
stand for $s$-wave Cooper-pair,
$\eta$-singlet pair with a total momentum $Q$
and total spin $S=0$,\cite{Eta1,Eta2}
and uniform charge density, respectively.
The most divergent $R_{PS}$ is interpreted
to describe the phase separation (PS).
The relationship among these order parameters is discussed in
Appendix \ref{sym}.
We note that each correlation is treated independently
in the above procedure.
Therefore,
we can determine the most dominant susceptibility in the normal state,
and cannot assess the coexistence of different orders.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=5cm
\epsfbox{G1loop.eps}
\end{center}
\caption{Diagrams contributing to the one-loop order correction to
coupling constants.}
\label{G1loop}
\end{figure}
\begin{figure}
\begin{center}
\leavevmode\epsfysize=2cm
\epsfbox{R1loop.eps}
\end{center}
\caption{Diagrams contributing to the one-loop order correction to
response functions.}
\label{R1loop}
\end{figure}
\begin{figure}
\begin{center}
\leavevmode\epsfysize=4.5cm
\epsfbox{tab.eps}
\end{center}
\caption{Operator $\hat{O}_{\nu}$, bare response function
$R_{\nu}^{0}(>0)$ and
coupling constant $g_{\nu}$ for each correlation $\nu$.
Each response function can be divergent (or there exists mean-field solution
$<\hat{O}_{\nu}>\neq 0$)
for $g_{\nu}>0$. $g$ is defined as $g\equiv g_{1\parallel}-g_{2\parallel}$.
In the RG method where only electrons near the saddle points
are taken into account,
the sum over $p$ is restricted to $p\sim(\pi,0),\:(0,\pi)$
and the $p$-dependence of the $d_{x^{2}-y^{2}}$-wave factor
$w_{p}\propto \cos p_{x}-\cos p_{y}$ is ignored.
In our calculation, we take $w_{p}={\rm sgn}(\cos p_{x}-\cos p_{y})$.}
\label{gn}
\end{figure}
\subsection{Phase Diagram}
We solve the scaling equations, eq. (\ref{intflow}) and (\ref{Rnflow}),
with the initial conditions,
\begin{subequations}
\begin{eqnarray}
g^{0}_{1\perp}=g^{0}_{3\perp}&\equiv& U+2V_{Q}=U-4V,\\
g^{0}_{2\perp}=g^{0}_{4\perp}&\equiv& U+2V_{0}=U+4V,\\
g^{0}_{1\parallel}=g^{0}_{3\parallel}&\equiv& 2V_{Q}=-4V,\\
g^{0}_{2\parallel}=g^{0}_{4\parallel}&\equiv& 2V_{0}=4V,
\end{eqnarray}
\end{subequations}
at $x=x_{i}$.
Here, we take $x_{i}\equiv 0$ and $R_{\nu}(x_{i})\equiv 0$ for simplicity,
although the solution of the scaling equations
depends on the value of $x_{i}$ and $R_{\nu}(x_{i})$.
Before we show our results,
we refer to previous results for $U>0$ and $V=0$ obtained by many authors.
$H_{U}$ can be rewritten as follows,
\begin{equation}
H_{U}=
\frac{U}{2}\sum_{i}n_{i}n_{i}-\frac{U}{2}\sum_{i}n_{i}.\label{nini}
\end{equation}
If we regard the second term in the r.h.s of eq. (\ref{nini})
as the chemical potential shift,
we can take $g^{0}_{i\perp}=g^{0}_{i\parallel}=U$ as the initial conditions
and therefore use eq. (\ref{intflow2}) as the scaling equations of the
coupling constants.
For the perfect nesting case $r\equiv |t'|/t=0$,
Schulz\cite{Schulz} and { Dzyaloshinski\u\i}\cite{scvHs}
showed that SDW occurs,
and pointed out that small deviations from half filling lead to dSC.
Lederer {\em et al.}\cite{Lederer} and
Furukawa {\em et al.}\cite{FurukawaRice} solved the scaling equations
for $r\ll 1$
and the same results.
Especially, Furukawa {\em et al.}\cite{FurukawaRice}
indicated that the correlation of $\pi$-triplet pair is suppressed to zero.
In these calculations, however,
$P_{1}$ and $K_{2}$ are neglected in eq. (\ref{intflow2})
for the reason that they are less singular than $P_{2}$,
i.e., $c,c''\ll c'$ for $r\ll 1$, in eq. (\ref{bubbles2}) and
(\ref{coefc}).
On the other hand,
Alvarez {\em et al.}\cite{ttu} solved the flow equations
with the initial condition $g^{0}_{i\perp}=U$ and $g^{0}_{i\parallel}=0$,
by neglecting the generation of $g_{i\parallel}$ and
omitting particle-particle diagram $K_{1}$ and $K_{2}$
in eq. (\ref{intflow}), and showed that dSC, SDW and FM can be stabilized
depending on $U/t$ and $r$.
Now, we show the phase diagram in the plane of $U>0$ and $V$
for $t'=-1/5$
in Fig.~\ref{RGpd},
in which the ordered state with highest onset temperature is shown.
First, we discuss the case that
each ordered state is treated independently,
because we cannot examine the coexistence of different orders
in the RG method.
The crucial difference from our mean field result
is that superconductivity appears even for $U>0$ and $V\geq 0$, i.e.,
dSC for $U>4V$ and sSC for $U<4V$.
This result that dSC is possible for small $V>0$ as well as $V=0$
near half filling is consistent with a recent calculation
based on the fluctuation-exchange (FLEX) approximation.\cite{scLRC}
Except for superconductivity for $U>0$ and $V\geq 0$,
the RG phase diagram is qualitatively
same as the mean field one when we do not take the coexistence of different
orders into account, i.e.,
SDW and CDW appear for $U>4V$ and $U<4V$, respectively,
and attractive $V<0$ favors dSC for small $|V|$ and PS for large $|V|$,
respectively.
With regard to the correlation of $\pi$-triplet pair,
our RG calculation has shown that
it can be divergent for attractive $V<0$ and large $|V|$
but is always subdominant.
Similarly, FM cannot be the most dominant solely.
These results are also consistent with our mean field ones.
Next, we discuss the possibility
of the coexistence of different orders at low temperature,
especially at $T=0$.
It is very important that our RG calculation shows the existence of
a region where the onset temperature of SDW or CDW
becomes highest, as shown in Fig.~\ref{RGpd}.
Since our mean field calculation in \S \ref{vnega} or \S \ref{vposi}
shows that the Fermi surface remains in the SDW and CDW states
near half filling,
we might expect to find a second phase transition at lower temperature
in such SDW and CDW states.
Therefore,
at lower temperature in the SDW region in Fig.~\ref{RGpd},
(1) the coexistent state with dSC, SDW {\em and} $\pi$-triplet pair
found for $V<0$ in the mean field approximation
might be expected to survive
for not only $V<0$ but also $V\geq 0$,
and (2) the ferrimagnetic coexistent state with CDW, SDW {\em and} FM
found for $U>4V>0$ in the mean field approximation
might be expected to survive.
In fact,
as we have pointed out in \S \ref{vnega},
VMC calculations for $U>0$ and $V=0$ show the coexistence of
dSC and SDW at low temperature,\cite{giar,yamaji}
although $\pi$-triplet pair has been neglected.
Similarly,
at lower temperature in the CDW region in Fig.~\ref{RGpd},
(3) the ferrimagnetic coexistent state with CDW, SDW {\em and} FM
found for $4V>U>0$ in the mean field approximation
might be expected to survive.
Especially,
$\pi$-triplet pair (FM),
which cannot be stabilized solely
in the parameter region considered
in the present mean field and RG approximation,
might be expected to arise as a result of the coexistence of
dSC and SDW (CDW and SDW).
In order to assess the effect of
$\pi$-triplet pair (FM) on the coexistence of dSC and SDW (CDW and SDW),
we need another theoretical treatment.
\begin{figure}
\begin{center}
\leavevmode\epsfysize=5cm
\epsfbox{pdRGr02.eps}
\end{center}
\caption{The RG phase diagram for $t'/t=-1/5$ and $\mu=4t'$.}
\label{RGpd}
\end{figure}
Here, we refer to the possibility of the coexistence of sSC and CDW.
In the g-\'ology model in I,
it can be easily shown that
the coexistent state with sSC, CDW {\em and} $\eta$-singlet pair
can be stabilized near half filling at low temperature
for $g_{3\perp}<0$ and $g_{1\perp}<0$.\cite{oreD}
As seen from eq. (\ref{tolattice}),
this case corresponds to that of $U<0$ and $V>0$
in the present extended Hubbard model.
Therefore, this coexistent state might be expected to be stabilized
also in the extended Hubbard model for $U<0$ and $V>0$
in the mean field approximation.
Moreover, based on the above discussion,
it might be expected to survive in the presence of fluctuation
for not only $U<0$ but also $U\geq 0$,
at lower temperature in the CDW region in Fig.~\ref{RGpd}.
This will be reported elsewhere.
Finally, we refer to the ambiguities of the above RG method.
Since there exist not only $\log$- but also
$\log^{2}$-divergence in the particle-particle and particle-hole
correlation functions, eq. (\ref{bubbles2}),
we cannot safely take the limit $\omega\rightarrow 0$
in the scaling equations of coupling constants and
response functions, eq. (\ref{intflow}) and (\ref{Rnflow}),
i.e., it is not clear at all whether the above RG treatment is
valid or not.
In fact, the solution of eq. (\ref{intflow}) and (\ref{Rnflow})
depends on the initial value $x_{i}$.
If we consider only the most singular $\log^{2}$ term
in $K_{1}$ (and $P_{2}$ for $r=0$)
and take $y\equiv x^{2}=\log^{2}\frac{E_{c}}{\omega}$
as a new scaling variable,\cite{Schulz}
we can safely take $\omega\rightarrow 0$ limit in the scaling equations of
coupling constants.
In this case, the above RG method
might correspond to a parquet summation of leading $\log^{2}$
divergences, rather than renormalization procedure.\cite{newVHS}
\section{Conclusion and Discussion}
We have studied in detail possible ordered states,
especially possible coexistence of different orders,
near half filling in the 2D extended Hubbard model with
on-site repulsion $U>0$ and
nearest-neighbor interaction $V$,
with emphasis on
electrons around the saddle points
($\pi$,0) and (0,$\pi$).
First, we have determined the phase diagram at $T=0$
in the mean field approximation.
For $V<0$, we have shown
that the coexistent state with dSC, SDW {\em and} $\pi$-triplet pair
can be stabilized near half filling.
Here, we have indicated the following important fact
which has often been neglected in previous studies:
{\em
when we discuss the coexistence of dSC and SDW,
it is necessary to take $\pi$-triplet pair into account from the outset,
because in general the coexistence of dSC and SDW results in
$\pi$-triplet pair
and is affected by $\pi$-triplet pair.}\cite{ore2and3}
Especially,
$\pi$-triplet pair, which cannot condensate solely
in the present model,
can arise through the coexistence of dSC and SDW.
Since the phase separation (PS) is generally expected to occur
in the presence of finite-range attractive interaction such as $V<0$,
we have examined the effect of PS on the mean field ground state
in the random phase approximation (RPA).
The coexistent state with dSC, SDW and $\pi$-triplet pair
is severely hampered by PS but survives,
and that the long-range Coulomb repulsion such as
next-nearest-neighbor density-density repulsion
leads to the suppression of PS.
On the other hand, for $V>0$, we showed that
a {\em ferrimagnetic} coexistent state with
commensurate charge-density-wave (CDW), SDW {\em and} ferromagnetism (FM)
can be stabilized near half filling.
Here, we have indicated the following important fact:
{\em
when we discuss the coexistence of CDW and SDW,
it is necessary to take FM into account from the outset,
because in general the coexistence of CDW and SDW results in FM
and is affected by FM.}
Especially, FM, which cannot be stabilized solely
with the present choice of parameters,
can arise through the coexistence of CDW and SDW.
It is to be noted that
the above mean field coexistent states near half filling
can be stabilized at low temperature, especially at $T=0$,
when CDW or SDW, in which the Fermi surface remains near half filling,
arises first at high temperature.
In order to examine the effects of fluctuation on
the mean field ordered states,
we have adopted the RG method
for the {\em special} case that
the Fermi level lies just on the saddle points.
We have shown that
the crucial difference from our mean field result is
that superconductivity can arise even for $U>0$ and $V\geq0$;
dSC and sSC for $U>4V$ and $U<4V$, respectively.
Except for this difference,
the RG phase diagram is qualitatively
same as the mean field one when we do not take the coexistence of different
orders into account, {\em e.g.},
SDW and CDW can arise for $U>4V$ and $U<4V$, respectively.
Especially, the correlation of $\pi$-triplet pair or FM
cannot be the most dominant solely.
Here, it is very important that
a region where the onset temperature of SDW or CDW
becomes highest is found in the RG phase diagram.
Since the Fermi surface remains near half filling
in these SDW and CDW states,
we might expect to find a second phase transition at lower temperature.
In the RG method, however,
we cannot assess such possibilities.
On the other hand,
the mean field approximation,
which is often questionable for the 2D case,
is of great advantage in that
we can study the stability of coexistent states with
different order parameters quantitatively.
Therefore,
we can conclude that our mean field calculations
indicate the possibilities that
(1) SDW, which has been shown in our RG calculation
to arise first at high temperature for $U>4V$,
coexists with dSC {\em and} $\pi$-triplet pair, or with
CDW {\em and} FM, at lower temperature, and that
(2) CDW, which has been shown in our RG calculation to arise
first at high temperature for $U<4V$,
coexists with SDW {\em and} FM, at lower temperature.
Throughout this letter,
we have assumed YBCO-type Fermi surface by introducing $t'$ and
consider only {\em commensurate} (C) SDW or CDW.
Near half filling, however, {\em incommensurate} (IC) ordering
or stripe formation can be expected, especially for $t'=0$
in the repulsive Hubbard model.\cite{schulzIC,riceIC,machida}
Recently, the effect of $V$ on such stripe states
has been examined.\cite{UVIC}
The effect of IC ordering on the stability of
the coexistent states
(with dSC, C-SDW and $\pi$-triplet pair,
and with C-CDW, C-SDW and FM) is beyond the scope of the present study.
In the repulsive Hubbard model with $t'=0$,
Giamarchi {\em et al.}\cite{giarICHU} have shown that
the coexistent state with dSC and C-SDW state
have higher energy than that with dSC and IC-SDW.
With regard to dSC,
we have assumed that the superconducitng gap symmetry
is purely of $d_{x^{2}-y^{2}}$-wave.
However, there are a few indications that
dSC mixed with components of other symmetry
can be stabilized,\cite{onlyV,laughlin,ogata,kuroki}
dependent on interaction, electron density, etc.
Such mixed pairing states leave much room for future studies.
The author thanks to H. Fukuyama, H. Kohno and M. Ogata
for valuable discussions.
|
\section{Introduction}
The use of gravitational lensing statistics as a cosmological tool was
first considered in detail by \citet{ETurnerOG84a}; the influence of the
cosmological constant was investigated thoroughly by
\citet{MFukugitaFKT92a}, building on the work of \citet{ETurner90a} and
\citet{MFukugitaFK90a}. \citet[and references therein]{CKochanek96a}
and, more recently, \citet[hereafter FKM]{EFalcoKM98a} have laid the
groundwork for using gravitational lensing statistics for the detailed
analysis of extragalactic surveys. \citet[hereafter
Paper~I]{RQuastPHelbig99a} reanalysed optical surveys from the
literature, for the first time exploring a range of the
$\lambda_{0}$-$\Omega_{0}$ parameter space large enough to enable a
comparison with other cosmological tests. Here, we use the formalism
outlined in Paper~I to analyse the Jodrell Bank-VLA Astrometric Survey
(JVAS), the largest completed gravitational lens survey to date.
Radio surveys offer several advantages over optical surveys (see, e.g.,
FKM): one doesn't have to worry about systematic errors due to
extinction or a lens galaxy of apparent brightness comparable to that of
the lensed images of the source, the resolution (of followup
observations if not of the survey proper) is much smaller than the
typical image separation, parent catalogues in the form of large-area
surveys exist from which unbiased samples can be selected and relatively
easily observed. Disadvantages in the radio are due to our relatively
poor knowledge of the flux density-dependent redshift distribution or
equivalently the redshift-dependent number-magnitude relation.
For a description of our method see Paper~I. The plan of this paper is
as follows. Sect.~\ref{JVAS} describes the JVAS gravitational lens
survey. In Sect.~\ref{calculations} we describe the calculations we
have done based on the JVAS data. Sect.~\ref{results} presents our
results, using both the JVAS data alone and in combination with the
results from the optical surveys analysed in Paper~I. Finally in
Sect.~\ref{conclusions} we compare our results to those of Paper~I and
present our conclusions and our prognosis for the analysis of future
large surveys such as CLASS.
\section{The JVAS Gravitational Lens Survey}
\label{JVAS}
\subsection{The sample}
The Jodrell Bank-VLA Astrometric Survey (JVAS) is a survey for
flat-spectrum radio sources with a flux density greater than 200\,mJy at
5\,GHz. Flat-spectrum radio sources are likely to be compact, thus
making it easy to recognise the lensing morphology. In addition, they
are likely to be variable, making it possible to determine $H_{0}$ by
measuring the time delay between the lensed images. (See
\citet{ABiggsBHKWP99a} for the description of a time delay measurement
in a JVAS gravitational lens system.) JVAS is also a survey for MERLIN
phase-reference sources and as such is described in
\citet{APatnaikBWW92a}, \citet{IBRownePWW98a} and
\citet{PWilkinsonBPWS98a}. JVAS as a gravitational lens survey, the lens
candidate selection, followup process, confirmation criteria and a
discussion of the JVAS gravitational lenses is described in detail in
\citet{LKingBMPW99a} \citep[see also][]{LKingIBrowne96a}.
In order to have a parent sample which is as large as possible and as
cleanly defined as practical, our ``JVAS gravitational lens survey
sample'' is slightly different than the ``JVAS phase-reference calibrator
sample''. For the former, the source must be a point source and must
have a good starting position (so that the observation was correctly
pointed) while its precise spectral index is not important. For the
latter, only the spectral index is important, as the source can be
slightly resolved or the observation can be less than perfectly pointed.
Thus, the JVAS astrometric sample
\citep{APatnaikBWW92a,IBRownePWW98a,PWilkinsonBPWS98a} contains 2144
sources. To these must be added 103 sources which were too resolved to
be used as phase calibrators and 61 sources which had bad starting
positions (thus the observations were too badly pointed to be useful for
the astrometric sample), bringing the total to 2308. This formed our
gravitational lens sample, since these additional sources were also
searched for gravitational lenses \citep{LKingBMPW99a} (none were found
meeting the JVAS selection criteria).
\subsection{The lenses}
We have used the gravitational lens systems in Table~\ref{ta:lenses} in
this analysis.
\begin{table*}
\caption[]{JVAS lenses used in this analysis. Of the information in the
table, for this analysis we use only the source redshift $z_{\mathrm{s}}$
and the image separation $\Delta\theta$}
\label{ta:lenses}
\begin{tabular*}{\textwidth}{@{\extracolsep{\fill}}llllll}
\hline
\hline
Name &
\# images &
$\Delta\theta ['']$ &
$z_{\mathrm{l}}$ &
$z_{\mathrm{s}}$ &
lens galaxy \\
\hline
\object{B0218$+$357} &
2 + ring &
0.334 &
0.6847 &
0.96 &
spiral \\
\object{MG0414$+$054} &
4 &
2.09 &
0.9584 &
2.639 &
elliptical \\
\object{B1030$+$074} &
2 &
1.56 &
0.599 &
1.535 &
spiral \\
\object{B1422$+$231} &
4 &
1.28 &
0.337 &
3.62 &
\textbf{?} \\
\hline
\end{tabular*}
\end{table*}
The JVAS lens \object{B1938$+$666} \citep{LKingJBBBdBFKMNW98a} was not
included because it is not formally a part of the sample, having a too
steep spectral index and having been recognised on the basis of a lensed
extended source as opposed to lensed compact components. Also, the JVAS
lens \object{B2114$+$022} \citep{PAugustoBWJFM99a} was not included
because it is not a single-galaxy lens system.
\section{Calculations}
\label{calculations}
A major difference between the analysis of an optical survey (see
Paper~I and references therein) and a radio survey is that in the latter
one does not know the redshifts of all the unlensed sources. One can
still use the formalism of Paper I, however, substituting for the
non-lensed objects in the sample a subsample with known redshifts,
multiplying the logarithm of this contribution from the non-lenses to
the likelihood by the ratio of the size of the parent sample to that of
the subsample. Alternatively, one can take the redshifts from a sample
selected according to the same criteria, assigning these randomly to
objects in (a subsample of) the parent sample for a similar flux density
range. Similarly, one does not know the number-magnitude relation for
the sample and for its extension to fainter flux densities (needed to
allow for the lens amplification). Again, this can be estimated from
either a subsample (through extrapolation) or from another sample
selected according to the same criteria (either through extrapolation or
by having a fainter flux density limit in this other sample; in the
latter case obviously the selection criteria should be identical to that
of the original sample except for the lower flux density limit).
For this analysis, due to the paucity of the observational data, we have
made rather stark assumptions: the redshift distribution of the sample
is assumed to be identical to that of the CJF sample
\citep{GTaylorVRPHW96a}, independent of flux density, and the
number-magnitude relation is assumed to be identical to that of CLASS
\citep[Cosmic Lens All-Sky Survey,][]{SMyersetal99a}, independent of
redshift.
Otherwise, we have followed the procedure outlined in Paper~I,
calculating the a priori likelihood of obtaining the observational data
as a function of $\lambda_{0}$ and $\Omega_{0}$ and the a posteriori
likelihood for the three different choices of prior information used in
Paper I. We present results both for the JVAS lens survey and for the
combination of the JVAS results with those from the optical surveys
analysed in Paper~I.
\section{Results and discussion}
\label{results}
\begin{figure*}
\resizebox{0.375\textwidth}{!}{\includegraphics{rdlik.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdlik.eps}}
\caption[]{\emph{Left panel:} The likelihood function
$p(D|\lambda_0,\Omega_0,\vec{\xi}_0)$ based on the JVAS lens sample.
All nuisance parameters are assumed to take precisely their mean values.
The pixel grey level is directly proportional to the likelihood ratio,
darker pixels reflect higher ratios. The pixel size reflects the
resolution of our numerical computations. The contours mark the
boundaries of the minimum $0.68$, $0.90$, $0.95$ and $0.99$ confidence
regions for the parameters $\lambda_0$ and $\Omega_0$. \emph{Right
panel:} Exactly the same as the left panel, but the joint likelihood
from the JVAS lens sample and the optical samples from
\citet[Paper~I]{RQuastPHelbig99a}}
\label{fi:likelihood
\end{figure*}
The left panel of Fig.~\ref{fi:likelihood} shows the constraints on the
cosmological parameters $\lambda_{0}$ and $\Omega_{0}$ based only on the
information obtained from the JVAS lens statistics, while the right
panel shows the joint constraints from the JVAS lens sample and the
optical samples from Paper~I. Fig.~\ref{fi:xlikelihood} is identical
except that one of
the input parameters, the normalisation of the galaxy luminosity
function, was increased by two standard deviations. This gives an idea
of the magnitude of systematic uncertainties. (See the discussion in
Paper~I.)
\begin{figure*}
\resizebox{0.375\textwidth}{!}{\includegraphics{rxlik.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jxlik.eps}}
\caption[]{Exactly the same as Fig.~\ref{fi:likelihood}, but the
parameter $n_{\mathrm{e}}$ is increased by two standard deviations.
This parameter, the normalisation of the luminosity function of the lens
galaxies, is one of the more uncertain input parameters, thus one can
get a rough estimate of the overall uncertainty by comparing this figure
and Fig.~\ref{fi:likelihood}. See the discussion in Paper~I}
\label{fi:xlikelihood}
\end{figure*}
\begin{figure*}
\resizebox{0.375\textwidth}{!}{\includegraphics{rdpost1.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdpost1.eps}}
\noindent
\resizebox{0.375\textwidth}{!}{\includegraphics{rdpost2.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdpost2.eps}}
\noindent
\resizebox{0.375\textwidth}{!}{\includegraphics{rdpost3.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdpost3.eps}}
\caption[]{\emph{Left column:} The posterior probability density
functions $p_1(\lambda_0,\Omega_0|D)$ (top panel),
$p_2(\lambda_0,\Omega_0|D)$ (middle panel) and
$p_3(\lambda_0,\Omega_0|D)$ (bottom panel). All nuisance parameters are
assumed to take precisely their mean values. The pixel grey level is
directly proportional to the likelihood ratio, darker pixels reflect
higher ratios. The pixel size reflects the resolution of our numerical
computations. The contours mark the boundaries of the minimum $0.68$,
$0.90$, $0.95$ and $0.99$ confidence regions for the parameters
$\lambda_0$ and $\Omega_0$. The respective amounts of information
obtained from our sample data are $I_1=1.42$, $I_2=1.32$ and $I_3=1.45$.
\emph{Right column:} Exactly the same as the left panel, but the joint
likelihood from the JVAS lens sample and the optical samples from
\citet{RQuastPHelbig99a}. The respective amounts of information
obtained from our joint sample data are $1.98$, $1.95$ and $1.96$. See
Paper~I for definitions}
\label{fi:posterior
\end{figure*}
The left plot in the top row of Fig.~\ref{fi:posterior} shows the joint
likelihood of our lensing statistics analysis and that obtained by using
conservative estimates for $H_{0}$ and the age of the universe (see
Paper~I). Although neither method alone sets useful constraints on
$\Omega_{0}$, their combination does, since the constraint from
$H_{0}$ and the age of the universe
only allows large values of $\Omega_{0}$ for $\lambda_{0}$ values which
are excluded by lens statistics. Even though the 68\% confidence
contour still allows almost the entire $\Omega_{0}$ range, it is obvious
from the grey scale that much lower values of $\Omega_{0}$ are favoured
by the joint constraints. The upper limit on $\lambda_{0}$ changes only
slightly while, as is to be expected, the lower limit becomes tighter.
Right plot: exactly the same, but including optical constraints from
Paper~I. The upper limits on $\lambda_{0}$ decrease slightly, while the
lower limits improve considerably. The latter is probably due to the
fact that, in addition to just using more data the JVAS sources are at
significantly different redshifts than those from the optical surveys
analysed in Paper~I (the JVAS sources are generally at lower redshift).
The former is consistent with the slightly higher optical depth for
radio surveys found by FKM and will be discussed more below.
The middle row of Fig.~\ref{fi:posterior} shows the effect of including
our prior information on $\Omega_{0}$ (see Paper I). As is to be
expected, (for both the JVAS and combination data sets) lower values of
$\Omega_{0}$ are favoured. This has the side effect of weakening our
lower limit on $\lambda_{0}$ (though only slightly affecting the upper
limit). This should not be regarded as a weakness, however, since
including prior information for $\lambda_{0}$ and $\Omega_{0}$ from the
constraint from $H_{0}$ and the age of the universe as well as for
$\Omega_{0}$ itself, as illustrated in the bottom row of
Fig.~\ref{fi:posterior}, tightens the lower limit again (without
appreciably affecting the upper limit).
We believe that the right plot of the bottom row of
Fig.~\ref{fi:posterior} represents very robust constraints in the
$\lambda_{0}$-$\Omega_{0}$ plane. The upper limits on $\lambda_{0}$
come from gravitational lensing statistics, which, due to the extremely
rapid increase in the optical depth for larger values of $\lambda_{0}$,
are quite robust and relatively insensitive to uncertainties in the
input data (cf.~Fig.~\ref{fi:xlikelihood} and the discussion of the
effect of changing the most uncertain input parameter by 2\,$\sigma$ in
Paper~I) as well as to the prior information used (compare the upper,
lower and middle rows of Fig.~\ref{fi:posterior}). The combination of
data from JVAS and optical surveys leads to much tighter lower limits on
$\lambda_{0}$ than using either alone. The upper and lower limits on
$\Omega_{0}$ are based on a number of different methods and appear to be
quite robust (see Paper I). The combination of the relatively secure
knowledge of $H_{0}$ and the age of the universe combine with lens
statistics to produce a good lower limit on $\lambda_{0}$, although this
is to some extent still subject to the caveats mentioned above.
If one is interested in the allowed range of $\lambda_{0}$, one can
marginalise over $\Omega_{0}$ to obtain a probability distribution for
$\lambda_{0}$. This is illustrated in Fig.~\ref{fi:marginal}
\begin{figure*}
\resizebox{0.375\textwidth}{!}{\includegraphics{rdmar.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdmar.eps}}
\noindent
\resizebox{0.375\textwidth}{!}{\includegraphics{rdcum.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdcum.eps}}
\caption[]{\emph{Left column:} The top panel shows the normalised
marginal likelihood function $p(\lambda_0|D)$ (light gray curve) and the
marginal posterior probability density functions $p_1(D|\lambda_0)$
(medium gray curve), $p_2(D|\lambda_0)$ (dark gray curve) and
$p_3(D|\lambda_0)$ (black curve) derived from the JVAS analysis. All
nuisance parameters are assumed to take precisely their mean values.
The bottom panel shows the respective cumulative distribution functions.
\emph{Right column:} Exactly the same as the left panel, but the joint
likelihood from the JVAS lens sample and the optical samples from
\citet{RQuastPHelbig99a}}
\label{fi:marginal
\end{figure*}
and Table~\ref{ta:results}.
\begin{table*}
\caption[]{Marginal mean values, standard deviations and $0.95$
confidence intervals for the parameter~$\lambda_0$ on the basis of the
marginal distributions shown in the top row of Fig.~\ref{fi:marginal}}
\label{ta:results}
\begin{tabular*}{\linewidth}{@{\extracolsep{\fill}}llccccc}
\hline
\hline
Sample & Distribution & Mean & standard deviation &
\multicolumn{2}{l}{95\% c.l.~range} & information \\
\hline
JVAS & $p(D|\lambda_0)$ & $0.13$ & $1.08$ & $-2.08$ & $1.91$ & \\
JVAS & $p_1(\lambda_0|D)$ & $0.44$ & $0.77$ & $-1.05$ & $1.87$ & $1.42$ \\
JVAS & $p_2(\lambda_0|D)$ & $-0.29$ & $0.98$ & $-2.38$ & $1.17$ & $1.32$ \\
JVAS & $p_3(\lambda_0|D)$ & $0.11$ & $0.64$ & $-1.20$ & $1.16$ & $1.45$ \\
joint & $p(D|\lambda_0)$ & $0.19$ & $0.70$ & $-1.17$ & $1.48$ & \\
joint & $p_1(\lambda_0|D)$ & $0.24$ & $0.63$ & $-0.96$ & $1.46$ & $1.98$ \\
joint & $p_2(\lambda_0|D)$ & $-0.25$ & $0.59$ & $-1.46$ & $0.77$ & $1.95$ \\
joint & $p_3(\lambda_0|D)$ & $-0.09$ & $0.48$ & $-1.08$ & $0.77$ & $1.96$ \\
\hline
\end{tabular*}
\end{table*}
The comparison values from this work corresponding to those in Tables~3
and 4 of Paper~I are presented in Tables~\ref{ta:specialo} and
\ref{ta:specialk}.
\begin{table*}
\caption[]{Mean values and ranges for assorted confidence levels for the
parameter $\lambda_{0}$ for our a priori and various a posteriori
likelihoods from this work for $\Omega_{0}=0.3$. This should be
compared to Table~3 in Paper~I}
\label{ta:specialo}
\begin{tabular*}{\linewidth}{@{\extracolsep{\fill}}lrrrrrrrr}
\hline
\hline
Cosmological test&
\multicolumn{2}{c}{68\% c.l.~range} &
\multicolumn{2}{c}{90\% c.l.~range} &
\multicolumn{2}{c}{95\% c.l.~range} &
\multicolumn{2}{c}{99\% c.l.~range} \\
\hline
JVAS, $p(D|\lambda_0)$ &
$-0.66$ & $0.72$ & $-1.68$ & $0.87$ & $-2.36$ & $0.96$ & $-3.91$ & $1.08$ \\
JVAS, $p_1(\lambda_0|D)$ &
$-0.44$ & $0.80$ & $-1.00$ & $0.92$ & $-1.38$ & $1.00$ & $-2.27$ & $1.09$ \\
JVAS, $p_2(\lambda_0|D)$ &
$-1.38$ & $0.86$ & $-2.81$ & $1.00$ & $-3.70$ & $1.06$ & $<-5.00$ & $1.15$ \\
JVAS, $p_3(\lambda_0|D)$ &
$-0.69$ & $0.86$ & $-1.45$ & $0.99$ & $-1.89$ & $1.03$ & $-2.91$ & $1.15$ \\
JVAS \& optical, $p(D|\lambda_0)$ &
$-0.54$ & $0.26$ & $-1.08$ & $0.44$ & $-1.41$ & $0.54$ & $-2.15$ & $0.70$ \\
JVAS \& optical, $p_1(\lambda_0|D)$ &
$-0.63$ & $0.40$ & $-0.95$ & $0.53$ & $-1.18$ & $0.62$ & $-1.72$ & $0.73$ \\
JVAS \& optical, $p_2(\lambda_0|D)$ &
$-1.02$ & $0.44$ & $-1.72$ & $0.63$ & $-2.08$ & $0.72$ & $-2.95$ & $0.80$ \\
JVAS \& optical, $p_3(\lambda_0|D)$ &
$-0.77$ & $0.52$ & $-1.23$ & $0.63$ & $-1.52$ & $0.70$ & $-2.15$ & $0.79$ \\
\hline
\end{tabular*}
\end{table*}
\begin{table*}
\caption[]{Mean values and ranges for assorted confidence levels for the
parameter $\lambda_{0}$ for our a priori and various a posteriori
likelihoods from this work for $k=0$. This should be compared to
Table~4 in Paper~I}
\label{ta:specialk}
\begin{tabular*}{\linewidth}{@{\extracolsep{\fill}}lrrrrrrrr}
\hline
\hline
Cosmological test&
\multicolumn{2}{c}{68\% c.l.~range} &
\multicolumn{2}{c}{90\% c.l.~range} &
\multicolumn{2}{c}{95\% c.l.~range} &
\multicolumn{2}{c}{99\% c.l.~range} \\
\hline
JVAS, $p(D|\lambda_0)$ &
$-0.11$ & $0.70$ & $-0.83$ & $0.78$ & $<-1.00$ & $0.82$ & $<-1.00$ & $0.86$ \\
JVAS, $p_1(\lambda_0|D)$ &
$0.13$ & $0.75$ & $-0.15$ & $0.82$ & $-0.33$ & $0.85$ & $-0.69$ & $0.89$ \\
JVAS, $p_2(\lambda_0|D)$ &
$0.35$ & $0.77$ & $0.13$ & $0.83$ & $0.02$ & $0.85$ & $-0.21$ & $0.88$ \\
JVAS, $p_3(\lambda_0|D)$ &
$0.41$ & $0.79$ & $0.25$ & $0.83$ & $0.16$ & $0.85$ & $-0.04$ & $0.88$ \\
JVAS \& optical, $p(D|\lambda_0)$ &
$-0.15$ & $0.45$ & $-0.49$ & $0.55$ & $-0.69$ & $0.60$ & $<-1.00$ & $0.67$ \\
JVAS \& optical, $p_1(\lambda_0|D)$ &
$0.02$ & $0.54$ & $-0.12$ & $0.61$ & $-0.29$ & $0.64$ & $-0.60$ & $0.70$ \\
JVAS \& optical, $p_2(\lambda_0|D)$ &
$0.39$ & $0.39$ & $0.09$ & $0.59$ & $0.00$ & $0.64$ & $<-0.22$ & $0.70$ \\
JVAS \& optical, $p_3(\lambda_0|D)$ &
$0.39$ & $0.51$ & $0.18$ & $0.63$ & $0.09$ & $0.66$ & $-0.09$ & $0.72$ \\
\hline
\end{tabular*}
\end{table*}
For a ``likely'' $\Omega_{0}$ value of 0.3 we have calculated the
likelihood with the higher resolution $\Delta\lambda_{0}=0.01$. This is
show in Fig.~\ref{fi:r03}.
\begin{figure*}
\resizebox{0.375\textwidth}{!}{\includegraphics{rdlik_03.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{rdcum_03.eps}}
\caption[]{\emph{Left panel:} The likelihood function as a function of
$\lambda_{0}$ for $\Omega_{0}=0.3$ and with all nuisance parameters
taking their default values, using just the JVAS data. \emph{Right
panel:} The same but plotted cumulatively}
\label{fi:r03}
\end{figure*}
\begin{figure*}
\resizebox{0.375\textwidth}{!}{\includegraphics{jdlik_03.eps}}
\hfill
\resizebox{0.375\textwidth}{!}{\includegraphics{jdcum_03.eps}}
\caption[]{As Fig.~\ref{fi:r03} but combining optical and radio data.
\emph{Left panel:} The likelihood function as a function of
$\lambda_{0}$ for $\Omega_{0}=0.3$ and with all nuisance parameters
taking their default values. \emph{Right panel:} The same but plotted
cumulatively}
\label{fi:j03}
\end{figure*}
From these calculations one can extract confidence limits which, due to
the higher resolution in $\lambda_{0}$, are more accurate. These are
presented in Table~\ref{ta:03} and should be compared to those for
$p(D|\lambda_0)$ from Table~\ref{ta:specialo}.
\begin{table*}
\caption[]{Confidence ranges for $\lambda_{0}$ assuming
$\Omega_{0}=0.3$. Unlike the results presented in
Table~\ref{ta:specialo}, these figures are for a specific value of
$\Omega_{0}$ and not the values of intersection of particular contours
with the $\Omega_{0}=0.3$ line in the $\lambda_{0}$-$\Omega_{0}$ plane.
These are more appropriate if one is convinced that$\Omega_{0}=0.3$ and
have been calculated using ten times better resolution than the rest of
our results presented in this work. See Figs.~\ref{fi:r03} and
\ref{fi:j03}}
\label{ta:03}
\begin{tabular*}{\linewidth}{@{\extracolsep{\fill}}lrrrrrrrr}
\hline
\hline
data set &
\multicolumn{2}{c}{68\% c.l.~range} &
\multicolumn{2}{c}{90\% c.l.~range} &
\multicolumn{2}{c}{95\% c.l.~range} &
\multicolumn{2}{c}{99\% c.l.~range} \\
\hline
JVAS &
$-0.69$ & $0.72$ & $-1.72$ & $0.91$ & $-2.39$ & $0.98$ & $-3.83$ & $1.06$ \\
JVAS+optical &
$-0.65$ & $0.30$ & $-1.17$ & $0.49$ & $-1.48$ & $0.57$ & $-2.22$ & $0.70$ \\
\hline
\end{tabular*}
\end{table*}
As mentioned in Paper~I, to aid comparisons with other cosmological
tests, the data for the figures shown in this paper are available at
\begin{quote}
\verb|http://multivac.jb.man.ac.uk:8000/ceres|\\
\verb|/data_from_papers/JVAS.html|
\end{quote}
and we urge our colleagues to follow our example.
\section{Conclusions and outlook}
\label{conclusions}
We have used the method outlined in \citet{RQuastPHelbig99a} to
measure the cosmological constant $\lambda_{0}$ from the lensing
statistics of the Jodrell Bank-VLA Astrometric Survey. At 95\%
confidence, our lower and upper limits on $\lambda_{0}$-$\Omega_{0}$,
using the JVAS lensing statistics information alone, are respectively
$-2.69$ and $0.68$. For a flat universe, these correspond to lower and
upper limits on $\lambda_{0}$ of respectively $-0.85$ and $0.84$. Using
the combination of JVAS lensing statistics and lensing statistics from
the literature as discussed in \citet{RQuastPHelbig99a} the
corresponding $\lambda_{0}-\Omega_{0}$ values are $-1.78$ and $0.27$.
For a flat universe, these correspond to lower and upper limits on
$\lambda_{0}$ of respectively $-0.39$ and $0.64$. Note that the lower
limit is affected more than the upper limit with respect to the
difference between the JVAS results and those in Paper~I and with
respect to combining the JVAS results with those from Paper~I.
Our determination is consistent with other recent measurements of
$\lambda_{0}$, both from lensing statistics and from other cosmological
tests \citep[see][Paper I, for a discussion]{RQuastPHelbig99a}. We
confirm the result of \citet[FKM]{EFalcoKM98a} that radio surveys give
higher values of $\lambda_{0}$ than optical surveys.
\citet{ACoorayQC99a} and \citet{ACooray99a} obtain a 95\% confidence upper limit on
$\lambda_{0}$ in a flat universe of 0.79 from analyses of the Hubble
Deep Field and CLASS. However, these analyses suffer from systematic
effects due to our ignorance of the underlying flux density-dependent
redshift distribution (or, equivalently, the redshift-dependent
luminosity function) of the unlensed parent population. As discussed in
\citet{ACooray99a}, the value of $\lambda_{0}$ obtained from CLASS will
decrease if the mean redshift of the sample is lower than presumed.
Thus, although there is no real conflict at present as the \emph{lower}
limits on $\lambda_{0}$ are not as tight, it seems not unlikely that a
more detailed analysis of CLASS, incorporating more information about
the unlensed parent population, will result in a value more in line with
our value obtained from the JVAS analysis. Of course, the JVAS analysis
also suffers from systematic effects, but the general agreement between
the results obtained from the analysis of optical surveys (cf.~Paper~I
and references therein) and radio surveys as presented here and in FKM
suggests that these are not overwhelming. Also, the difference, a
higher value of $\lambda_{0}$ from radio surveys, is what one would
expect, as lens systems which go unnoticed will, all other things being
equal, reduce the value of $\lambda_{0}$. This could be the case in
optical surveys since it is possible that extinction in the lens galaxy
and the fact that the resolution is only slightly better than the image
separation could lead to lens systems being missed. Again, the general
agreement does suggest though that these effects are not overwhelming.
Of course, one could imagine that the agreement is coincidental, the
optical surveys being heavily affected by extinction and resolution bias
and the radio surveys by our ignorance of the unlensed parent
population. However, the fact that lens statistics in general gives
results which are not in conflict with other cosmological tests
(cf.~Paper~I) suggests that this is not the case. Moreover,
extinction would bias the results from lens
statistics and the $m$-$z$ relation (e.g.~for type Ia supernovae,
cf.~the results in Tables 3 and 4 of paper I and in the references
mentioned there) in the opposite direction. Thus, their agreement
suggests that both methods have their systematics more or less under
control.
The major source of uncertainty in radio lens surveys is the lack of
knowledge about the redshift distribution and number-magnitude relation
of the source sample \citep[e.g.][]{CKochanek96b}. We are currently
undertaking the necessary observations to reduce this source of
systematic error. Since the time scale for this project is comparable
to that for the followup of the CLASS survey, there seems little point
in doing a better analysis of JVAS in the future, especially since CLASS
is defined so that JVAS is essentially a subset of it.\footnote{The
definition of both is flat-spectrum between L-band and C-band,
i.e.~$\alpha>-0.5$ where $s_{f}\sim f^{\alpha}$, the essential
difference being the lower flux density limit of 200\,mJy for JVAS and
30\,mJy for CLASS. However, since CLASS is defined based on newer
catalogues \citep[GB6 and NVSS:][]{PGregorySDJ96a,JCondonCGYPTB98a} than
JVAS, there will be some essentially random differences due to differing
quality of observations and variability of the sources. All the JVAS
lenses mentioned in Table~\ref{ta:lenses} are in the new CLASS sample,
which, having no upper flux density limit, subsumes JVAS. The previous
samples CLASS-I and CLASS-II will be similarly subsumed in the same
sense as JVAS, though the differences here will be slightly larger since
bands other than L and C were used in the preliminary definition of
these samples.} The larger size of the CLASS survey, coupled with better
knowledge of the redshift distribution and number-magnitude relation of
the source sample, should reduce both the random and systematic errors
on our value of $\lambda_{0}$.
\begin{acknowledgements}
We thank our collaborators in the JVAS, CJF and CLASS surveys for useful
discussions and for providing data in advance of publication and many
colleagues at Jodrell Bank for helpful comments and suggestions. We
also thank John Meaburn and Anthony Holloway at the Department of
Astronomy in Manchester and the staff at Manchester Computing for
providing us with additional computational resources. RQ is grateful to
the CERES collaboration for making possible a visit to Jodrell Bank
where this collaboration was initiated. This research was supported in
part by the European Commission, TMR Programme, Research Network
Contract ERBFMRXCT96-0034 ``CERES''.
\end{acknowledgements}
\bibliographystyle{aa}
|
\section{Introduction and Summary}\label{sec:intro}
Explaining the thermodynamic nature of black holes was recognized
as an essential
hallmark of any complete quantum theory of gravity long before such a
theory was constructed. The semi--classical approach to quantum
gravity, which has become quite a mature subject over the
years\cite{gibbhawk,refined}, allows for the computation of a number
of physical quantities. These treatments ignore the details of how a
specific solution of Einstein's equations (regarded as the effective
low energy truncation of the complete quantum gravity) arises, and
instead perform a quantum treatment of field degrees of freedom
in a fixed classical space--time background.
In that way it was learned that the entropy of Bekenstein\cite{beken}
and the temperature of Hawking\cite{hawk}, for example, fit into an
elegant thermodynamic framework, with questions (such as scattering,
unitarity, {\it etc.},) concerning the underlying microscopic
description ---which we might use to construct the underlying
``Statistical Mechanics''--- best left for the future development of a
quantum theory of gravity.
That future is now here. String theory (and/or ``M--theory'') supplies
a microscopic description of the underlying degrees of freedom upon
which a statistical description of the laws of black hole
thermodynamics can be based. This is true even though we do not yet
have a satisfactory way of writing the theory in all regimes: the
``D--brane calculus''~\cite{dnotes} provides a robust
framework within which to describe many properties of black
holes\cite{amanda}, while in turn being firmly rooted in the dynamical
framework of string duality and, ultimately,
M--theory\cite{duality}. Typically, the description of black holes
(and other important geometrical backgrounds) proceeds by translating
their properties into properties of an auxiliary field theory,
identified as residing on the world--volume of some collection of (D--
or M--) branes.
One of the succinct ways of organizing this microscopic description of
the properties of black holes is via ``AdS
Holography''\cite{juan,gubkleb,edads,edadsii,hologram}. Then, the
thermodynamic properties of black holes in anti--de Sitter space--time
are dual to those of a field theory in one dimension
fewer\cite{edads,edadsii}. The fact that the thermodynamic properties
of the AdS black holes\cite{hawkpage} are organized by an effective
field theory is not implausible, in the light of the fact that AdS
acts like a natural ``box'' (with reflecting walls) which neutralizes
the tendency of gravitational interactions to render a canonical
thermodynamic ensemble unstable. The fact that the effective field
theory is one which does not contain gravity, and that it is actually
a ``holographically'' dual four dimensional gauge theory (with
suitable generalizations beyond $D{=}4$) is another striking example
of the fundamental role that gauge theory plays in duality in various
situations\footnote{See also refs.\cite{methroat,oferthroat,etudes}
for discussion of how this extends to relating the physics of linear
dilaton backgrounds to theories on the world--volumes of NS--branes.}.
That the AdS arena stabilizes the thermodynamics of black holes is
especially apparent when one discovers phase structures completely
analogous to familiar thermodynamic systems from elsewhere in
Nature. Such an example can be found in the
Reissner--Nordstrom--anti--de--Sitter (RNadS) systems in various
dimensions\cite{cejmii}. There, the $(Q,T)$ diagram showing the
thermally stable phases for a fixed charge (canonical) ensemble turns
out to be completely analogous to that of the $(T,P)$ phase diagram of
the liquid--gas system. The structure of the first order phase
transitions, {\it etc.}, are controlled by a ``cusp''
catastrophe\cite{catastrophe}, common in the theory of discontinuous
transitions in thermodynamics and many other fields\footnote{Recently,
the cusp catastrophe has appeared again the AdS/CFT context, in
ref.\cite{klemm}.}. Meanwhile the free energy as a function of
temperature $F(T)$, displays the characteristic ``swallowtail'' shape.
In this paper, we report the results of our further examination of
these structures, exploring the equilibrium thermodynamics more
closely, including the effects of considering electrical stability,
and thermal fluctuations. The similarities noted between the RNadS
physics and that of well--defined systems such as the liquid--gas are
more than mere analogies: {\it We find that everything has a very natural
place in classic equilibrium thermodynamics, as is consistent with a
holographic duality to thermal field theory without
gravity.} Accordingly, using the techniques of equilibrium thermodynamics,
we refine the phase diagrams which we found in ref.\cite{cejmii}
somewhat, and identify the generic physical properties which give rise
to the cusp and swallowtail structures.
As discussed in our previous paper~\cite{cejmii}, the thermodynamics of the
Reissner--Nordstr\"om black holes in the presence of negative
cosmological constant in various dimensions are pertinent (because of
the holographic map) to the thermodynamics of families of field
theories found on the common world--volume of collections of large
numbers of branes (for example M2-- and D3--branes), in the situation
where a global background current (or its canonical conjugate charge)
has been switched on and held fixed\footnote{See\cite{cveticetal} for
additional work on how to relate charged AdS black holes to
string/M-theory.}.
Geometrically this is performed by simply setting the M2-- and
D3--branes rotating equally in each of the available transverse
orthogonal two--planes. The higher dimensional angular momentum
becomes the Maxwell $U(1)$ charge after the Kaluza--Klein reduction on
the (now twisted) sphere, which yields the gauged supergravity.
Obtaining a pure Maxwell term in this way is not possible starting
with the M5--brane, and so the seven dimensional
Einstein--Maxwell--anti--de--Sitter (EMadS$_7$) theory defines at best
a close cousin to the field theory found on the M5--branes'
world--volumes. The dual theory relevant to EMadS$_6$ should be
considered in a similar manner.
A truly rich phase structure for the field theories (with transition
temperatures away from $T{=}0$) is obtained only for finite volume,
which is the case we concentrate on here. Our studies correspond to the
study of black holes with spherical horizons, $S^{n-1}$. The field
theory resides on ${\hbox{{\rm I}\kern-.2em\hbox{\rm R}}}{\times}S^{n-1}$. The case of infinite volume
corresponds to black holes with horizons ${\hbox{{\rm I}\kern-.2em\hbox{\rm R}}}^{n-1}$, and to field
theory on ${\hbox{{\rm I}\kern-.2em\hbox{\rm R}}}^{n}$. This is of course the case which comes from
taking directly the near horizon limit of explicit brane solutions.
As shown in ref.\cite{cejmii}, the results for infinite volume may be
easily obtained as a scaling limit of the results of finite volume,
and so we will not discuss them here. Of course, even though we are in
finite volume for much of our discussion, the thermodynamic limit is
still valid here, because the dual field theory is at large $N$, and a
positive power of $N$ measures the number of degrees of freedom in the
field theory (for example, $N^2$ in the case of gauge theory, for
$n{=}4$ here).
The structure of the paper is as follows: In
section~\ref{sec:theholes} we recall the charged black solutions of
the Einstein--Maxwell--anti--de--Sitter system. We also recall the
results of performing the Euclidean section and ensuring its
regularity. In section~\ref{sec:state}, we translate these results
into a statement of about the relation between the thermodynamic
variables of the black hole system in thermodynamic equilibrium {\it
i.e.,} the ``equation of state''. In section~\ref{sec:ensemble}, we
define the grand canonical and canonical thermodynamic ensembles and
compute the associated Gibbs and Helmholtz thermodynamic potentials,
contrasting the techniques used (and results obtained) to those of our
previous work. In particular, we note that we can obtain an intrinsic
definition of these quantities in Euclidean quantum gravity, by
sidestepping some of the technical subtleties ---encountered in the
``background subtraction'' technique for regularizing the action--- in
favour of the ``counterterm subtraction'' technique\cite{pervijay,ejm}. In
the rest of the section, we examine the features of these potentials
quite closely, in preparation for later detailed studies. In
section~\ref{sec:swallowtail}, we use the equation of state and the
first law of thermodynamics to identify the origins of the crucial
features of the shape of the Helmholtz potential (free energy). This
``swallowtail'' shape is responsible for the interesting phase
structure in the canonical ensemble. Section~\ref{sec:stability}
examines the conditions for thermodynamic stability of the black
holes, examining the specific heats and permittivity of the black
holes. In this way, we identify the stable regions of the solution
space of the equation of state. We use this stability information,
together with the information gained in earlier sections, to deduce
the refined phase diagrams exhibited in section~\ref{sec:phases}, and
some details of the phase diagrams (the slope and convexity of the
coexistence curves) are refined by using the Clayperon
equation in section~\ref{sec:coexist}.
As already stressed in this section, the thermodynamic quantities and
studies performed in those sections are rooted firmly in a microscopic
description. This is ensured by the fact that we can in principle
embed this entire discussion into a complete theory of quantum
gravity: string (and/or) M--theory. In practical terms, this
microscopic description ---the ``statistical mechanics'' underlying
the thermodynamics--- is summarized neatly in terms of the
holographically dual field theory. In this way, therefore we may carry
our calculations further and examine the nature and magnitude of the
microscopic fluctuations of the various thermodynamic quantities we
have computed, knowing that we have a description of their origin in
field theory. Thus, we find in section~\ref{sec:fluct} that the
fluctuations behave in a way consistent with the underlying
microscopic physics being supplied by the field theory: the size of
the (squared) fluctuations is controlled by a prefactor which
corresponds to precisely the inverse of the number of degrees of
freedom of the dual field theory. We observe that the size of the
fluctuations diverges as the system approaches a critical point in the
$(Q,T)$ plane.
Through most of the paper, we carried out our computations for the
four dimensional case, in order to keep many of our formulae
simple. Section~\ref{sec:higher} collects together some of the results
for the computation of various quantities. We stress that the
qualitative structure of the physics is the same for all dimensions
$d{\geq}4$, where $d{=}n{+}1$. Briefly, we also discuss in that
same section the issue of the meaning of the formal definition of
other thermodynamic ensembles by Legendre transform. It is not always
the case that the thermodynamic quantities thus defined may be arrived
at by (known) computations in Euclidean quantum gravity. Therefore,
interpretations of the physics of such ensembles are to be taken with
(at least) a pinch of salt, until such time as new technology becomes
available to compute the relevant quantities directly in quantum
gravity, as we have done here for the fixed potential (grand
canonical) and fixed charge (canonical) ensembles.
Section~\ref{sec:critical} discusses the underlying structure of the
phase structure of the canonical ensemble in the neighbourhood of the
critical point. In particular, the physics local to critical point is
universal for all of dimensions $d{\geq}4$. The critical point is a
second order phase transition point at the end of a coexistence line
of first order phase transitions. As such, it has a universal
description in terms of a Landau--Ginzburg model, with a quartic
potential ---$A_3$ in the $A$--$D$--$E$ classification of such
potentials. The deformation of this potential gives the classic
``cusp'' catastrophe which underlies the critical behaviour, as is
well known from the van--der Waals--Maxwell description of the
liquid--gas system, with which our black hole physics shares many
features, as originally reported in ref.~\cite{cejmii}.
In closing the introduction,
we would like to stress once again how elegantly the
properties of anti--de Sitter space yield charged black hole physics
so closely akin in structure to that of ordinary field theory--like
systems, with which we have more intuition.
{}From the point of view of the Maxwell part of the action, the black
holes are nothing more than spherical capacitors, and as such, the
amount of energy they can store grows with the charge on them, but
falls with increasing hole radius. From the point of view of the
Einstein--Hilbert action however, the black holes store an amount of
energy which grows with radius. After a little thought, one might
expect on general grounds, therefore, that there might be an
interesting phase structure resulting from a competition between these
two pieces of the action.
Such reasoning on its own would not be enough to genuinely fill out
the whole $(Q,T)$ phase diagram, as the equation of state needs
additional structure. It is the presence of (negative) cosmological
constant which provides this final part: First, it provides black hole
solutions which are thermally stable in ensembles involving fixed
temperature\cite{hawkpage}, but secondly, as it defines a new length
scale, is allows the system to distinguish, on the one hand, black
holes which are large from those which are small, and on the other
hand, black holes which have small charge, from those with large
charge.
It is because of these features that the charged black hole
thermodynamics has a chance to be similar to the van--der Waals model
of the liquid--gas. Recall that without the inclusion of the effects
of the length scales set by finite particle size, on the one hand, and
attractive inter--particle forces on the other, that system would have
only the much less interesting physics of the ideal gas: there would
be no competing effects, as a function of length scale, with which to
trigger a phase transition. These basic features of AdS give
holography a chance to work in a way which is consistent with our
intuition that the microscopic physics should be modelled by ordinary
field theory.
\section{\bf Charged AdS Black Holes}\label{sec:theholes}
For space--time dimension $n{+}1$, the
Einstein--Maxwell--anti--deSitter (EMadS$_{n+1}$) action may be
written as\footnote{We scale the gauge field $A_\mu$ so as to absorb
the prefactors involving the $U(1)$ gauge coupling into the action.}
\begin{equation}
I = -\frac{1}{16{\pi}G} \int_{M} d^{n+1}x \sqrt{-g} \left[R - F^2 +
\frac{n(n-1)}{l^2}\right]\ ,
\label{actionjackson}
\end{equation}
with ${\Lambda}{=}{-}\frac{n(n-1)}{2l^2}$ being the cosmological
constant associated with the characteristic length scale $l$. Then
the metric on the Reissner--Nordstr\"om--anti--deSitter (RNadS)
solution may be written in static coordinates
as\cite{romans,lee,cejmii}
\begin{equation}
ds^2 = -V(r)dt^2 + \frac{dr^2}{V(r)} +r^{2}d{\Omega}^2_{n-1}\ ,
\end{equation}
\noindent where $d{\Omega}^2_{n-1}$ is the metric on the round
unit $(n{-}1)$--sphere, and the function $V(r)$ takes the form
\begin{equation}
V(r) = 1 - \frac{m}{r^{n-2}} + \frac{q^2}{r^{2n-4}} + \frac{r^2}{l^2}\ .
\end{equation}
\noindent Here, $m$ is related to the ADM mass of the hole, $M$
(appropriately generalized to geometries asymptotic to
AdS\cite{adhh}), as
\begin{equation}
M={(n-1)\omega_{n-1}\over 16\pi G}m\ ,
\end{equation}
where $\omega_{n-1}$ is the volume of the unit $(n{-}1)$--sphere.
The parameter
$q$ yields the charge
\begin{equation}
Q=\sqrt{2(n-1)(n-2)}\left({\omega_{n-1}\over 8\pi G}\right)q\ ,
\label{thecharge}
\end{equation}
of the
(pure electric) gauge potential, which is
\begin{equation}
A=\left(-{1\over c}{q\over r^{n-2}}+\Phi\right)dt\ ,
\label{pure}
\end{equation}
where
\begin{equation}
c=\sqrt{2(n-2)\over n-1}\ ,
\end{equation}
and $\Phi$ is a constant (to be fixed below). If $r_+$ is the largest
real positive root of $V(r)$, then in order for this RNadS metric to
describe a charged black hole with a non--singular horizon at $r{=}r_+$, the
latter must satisfy
\begin{equation}\label{extbound}
\left({n\over n-2}\right) r_+^{2n-2}+l^2 r_+^{2n-4} \geq q^2 l^2\ .
\end{equation}
Finally, we choose
\begin{equation}
\Phi={1\over c}{q\over r_+^{n-2}}\ ,
\end{equation}
which then fixes $A_t(r_+){=}0$, as is required by (Euclidean)
regularity of the one--form potential (\ref{pure})
at the fixed point set of the Killing vector
$\partial_t$. The physical significance of the quantity $\Phi$, which
plays an important role later, is that it is the electrostatic
potential difference between the horizon and infinity.
If the inequality in eqn.~(\ref{extbound}) is saturated, the horizon is
degenerate and we get an extremal black hole. This inequality imposes
a bound on the black hole mass parameter of the form $m{\geq}m_e(q,l)$.
In passing to the thermodynamic discussion, we define the Euclidean
section ($t{\to}i\tau$) of the solution, and identify the period,
$\beta$, of the imaginary time with the inverse temperature. Using the
usual formula for the period, $\beta{=}{4\pi}/{V^{\prime}(r_{+})}$,
which arises from the requirement of regularity of the solution, we
obtain:
\begin{equation}
\label{betaform}
\beta = \frac{4{\pi}l^{2}r_{+}^{2n-3}}{nr_{+}^{2n-2} +
(n-2)l^{2}r_{+}^{2n-4} - (n-2)q^{2}l^{2}}\ .
\end{equation}
\noindent This may be rewritten in terms of the potential as:
\begin{equation}
\beta = \frac{4{\pi}l^{2}r_+}{(n-2)l^{2}(1-c^2\Phi^2)+nr_+^2}\ .
\label{betaformtwo}
\end{equation}
For simplicity, we will specialize to $n{=}3$ (therefore working with
EMadS$_4$) to avoid cluttering our expressions with complicated
dependences on $n$. Our results will remain qualitatively the same
for higher $n$ (see the comments in section~\ref{sec:critical}),
and we list some of the $n$--dependent formulae in
section~\ref{sec:higher}. Our analysis is further simplified by
adopting the following rescalings (once
we have set $n{=}3$):
\begin{equation} T \rightarrow {2\pi l\over \sqrt{3}}T\ ,
\qquad Q \rightarrow {\sqrt{3}\over l} GQ\ ,
\qquad \Phi \rightarrow \Phi\end{equation}
and for the various thermodynamic quantities used in ref.\cite{cejmii},
\begin{equation}\{W, F, E\} \rightarrow
{\sqrt{3}\over l} G \{W, F, E\}\ ,\qquad
S\rightarrow {3G\over2\pi l^2}S\ .
\end{equation}
\begin{equation} {\rm and}\qquad r_+\rightarrow{\sqrt{3}\over l}r_+\ .
\end{equation}
Essentially we are introducing a system of dimensionless quantities
in which everything is measured in units of the AdS scale $l$.
This scaling is chosen so that the thermodynamic formulae still
all have their standard form, {\it i.e.,} \begin{equation}
dE = TdS+\Phi dQ\ .
\qquad
dF=-SdT+\Phi dQ \qquad dW=-SdT-Qd\Phi\ ,\ \ {\it etc.}
\label{firstlaw}
\end{equation}
In the following, all of the quantities which follow are the rescaled
dimensionless quantities, unless stated otherwise.
\section{Equation of State}\label{sec:state}
The Euclidean regularity at the horizon discussed at
eqn.~(\ref{betaform}) is equivalent to the condition that the black
hole is in thermodynamical equilibrium. The resulting
equation~(\ref{betaform}) may therefore be written as an equation of
state $T{=}T(\Phi,Q)$ (analogous to $T{=}T(P,V)$ for, say, a gas at
pressure $P$ and volume $V$). For $n{=}3$, one finds:
\begin{equation}
T={\Phi^2 (1-\Phi^2) +Q^2 \over 2Q\Phi}\ .
\end{equation}
One can also solve for $Q$ as
\begin{equation}
Q= T\Phi \pm \Phi \sqrt{T^2 +\Phi^2-1}\ .
\label{eqn:state}
\end{equation}
{}From this equation of state we see that for fixed $\Phi$ we get two
branches, one for each sign, when the discriminant under the square root is
positive. For fixed $Q$, $T(\Phi)$ has three branches for $Q{<}Q_{\rm
crit}$ and one for $Q{>}Q_{\rm crit}$, where the critical charge is determined
solving for the ``point of inflection'' where $\left(\partial
Q/\partial\Phi\right)_T{=}\left(\partial^2 Q/\partial\Phi^2\right)_T{=}0.$
In the dimensionless units used here,
one finds $Q_{\rm crit} {=}1/(2\sqrt{3})$, $T_{\rm
crit}{=}2\sqrt{2}/3$, $\Phi_{\rm crit}{=}1/\sqrt{6}$, $E_{\rm
crit}{=}\sqrt{2}/3$, and $r_{\rm +(crit)}{=}1/\sqrt{2}$. It is useful
to plot the $(\Phi,Q)$ isotherms, {\it i.e.,} plot $Q(\Phi)$ for fixed
$T$, and we exhibit these in figure~\ref{fig:state}.
\begin{figure}[hb]
\hskip1cm
\psfig{figure=isotherm.eps,height=3.0in}
\hskip2cm
\psfig{figure=isotherm3D.eps,height=3.2in}
\caption{Plots of the equation of state of $\Phi$ {\it vs.}~$Q$,
showing isotherms above and below the critical temperature $T_{\rm
crit}$. For~$T{<}T_{\rm crit}$, there is only one branch of solutions,
while for $T{>}T_{\rm crit}$, there are three branches. The values of
$T$ for the isotherms plotted are (top down) $T{=}0,0.8,T_{\rm
crit},1.0,1.2$. The central (dotted) curve is at the critical
temperature.}
\label{fig:state}
\end{figure}
As $T$ goes to zero, we approach the extremal black holes. Their
equation of state is
\begin{equation}
\Phi_e^2 ={1\over 2} (1+\sqrt{1+4 Q^2}),\qquad {\rm for\; arbitrary}\, T.
\end{equation}
For some later computations, it is often convenient to use as an
additional, non--independent parameter, the black hole radius $r_+$,
in terms of which
\begin{equation}
Q=r_+\sqrt{r_+^2-2r_+T+1},\quad 2T=r_+ +{1\over r_+}-{Q^2\over r_+^3}\ ,
\label{param1}
\end{equation}
and
\begin{equation}
\Phi={Q\over r_+}=\sqrt{r_+^2-2r_+T+1}\ .
\label{param2}
\end{equation}
\section{The grand canonical and canonical ensembles}\label{sec:ensemble}
In thermodynamic parlance, the ``grand canonical ensemble'' is defined
by coupling the system to energy and charge reservoirs at fixed
temperature $T$ and potential $\Phi$ (an intensive variable). The
associated thermodynamic potential is the Gibbs free energy,
$W[T,\Phi]{=}E{-}TS{-}\Phi Q$. Holding the extensive variable, $Q$,
fixed, on the other hand, defines the canonical ensemble, with its
associated thermodynamic potential is the Helmholtz free energy
$F[T,Q]{=}E{-}TS$. See section~\ref{sec:higher} for a brief
discussion of other ensembles.
In ref.~\cite{cejmii}, the calculation at fixed potential was carried
out by computing the action {\it \'{a} la} Gibbons--Hawking. With that
technique, one must regularize the computation (as the action is
formally infinite) by subtracting a contribution from a ``reference''
background which matches the solution of interest asymptotically,
giving a definition of the action relative to that of the reference
spacetime. In this case it is appropriate to use AdS ---with a fixed
(pure gauge) potential at infinity--- as the reference background.
Remarkably AdS spacetime provides for another regularization which
yields an {\it intrinsic} definition of the action. In other words,
the computation makes no reference to any other solution of the
equations of motion. Instead, the method\cite{pervijay,ejm} proceeds by
adding a series of boundary counterterms to the action.
We refer to this as the
``counterterm subtraction'' method of defining the action, a technique
tailored to spacetimes which are locally asymptotic to anti--de
Sitter, as the counterterms are defined on the natural boundary, with
which such spaces are endowed, using the AdS scale $l$.
Also note that the inclusion of additional sectors
to the gravitational and cosmological parts of the action, such as Maxwell
terms, does not affect the definitions and therefore we can still use the
same counterterms in the present context.
The results, using either the reference background or the counterterm
subtraction methods, are identical for the particular case in which we
want to fix the potential\footnote{For even values of $n$ there
appears a Casimir energy term \cite{pervijay}, which is immaterial for
the discussion of thermodynamics here.}, since it is possible to have
AdS space as a background solution at arbitrary temperature and
(constant) potential (but, crucially, see later). In the present
notation, the answer is:
\begin{equation}
W[\Phi,T]={1\over 12} \left[ 3 {Q\over
\Phi}(1-\Phi^2)-\left({Q\over\Phi}\right)^3\right]\ .
\label{eqn:gibbs}
\end{equation}
Here, $Q$ is given as $Q(\Phi,T)$ by the equation of
state~(\ref{eqn:state}). In terms of $r_+$, this is
\begin{equation}
W={1\over 12} \left[ 3 r_+(1-\Phi^2)-r_+^3\right]\ ,
\end{equation}
\noindent
and it is plotted in figure~\ref{fig:gibbs3d}, with choice slices
displayed in figure~\ref{fig:gibbssnaps}.
\begin{figure}[hb]
\hskip5cm
\psfig{figure=Gibbs3D.eps,height=3.2in}
\caption{Plots of the Gibbs potential $W[\Phi,T]$ in three dimensions.}
\label{fig:gibbs3d}
\end{figure}
\begin{figure}[hb]
\hskip-0.5cm
\psfig{figure=Gibbs2D1.eps,height=2.8in}
\psfig{figure=Gibbs2D2.eps,height=2.8in}
\psfig{figure=Gibbs2D3.eps,height=2.8in}
\caption{Slices of the Gibbs potential $W[\Phi,T]$, for $\Phi{=}0$, $\Phi{=}0.7$
and
$\Phi{=}1$.}
\label{fig:gibbssnaps}
\end{figure}
Turning to the canonical (fixed charge) ensemble, we wish to compute
the Helmholtz potential $F[Q,T]$ (a.k.a, the ``free energy''). In
ref.\cite{cejmii}, where we used the reference background method to
compute this, it was necessary to compute the action using an
extremal black hole as the reference background. This is because
anti--de Sitter space with a fixed charge $Q$, as measured at infinity
is {\it not} a solution of the equations of motion and so is not an
appropriate background. In order to get an intrinsic definition of
the action for fixed charge therefore, we employ the method of
counterterm subtraction, yielding:
\begin{equation}
F[Q,T]={1\over 12} \left[ 3 {Q\over
\Phi}-\left({Q\over\Phi}\right)^3+9Q\Phi\right]\ ,
\end{equation}
where $\Phi$ is given as $\Phi(Q,T)$ by the equation of
state~(\ref{eqn:state}). In terms of $r_+$, $F[Q,T]$ may be written
\begin{equation}
F[Q,T]={1\over 12} \left[ 3 r_+ -r_+^3 + 9{Q^2\over r_+}\right]\ .
\label{param3}
\end{equation}
As a consistency check that we have performed the computation
correctly, note that this result may be obtained from the result for
the Gibbs potential by formally calculating the Legendre transform
$F[Q,T]{=}W[\Phi,T]{+}Q\Phi$. When computing $F$ from a Euclidean
action, the additional $Q\Phi$ term has its origin in the boundary term
introduced so as to recover the correct variational problem from the
action. It is especially satisfying to see that the counterterm
subtraction method places such intuitive relationships from
equilibrium thermodynamics on a firm footing. We shall have more to
say about this in section~\ref{sec:higher}.
In ref.\cite{cejmii}, where we computed the action using an extremal
reference background, we obtained the following expression for the free
energy (which we denote here as $\bar F$):
\begin{equation}
{\bar F}[Q,T]={1\over 12} \left[ 3 {Q\over \Phi}-\left({Q\over\Phi}\right)^3 +
9Q\Phi-4{Q\over \Phi_e}
-8Q\Phi_e\right]\ .
\end{equation}
Note that in this case, one should consider ${\bar
\Phi}{=}\Phi{-}\Phi_e$ as the state variable, instead of $\Phi$.
Then, the first law is in this case $dE{=}TdS{+}{\bar\Phi}dQ$, and $E$
measures the energy above the extremal state. Furthermore, it is with
$\bar\Phi$ that $W[T,\Phi]$ of eqn.~(\ref{eqn:gibbs}) and ${\bar F}[T,Q]$ are
Legendre
transforms of each other, as they should be. While $\bar F[Q,T]$ as
computed in ref.\cite{cejmii} using the extremal background is in no
way problematic, we shall not examine it further here, as the new
technology of the counterterm subtraction method has supplied us with
an intrinsic definition of the Helmholtz potential, which is the more
natural Legendre--transform partner of the Gibbs potential (\ref{eqn:gibbs})
found earlier.
We shall see that the qualitative features of the results obtained in
ref.\cite{cejmii} for the canonical ensemble using $\bar F[Q,T]$ will
persist here, as the extremal background subtraction essentially
redefines the absolute normalization of some results. (The later
analysis of intrinsic stability which we do in section~\ref{sec:stability} would
have
to be somewhat modified before direct comparison to the extremal
subtraction results, however, as we will make heavy use of the
equation of state in terms of variables $(\Phi,Q,T)$, and not the
triple $({\bar\Phi},Q,T)$ appropriate to that case.)
We now return to the analysis of the intrinsically defined Helmholtz
potential $F[Q,T]$. It was noticed in ref.~\cite{cejmii} that a plot
of ${\bar F}(T)$ for various values of $Q$ reveals (below a $Q_{\rm
crit}$) a section of a``swallowtail'' shape, which controls much of
the phase structure (in the canonical ensemble)
discussed there, and to be discussed later
here. (See figs.~5 and~6 of ref.~\cite{cejmii} and associated text for
details.) The same may be observed here for $F(T)$ for varying $Q$, as
shown in figure~\ref{fig:Tfreesnaps}.
\begin{figure}[ht]
\hskip-0.5cm
\psfig{figure=Tfree1.eps,height=2.8in}
\psfig{figure=Tfree2.eps,height=2.8in}
\psfig{figure=Tfree3.eps,height=2.8in}
\caption{The free energy {\it vs.}~temperature for the fixed charge
ensemble, in a series of snapshots for varying charge, for values
$Q{=}0,0.15$ and $Q{=}0.299$. Note that $Q_{\rm crit}{=}0.289$, so in
the last plot, the bend (near $T_{\rm crit}{=}0.943$, is in the
neighbourhood of the critical point of second order.}
\label{fig:Tfreesnaps}
\end{figure}
It may be further observed that a plot of $F(Q)$ for fixed $T$
reveals (above a $T_{\rm crit}$), a similar swallowtail section, as
shown in figure~\ref{fig:Qfreesnaps}.
\begin{figure}[ht]
\hskip-0.5cm
\psfig{figure=Qfree1.eps,height=2.8in}
\psfig{figure=Qfree2.eps,height=2.8in}
\psfig{figure=Qfree4.eps,height=2.8in}
\psfig{figure=Qfree5.eps,height=2.8in}
\psfig{figure=Qfree3.eps,height=2.8in}
\psfig{figure=Qfree6.eps,height=2.8in}
\caption{The free energy {\it vs.}~charge for the fixed charge
ensemble, in a series of snapshots for varying temperature, for values
$T{=}0.943,0.997,1.00,1.10$, and (for the ``Zorro'' plot) $T{=}T_{\rm
HP}{=}1.154$, and finally $T{=}1.20$. Note that $T_{\rm
crit}{=}0.943$, and so in the first plot, the bend (near $Q_{\rm
crit}{=}0.289$), is in the neighbourhood of the critical point of
second order.}
\label{fig:Qfreesnaps}
\end{figure}
The full three dimensional shape of $F[Q,T]$ is plotted in
figure~\ref{fig:free3d}.
\begin{figure}[hb]
\hskip5cm
\psfig{figure=Free3D.eps,height=3.5in}
\caption{Plots of the Helmholtz potential $F[Q,T]$, in three
dimensions, clearly showing the swallowtail shape for $T{>}T_{\rm crit}$
and $Q{<}Q_{\rm crit}$.}
\label{fig:free3d}
\end{figure}
That such a shape appears in the thermodynamics (above $T_{\rm crit}$
or below a $Q_{\rm crit}$) can be shown to follow from the first law
of thermodynamics, the definition of the thermodynamic potentials, and
the form of the particular equations of state which the black holes
obey. We will show how this comes about next.
\section{The swallow tales}\label{sec:swallowtail}
The sections of swallowtails in the $F(Q)$ and $F(T)$ plots (above a
$T_{\rm crit}$ or below a $Q_{\rm crit}$) can be seen to come from the
existence of the previously mentioned
three branches of solutions to the equation of state. We have from the first
law,
and the definition of the thermodynamic
potential, that $dF{=}{-}SdT{+}\Phi dQ$. Therefore, for fixed
$T$ we find
\begin{equation}
F(T)=\int\Phi(Q)\,dQ + f(T)\ ,
\label{fixedt}
\end{equation}
where $f(T)$ is an arbitrary function of $T$. The integral function
can be obtained by looking at the plot of isotherms. When we have
three branches ({\it i.e.,} $T{>}T_{\rm crit}$), the curve $\Phi(Q)$
winds back and forth in a way that the integral describes a shape with
three connected branches, constituting a section of the
``swallowtail'' shape. This can be seen by examination of the plots of
the equation of state in figure~\ref{fig:state} and the plots of the
slices $F(T)$ displayed in figure~\ref{fig:Tfreesnaps}\footnote{As
visual differentiation is often easier to perform than integration, we
gently remind the reader that the defining relation
$\Phi{\equiv}(\partial F/\partial Q)_T$ may be of use here, in
conjunction with the snapshots of $F(Q)$ for fixed $T$ given in
figure~\ref{fig:Qfreesnaps}.}.
Equation~(\ref{fixedt}) is usually employed to formulate an ``equal
area law" governing the phase transitions of the system. The latter
occur at the point where the free energies of two branches, (say A and
B), are equal: $F_{\rm A}{=}F_{\rm B}$. From eqn.~(\ref{fixedt}) this
equality may be translated
into a statement about the equality of the areas enclosed by the
isotherm curves and a line of constant $Q$ in the~$(\Phi,Q)$ plane, as
shown in the sketch on the left in figure~\ref{fig:arealaw}. There is a
subtlety, though, in using eqn.~(\ref{fixedt}) with the isotherm curves
of eqn.~(\ref{eqn:state}) for $T{>}1$ (recall that isotherms with
$T{\geq}1$ go through the origin $\Phi{=}Q{=}0$. See
figure~\ref{fig:state}.). Given that the transition is governed by the
equal area law, it would seem from the curves on the right in
figure~\ref{fig:arealaw} and the area law deduced from
eqn.~(\ref{fixedt}) that even for $T{>}1$, for which a minimum value of
$\Phi$ ceases to exist, one can always find a phase transition point
for arbitrarily large temperatures and small enough charge. This must
be wrong since it contradicts what we know about the phase transition
from the curves of $F$ for constant $Q$, namely, that the phase
transition takes place at a temperature that is smaller (or equal, at
$Q{=}0$) than the Hawking--Page temperature $T_{\rm HP}$. (See
ref.\cite{cejmii} and the upcoming section~\ref{sec:phases} for
detailed discussion of the phase structure.)
\begin{figure}[ht]
\hskip3.0cm
\psfig{figure=arealaw.eps,height=2.5in}
\caption{The figure on the left shows how the condition for a phase
transition may be interpreted in terms of an ``equal area law''
analogous to that due to Maxwell for the van--der Waals liquid--gas
model. For $T{\geq}1$, though, the isotherms have a very different
qualitative structure. The equal area law one might formulate,
deducing a phase transition for arbitrarily high $T{>}1$, for small
enough $Q$, actually is incorrect. See text for the resolution of the
puzzle.}
\label{fig:arealaw}
\end{figure}
The resolution of this puzzle is instructive, and is made manifest
most clearly by working in terms of the parameter~$r_+$, using
eqns.~(\ref{param1}),~(\ref{param2}) and~(\ref{param3}). We can
explicitly compute $F$ using eqn.~(\ref{fixedt}) as
\begin{equation}
F=\int_0^{r_+} \Phi(\bar r_+) (dQ/d\bar r_+) d\bar r_+
\end{equation}
to recover precisely eqn.~(\ref{param3}). So, what is the reason that
``naive integration'' using the ``the equal area law'' yields a
different result?
The point is that for $T{>}1$ the function $F(Q)$ is discontinuous at
$Q{=}0$, where branches 2 and 3 separate (see for example, the last
plot of figure~\ref{fig:Qfreesnaps}). For those isotherms, there is a
range of values for $r_+$,
$T{-}\sqrt{T^2{-}1}{<}r_+{<}T+\sqrt{T^2{-}1}$ for which $Q$ and $\Phi$
become imaginary. Nonetheless, the product $\Phi dQ$ is real
throughout, and so is $F$. Then, $F(Q)$ would be a continuous function
if we plotted it in the complex $Q$ plane. In performing the
integration above for $T{>}1$ we have implicitly included the points
where $\Phi$ and $Q$ are imaginary. Notice that it is by including
these points that we recover sensible physics, since we want the
critical line to end at $Q{=}0$ at the point of the Hawking--Page
phase transition. The ``equal area law'', as it is, fails in this
instance.
Let us now turn to the study of the free energy for fixed $Q$. We have
\begin{equation}
F(Q)=-\int S(T) dT + g(Q)
\end{equation}
In this case we need $S(T)$. Since
\begin{equation}
S={1\over2} \left({Q\over \Phi}\right)^2= {r_+^2\over2}
\end{equation}
we can use the equation of state to plot $S(T)$ for fixed $Q$, which
is shown in figure~\ref{fig:stateS}.
\begin{figure}[hb]
\vskip-1.5cm
\hskip3cm
\psfig{figure=isocharge.eps,height=3.5in,width=3.5in}
\caption{Plots of the equation of state of $S$ {\it vs.}~$T$, showing
isocharge lines above and below the critical charge $Q_{\rm
crit}$. For $Q{>}Q_{\rm crit}$, there is only one branch of solutions,
while for $Q{<}Q_{\rm crit}$, there are three branches. The values of
$Q$ for the isocharge curves plotted are (top down) $Q{=}0,0.20,Q_{\rm
crit},0.45,0.80$. (The uppermost curve shows the $Q{=}0$ case, which
has two branches. The central (dotted) curve is at the critical
charge.}
\label{fig:stateS}
\end{figure}
It can be readily seen that for $Q{<}Q_{crit}$ we get three branches
(notice that the qualitative features of the plot of~$S(T)$ follow
from those of $r_+(T)$ or $r_+(\beta)$ plotted in figure~3 of
ref.\cite{cejmii}, where it has a resemblance to the van der Waals
$P(V)$ curve). A section of the swallowtail again
follows\footnote{Again, one can use figure~\ref{fig:stateS} to
reconstruct $F(Q)$ visually using the integral relation, or one may
use the definition of the entropy $S{\equiv}(\partial F/\partial T)_Q$
to reconstruct figure~\ref{fig:stateS} from
figure~\ref{fig:Tfreesnaps}.}.
The astute reader may wonder why the swallowtail shape (and the
resulting liquid--gas--like) phase diagram occurs in the canonical
ensemble, where in addition to $T$, the extensive variable $Q$ is an
external control parameter, and not in the grand canonical ensemble,
where the intensive variable $\Phi$ would be the control. This is of
course what happens in the case of the van--der--Waals--Maxwell
system, where the phase diagram is in $(P,T)$ space, and not $(V,T)$
space\cite{van,stanley}. The swallowtail shapes occur there in the
Gibbs potential. It is now hopefully clear that the answer follows
from the fact that our equation of state yields three branches of
solutions for the intensive variable $\Phi$ (or T) as a function of
the fixed extensive variable $Q$ (or $S$), as can be seen by examining
the curves displayed in figs.~\ref{fig:state} and~\ref{fig:stateS}.
That there are no swallowtail shapes in any of the other
ensembles follows from the fact that no more than two branches occur
for the equation of state written in terms of other variables.
\section{intrinsic stability}\label{sec:stability}
Given that we have the full power of the thermodynamic framework at
our disposal, (thanks to the stabilizing influence of a negative
cosmological constant), it is interesting to consider the
thermodynamic stability of our various solutions against microscopic
fluctuations\footnote{See refs.\cite{jorma,SandM} for analyses which
overlap with those presented here, in a similar context.}. Notice that
one can always formally compute the relevant macroscopic quantities
(like specific heats, {\it etc.,}) which we discuss here, without any
reference to an underlying microscopic description. This has been done
in the context of black hole thermodynamics since time
immemorial. {\it The difference here is that we know the nature of the
microscopic degrees of freedom which supply the underlying
``Statistical Mechanics'' which gives rise to these macroscopic
thermodynamics quantities.} The underlying physics is that of the
gauge theory to which this system is holographically dual, which in
turn is the physics of coincident branes. This will become more
apparent in section~\ref{sec:fluct} when we explicitly study the
fluctuations themselves.
Thermodynamic stability may be phrased in many different
ways\cite{callen,landau}, depending on which thermodynamic function we
choose to use, and how obscure we are attempting to seem. For
example, it can be seen as minimization of the energy, $E$, as a
function of $(S,Q)$, or maximization of the entropy $S$, as a function
of $(E,Q)$, {\it etc}. In any case, one is considering an
infinitesimal variation of the state function away from
equilibrium. The first law (\ref{firstlaw}) will ensure that the first
order terms vanish. Stability is then a statement about the second
order variations. Generally then the stability conditions are phrased
in terms of the restriction that the Hessian of the state function is
positive (or negative, depending on the context) semi--definite.
An equivalent but physically more transparent way of writing the
stability conditions is in terms of specific heats and other
``compressibilities'', to wit:
\begin{equation}
C_Q\equiv T\left({\partial S\over\partial T}\right)_Q \geq 0,\quad
C_\Phi\equiv T\left( {\partial S\over\partial T}\right)_\Phi \geq
0,\quad \varepsilon_T\equiv\left({\partial Q\over \partial
\Phi}\right)_T\geq 0\ .
\label{stabcrit}
\end{equation}
The first two, the specific heats at constant electric charge and
potential, are familiar analogues of the specific heats at constant
volume and pressure in fluid systems. In the case in hand, they
determine the thermal stability of the black holes, indicating whether
a thermal fluctuation results in an increase or decrease in the size
of the black hole. (This follows from the fact that the entropy is
proportional to the size of the black hole,.) Stability follows from
$C{\geq}0$, given the fact that black holes radiate at higher
temperatures when they are smaller.
The last quantity, $\varepsilon_T$, has the following physical
interpretation. It is negative if the black hole is electrically unstable
to electrical fluctuations (if they are possible, see later
discussion). This happens if the potential of the black hole {\it
decreases} as a result of placing more charge on it. The potential
should of course {\it increase}, in an attempt to make it harder to
move the system from equilibrium\footnote{This follows from common
sense, or more formally, Le Chatelier's principle.}. $\varepsilon_T$
therefore deserves to be called the ``isothermal (relative)
permittivity'' of the black hole.
There are of course other interesting ``response functions'' for the
system, such as the adiabatic permittivity, $(\partial
Q/\partial\Phi)_S$ or the quantity analogous to the coefficient of
thermal expansion in liquid--gas systems, $\alpha_\Phi{=}(\partial
Q/\partial T)_\Phi$, which are not all independent. The ones which we
have discussed above will suffice for the physics that we study in
this paper.
We may examine the plots of the isotherms in figure \ref{fig:state}
and deduce that the negatively sloped branches are electrically
unstable if there are electrical fluctuations possible. Similarly, we
may deduce that the negatively sloped branches of the $(S,T)$
isocharge curves in figure~\ref{fig:stateS}, are thermally unstable,
and so on.
Stability follows, equivalently, from the concavity/convexity of the
plots of $F$ and $W$ as functions of $T$. In fact, the specific heat
conditions are equivalent to
\begin{equation}
\left({\partial^2 F\over \partial T^2}\right)_Q\leq 0\quad {\rm
and}\quad\left({\partial^2 W\over \partial T^2}\right)_\Phi\leq 0\ ,
\label{condition1}
\end{equation}
whereas the permittivity condition is
\begin{equation}
\left({\partial^2 F\over \partial Q^2}\right)_T\geq 0, \quad {\rm or}\quad
\left({\partial^2 W\over \partial \Phi^2}\right)_T\leq 0\ .
\label{condition2}
\end{equation}
By examining the isotherms displayed in figure~\ref{fig:state}, we see
that there are a number of features in the $(Q,T)$ plane which govern
electrical stability. Generically, let us describe the three branches
of an isotherm as follows: We call ``branch 3'' the branch of
solutions which extends all the way from $Q{=}\infty$, terminating
where $dQ/d\Phi{=}0$. From there, ``branch 2'' takes over, terminating
where again $dQ/d\Phi{=}0$. The isotherm continues with ``branch 1''
until the point $Q{=}0,\Phi{=}1$ is reached. This terminology matches
that of ref.\cite{cejmii}.
{}From this definition, then, branch 3 is electrically stable for most
of its extent, except for a small region near the join with branch
2. In this case, before reaching the point where $dQ/d\Phi{=}0$ the
permittivity changes sign at a point where $dQ/d\Phi{=}\infty$ and
renders branch 3 electrically unstable thereafter. This is a feature
that is absent from the standard van--der--Waals--Maxwell system (in the
latter there are no points in the isotherms where $dP/dV=\infty$), and
which will introduce a significant modification of the phase diagram.
Branch 2, being between two places where $dQ/d\Phi{=}0$, has positive
definite slope and hence is electrically stable everywhere, while
branch~1 is electrically unstable everywhere, having negative definite
slope. To compute precisely where the electrical instability begins,
we need only find the location of the minimum of the isotherms, that
is, the above mentioned point where $dQ/d\Phi{=}\infty$, which is given
by the equation $Q{=}T\sqrt{1{-}T^2}$. With the segment of the
$T$--axis from 0 to 1, this forms a region in the $(Q,T)$ plane within
which branch~3 and branch~1 are unstable to electric
fluctuations. Branch~2 is electrically stable everywhere, as mentioned
before, but as already pointed out in ref.\cite{cejmii}, and as a
quick examination of the figure~\ref{fig:stateS} of the isocharge
$(S,T)$ curves reveals, branch~2 is unstable to thermal fluctuations,
and so never plays a role in the canonical and grand canonical
ensembles.
It is also entertaining to subject by eye the snapshots of $F$ and $W$
taken in figures~\ref{fig:gibbssnaps},~\ref{fig:Tfreesnaps}
and~\ref{fig:Qfreesnaps} to the convexity and concavity
conditions~(\ref{condition1}) and~(\ref{condition2}). We find that the
shapes of $F$ and $W$ do indeed confirm our conclusions about the
stability of the various branches.
It is very instructive to plot the boundaries of the various branches
in the $(Q,T)$ plane:
\begin{figure}[hb]
\hskip2cm
\psfig{figure=branches.eps,width=6in,angle=270}
\caption{The demarkation of the various branches of black holes in the
$(Q,T)$ plane. Points on branches~1 and~3 which lie inside the solid
curved line are unstable to electric fluctuations. Branch~2 is
electrically stable but thermally unstable everywhere.}
\label{fig:branches}
\end{figure}
It is particularly interesting to note that the figures in the
previous plot are simply the three sheets of an underlying ``cusp
catastrophe'' shape, as can be seen by assembling them in three
dimensions to reconstruct the equation of state in
figure~\ref{fig:state}. Indeed, it is highly instructive to align the
surface $\Phi(Q,T)$ describing equation of state and the surface
$F[Q,T]$ giving the swallowtail shape of the free energy, in such a
way as to project some of their important features to
the $(Q,T)$ plane, as done in figure~\ref{fig:triple}. This gives rise
to the critical phase diagram which we will discuss in the next
section.
\begin{figure}[hb]
\hskip2cm
\psfig{figure=triple.eps,height=6.5in}
\caption{The swallowtail shape (free energy) and cusp shape
(equation of state) for the charged black hole thermodynamic
system. Notice the features which result in the critical line and
point in the $(Q,T)$ plane.}
\label{fig:triple}
\end{figure}
As anticipated, the shape formed by the equation of state in the
neighbourhood of the critical point is merely a distortion of the
standard cusp shape, which was encountered in the variables
$(r_+,Q,\beta)$ in our previous
paper~\cite{cejmii}. Figure~\ref{fig:cusp} shows this standard shape
with two sample trajectories in state space. It will be discussed
further in section~\ref{sec:critical}.
\begin{figure}[hb]
\hskip3cm
\psfig{figure=cusp.eps,height=2.5in}
\caption{A sketch of the cusp catastrophe in action (in
$(r_+,Q,\beta)$ space). Two sample
trajectories are shown, one ($Q{<}Q_{\rm crit}$) encountering a phase
transition, the other ($Q{>}Q_{\rm crit}$) does not. The precise
location of the line across which the transition happens is given by
the minimum free energy condition, or equivalently, an appropriately
formulated ``equal area law''.}
\label{fig:cusp}
\end{figure}
As a final comment that in cases where one of the local stability
criteria (\ref{stabcrit}) are violated, we are not always able
to determine the stable ground state. However, the precise nature
of the stability violation is providing information about how the
system will relax to a new stable configuration. For example, one has
$\delta^2E{\propto}\varepsilon_T\,\delta Q^2$ and so $\varepsilon_T{<}0$
indicates that the black hole should relax by reducing its charge,
{\it i.e.,} it will emit charged particles (if possible).
\section{Phase structure}\label{sec:phases}
Figures~\ref{fig:triple} and~\ref{fig:free3d}, together with the
slices displayed in figures~\ref{fig:Tfreesnaps}
and~\ref{fig:Qfreesnaps}, show how the free energy curve determines
the phase structure of the black holes as one moves around on the
state curve in the canonical ensemble, while figures~\ref{fig:gibbs3d}
and the slices displayed in figure~\ref{fig:gibbssnaps} determine the
phase diagram for the grand canonical ensemble. We performed this
analysis in ref.~\cite{cejmii}, and we recall it here for completeness,
before going on to refine the resulting phase diagram using the
information uncovered in this paper.
The dashed line in the $(Q,T)$ plane shows the boundary of the region
multiply covered by $\Phi$, in the state curve, and correspondingly,
the free energy has three possible values in that region also (see
fig.~\ref{fig:triple}), which constitutes the swallowtail region. The
free energy of branch~2 is always greater than that of either
branches~1 or~3, however, and so there is no transition along the
dashed lines. Along the solid line, the free energies of branches~1
and~3 are equal, and there is a first order phase transition (the
first derivative of the free energy is discontinuous) along this line.
Also note that the one dimensional $Q{=}0$ situation is the familiar
Hawking--Page transition\cite{hawkpage} between AdS and
AdS--Schwarzschild, which happens (in our units) at
$T{=}T_{HP}{=}2/\sqrt{3}{\approx}1.154$, for $n{=}3$.
The solid line is the ``coexistence curve'' of the two phase of
allowed black hole. The line ends in a critical point. Above this
point, there is no transition, and one goes from large to small black
holes continuously (the distinction between branch~1 and branch~3 is
removed). (The reader should compare this to the physics of the
liquid--gas system for an exact analogue in classic thermodynamics.)
As the first derivative (but not the second) of the free energy $F$ is
continuous at the critical point, there is a second order phase
transition there, about which we will have some more to say in
sections~\ref{sec:fluct} and~\ref{sec:critical}.
This physics is all summarized in figure~\ref{fig:phases}, where we
have also displayed the phase diagram in the grand canonical ensemble
(the $(\Phi,T)$ plane), which is straightforward to determine. Some of
the details of the shape of these curves will be confirmed by
calculations in section~\ref{sec:coexist}. For most of the rest of the
paper, we will not have much more to say about the phase diagram in
the $(\Phi,T)$ plane, and refer the reader to ref.\cite{cejmii} for
discussions of its features\footnote{Discussed in ref.\cite{cejmii},
for example is the issue of the line of extremal black holes for
$T{=}0$ and $\Phi{>}1$. The calculation of $W[T,\Phi]$ yields a
non--zero result on this line, which is the contribution from the
extremal black holes. We expect that this does not represent the
equilibrium situation, because they will decay due to
``super--radiance'' effects on the approach to zero temperature, as the
charge in them is not fixed in this ensemble. This is an artifact of
the failure of the Euclidean quantum gravity techniques that we have
used to take into account such processes.}.
Note, however, that the boundary in this figure marks the line where
the Gibbs free energy of the black holes equals that of AdS. That is
the boundary does {\it not} denote a curve where one of the local stability
criteria begins to be violated.
\begin{figure}[hb]
\hskip1cm
\psfig{figure=qphases.eps,height=2.5in}
\hskip1cm
\psfig{figure=fphases.eps,height=2.8in}
\caption{Sketches of the thermally stable phases in the canonical
ensemble, and in the grand canonical ensemble, respectively.}
\label{fig:phases}
\end{figure}
Depending upon the situation, there may or may not be the possibility
of electrical fluctuations. This depends very much upon the setting
within which we are considering these black holes. In a theory without
charged particles, the black hole charge would be fixed and electrical
stability need not be considered.
In general, however, if
there are fundamental charged quanta in the theory, then there is the
possibility of the black holes emitting or absorbing such quanta,
introducing the possibility of electrical fluctuations. Such a
possibility must be considered in (for example) the case when the EMadS
system is considered to be a Kaluza--Klein truncation of some higher
dimensional theory, as discussed in our previous
work\cite{cejmii}. Then, the electrically charged black hole can in
principle emit or absorb electrically charged Kaluza--Klein particles
in order to allow its charge to fluctuate.
In the particular case of four dimensions, however, there is also the
possibility that we can exchange, by electric--magnetic duality, the
electric charge (and vector potential) that we have been considering
here for a magnetic charge (and vector potential). In this case, we
have instead that the only way for the magnetically charged black
holes to change their charge is to emit or absorb Kaluza--Klein
monopoles, which are not fundamental quanta, as they are very massive,
the further we are below the Kaluza--Klein scale.
In general, when there are allowed electrical fluctuations (by whatever
mechanism is appropriate to the situation in hand) we must also take
into account on the phase diagram, the electrical stability of the
solutions as determined in the previous section. Including those
regions, we obtain the following phase diagram:
\begin{figure}[hb]
\hskip4cm
\psfig{figure=qphase2.eps,height=3.0in}
\caption{The phase diagram in the canonical ensemble, showing the
disallowed (shaded) regions where the solutions are unstable to
electrical fluctuations. Note that the
critical point and part of the coexistence line lies within the
unstable region.}
\label{fig:critical}
\end{figure}
The question arises as to what the equilibrium system is which resides
in the shaded regions. The electrically unstable black holes cannot
reside there, and so we must search for other possibilities. One
formal possibility is that extremal black holes reside there, because
formally they can exist at any temperature for any charge. However, we
do not find this possibility very attractive. We expect that the
permission that the Euclidean computation appears to give them to
exist at any temperature is an artifact, and that they should
naturally be associated with zero temperature, in which case they can
only occupy the line $T{=}0$ on our phase diagram in the canonical
ensemble, which they do. In any event, one can infer from the
calculations of ref.~\cite{cejmii} that extremal black holes
actually have a higher free energy than the unstable nonextremal
black holes. Another possibility is that the preferred
state is simply anti--de Sitter space (which can also exist at
arbitrary temperature) filled with a charged gas. This is certainly
the case at $Q{=}0$ \cite{edadsii,hawkpage}. However, when the gas carries a
non--negligible charge (and hence mass), its backreaction on the AdS
geometry can not be neglected in determining the free energy.
Another interesting possibility is that of black hole surrounded
by a gas of particles. Again, if the gas component carries a sizable
fraction of the charge and mass, its backreaction on the geometry
would modify the equation of state and may then re--establish
thermodynamic stability. Pursuing either of these possibilities
lies beyond the scope of the present paper, and so
we will leave
settling of this interesting issue to a future date. Hence we must
simply regard
the shaded region as a sort of {\it terra incognita} with regard to
black hole physics. As a final note, we remind the reader that this
is only the region in which we are certain that the black holes do not
minimize the free energy due to its thermodynamic instability.
It may be that the onset of a phase transition to a state of lower
free energy actually occurs outside of the boundary in figure 13,
just as it does for the grand canonical ensemble in figure 12.
\section{Coexistence of phases and the Clapeyron equation}\label{sec:coexist}
Let us study further the coexistence lines which we discovered in the
phase diagrams, both in the canonical ensemble and in the grand
canonical ensemble (see figure~\ref{fig:phases}). We can use
straightforward thermodynamics to determine the shape of the lines.
Let us start with the grand canonical ensemble, with Gibbs potential
$W[\Phi,T]$, with $dW{=}{-}SdT{-}Qd\Phi$. An equation can be derived
for the line separating two phases A and B in the $(\Phi,T)$ phase
diagram as follows. Along such a coexistence line the phases (for
given $(\Phi,T)$) have the same $W$, and so the slope of the curve
$Q(T)$ is related to the change in entropy and $Q$ by:
\begin{equation}
{d\Phi\over dT} =-{S_{\rm A}-S_{\rm B} \over Q_{\rm A}-Q_{\rm B}}\ .
\end{equation}
In the case at hand, one of the phases is AdS which has zero entropy
and zero charge. So we find that (for all $n$):
\begin{equation}
{d\Phi\over dT} =-{S_{\rm bh} \over Q_{\rm bh}}=-\left({2\pi\over
n-1}\right){1\over cq}\left({q\over c\Phi}\right)^{n-1\over n-2}\ .
\label{ccequation}
\end{equation}
Here, $q(T,\Phi)$ is obtained from the equation of state for the
corresponding branch. Equation~(\ref{ccequation}) is the precise
analogue of the Clapeyron equation. From it, we see that the
slope of the curve is negative. For the case of $n{=}3$ we can give
explicit expressions. In the rescaled units, we have:
\begin{equation}
{d\Phi\over dT} =-{Q \over 2\Phi^2}=-{1\over 2}\left({T\over
\Phi}+\sqrt{{T^2\over \Phi^2}+1 -{1\over \Phi^2}}\right)\ .
\end{equation}
We see from here that the curve intersects the axes orthogonally, and
its convexity, sketched in the figure~\ref{fig:phases}, follows from
the fact that ${d^2\Phi/ dT^2}{<}0$.
Next (assuming the issue of electrical stability can be ignored),
we consider the canonical ensemble, defined by the Helmholtz
thermodynamic potential $F[Q,T]$, with $dF{=}SdT{-}\Phi dQ$. Along any
line of coexistence of two phases, we have:
\begin{equation}
{dQ \over dT} ={S_{\rm A}-S_{\rm B} \over \Phi_{\rm A}-\Phi_{\rm B}}
\end{equation}
The phase diagram is sketched in figure~\ref{fig:phases}.
The Clapeyron equation can be used to find the slope of the
curve at $Q{=}0$ and $Q{=}Q_{\rm crit}$ for the line separating the
two black holes phases (we show the expressions for all $n$):
\begin{equation}
\left.{dQ \over dT}\right|_{Q_{\rm crit}} =-{\omega_{n-1}\over 4
G}\left({n-1\over n-2}\right) {r_{+(\rm crit)}^{2n-3}\over q_{\rm crit}}\ ,
\end{equation}
\begin{equation}
\left.{dQ \over dT}\right|_{Q{=}0}=-{\omega_{n-1}\over 4 G} r_{+(3)}^{n-1}\ ,
\end{equation}
(where we have used here that $r_{+(1)}^{n-2}\simeq q$ near $q=0$, as
is easy to obtain). On the scale at which we have sketched the
coexistence curve in the previous section, it is essentially a
straight line, and we have drawn it as such in
figure~\ref{fig:phases}.
\section{Fluctuations for Charged AdS Black Holes}\label{sec:fluct}
In section~\ref{sec:stability}, we discussed and computed the
thermodynamic quantities (specific heats and permitivity) which signal
the stability (or not) of a black hole against fluctuations. While
these quantities pertain to the response of the system to macroscopic
thermodynamic processes which may be performed, in Euclidean Quantum
Gravity, where we ordinarily do not have a description of the
microscopic degrees of freedom, we usually cannot relate them directly
to microscopic fluctuations, as we can in ordinary thermal physics.
However, we can go further in this paper. Many of the adS models which
we have here can be embedded into a full theory of quantum gravity
---string and/or M--theory--- and where the holographic duality tells
us precisely that the microscopic description is organized neatly in
terms of a dual (gauge) field theory.
So we may go and boldly study the fluctuations of the
thermodynamic quantities in our theory, and we should see earmarks of
the underlying (gauge) theory in our quantities, connecting the
microscopic to the macroscopic.
Here, one uses the entropy to define a probability distribution
on the space of independent thermodynamic quantites~\cite{landau}:
$p(X_i){\propto}\exp(S(X_i))$. With the assumption that fluctuations
are small, we can work with a quadratic expansion of the entropy
in deviations from the equilibrium values. The stability analysis
of section VI establishes that the Hessian of $S$ is negative
semi--definite, and so we have a normalizable Gaussian distribution
within this approximation. One then finds that the fluctuations
are given by
\begin{equation}
\langle\delta X_i\,\delta X_k\rangle=
-\left({\partial^2S\over\partial X_i\partial X_k}\right)^{-1}
\label{gauss}
\end{equation}
where $\delta X_i$ denotes the deviation of $X_i$ from its equilibrium
value, and notation of the left--hand side denotes a matrix inverse.
Implicit above is the assumption that we have a closed system can be divided
into a number of subsystems. In the AdS context, the natural decomposition
is the black hole and the thermodynamic reservoirs\footnote{We are
neglecting the contributions of any gas component around the black hole
in all of our calculations in this paper. Further we should be able
to consider smaller subdivisions with the dual field theory in mind.}.
In this situation where
the subsystem of interest is really the entire object under study,
the most reasonable approach is to consider fluctuations in only
the extensive variables that are free to vary in the thermodynamic
ensemble \cite{callen}. Hence we denote the
general extensive variables that are free to vary
as $X_i$, and the $F_i$'s are their conjugate
intensive variables defined by $dS{=}F_k\,dX_k$.
Eqn.~(\ref{gauss}) then becomes\cite{callen,landau}
\begin{equation} \langle
\delta X_i\,\delta X_k\rangle=-{\partial X_i\over\partial F_k}
=-{\partial X_k\over\partial F_i}\ .
\label{fluct} \end{equation}
Now, these fluctuations are given practical
meaning when they are compared to for example, their equilibrium
values. For example, the relative root mean square of the
fluctuations: \begin{equation} {\sqrt{\langle\delta X_i^2\rangle}\over X_i} \end{equation}
which tells us about the sharpness of the distribution in $X_i$.
Note, that by the formula (\ref{fluct}) the above ratio goes roughly
as the extensive parameters to the ${-}1/2$, and therefore the
distribution is increasingly sharp as the size of the system
increases.
Now in the present problem of charged black holes,
\begin{equation} dS = (1/T)
dE - (\Phi/T) dQ\ . \end{equation}
Hence
for the canonical (fixed $Q$) ensemble (our analogue of a fixed volume system),
the only free extensive variable is the energy, and the above formulae
yield
\begin{equation} \langle \delta E^2 \rangle= -\left({\partial E\over
\partial\beta}\right)_Q =T^2\left({\partial E\over\partial T}\right)_Q=T\,C_Q\ .
\end{equation}
For the grand canonical (fixed $\Phi$) ensemble (analogue of a fixed
pressure system), the energy and the charge are free to vary, and one
has
\begin{eqnarray}
\langle \delta E^2\rangle &=& -\left({\partial
E\over\partial\beta}\right)_{\Phi/T} =T^2\left({\partial
E\over\partial T}\right)_{\Phi/T} \nonumber\\
&=&TC_\Phi+T\Phi\left({\partial E\over\partial \Phi}\right)_T\ . \\
\langle\delta Q^2\rangle&=&T\left({\partial Q\over\partial
\Phi}\right)_T =T\varepsilon_T\ . \\ \langle\delta E\,\delta
Q\rangle&=&T\left({\partial E\over\partial\Phi}\right)_T \nonumber\\
&=&T^2\left({\partial Q\over\partial T}\right)_\Phi
+T\Phi\left({\partial Q\over\partial \Phi}\right)_T=T^2\alpha_\Phi+T\Phi
\varepsilon_T\ .
\end{eqnarray}
We have recovered the fact that the thermodynamic fluctuations are
controlled by the same generalized compressibilities ---specific heats
permittivity, {\it etc.,}--- that determine the intrinsic stability in
section~\ref{sec:stability}. This follows since both analyses can be
phrased in terms of the Hessian of the entropy.
Above we have presented a general thermodynamic discussion.
Let us now focus on the case $n{=}3$ and present our results in terms
of the dimensionless variables introduced in section ~\ref{sec:intro}.
Note that translating the above thermodynamic formulae to the dimensionless
variables, there are extra factors, giving {\it e.g.,}
\begin{equation} \langle\delta E^2\rangle =
{3G\over2\pi l^2} T^2{\partial
E\over\partial T}\ . \label{flutter}\end{equation}
For the fixed charge ensemble: \begin{equation}
{\langle\delta E^2\rangle\over E^2}= {3G\over2\pi l^2}{1\over r_+^2}
{\left({r_+^2}+1-{q^2\overr_+^2}\right)^3\over
\left({r_+^2}-1+3{q^2\overr_+^2}\right)
\left({r_+^2}+1+{q^2\overr_+^2}\right)^2}\ . \end{equation} For the fixed
potential ensemble: \begin{equation} {\langle\delta E^2\rangle\over E^2}=
{3G\over2\pi l^2}{1\over r_+^2} {\left({r_+^2}+1-\Phi^2\right)\over
\left({r_+^2}-1+\Phi^2\right)}
\left[{\left({r_+^2}+1-\Phi^2\right)^2+4\Phi^2(1-\Phi^2)\over
\left({r_+^2\over 3}+1-\Phi^2\right)^2}\right]\ , \end{equation}
\begin{equation}
{\langle\delta Q^2\rangle\over Q^2}= {3G\over4\pi l^2}{1\over\Phi^2
r_+^2}\left[ {\left({r_+^2}+1-\Phi^2\right)
\left({r_+^2}-1+3\Phi^2\right)\over \left({r_+^2}-1+\Phi^2\right)}\right]\ .
\end{equation}
\begin{equation} {\langle\delta E\delta Q\rangle\over{E\,Q}}= {3G\over\pi
l^2}{1-\Phi^2\over r_+^2}\left[ {\left({r_+^2}+1-\Phi^2\right)\over
\left({r_+^2}-1+\Phi^2\right) \left({r_+^2\over 3}+1-\Phi^2\right)}\right]\ .
\end{equation}
Notice that all of these results are proportional to
$G/l^2{\sim}N^{-3/2}$, so for large $N$ the fluctuations are
suppressed. For $n{=}3$, the dual field theory (supplying our
microscopic description) is the field theory of ref.\cite{exotic},
associated to $N$ coincident M2--branes. The number of degrees of
freedom in this theory grows as $N^{3/2}$ (as seen for example in the
black hole entropy at high temperature). So the squared fluctuations
are controlled by the inverse of the number of degrees of freedom of
the field theory, which is precisely what we expect from standard
kinetic theory connecting the microscopic to the macroscopic! Note
here that we see these unconfined degrees of freedom appearing in our
formulae at arbitrary temperature in this ensemble. This is because
black holes dominate the thermodynamics for all values of the
temperature: the presence of charge effects a deconfinement of the
theory at all temperatures, even in finite volume. (This is to be
contrasted to the case of $Q{=}0$, where AdS dominates the physics for
some $T{<}T_{\rm HP}$, representing the ``confined'' phase.)
To gain more insight into these results, let us
rewrite eqn.~(\ref{flutter}) for the energy fluctuations as: \begin{equation}
{\langle\delta E^2\rangle\over E^2}={3G\over2\pi l^2}{T^3\over r_+^2
\left(T+{Q\over r_+^3}\right)^2}{1\over\left({\partial T\over \partial
r_+} \right)_Q} \end{equation}
{}From this form, one can pick out some of the
interesting behaviour. The fluctuations go to zero at zero temperature
as~$T^3$. For large $T$ (and hence large $E$), the fluctuations also
go to zero now as $1/T$ (since for large $r_+$, $2T{\simeq}r_+$). An
interesting factor is $(\partial T/\partial r_+)^{-1}$ which can
change sign for $Q{<}Q_{\rm crit}{=}1/(2\sqrt{3})$.
So for $Q{>}Q_{\rm crit}$, the fluctuations rise from zero at the
extremal black ($T{=}0$) go through a maximum and then die down for
large temperatures. As $Q$ approaches $Q_{\rm crit}$, the maximum
grows larger and larger, and actually becomes a divergence at
$Q{=}Q_{\rm crit}$! (We have plotted these squared fluctuations in
figure~\ref{fig:fluct}, where they are denoted $f(T)$.) This is
actually the same divergence as that in $C_Q$ at the critical point,
which was commented on in ref\cite{cejmii}. Hence one finds from
there that near the critical point, \begin{equation} {\langle\delta
E^2\rangle\over E^2}\sim (T-T_{\rm crit})^{-2/3}\ . \end{equation}
\begin{figure}[hb]
\hskip4cm
\psfig{figure=Efluct.eps,height=3.5in,width=3.5in}
\vskip1cm
\caption{The squared fluctuations $f(T)$ in the energy, relative to the
equilibrium energy, for varying $Q{\geq}Q_{\rm crit}$. The values of
$Q$ plotted here are (bottom up) $Q{=}0.49,0.44,0.39,0.34,Q_{\rm
crit}$. The dotted
curve shows that the fluctuations diverge at
$Q{=}Q_{\rm crit}$, at the critical temperature $T{=}T_{\rm crit}$.}
\label{fig:fluct}
\end{figure}
This divergence of the energy fluctuations signals the breakdown
of the Gaussian approximation considered in these calculations.
It is also the classic behaviour of a
second order phase transition point, where correlation lengths, {\it
etc.,} diverge as an order parameter vanishes. Here, the order
parameter can be taken to be a homogenous function of
$r_{+(3)}{-}r_{+(1)}$, the difference between horizon radii of the
branches~3 and~1.
For $Q{<}Q_{\rm crit}$, the fluctuations rise from zero at the
extremal black hole and diverge at the first zero of $\partial
T/\partial r_+$. Between the two zeros of $\partial T/\partial r_+$,
$\langle\delta E^2\rangle/ E^2$ is negative. This is simply an
indication that we are in the thermally unstable regime, otherwise
known as ``branch~2''. For $T$ larger than the second zero of
$\partial T/\partial r_+$, the fluctuations are monotonically
decreasing (from infinity at the zero, to zero as
$T\rightarrow\infty$). As we know from the minimum free energy
condition, we are protected from the unstable regime by making a phase
transition from branch 1 and 3. So in a $\langle\delta E^2\rangle/
E^2$ versus $T$ plot, the fluctuations rise from zero to the phase
transition point and the (discontinuously) jump to the decreasing
curve.
\section{Of Higher dimensions and Other thermodynamic functions
and ensembles}\label{sec:higher}
In this section, we collect together some results for various
thermodynamic quantities computed for arbitrary $n$, with all of the
factors explicitly included. The thermodynamic functions are written
in terms of their canonical state variables. We do not use the
physical charge $Q$ instead of $q$, for simplicity of presentation. In
any expression, $Q$ may be restored by recalling that
\begin{equation}
Q={\omega_{n-1}\over 8\pi G}(n-1) c q\quad {\rm and}\quad
c=\sqrt{2(n-2)\over n-1}\ .
\end{equation}
Similarly, we also introduce the parameter $s$ as
\begin{equation}
S={\omega_{n-1}\over 4G}s\ .
\end{equation}
Notice that it does make sense to write physical quantities in terms
of $q$ and $s$, since they are related to the charge and entropy
densities. This follows from the fact that we may replace
$l^{n-1}\omega_{n-1}$ by the field theory's volume $V_{n-1}$.
The equation of state, following from eqn.~(\ref{betaform}) is:
\begin{equation}
T={(n-2)l^2 (1-c^2\Phi^2)(c\Phi)^{2\over n-2}+ nq^{2\over n-2} \over
4\pi l^2 (cq\Phi)^{1\over n-2}}
\end{equation}
Equation of state for extremal black holes is:
\begin{equation}
q^{2\over n-2} = {n-2\over n} l^2 (c^2\Phi_e^2-1) (c\Phi_e)^{2\over
n-2},\qquad{\rm for}\quad T\,\, {\rm arbitrary.}
\end{equation}
The Gibbs thermodynamic potential for the grand canonical ensemble
is\cite{cejmii}:
\begin{equation}
W[T,\Phi]={\omega_{n-1}\over 16\pi G l^2}\left[ l^2 {q\over
c\Phi}(1-c^2\Phi^2) -\left({q\over c\Phi}\right)^{n\over n-2}\right]
\end{equation}
where $Q{=}Q(T,\Phi)$ is obtained from the equation of state. Notice
that $W[T,\Phi]$ vanishes for anti--de Sitter spacetime (which has
$Q{=}0$), and so AdS may be thought of as the reference background for
the calculation of the action, and indeed $W[T,\Phi]$ was computed in
this way in ref.\cite{cejmii}, using the background subtraction
method, which we see (in the present work) gives the same result as
the intrinsic definition by ``counterterm subtraction'' methods.
The Helmholtz free energy which is the Legendre transform of
$W[T,\Phi]$ may be computed with an explicit action calculation, using
the counterterm subtraction method to give an intrinsic
(``backgroundless'') definition. The result is:
\begin{equation}
F[T,Q]={\omega_{n-1}\over 16\pi G l^2}\left[ l^2 {q\over c\Phi}
-\left({q\over c\Phi}\right)^{n\over n-2}+(2n-3)l^2 cq\Phi
\right]\ .
\end{equation}
Again, $\Phi{=}\Phi(T,Q)$ is obtained from the equation of
state. Notice that AdS with with non--zero charge is not a solution of
the equations of motion and so cannot be considered as the ``ground
state'' or reference background for this result. Indeed, this result
cannot be obtained by an action calculation which uses a matching to a
background, precisely for this reason. The counterterm subtraction
technique is therefore necessary here to supply the honest action
computation for this thermodynamic potential. It is satisfying to note
that $W[T,\Phi]$ and $F[T,Q]$ are Legendre transforms of each other,
$W{=}F{+}Q\Phi$, as they should be.
We can arrive at a variety of other ensembles, with their
corresponding associated potentials, by formal Legendre transforms.
For example, we can consider the enthalpy $H[S,\Phi]$, a function of
entropy and potential (this notation is not to be confused with the
Hamiltonian!). Starting from $W[T,\Phi]$ we can construct
$H{=}W{+}TS$, finding
\begin{equation}
H[S,\Phi]={(n-1)\omega_{n-1}\over 16\pi G l^2}\left[ s^{n\over n-1}
+ l^2 s^{n-2 \over n-1}(1-c^2\Phi^2)\right]\ .
\end{equation}
Notice that this function can {\it not} be obtained by performing a
proper background subtraction in Euclidean gravity, since for any
given solution we cannot find another regular solution with the same
values of the entropy and the potential. However, the enthalpy
vanishes for AdS, which could therefore be regarded as the ground
state or reference background here.
Another thermodynamic function in terms of its canonical variables is
the (internal) energy $E{=}W{+}TS{+}Q\Phi$,
\begin{equation}
E[S,Q]={(n-1)\omega_{n-1}\over 16\pi G l^2}\left[ s^{n\over n-1}
+ l^2 s^{n-2 \over n-1}+ q^2 l^2s^{-{n-2 \over n-1}}\right]\ ,
\end{equation}
which vanishes as well for AdS. This function would define the
thermodynamic
potential for the microcanonical ensemble, and as for the enthalpy
above, a calculation from
Euclidean gravity should proceed by fixing the entropy of the
state--the black hole area, if we neglect the entropy of the charged
gas in AdS\footnote{See\cite{microc} for work on defining the microcanonical
ensemble in gravity.}.
\section{The Universal Neighbourhood of the Critical Point}
\label{sec:critical}
In the section~\ref{sec:fluct}, we saw that fluctuations diverge as we
approach the critical point in the canonical ensemble. This point
represents a second order phase transition, as can be seen from the
fact that the free energy's first derivative ceases to have a
discontinuity there (see figures~\ref{fig:Tfreesnaps}
and~\ref{fig:Qfreesnaps} for visual confirmation), while the
divergences of the last section signals a discontinuity in the second
derivative.
While much of the detailed discussion of the paper has been for
$n{=}3$, we emphasize here again that the results extend to all~$n{>}2$.
This is most clearly seen from the important features
of the equation of state. Let us examine some of these more closely.
Consider equation~(\ref{betaform}). Originating as the condition for
Euclidean regularity, and hence thermodynamic equilibrium, the
qualitative features of $\beta(r_+)$ for varying $q$ are shown in
figure \ref{fig:isocharges}. These features are the same for all $n$:
There is a critical charge, $q_{\rm crit}$, below which there are
three solutions for $r_+$ for a range of values of $\beta$,
corresponding to the small (branch 1), branch 2, and large (branch 3)
black holes, in the language of ref.\cite{cejmii}, and in this paper.
\begin{figure}[hb]
\hskip4cm
\psfig{figure=isocharges.eps,height=2.8in,width=3.0in}
\caption{A family of isocharge curves for the $(\beta,r_{\rm +})$ form
of the equation of state. Note that the middle curve is for the
critical value of the charge, $Q_{\rm crit}$, below which multiple
branches of $r_{\rm +}$ solutions appear. The neighbourhood of the
critical point is a universal cubic, true for all dimensions.}
\label{fig:isocharges}
\end{figure}
That this shape persists for arbitrary $n$ can be seen as follows.
First, note that for large $r_+$, $\beta(r_+)$ goes as
${\sim}1/r_+$. Secondly, note that the denominator of the right hand
side of eqn.~(\ref{betaform}) after choosing scalings similar to those
done for $n{=}3$ at the beginning of section~\ref{sec:state},
is\footnote{That is, we absorb a factor of $Gl^{-1}\sqrt{n/(n{-}2)}$
into $Q$ and $l^{-1}\sqrt{n/(n{-}2)}$ into $r_+$, {\it etc.}}
\begin{equation}
r_+^{2n-2}+r_+^{2n-4}-q^2=0\ ,
\end{equation} which has a single positive root,
$r_e$, where $\beta$ diverges. This corresponds to the $T{=}0$
situation, and $r_e$ is the radius of the corresponding extremal black
hole. Given the above, any turning points for finite $r_+$ must come
in pairs, and the condition $\partial\beta/\partial r_+{=}0$ shows
that there are only two real, positive such solutions, which we call
$r_{+(1)}$ and $r_{+(3)}$, labeling where branch 1 ends and,
respectively, where branch 3 begins. Branch 2 lies between these
roots. The equations determining those roots also have an elegant
form (for the same rescaling as before): \begin{eqnarray}
r_+^{2n-2}-r_+^{2n-4}+(2n{-}3)q^2=0\ . \end{eqnarray}
The two roots coalesce at the critical point ({\it i.e.,}
$\partial^2\beta/\partial r_+^2{=}0$ also) where $q{=}q_{\rm
crit}$. The value of the radius of this critical black hole is
$r_{+({\rm crit})}$ and it is at (inverse) temperature $\beta_{\rm
crit}$. For example, in the case $n{=}3$, the quantities $\{q_{\rm
crit},r_{+({\rm crit})},\beta_{\rm crit}\}$ take the values
$\{1/\sqrt{12},1/\sqrt{2},3/(2\sqrt{2})\}$, while for $n{=}4$, they
have the values $\{2/\sqrt{135},\sqrt{(2/3)},5/(4\sqrt{6})\}$. The
basic point here is that while the critical values themselves vary,
the important structures do not depend upon $n$ in any essential way.
The neighbourhood of the critical point is extremely
interesting. Because of the fact that for all $n$, there at most two
turning points below $q_{\rm crit}$, it is clear that this
neighbourhood can be better written as a cubic, in terms of local
coordinates near the point. To this end, write $\rho{=}r_+{-}r_{+({\rm
crit})}$, ${\hat\beta}{=}\beta{-}\beta_{\rm crit}$, and ${\hat
q}{=}q{-}q_{\rm crit}$, and rewrite the equation of state in these
coordinates. The neighbourhood of the critical point is found by
taking these coordinates $(\rho,{\hat\beta},{\hat q})$ to be small.
For the example of $n{=}3$, after some algebra, we obtain \begin{equation}
0=\left(\sqrt{2}-{1\over\beta}\right)\rho^3
+2{\hat\beta\over\beta}\rho^2 +\sqrt{2}{\hat\beta\over\beta}\rho-{\hat
q\over2\sqrt{3}}+{1\over 3}{\hat\beta\over\beta}\ .
\label{thiscubic}
\end{equation} Note that the quadratic and linear terms vanish with at the
approach to the critical temperature ${\hat\beta}{\rightarrow}0$, and
the term which contains $\rho^3$ does not vanish in this way, and so
we neglect higher powers of $\rho$ in favour of this one in order to
study the near--critical behaviour. Here, and in what follows, we will
also neglect terms which are not linear in ${\hat q}$ and $\hat\beta$.
This cubic form~(\ref{thiscubic}) may always be obtained in this limit
for all $n$, because of the observations made in the preceding few
paragraphs. From this, certain universal behaviour can be easily
deduced, such as the critical exponent characterizing how fast our
``order parameter'', $\rho$, vanishes. (Recall that $\rho$ represents
the difference in equilibrium radius between the black holes of branch
1 and that of branch 3; it measures the distance from the analogue of
the ``fluid'' phase in liquid--gas language, where the two forms are
indistinguishable.) Setting ${\hat\beta}{=}0$, we see that the
critical exponent is $1\over3$, since \begin{equation}
\rho\simeq\left({3\over8}\right)^{1\over6}{\hat
q}^{1\over3}\sim(Q-Q_{\rm crit})^{1\over3}\ . \end{equation} Performing this
computation for other $n$ will change the numerical prefactor, but not
the exponent, which in this sense deserves to be called universal.
That a cubic equation controls the phase structure can be traced back
a step further. First, notice that the three dimensional plot of the
curve in $(\beta,r_+,Q)$ space is the cusp catastrophe, as drawn in
figure~\ref{fig:cusp}, with some sample state space trajectories. We
can remove the quadratic term in our cubic polynomial by shifting
$\rho$ by an appropriate amount. Multiplying overall by a normalizing
factor, our cubic may be written as: \begin{eqnarray}
\label{cubic}
0&=&\rho^3+A\rho+B\ , \\ \nonumber
{\rm with}\quad A&\simeq&4\sqrt{2}{\hat\beta}\quad {\rm and}\quad
B\simeq{4\over3}{\hat\beta}-\sqrt{3\over2}{\hat q}\ .
\end{eqnarray}
Equation~(\ref{cubic}) is actually telling us about the location
of the turning points
of a {\it quartic} function
\begin{equation}
{\cal V}(\rho)={1\over4}\rho^4+{A\over2}\rho^2+B\rho\ ,
\end{equation}
where we have discarded an arbitrary additive constant. Treated as a
{\it potential}, (for reasons which will be clear below), it is the
generic form of ${\cal V}(\rho)$ as $A$ and $B$ vary that controls
much of the critical behaviour in the neighbourhood of the critical
point. (As $A$ and $B$ are functions of $Q$ and $T$, this critical
behaviour in $(A,B)$ space translates directly into the earlier
discussed critical behaviour in $(Q,T)$ space.)
The function ${\cal V}(\rho)$ deserves to be treated an effective
potential which organizes the description of much of the local
physics. In particular, away from the critical point, where $A$ and
$B$ are both non--zero, the potential generically has two minima and
one maximum, the location of which are given by the solutions of our
universal cubic. These locations may be smoothly visualized in the
form of the cusp, sketched in figure~\ref{fig:cusp}. The location of
the minima in $\rho$ are the values $r_{+(1)}$ and $r_{+(3)}$, of the
equilibrium black hole radii of branch 1 and branch 3, while the
location of the maximum is $r_{+(2)}$, the branch 2 black hole
radius. The thermal stability of the branches correlates with the
whether the turning point is a maximum or a minimum of ${\cal
V}(\rho)$, further justifying its treatment as a potential.
The boundary of the region where there are three solutions marks the
situation where one of the minima of the potential ${\cal V}(\rho)$
merges with the maximum and disappears. This boundary is simply given
by the values of $A$ and $B$ where the cubic's discriminant,
$\Delta{=}27B^2{+}4A^3$, vanishes. (This can only happen for $A{<}0$,
therefore telling us that we have the distinct branches {\it below}
$\beta_{\rm crit}$.) The interior of this region may be translated
into $(Q,T)$ space, where it gives the shaded region in the third
diagram of figure~\ref{fig:branches} where branch 2 resides.
Along the line in the $({\hat q},{\hat\beta})$ plane (or the $(Q,T)$
plane) where $B$ vanishes, given by ${\hat
q}{=}\sqrt{(32/27)}{\hat\beta}$, the two minima of the potential
${\cal V}(\rho)$ are degenerate. This is the point at which there is
phase transition, as the system moves from one minimum of the
potential to the other.
At the critical point, $(A{=}0,B{=}0)$, the maximum and the two minima
merge into a single minimum of the potential. Notice that the well
formed by the potential is very flat there, and so the range of
allowed fluctuations within it is larger at this point than at any
other point in the plane, as they are less confined. We have seen
this physics before as the divergence of the fluctuations of the
microscopic degrees of freedom at the critical point. The potential
${\cal V}(\rho)$ is an effective potential for the uncharged
microscopic degrees of freedom of the theory in the neighbourhood of
the critical point. (See figure~\ref{fig:LGpot} for a summary of these
critical points of the potential.)
\begin{figure}[hb]
\hskip4.5cm
\psfig{figure=LGpotential.eps,height=3.5in}
\caption{The behaviour of the Landau--Ginzburg $A_3$ potential at various
points in the $(A,B)$ plane. This plane maps to the $(Q,T)$ plane of the
charged black holes system. The line $(A{<}0,B{=}0)$, maps to the critical
coexistence line found in that system.}
\label{fig:LGpot}
\end{figure}
Also, in this language, the meaning of the swallowtail shape for the
thermodynamic potential $F[Q,T]$ is now clear: It is simply the actual
{\it value} of the potential ${\cal V}(\rho;{\hat\beta},{\hat q})$ at
its maxima and minima: the critical line is the place where these two
values at the minima are equal, the place where ${\cal V}$ has
degenerate minima.
This function ${\cal V}(\rho)$ is the $A_3$ Landau--Ginzburg
potential. The effective Landau--Ginzburg theory which we can write
here is an effective theory of the uncharged microscopic degrees of
freedom underlying the thermodynamics. Kinetic terms to complete the
Landau--Ginzburg model would have their origins in the holographically dual
field theory. One can in principle derive additional potential terms
governing the charged degrees of freedom as well, in order to model
the stability structure uncovered in section~\ref{sec:stability}, but
we will not do that here.
In the language of catastrophe theory\cite{catastrophe}, the term
$\rho^4$ is the basic ``germ'' of the cusp catastrophe, and $A$ and
$B$ are the ``unfolding parameters'' which deform the potential,
giving a line of first order phase transition points along the line
$(B{=}0,A{<}0)$ where its mimima are degenerate. The ($A$--$D$--$E$)
classification of such potentials is isomorphic to that of certain
geometrical singularities\cite{arnold}. We cannot help but wonder if
this story marks the beginning of a richer tale involving a more
profound underlying geometrical structure into which this physics is
all embedded. As all of the physics of this paper is intimately
connected to the physics of branes, perhaps the possibility of such a
connection should be pursued.
\section*{Acknowledgments}
AC is supported by Pembroke College, Cambridge. RE is supported by
EPSRC through grant GR/L38158 (UK), and by grant UPV 063.310--EB187/98
(Spain). Support for CVJ's research was provided by an NSF Career
grant, \# PHY9733173 (UK). RCM's research was supported by NSERC
(Canada) and Fonds FCAR du Qu\'ebec. This paper is report \#'s
DAMTP--1999--54, EHU-FT/9907, DTP--99/25, UK/99--5 and
McGill/99--15. RCM would like to thank Martin Grant for interesting
conversations.
\section*{References}
|
\section{Introduction}
\def \frac{ N_f N_cg^2_{\pi q\bar q}}{8\pi} {\frac{ N_f N_cg^2_{\pi q\bar q}}{8\pi} }
\def N_f N_cg^2_{\pi q\bar q}/(8\pi) {N_f N_cg^2_{\pi q\bar q}/(8\pi) }
\def \begin{equation} {\begin{equation}}
\def \begin{eqnarray} {\begin{eqnarray}}
\def \end{equation} {\end{equation}}
\def \end{eqnarray} {\end{eqnarray}}
\def {\rm gap} {{\rm gap}}
\def {\rm {gaps}} {{\rm {gaps}}}
\def {\rm \overline{\overline{gap}}} {{\rm \overline{\overline{gap}}}}
\def {\rm Im} {{\rm Im}}
\def {\rm Re} {{\rm Re}}
\def {\rm Tr} {{\rm Tr}}
\def $0^{-+}$ {$0^{-+}$}
\def $0^{++}$ {$0^{++}$}
\def $u\bar u+d\bar d$ {$u\bar u+d\bar d$}
\def $s\bar s$ {$s\bar s$}
In this talk\footnote{Invited talk at
the Workshop on Hadron Spectroscopy, March 8-13, 1999 Frascati, Italy.
To appear in the Frascati Physics Series.}
I shall discuss mainly the light and
broad $\sigma$, which was picked by Matts Roos and myself\cite{NAT}
from the particle data group wastebasket 4 years ago, having been there for
over 20 years. Today an increasing number of papers,
many of which have been reported at this meeting\cite{sighere},
are quoting its
parameters, with a pole position near
500-i250 MeV (See the table 1).
An important question today is: Does this broad resonance really exist?
And if so, what is its nature, together with the other light scalar mesons in
the 1 GeV region? The naive quark model (NQM), which works reasonably
well for the vectors and heavier multiplets,
definitely fails for the scalars taken as
$\sigma(500), f_0(980), a_0(980)$ and $K^*_0(1420)$.
\begin{table}
\centering
\caption{ \it $\sigma$ pole position.
}
\vskip 0.1 in
\begin{tabular}{|l|c|} \hline
Reference & pole position (MeV \\
\hline
\hline
Kaminski et al.\cite{sighere}& $532\pm 12-i(259\pm 7)$ \\
Locher et al.\cite{sighere} & $424-i213$ \\
Harada et al.\cite{sighere} & $\approx 500-i250$ \\
Lucio et al.\cite{sighere} & $\approx 400-i200$ \\
Ishida et al.\cite{pdg98} & $ 602\pm26-i(196\pm27)$ \\
Kaminski et al.\cite{pdg98} & $537\pm20-i(250\pm 17)$ \\
Oller et al.\cite{pdg98} & $469.5-i178.6$ \\
T\"ornqvist et al.\cite{pdg98} & $470-i250$ \\
Amsler et al.\cite{pdg98} & $1100-i300$ \\
Amsler et al.\cite{pdg98} & $400-i500$ \\
Janssen et al.\cite{pdg98} & $387-i305$ \\
Achasov et al.\cite{pdg98} & $525-i269$ \\
Zou et al.\cite{pdg98} & $370-i356$ \\
Zou et al.\cite{pdg98} & $408-i342$ \\
Au et al.\cite{pdg98} & $870-i370$ \\
Beveren et al.\cite{pdg98} & $470-i208$ \\
Estabrooks\cite{pdg98} & $750\pm50-i(450\pm 50)$ \\
Protopescu et al.\cite{pdg98} & $660\pm100-i(320\pm 70)$ \\
Basdevant et al.\cite{pdg98} & $650-i370$ \\
Scadron et al.\cite{scad}& $\approx 500-i250$ \\
Lucio et al.\cite{lucio} &$600^{+200}_{-100}-i350 $\\
Igi et al.\cite{igi}&$\approx 760$\\
\hline
\end{tabular}
\label{extab}
\end{table}
We all believe the vectors ($\rho, \omega,\phi,K^*)$ and heavier well
established multiplets
are $q\bar q$ states because with a few parameters,
such as an equally spaced bare mass spectrum, a small OZI rule violating
parameter and SU3$_f$ related coupligs, we can describe
the masses, widths and couplings of the whole nonet.
If we had data only on the $\rho(770)$
we could not conclude that it is $q\bar q$. But with the
successful SU3$_f$ predictions for the whole nonet we strongly believe it is
$q\bar q$.
The same is true for the $\sigma$. No single analysis of the $\pi\pi$
S-wave, however refined, could ever decide on what is the nature of the
$\sigma$.
Even the decision as to
whether it really exists, cannot be done using data on the
$\pi\pi$ S-wave alone, since there are inherent, model dependent,
ambiguities as to how to continue analytically to a pole which is
far from the physical region, as is the case for the broad $\sigma$.
It is also obvious why the NQM fails for the scalars:
Chiral symmetry is absent in the NQM, but is crucial for the scalars.
Chiral symmery is believed to be broken in the vacuum, and
two of the scalars ($\sigma$ and $f_0$) have the same quantum
numbers as the vacuum.
Thus to understand the scalar nonet in the same way as we
believe we understand the vectors,
and to make a sensible comparison with experiment,
one must include chiral symmetry in
addition to flavour symmetry in the quark model.
The simplest such chiral quark model is the
linear U3$\times$U3 sigma model with 3 flavours.
Then we can treat both the scalar and pseudoscalar nonets simultaneously, and
on the same footing, getting automatically small masses for the pseudoscalar
octet,
and symmetry breaking through the vacuum expectation values (VEV's) of the
scalar fields.
As an extra bonus we have in principle a
renormalizable theory, i.e. ``unitarity
corrections'' are calculable.
In fact, in the flavour symmetric limit the unitarity corrections can
be thought to be already included
into the mass parameters of the theory, once the original 4-5 parameters are
replaced by the 4 physical
masses for the singlet and octet $0^{-+}$ and $ 0^{++}$ masses
and the $\sigma$ VEV.
Unfortunately this over 30 years old
model\cite{sigma} has had very few phenomenological
applications. An important exception is the intensive efforts of
M. Scadron and collaborators. %
\section{The Linear sigma model with 3 flavours}
The well known linear sigma model\cite{sigma} generalized to 3 flavours
with complete scalar ($s_a$) and pseudoscalar ($p_a$)
nonets has at the tree-level the
Lagrangian the same flavour and chiral symmetries
as massless QCD. The U3$\times$U3 Lagrangian with a symmetry breaking term
${\cal L}_{SB}$ is
\begin{equation}
{\cal L}=
\frac 1 2 {\rm Tr} [\partial_\mu\Sigma \partial_\mu\Sigma^\dagger]
-\frac 1 2 \mu^2{\rm Tr} [\Sigma \Sigma^\dagger] -\lambda {\rm Tr}[\Sigma\Sigma^\dagger
\Sigma\Sigma^\dagger]\ -\lambda'
({\rm Tr}[\Sigma\Sigma^\dagger])^2+{\cal L}_{SB} \ .
\label{lag}
\end{equation}
Here $\Sigma$ is a $3\times 3$ complex
matrix, $\Sigma=S+iP= \sum_{a=0}^8(s_a+ip_a)\lambda_a/\sqrt 2$, in which
$\lambda_a$ are the Gell-Mann matrices, normalized as ${\rm Tr}[\lambda_a\lambda_b]=
2\delta_{ab}$, and where for the singlet
$\lambda_0 = (2/N_f)^{1/2} {\bf 1}$ is
included. Each meson in Eq. (1) has a definite SU3$_f$ symmetry
content, which in the quark model means that it has the same flavour
structure as
a $q\bar q$ meson. Thus the fields $s_a$ and $p_a$ and
potential terms in Eq.~(1) can be given
a conventional quark line structure\cite{NATPL}.
The symmetry breaking terms are most simply:
\begin{equation}
{\cal L}_{SB}=\epsilon_\sigma \sigma_{u\bar u+d\bar d} +
\epsilon_{s\bar s} \sigma_{s\bar s} +c[\det \Sigma +\det \Sigma^\dagger]\ ,
\end{equation}
which give the pseudoscalars mass and break the
flavour and $U_A(1)$ symmetries. The small parameters $\epsilon_i$
can be expressed in terms of
the pion and kaon decay constants and masses:
$\epsilon_\sigma = m_\pi^2f_\pi$, $\epsilon_{s\bar s}=
(2m_K^2f_K-m_\pi^2f_\pi)/\sqrt 2$.
My fit to the scalars with the unitarized quark model (UQM)\cite{NAT}
is essentially a unitarization of
eq.\ref{lag} with $\lambda\approx 16$ and $\lambda '= 0$,
and with the main symmetry breaking generated by
putting the pseudoscalar masses at their physical values.
The model was used as an effective theory
with a symmetric smooth 3-momentum cutoff 0.54 GeV/c given by a gaussian
form factor. Such a form factor is natural, since physical mesons are of
course not pointlike, but have finite size of 0.7-0.8 fm.
The fit included the Adler zeroes which follow from eq.1,
but only approximate crossing symmetry.
Eq.~(1) without ${\cal L}_{SB}$
is clearly invariant under $\Sigma \to U_L\Sigma U_R^\dagger$ of
U3$\times$U3.
After shifting the flavourless scalar fields by
the VEV's ($\Sigma \to \Sigma+ V$) to the minimum of the potential,
the scalars aquire masses and also
the pseudoscalars obtain a (small) mass because of
${\cal L}_{SB}$. Then the $\lambda$ and $\lambda'$ terms
generate trilinear $spp$ and $sss$
couplings, in addition to those coming from the $U_A(1)$ symmetry breaking
determinant term. The $\lambda$ term, which turns out to be the largest,
obeys the OZI rule, while the small $\lambda '$ term, and $c$ violate this rule.
\section{ Tree-level masses and couplings.}
It is an ideal problem for a symbolic program like Maple V to calculate the
predicted masses, and couplings from the Lagrangian, which has
6 parameters $\mu,\lambda,\lambda',c$
and the VEV parameters $u=d$ and $s$, which define a diagonal
matrix for the flavourless meson VEV's: $V=diag[u,d,s]$. These are at the
tree level related to the pion and kaon decay constants through
$u=d=<\sigma_{u\bar u+d\bar d}>/\sqrt 2=f_\pi/\sqrt 2$ (assuming
isospin exact) and $s=<\sigma_{s\bar s}>=(2f_K-f_\pi)/\sqrt 2$:
One finds denoting the often occurring combination
$\mu^2+4\lambda'(u^2+d^2+s^2)$ by $\bar \mu^2$, and expressing the flavourless mass matrices in the ideally mixed frame:
\eject
\begin{eqnarray}
m^2_{\pi^+}\ &=&\bar \mu^2 + 4\lambda(u^2+d^2-ud)+2cs \cr
m^2_{K^+} \ &=&\bar \mu^2 + 4\lambda(u^2+s^2-su)+2cd \cr
\begin{array}{c} m^2_\eta \\ m^2_{\eta'} \end{array} &=& diag
\left( \begin{array}{cc} \bar \mu^2+2\lambda(u^2+d^2)-2cs & -c\sqrt 2 (u+d) \\
-c\sqrt 2 (u+d)& \bar \mu^2+4\lambda s^2 \end{array}\right)
\\
m^2_{a_0^+}\ &=&\bar \mu^2 + 4\lambda(u^2+d^2+ud)-2cs \cr
m^2_{\kappa^+} \ &=&\bar \mu^2 + 4\lambda(u^2+s^2+su)-2cd \cr
\begin{array}{c} m^2_\sigma \\ m^2_{f_0} \end{array} &=& diag
\left( \begin{array}{cc} \bar \mu^2+4\lambda'(u+d)^2+6\lambda(u^2+d^2)+2cs
& (4\lambda' s+c)\sqrt 2 (u+d) \\
(4\lambda' s+c)\sqrt 2 (u+d)& \bar \mu^2+
8\lambda' s^2+12\lambda s^2 \end{array}\right) \nonumber
\cr
\end{eqnarray}
\begin{table}
\centering
\caption{ \it Predicted masses in MeV and mixing angles for two values of the
$\lambda'$ parameter. The
asterix means that $m_\pi,m_K$
and $m_\eta^2+m^2_{\eta'}$ are fixed by experiment together with $f_\pi$ and
$f_K$. }
\vskip 0.1 in
\begin{tabular}{|l|c|c|c|} \hline
Quantity & Model $\lambda '=1$& Model $\lambda '=3.75$& Experiment \\
\hline
\hline
$m_\pi$ & 137$^{*)} $& 137$^{*)} $ &137 \\
$m_K$ & 495$^{*)} $& 495$^{*)} $ &495 \\
$m_\eta $& 538$^{*)} $& 538$^{*)} $ &547.3 \\
$m_{\eta'}$ &963$^{*)} $& 963$^{*)} $ &957.8 \\
$\Theta^{\eta'-singlet}_P$ &-5.0$^\circ$ &-5.0$^\circ$ &(-16.5$\pm6.5)^\circ$\cite{pdg98} \\
$m_{a_0}$ &1028 &1028 & 983 \\
$m_{\kappa}$ &1123 &1123& 1430 \\
$m_{\sigma}$ &651 &619 & 400-1200 \\
$m_{f_0}$ &1229 &1188 &980 \\
$\Theta^{\sigma-singlet}_S$
& 21.9$^\circ$ & 32.3$^\circ$ &(28-i8.5)$^\circ$\cite{NAT} \\
\hline
\end{tabular}
\label{extab2}
\end{table}
Now we fix 5 of the 6 parameters except $\lambda'$ by the 5 experimental
quantities
$m_\pi$, $m_K$, $m^2_\eta+m^2_{\eta'}$,
$f_\pi=92.42$ MeV and $f_K=113$ MeV\cite{pdg98}. One finds at the tree level
$\lambda=11.57$,
$\bar \mu^2=0.1424$ GeV$^2$, $c=-1701$ MeV, $u=d=65.35 $MeV
and $s=94.45$MeV.
The remaining $\lambda'$ paramerer affects only the $\sigma$ and $f_0$ masses
and their trilinear
couplings. This dependence is rather weak for the masses, but is very sensitive
to the couplings as it changes the ideal mixing angle for the scalars.
It turns out below that this must be small to fit the tri-linear couplings.
By putting $\lambda'=1$ one gets a reasonable compromise for most of
these couplings. With $\lambda'\approx 3.75$ one almost
cancels the OZI rule breaking
coming from the determinant term, and the scalar mixing becomes near ideal
(for $\lambda '=-c/(4s)=4.5$ the cancellation is exact).
As can be seen from Table 2 the predictions
are not far from the experimental values.
Considering that one expects from our previous analysis\cite{NAT}
that unitarity
corrections can easily be more than 20\% , and should
go in the right direction, one must
conclude that these results are even better than expected.
The trilinear coupling constants follow from the Lagrangian,
and are at the tree level:
\begin{eqnarray}
g_{\sigma\pi^+\pi^-}&=&\cos \phi_S^{id}(m^2_\sigma-m^2_\pi)/f_\pi \cr
g_{\sigma K^+K^-}&=&-\sqrt 3\sin( \phi_S^{id}-35.26^\circ )
(m^2_{\sigma}-m^2_K)/(2f_K) \cr
g_{f_0\pi^+\pi^-}&=&\sin \phi_S^{id}(m^2_{f_0}-m^2_\pi)/f_\pi \cr
g_{f_0 K^+K^-}&=&\sqrt 3\cos( \phi_S^{id}-35.26^\circ )
(m^2_{f_0}-m^2_K)/(2f_K) \\
g_{a_0\pi\eta}&=&\cos \phi_P^{id}(m^2_{a_0}-m^2_\eta)/f_\pi \cr
g_{a_0\pi\eta'}&=&\sin \phi_P^{id}(m^2_{a_0}-m^2_{\eta'})/f_\pi \cr
g_{a_0 K^+K^-}&=&(m^2_{a_0}-m^2_K)/f_K \cr
g_{\kappa^+K^0\pi^+}&=&(m^2_\kappa-m^2_\pi)/(\sqrt 2 f_K) \cr
g_{\kappa K^+\eta}&=&-\sqrt 3\sin( \phi_P^{id}-35.26^\circ )
(m^2_{\kappa}-m^2_\eta)/(2f_K) \cr
g_{\kappa K^+\eta'}&=&\sqrt 3\cos( \phi_P^{id}-35.26^\circ )
(m^2_{\kappa}-m^2_{\eta'})/(2f_K) \ .\nonumber \cr
\end{eqnarray}
Here $\phi^{id}_P=54.73^\circ+\Theta^{\eta'-singlet}_P$ i.e.
the angle between $s\bar s$ and $\eta'$
and $\phi^{id}_S=-35.26^\circ+\Theta^{\sigma-singlet}_S
$,
i.e. the angle between $u\bar u+d\bar d$ and $\sigma$.
\begin{table}
\centering
\caption{ \it Predicted couplings $\sum_i\frac{g_i^2}{4\pi}$ (in GeV$^2$)
, when $\lambda'=1$,
compared with experiment and predicted widths with experiment (in MeV). (We
have used isospin invariance to get the sum over charge channels, when data is
for one channel only.) The
predicted $f_0\to \pi\pi$ width is extremely
sensitive to the value of $\lambda^\prime$ (for $\lambda '=
3.75 $ it nearly vanishes)
and unitarity effects as discussed in the text. }
\vskip 0.1 in
\begin{tabular}{|l|c|c|c|c|} \hline
Process & $ \sum_i\frac{g_i^2}{4\pi}$ & $ \sum_i\frac{g_i^2}{4\pi}$ &
$\sum_i\Gamma_i$ & $\sum_i\Gamma_i$ \\
& in model &in experiment & model&experiment\\
\hline
\hline
$\kappa^+\to K^0\pi^++K^+\pi^0$ & 7.22 & - & 678 & $278\pm23$\cite{pdg98} \\
$\kappa^+\to K^+\eta $ & 0.28 & $\approx$ 0 & & \\
$\sigma\to \pi^+\pi^-+\pi^0\pi^0$ & 2.17 & 1.95\cite{ach} & 574
&300-1000\cite{pdg98} \\
$\sigma\to K^+K^-+K^0\bar K^0$ & 0.16 & 0.004\cite{ach} & 0 & 0\\
$f_0\to \pi^+\pi^-+\pi^0\pi^0$
& 1.67 & 0.765$^{+0.20}_{-0.14}$\cite{nov} & see text & 40 - 100\cite{pdg98}\\
$f_0\to K^+K^-+K^0\bar K^0$ & 6.54 & 4.26$^{1.78}_{-1.12}$\cite{nov} & 0 & 0\\
$a_0^+\to \pi^+\eta$ & 2.29 & 0.57\cite{nov} & 273 see text & 50 - 100\cite{pdg98}\\
$a_0^+\to K^+\bar K^0$ & 2.05 & 1.34$^{+0.36}_{-0.28}$\cite{nov} & 0 &0 \\
\hline
\end{tabular}
\label{extab2}
\end{table}
In some of the channels of table 3 the resonance is
below threshold and the widths vanish at the resonance mass.
However, the coupling
constants have then been determined through a loop diagram from
$\phi\to\gamma\pi\pi$ and $\phi\to\gamma\pi\eta$ (albeit in a somewhat model
dependent way) by the Novosibirsk group.
For channels where the phase space is large, it is important that one includes
a form factor related to the finite size of physical mesons. In the quark
pair creation ($^3P_0$) model a radius of 0.8 fm
leads to a gaussian form factor,
as in the formula below, where $k_0 \approx 0.56$ GeV/c (as was found
in the UQM\cite{NAT}).
Thus the widths are computed from the formula:
\begin{equation}
\Gamma(m)=\sum_{isospin}\frac{g_i^2}{8\pi}
\frac{k^{cm}(m)}{m^2}e^{-[k^{cm}(m)/k_0]^2}\ .
\end{equation}
As can be seen from table 3 most of the couplings are not far from experiment.
The main exception is the $f_0\to\pi\pi$ coupling and width, but this is
extremely sensitive to the ideal mixing angle. If one choses
$\lambda'= 3.75$ this mixing angle nearly vanishes
($\phi^{id}_S=-3.0^\circ$) together with the $f_0\to\pi\pi$ coupling
(c.f. eq.4).
From our experience with the UQM\cite{NAT} the couplings, when unitarized,
are very sensitive to especially
the nearby $K\bar K$ threshold. Similarily the $a_0\to\pi\eta$ peak width
is reduced, because of
the $K\bar K$ theshold, by up to a factor 5. Therefore one cannot
expect that the tree level couplings should agree better with data
than what those of Table 3 do. In fact, I was myself astonished by the fact
that the agreement turned out to be this good. After all,
this is a very strong
coupling model ($\lambda=11.57$, leading to large $g_i^2/4\pi$)
and higher order effects should be important.
\section{Conclusions.}
In summary, I find that the linear sigma model
with three flavours works, at the tree level, much
better than expected. It
works, in my opinion, just as well as the
naive quark model works for the heavier nonets.
One should of course include higher order effects, i.e., the model
should be unitarized phenomenologically,
e.g., along the lines of the UQM\cite{NAT},
whereby a more detailed data comparison becomes meaningful.
Those working on chiral perturbation theory and nonlinear sigma models
usually point out that
the linear model does not predict all low energy constants
correctly. However, one should remember that the energy regions of validity
are different for the two approaches.
Chiral perturbation theory usually breaks down when one approaches
the first scalar resonance.
The linear sigma model, on the other hand,
includes the scalars from the start
and can be a reasonable interpolating model in the intermediate
energy region near 1 GeV, where QCD is too difficult to solve.
These results strongly favour the interpretation that the $\sigma(500)$, $
a_0(980)$, $f_0(980)$, $ K^*_0(1430)$ belong to the same nonet,
and that they are the chiral partners of the $\pi$, $\eta$, $ K$, $\eta '$.
If the latter are believed to be unitarized $q\bar q$ states,
so are the light scalars $\sigma(500),$ $
a_0(980),$ $ f_0(980),$ $ K^*_0(1430)$, and the broad
$\sigma(500)$ should be interpreted as an existing resonance.
The $\sigma$ is a very important hadron indeed,
as is evident in the sigma model, because this is the boson
which gives the constituent quarks most of their mass
and thereby it gives also
the light hadrons most of their mass. It is the
Higgs boson of strong interactions.
|
\section{Introduction}
\label{intro}
The numerical evolution of neutron stars in full General Relativity
has been the focus of many research groups in recent years
\cite{font2,shibata,mathews,nakamura}. So far, these studies have
been limited to initially non-rotating stars. However, the numerical
investigation of many interesting astrophysical applications, such as
the rotational evolution of proto-neutron stars and merged neutron
stars or the simulation of gravitational radiation from unstable
pulsation modes, requires the ability of accurate long-term evolutions
of rapidly rotating stars. We thus present here the first study of
hydrodynamical evolutions of rotating neutron stars in the
approximation of a static spacetime. This approximation allows us to
evolve relativistic matter for a much longer time than present coupled
spacetime plus hydrodynamical evolution codes. Since the pulsations of
neutron stars are mainly a hydrodynamical process, the exclusion of
the spacetime dynamics has only a limited effect and allows for
qualitative conclusions to be drawn.
The rotational evolution of neutron stars can be affected by several
instabilities (see \cite{nik1} for a recent review). If hot
protoneutron stars are rapidly rotating, they can undergo a dynamical
bar-mode instability \cite{centrella}. When the neutron star has
cooled to about $10^{10}$K after its formation, it can be subject to
the Chandrasekhar-Friedman-Schutz instability \cite{CFS1,CFS2} and it
becomes an important source of gravitational waves. It was recently
found that the $l=m$ $r$-mode has the shortest growth time of the
instability \cite{Andersson1,Friedman} and it can transform a rapidly
rotating newly-born neutron star to a Crab-like slowly-rotating pulsar
within about a year after its formation \cite{Andersson2,Lindblom}. In
this model, there are two important questions still to be answered
\cite{Owen}: What is the maximum amplitude that an unstable $r$-mode
can reach (limited by nonlinear saturation) and is there any transfer
of energy to other stable or unstable modes via non-linear couplings?
Such questions cannot be answered by computations of normal modes of
the linearized pulsation equations, but require non-linear effects to
be taken into account. We therefore need to develop the capability of
full non-linear numerical evolutions of rotating stars in General
Relativity.
Our present 2-D (axisymmetric) code uses high-resolution
shock-capturing (HRSC) finite-difference schemes for the numerical
integration of the general relativistic hydrodynamic equations
\cite{Leveque92} (see \cite{jcam} for a recent review of applications
of HRSC schemes in relativistic hydrodynamics). In a similar context
to the one presented here let us note that such schemes have been
succesfully used before in the study of the numerical evolution and
gravitational collapse of non-rotating neutron stars in 1-D
\cite{Romero}. An alternative approach, based on pseudospectral
methods, has been presented in \cite{eric1}.
Using our code in 1-D time-evolutions we can accurately identify
specific normal modes of pulsation. In 2-D the code is suitable for
the evolution of rotating stars, with the additional complication of
having to pay special attention to an angular momentum-loss at the
(non-spherical) surface of the star, as we will show below.
\section{Initial Configurations}
\label{initial}
Our initial models are fully relativistic, stationary and axisymmetric
configurations, rotating with uniform angular velocity $\Omega$. The
metric in quasi-isotropic coordinates \cite{Butterworth} is
\begin{eqnarray}
ds^2=-e^{2\nu}dt^2 + B^2e^{-2\nu}r^2\sin^2\theta(d\phi-\omega dt)^2
+e^{2\alpha}(dr^2+r^2d\theta^2),
\label{metric}
\end{eqnarray}
where $\nu$, $B$, $\alpha$ and $\omega$ are metric functions
(gravitational units
are implied). In the non-rotating limit the above metric reduces to
the metric of spherical relativistic stars in isotropic coordinates.
We assume a perfect fluid, zero-temperature equation of
state (EOS), for which the energy density is a function of pressure
only. The following relativistic generalization of the Newtonian
polytropic EOS is chosen:
\begin{eqnarray}
p&=&K\rho_0^{1+1/N}
\\
\epsilon&=& \rho_0 +Np,
\end{eqnarray}
\noindent
where $p$ is the pressure, $\epsilon$ is the energy density, $\rho_0$ is the
rest-mass density, $K$ is the polytropic constant and $N$ is the
polytropic exponent.
The initial equilibrium models are computed using a numerical code by
Stergioulas \& Friedman \cite{nik2} which follows the Komatsu,
Eriguchi \& Hatchisu \cite{keh} method (as modified in
\cite{cst}) with some changes for improved accuracy (see \cite{NSGE}
for a comparison with other codes). The code is freely available and
can be downloaded from the following URL address:
http://www.gravity.phys.uwm.edu/Code/rns.
\section{Relativistic Hydrodynamic Equations}
\label{hydro}
The equations of (ideal) relativistic hydrodynamics are obtained from
the local conservation laws of density current, $J^{\mu}$ and
stress-energy, $T^{\mu\nu}$
\begin{eqnarray}
\nabla_{\mu}J^{\mu}&=&0
\\
\nabla_{\mu}T^{\mu \nu}&=&0
\end{eqnarray}
\noindent
with
\begin{eqnarray}
J^{\mu}&=&\rho_0u^{\mu}
\\
T^{\mu \nu}&=&\rho_0hu^{\mu}u^{\nu} + p g^{\mu \nu},
\end{eqnarray}
\noindent
for a general EOS $p=p(\rho,\varepsilon)$. This choice of the
stress-energy tensor limits our study to perfect fluids.
In the previous expressions $\nabla_{\mu}$ is the covariant derivative,
$u^{\mu}$ is the fluid 4-velocity and $h$ is the specific enthalpy
\begin{eqnarray}
h=1+\varepsilon+\frac{p}{\rho_0}
\end{eqnarray}
\noindent
with $\varepsilon$ being the specific internal energy, related to
the energy density $\epsilon$ by
\begin{eqnarray}
\varepsilon=\frac{\epsilon}{\rho_0}-1.
\end{eqnarray}
\noindent
With an appropriate choice of matter fields the equations of
relativistic hydrodynamics constitute a (non-strictly) hyperbolic
system and can be written in a flux conservative form, as was first
shown in \cite{mim} for the
one-dimensional case. The knowledge of the characteristic fields of
the system allows the numerical integration to be performed by means
of advanced high-resolution shock-capturing (HRSC) schemes,using
approximate Riemann solvers (Godunov-type methods). The
multidimensional case was studied in \cite{betal97},
within the framework of the 3+1 formulation. Further
extensions of this work to account for {\it dynamical} spacetimes,
described by the full set of Einstein's non-vacuum equations, can be
found in \cite{font2}. Fully {\it covariant}
formulations of the hydrodynamic equations (i.e., not restricted to
{\it spacelike} approaches) and also adapted to Godunov-type methods,
are presented in \cite{pf,fp}.
In the present work we use the hydrodynamic equations as formulated in
\cite{betal97}. Specializing for the metric given by
Eq.~(\ref{metric}), the 3+1 quantities read
\begin{eqnarray}
\tilde{\alpha} &=& e^{\nu}
\\
\beta_{\phi} &=& -\omega B^2 e^{-2\nu} r^2 \sin^2\theta
\\
\gamma_{rr} &=& e^{2\alpha}
\\
\gamma_{\theta\theta} &=& r^2 e^{2\alpha}
\\
\gamma_{\phi\phi} &=& B^2 e^{-2\nu} r^2 \sin^2\theta
\end{eqnarray}
\noindent
where $\tilde{\alpha}$ is the lapse function (the tilde is
used to avoid confussion with the metric potential $\alpha$) and
$\beta_{\phi}$ is the azimuthal shift.
The hydrodynamic equations are written as a first-order flux
conservative system of the form
\begin{eqnarray}
\frac{\partial {\bf u}}{\partial t} +
\frac{\partial \tilde{\alpha} {\bf f}^r} {\partial r} +
\frac{\partial \tilde{\alpha} {\bf f}^{\theta}} {\partial \theta} =
{\bf s}
\label{hydro_system}
\end{eqnarray}
\noindent
where ${\bf u}, {\bf f}^r, {\bf f}^{\theta}$ and ${\bf s}$ are,
respectively, the state vector of evolved quantities, the radial and polar
fluxes and the source terms. More precisely, they take the form
\begin{eqnarray}
{\bf u} &=& (D,S_r,S_{\theta},S_{\phi},\tau)
\\
{\bf f}^r &=& (Dv^r, S_rv^r+p, S_{\theta}v^r, S_{\phi}v^r, (\tau+p)v^r)
\\
{\bf f}^{\theta} &=& (Dv^{\theta}, S_rv^{\theta}, S_{\theta}v^{\theta}+p,
S_{\phi}v^{\theta}, (\tau+p)v^{\theta}).
\end{eqnarray}
\noindent
The source terms can be decomposed in the following way
\begin{eqnarray}
{\bf s} = \tilde{\alpha} {\bf s}^{\star} -
\tilde{\alpha} {\bf f}^r \frac{\partial \log\sqrt{\gamma}}{\partial r}
-\tilde{\alpha} {\bf f}^{\theta} \frac{\partial \log\sqrt{\gamma}}
{\partial \theta}
\end{eqnarray}
\noindent
with $\gamma=\det\gamma_{ij}$ and
\begin{eqnarray}
{\bf s}^{\star} = \left(0, T^{\mu\nu}\left[\frac{\partial g_{\nu j}}
{\partial x^{\mu}} - \Gamma^{\delta}_{\mu\nu}g_{\delta j}\right],
\tilde{\alpha}\left[T^{\mu t}\frac{\partial\log\tilde{\alpha}}
{\partial x^{\mu}} - T^{\mu\nu}\Gamma^t_{\mu\nu}\right]\right)
\end{eqnarray}
\noindent
with $j=r,\theta,\phi$. The definitions of the evolved quantities in terms
of the ``primitive" variables ($\rho,v_j,\varepsilon$) are
\begin{eqnarray}
D &=& \rho_0 W
\\
S_j &=& \rho_0 h W^2 v_j
\\
\tau &=& \rho_0 h W^2 - p - D
\end{eqnarray}
\noindent
where $W$ is the relativistic Lorentz factor
\begin{eqnarray}
W \equiv \tilde{\alpha} u^t = \frac{1}{\sqrt{1-v^2}}
\end{eqnarray}
\noindent
with $v^2=\gamma_{ij}v^iv^j$. The 3-velocity components are obtained
from the spatial components of the 4-velocity in the following way
\begin{eqnarray}
v^i=\frac{u^i}{W}+\frac{\beta^i}{\tilde{\alpha}}.
\end{eqnarray}
Explicit expressions for the
non-vanishing Christoffel symbols for metric~(\ref{metric}), appearing
in the source terms of the hydrodynamic equations, are presented in \cite{fsk}.
\section{Numerical Methods}
As stated before, our numerical integration of system
(\ref{hydro_system}) is based on Godunov-type methods (also known as
HRSC schemes). In a HRSC scheme, the knowledge of the characteristic
fields (eigenvalues) of the equations, together with the corresponding
eigenvectors, allows for accurate integrations, by means of either
exact or approximate Riemann solvers, along the fluid characteristics.
These solvers, which constitute the kernel of our numerical algorithm,
compute, at every interface of the numerical grid, the solution of
local Riemann problems (i.e., the simplest initial value problem with
discontinuous initial data). Hence, HRSC schemes automatically
guarantee that physical discontinuities appearing in the solution,
e.g., shock waves, are treated consistently (the {\it shock-capturing}
property). HRSC schemes are also known for giving stable and sharp
discrete shock profiles. They have also a high order of accuracy,
typically second order or more, in smooth regions of the solution.
We perform the time update of system (\ref{hydro_system})
according to the following conservative algorithm:
\begin{eqnarray}
{\bf u}_{i,j}^{n+1} = {\bf u}_{i,j}^{n}
& - & \frac{\Delta t}{\Delta r}
(\widehat{{\bf f}}_{i+1/2,j}-\widehat{{\bf f}}_{i-1/2,j})
\nonumber \\
& - & \frac{\Delta t}{\Delta \theta}
(\widehat{{\bf g}}_{i,j+1/2}-\widehat{{\bf g}}_{i,j-1/2})
+ \Delta t {\bf s}_{i,j} \, .
\end{eqnarray}
\noindent
Index $n$ represents the time level and the time (space)
discretization interval is indicated by $\Delta t$ ($\Delta r, \Delta\theta$). The
``hat" in the fluxes is used to denote the so-called numerical fluxes
which, in a HRSC scheme, are computed according to some generic
flux-formula, of the following functional form (suppressing index
$j$):
\begin{eqnarray}
\widehat{{\bf f}}_{i\pm{1\over 2}} = \frac{1}{2}
\left( {\bf f}({\bf u}_{i\pm{1\over 2}}^{L}) +
{\bf f}({\bf u}_{i\pm{1\over 2}}^{R}) -
\sum_{\alpha = 1}^{5} \mid \widetilde{\lambda}_{\alpha}\mid
\Delta \widetilde {\omega}_{\alpha}
\widetilde {r}_{\alpha} \right) \, .
\end{eqnarray}
\noindent
Notice that the numerical flux is computed at cell interfaces
($i\pm1/2$). Indices $L$ and $R$ indicate the left and right sides of a
given interface. Quantities $\lambda$, $\Delta\omega$ and $r$ denote the
eigenvalues, the jump of the characteristic variables and the
eigenvectors, respectively, computed at the cell interfaces according
to some suitable average of the state vector variables. Generic
expressions for the characteristic speeds and eigenfields can be found
in \cite{font2}. Our code has the ability of using different
approximate Riemann solvers: the Roe solver \cite{Roe81}, widely
employed in fluid dynamics simulations, with arithmetically averaged
states and the Marquina solver \cite{donat96}, which has been extended
to Relativity in \cite{donat98}. The computations presented here were
obtained using Marquina's scheme.
A technical remark: the equilibrium star is supplemented by a
low-density uniform atmosphere, which is necessary for computing
non-singular solutions of the hydrodynamic equations everywhere in the
computational domain. After each time-step we reset the atmosphere's
density and pressure to their initial values, avoiding unwanted
accretion of matter onto the star. The influence of the atmosphere is
thus restricted to the surface grid-cells.
\section{Pulsations of Non-rotating Stars}
\label{nonrot}
Since our code uses spherical polar coordinates, it can also be employed
to study the evolution of non-rotating stars in 1-D. In the evolution
of initially static non-rotating stars, we observe the following
properties (note that our numerical grid is Eulerian):
\begin{enumerate}
\item Small-amplitude radial pulsations are triggered by the
truncation errors of the finite-differencing scheme.
\item The radial pulsations are dominated by a set of discrete
frequencies, which correspond to the normal modes of pulsation of
the star.
\item The numerical viscosity of the finite-difference scheme damps
the pulsations and the damping is stronger for the higher frequency
modes.
\item The presence of a constant density atmosphere affects the
finite differencing at the surface grid-cells, which increases
the numerical damping of pulsations and also causes a
continuous but very small drifting of the density distribution.
\end{enumerate}
The initial amplitude of the radial pulsations and the small drift in
density converge to zero at a second order rate with increasing resolution.
The value of the density in the atmosphere region has a large effect on the
damping of the pulsations. If it is too large, the damping is strong.
To minimize this effect, we typically set the density of the
atmosphere equal to $10^{-6}$ times the density of the last grid point
inside the star.
\subsection{1-D evolutions}
We study the numerical evolution of a nonrotating $N=1.5$ relativistic
polytrope with $M/R=0.056$. The star is immediately set into radial
pulsation, triggered by the finite difference truncation errors. The
time-evolution of the radial velocity $v_r$, computed at 25\% of the
star radius using a grid of 400 zones, is shown in Fig. \ref{fig_nonrot1}.
The vertical axis is dimensionless ($c=G=M_\odot=1$). The radial velocity
is initially a very complex function of time. As we will show next,
the pulsation consists mainly of a superposition of normal modes of
oscillation of the fluid. The high frequency normal modes are damped
quickly and after 20ms the star pulsates mostly in lowest frequency
modes. Because these oscillations are caused only by the truncation
errors, the magnitude of the radial velocity is extremely small. As
shown in Fig. \ref{fig_nonrot1} it is only a few times larger than
the non-zero residual velocity around which the star is oscillating.
This residual velocity converges to zero as second order with
increased resolution.
\begin{figure}
\centering
\epsfig{file=velr_sph1.eps,width=9cm}
\caption{Evolution of the radial velocity of an initially static,
non-rotating star, caused by the truncation errors of the
finite-difference scheme. The radial pulsations are mainly a
superposition of normal modes of the star.}\label{fig_nonrot1}
\end{figure}
The small-amplitude radial pulsations in the non-linear, fixed
spacetime evolutions correspond to linear normal modes of pulsation in
the relativistic Cowling approximation, in which perturbations of the
spacetime are ignored. A Fourier transform of the density or radial
velocity time-evolution can be used to identify the normal mode frequencies.
Fig. \ref{fig_nonrot2} shows the Fourier transform of the radial velocity
evolution shown in Fig. \ref{fig_nonrot1}. The normal
mode frequencies stand out as sharp peaks on a continuous background.
The width of the peaks increases with frequency. The frequencies of
radial pulsations identified from Fig. \ref{fig_nonrot2} are shown in
Table \ref{tab_radial1}.
\begin{figure}
\centering
\epsfig{file=velr_sph1_fft.eps,width=9cm}
\caption{Fourier transform of the evolution of the radial velocity in
Fig. \ref{fig_nonrot1}. The frequencies are in excellent agreement
with linear normal mode frequencies computed with an eigenvalue
code. The units of the vertical axis are arbitrary.}
\label{fig_nonrot2}
\end{figure}
To compare the obtained frequencies to linear normal mode frequencies,
we use a different code that solves the linearized relativistic
pulsation equations for the stellar fluid, in the Cowling
approximation \cite{McDermott,LindblomSplinter,YoshidaE}, as an
eigenvalue problem. In Table \ref{tab_radial1} we present the results
of this comparison. The typical agreement between frequencies
computed by the two methods is better than 0.5\% for the fundamental
$F$-mode and the lowest frequency harmonics $H_1-H_4$ and better than
0.8\% for the the higher harmonics $H_5-H_9$. This is a strong test
for the accuracy of the evolution code and our results can be used as
a testbed computation for other relativistic multi-dimensional
evolution codes.
\begin{table}[t]
\begin{center}
\begin{tabular}{*{4}{r}}
\multicolumn{4}{c}{}\\
\multicolumn{4}{c}{\large \bf Radial Pulsation Frequencies}\\
\multicolumn{4}{c}{}\\
\hline
Mode & non-linear code (kHz) & Cowling (kHz) & difference \\[0.5ex]
\hline
\\[0.5ex]
$F$ & 1.703 & 1.697 & 0.3\% \\[0.5ex]
$H_1$ & 2.820 & 2.807 & 0.5\% \\[0.5ex]
$H_2$ & 3.862 & 3.868 & 0.02\% \\[0.5ex]
$H_3$ & 4.900 & 4.910 & 0.2\% \\[0.5ex]
$H_4$ & 5.917 & 5.944 & 0.4\% \\[0.5ex]
$H_5$ & 6.930 & 6.973 & 0.6\% \\[0.5ex]
$H_6$ & 7.947 & 8.001 & 0.7\% \\[0.5ex]
$H_7$ & 8.960 & 9.029 & 0.8\% \\[0.5ex]
$H_8$ & 9.973 & 10.057 & 0.8\% \\[0.5ex]
$H_9$ & 11.030 & 11.086 & 0.5\% \\[0.5ex]
\end{tabular}
\vspace{3mm}
\caption{Comparison of small-amplitude radial pulsation frequencies
obtained with the present non-linear evolution code to linear
perturbation mode frequencies in the relativistic Cowling
approximation. The equilibrium model is a nonrotating $N=1.5$
relativistic polytrope with $M/R=0.056$.}
\label{tab_radial1}
\end{center}
\end{table}
\begin{table}[t]
\begin{center}
\begin{tabular}{*{4}{r}}
\multicolumn{4}{c}{}\\
\multicolumn{4}{c}{\large \bf Quadrupole Pulsation Frequencies}\\
\multicolumn{4}{c}{}\\
\hline
Mode & non-linear code (kHz) & Cowling (kHz) & difference \\[0.5ex]
\hline
\\[0.5ex]
$f$ & 1.28 & 1.286 & 0.5\% \\[0.5ex]
$p_1$ & 2.68 & 2.681 & 0.04\% \\[0.5ex]
$p_2$ & 3.65 & 3.699 & 1.3\% \\[0.5ex]
$p_3$ & 4.66 & 4.719 & 1.3\% \\[0.5ex]
$p_4$ & 5.66 & 5.742 & 1.4\% \\[0.5ex]
$p_5$ & 6.83 & 6.764 & 1.0\% \\[0.5ex]
$p_6$ & 7.80 & 7.788 & 0.2\% \\[0.5ex]
\end{tabular}
\vspace{3mm}
\caption{Comparison of small-amplitude quadrupole ($l=2$) pulsation
frequencies obtained with the present non-linear evolution code to
linear perturbation mode frequencies in the relativistic Cowling
approximation. The equilibrium model is a nonrotating $N=1.5$
relativistic polytrope with $M/R=0.056$.}
\label{tab_quad1}
\end{center}
\end{table}
\subsection{2-D evolutions}
In a similar way, small-amplitude non-radial pulsations can be studied
with the present evolution code and the obtained frequencies can be
compared to perturbation results. We find that the truncation errors
of the finite difference scheme do not excite non-radial pulsations to
a sufficiently large amplitude compared to the amplitude of radial
pulsations, so that one cannot identify them accurately in a Fourier
transform. Instead, one has to perturb the initial configuration,
using an appropriate eigenfunction for each nonradial angular index
$l$. Such a perturbation can be constructed using the eigenfunctions
of linear pulsation modes, computed with the perturbation code in the
Cowling approximation. The frequencies of the non-radial modes are
then found from a Fourier transform of the time-evolution of the
velocity component $v_\theta$.
Table \ref{tab_quad1} shows a similar comparison as in Table
\ref{tab_radial1} for the quadrupole ($l=2$) pulsations of the same
$N=1.5$ relativistic polytrope. Since the non-radial modes have to be
computed on a 2-D grid, we cannot use resolutions as high as in the
1-D computations. For a small grid-size of $80\times80$ zones and a total
evolution time of 6ms, the agreement between frequencies computed by
the two methods is better than 1.4\% for the fundamental $f$-mode and
the $p$-modes $p_1-p_6$. For this grid-size, frequencies higher than
the $p_6$ mode could not be computed accurately, because the grid is
to coarse to resolve their eigenfunctions (higher harmonic
eigenfunctions have a larger number of nodes in the radial direction).
\section{Rotating Stars}
We now turn to the evolution of initially stationary, uniformly
rotating neutron stars. In these evolutions, we observe the same
qualitative properties as for non-rotating stars (section
\ref{nonrot}) and an additional important property: {\it the angular
momentum of the star is not conserved at the surface layer}. This is
due to the fact that the velocity component $v_\phi$ of the fluid
has a maximum at the surface, while the numerical scheme (although
second-order accurate in smooth regions of the solution) is only
first-order accurate at local extrema. Moreover, the code evolves the
relativistic momenta, $S_i$, and the velocity components (as well as
the rest of ``primitive" variables) must be recovered through a root
finding procedure which involves dividing by the density. At the
surface of the star (where the density is very small) this contributes
to obtaining less than second-order accuracy.
\begin{figure}
\centering
\epsfig{file=velp_e_xl_proc.eps,width=9cm}
\caption{Time-evolution of the velocity component $v_\phi$ of a rotating star
(see text for details). The accuracy is second order in the interior
but only first order at the surface. This results in an angular
momentum loss of the surface layers.}\label{f_rot1}
\end{figure}
A representative example of the evolution of a rotating star is
presented in Fig. \ref{f_rot1}, which shows the evolution, at
different times, of the velocity component $v_\phi$. The star is again a $N=1.5$
polytrope with the same central density as the non-rotating star
presented in section \ref{nonrot} and rotating at 74\% of the
mass-shedding limit at same central density. The evolution was for
one rotation period on a $96\times60$ grid. The vertical axis is
dimensionless ($c=G=M_\odot=1$). The figure shows that the $\phi$-velocity
in the interior of the star remains close to its initial value, while
it decreases as a function of time in the outer layers. We find that
this is a generic property of the present numerical scheme for any
rotation rate and for any grid-size. By comparing evolutions with
different grid sizes, we verified that the loss of angular momentum at
the surface improves as first-order with resolution, while the
evolution of the $\phi$-velocity in the interior is second-order
accurate. However, as the evolution proceeds in time, the first-order
surface effect gradually affects the interior of the star.
\section{Quasi-radial Modes of Rotating Stars}
As a first application of our code, we compute quasi-radial modes
(i.e. modes that in the non-rotating limit reduce to radial modes) of
rapidly rotating relativistic stars in the Cowling approximation.
Previously, these modes have been computed for fully relativistic
stars only in the slow-rotation limit (but without the assumption of a
fixed spacetime) by Hartle \& Friedman \cite{Hartle} (see also
\cite{Datta}). We compute the three lowest-frequency quasi-radial
modes for a sequence of rotating stars of same central density. The
non-rotating member of the sequence is the non-rotating star of
section \ref{nonrot}. Table \ref{qr_tab} and Fig. \ref{f_rot2} show
our results for a low resolution grid of $100\times 80$ zones (note that our
computational grid assumes equatorial plane symmetry). For
this resolution we estimate the accuracy of the frequencies to be of
the order of $1-2$\%.
For the sequence of stars considered here, the frequencies of the
quasi-radial modes decrease with increasing rotation rate. This agrees
with previous slow-rotation computations which predict a decrease as
$\Omega^2$, where $\Omega$ is the angular velocity of the star. For fast
rotation, the change in the frequencies of quasi-radial modes is
affected by higher order terms in $\Omega$, because of the large
deformation of the equilibrium star. Also, for rapidly rotating stars
the quasi-radial mode frequencies are more ``closely packed'' than in
non-rotating stars.
\begin{table}[t]
\begin{center}
\begin{tabular}{*{4}{r}}
\multicolumn{4}{c}{}\\
\multicolumn{4}{c}{\large \bf Quasi-radial Pulsation Frequencies}\\
\multicolumn{4}{c}{}\\
\hline
$\Omega / \Omega_K$ & F (kHz) & H1 (kHz) & H2 (kHz) \\[0.5ex]
\hline
\\[0.5ex]
0.0 & 1.71 & 2.82 & 3.90 \\[0.5ex]
0.32 & 1.70 & 2.77 & 3.81 \\[0.5ex]
0.44 & 1.67 & 2.71 & 3.68 \\[0.5ex]
0.62 & 1.64 & 2.52 & 3.46 \\[0.5ex]
0.74 & 1.53 & 2.38 & 3.33 \\[0.5ex]
0.84 & 1.36 & 2.25 & 3.17 \\[0.5ex]
\end{tabular}
\vspace{3mm}
\caption{Frequencies of quasi-radial modes for different values of the ratio of
angular velocity
of the star $\Omega$ to the angular velocity at the mass-shedding limit
$\Omega_K$, for a sequence of rotating relativistic stars of same
central density. The non-rotating member of the sequence is the same
as in Table \ref{tab_radial1}.}
\label{qr_tab}
\end{center}
\end{table}
\begin{figure}
\centering
\epsfig{file=quasi_rad.eps,width=9cm}
\caption{Frequencies of the lowest three quasi-radial modes vs.
the ratio of angular velocity of the star $\Omega$ to the angular
velocity at the mass-shedding limit $\Omega_K$, for the sequence of
rotating relativistic stars in Table \ref{qr_tab}.}
\label{f_rot2}
\end{figure}
\section{Discussion}
Our axisymmetric relativistic hydrodynamical code is capable to evolve
rapidly rotating stars in a fixed spacetime. We find that, for non-rotating
stars, small amplitude oscillations have frequencies that agree with
linear normal mode frequencies in the Cowling approximation and we
compute the quasi-radial modes of rapidly rotating stars.
Modern HRSC numerical schemes (as the ones used in our code), satisfying the
``total variation diminishing" (TVD) property \cite{H84}, are
second-order accurate in smooth regions of the flow, but only first-order
accurate at local extrema. In our rotating stars runs we find that this
results in a loss of
angular momentum of the surface layers of the star, which gradually
also affects the interior of the star. This angular momentum loss only
vanishes as first-order with incresing resolution and we thus conclude
that for accurate long-term evolutions of rotating neutron stars it is
essential to use rather fine grids. Furthermore, to reduce the
computational cost, one could use surface-adapted coordinates or
fixed-mesh refinement. It would also be interesting to see whether the
loss of angular momentum per rotation period will be significantly
smaller in a frame co-rotating with the star. An alternative solution
to this problem, which we plan to investigate, could be the use of
``essentially non-oscillatory" (ENO) schemes, which maintain
high-order of accuracy even at local extrema \cite{HO87}.
All previous considerations are important for the study of the non-linear
dynamics of unstable toroidal oscillations ($r$-modes) in 3-D, which
have a long growth time and thus require highly accurate long-term
evolutions.
\section*{Acknowledgements}
We thank John L. Friedman, Curt Cutler, Philippos Papadopoulos and Tom
Goodale for helpful discussions. We also thank S. Yoshida for sending
us for comparison unpublished results on quasi-radial modes of
rotating stars in the Cowling approximation, computed with a linear
perturbation code. J.A.F acknowledges financial support from a TMR
grant from the European Union (contract nr. ERBFMBICT971902). K.D.K.
is grateful to the Max-Planck-Institut f\"ur Gravitationsphysik
(Albert-Einstein-Institut), Potsdam, for generous hospitality.
\section*{References}
|
\section{Introduction}
It is usual in cosmology to consider
the standard Universe as spatially homogeneous and isotropic
on the largest scales. In fact, there is very good observational support for
doing so. However, we also know that inhomogeneities
exist at almost all scales -- the smaller the scale the larger the
inhomogeneity.
All these inhomogeneities are apparently insignificant in cosmology, as long
as we are not interested in modelling structure formation. It is therefore
usual
and considered acceptable by the vast majority of researchers
to ignore them in investigating the dynamics and geometry of the
Universe as a whole. Nevertheless, Ellis \cite{Ellis} has given compelling
reasons why conceptually the large-scale cosmological metric should really be
an average over very large regions of space-time. According to this view,
the metric $g_{\mu \nu}$ and the Einstein field equations
\begin{equation}
G_{\mu \nu} \equiv R_{\mu \nu} - \frac{1}{2}g_{\mu \nu} R = 8 \pi T_{\mu \nu},
\end{equation}
where $R_{\mu \nu}$ is the Ricci tensor and $R$ is the Ricci scalar, both
depending on second-order derivatives of $g_{\mu \nu}$, and $T_{\mu \nu}$ is
the stress-energy tensor, must be averaged over
small and intermediate local
inhomogeneities on larger and larger scales to
obtain the average cosmological metric and the averaged
dynamical equations. \\
However,
averaging and operating with the Einstein differential operator on a metric
do not commute, because of the nonlinearity of the operator. Thus, the
solution to the averaged Einstein equations will {\it not} be the averaged
metric. The averaged metric will, therefore,
obey equations
different from the averaged Einstein equations.
In general, the smoothing-out operation will introduce
extra tensor terms in the field equations, which may affect
the dynamics and the energy conditions in the averaged universe.
Furthermore and just as importantly, it is a very complicated and unresolved
issue to define an adequate averaging
scheme with the necessary properties -- including the uniqueness of the
averaged objects and their at least approximate tensorial character.
This difficulty arises,
because, in general, integrating a tensor field does not yield another tensor
field in a curved space-time. \\
In spite of these difficulties and uncertainties,
several averaging procedures have been proposed
by Isaacson \cite{Isaac}, Noonan \cite{Noonan1,Noonan2},
Zotov and Stoeger \cite{Zotov}, and more recently
by Boersma \cite{Boersma}.
Their different philosophies and results, together with a comparison with the
averaging scheme we shall introduce here,
are briefly described and discussed
below. Although far from the object of the present paper,
it is also worth mentioning the
ideas developed by Zalaletdinov in constructing his theory
of macroscopic gravity \cite{Zalaletdinov}.
He has used the concept of duality -- the existence of two types of observers,
microscopic and macroscopic -- for the study of classical physical phenomena.
Relying on this, he constructed a theory based on averaging a curved
space-time itself
and then determined the geometrical objects (metric, connection and
curvature)which describe the averaged space-time \cite{Zalaletdinov}.
This approach is non-pertubative in nature.\\
In this investigation we define an averaging scheme,
which is a straight-forward
generalization of one often used in macroscopic electromagnetic
theory \cite{Jackson},
carefully examine its properties, particularly for weak fields
and for perturbations, and show that it is also an improvement of
Noonan's averaging procedure
\cite{Noonan1,Noonan2}. The behavior of our averaging scheme
in the cases in which the linear approximation
is sufficient is acceptable. In these
situations the noncommutability of averaging and operating on the
metric with the relevant differential operator is replaced by commutability,
yielding simple, almost trivial results. However, these provide an essential
reference for evaluating our averaging procedure and for reaching firm
conclusions concerning the
proper interpretation of and constraints on cosmological averaging in general,
as well as interesting and important applications.
Since much
of cosmology involves perturbation treatments, we believe it is crucial to
be clear about
the averaging operator and its results in this context first, in order to be
able to understand the cases in which nonlinear effects become important. \\
Next, we discuss in detail the general properties of our averaging
operator --
in particular its approximately tensorial character in cosmological
coordinates and for general coordinates in the weak field case, and
its apparent lack of uniqueness with respect to local coordinates. \\
Penultimately, we apply our averaging procedure to the important case of
perturbed
Friedmann-Lema\^{\i}tre-Robertson-Walker (FLRW)
cosmologies, showing that averaging an exact FLRW space-time yields, to first
order, in the local coordinates, an FLRW space-time, and further that
averaging a perturbed FLRW cosmology yields, as expected, another perturbed
FLRW cosmology. We then discuss in some detail, from the averaging point
of view, what type of inhomogeneities in an FLRW background may be considered
as perturbations of an FLRW metric. Our linear formalism is able to handle
deviations for which the local dynamics are not completely decoupled from the
general expansion of the universe. In contrast, if the local inhomogeneities
are not in the linear regime, then a first order perturbative approach cannot
be applied. Nevertheless, our general averaging scheme is still valid, but the
resulting metric is not a simple superposition of FLRW plus perturbations. \\
Finally, we show how our averaging procedure may be approximately implemented
in this nonperturbative case in simple situations. The procedure used by
Zotov and Stoeger \cite{Zotov}, applied throughout a space at each point,
essentially accomplishes in an easily implementable way what our averaging
procedure requires.
\section{The Averaging Procedure}
\subsection{Definition}
We define an averaging operator, or simply, an {\it averager}, acting on a
field $Q$ as
\begin{equation}
<Q(x)>=\frac{{\displaystyle \int \! Q(x+x')\, \sqrt{-g(x + x')}
\,d\Omega '}}{{\displaystyle \int \! \sqrt{-g(x + x')} \,d\Omega '}}.
\end{equation}
This is a simple extension of the definition used by Jackson
\cite{Jackson}, for macroscopic electromagnetic fields.
Here $Q$ is any field -- it may be a scalar, a vector, or a tensor. The
coordinate $x$
gives the cosmic, large-scale location of the averaging volume in space-time,
and $x'$ gives the small-scale
location of a
point within the averaging volume $\Omega'$ relative to its location $x $
(its center, if it is a sphere) in the space-time, both expressed in
Minkowski (rectangular) coordinates. Throughout this paper we often write
$x$ and $x'$ for the cosmological and local
coordinates, respectively, and
$x + x'$ for their sum, by which we mean $x^{\mu}$, $x'^{\mu}$ and
$x^{\mu} + x'^{\mu}$, respectively.
The metric of the space-time, for the observer
who sees the inhomogeneities, is $g_{\mu \nu}(x+x')$ at
the point $x + x'$. This is why the determinant of the metric $g$ in the
integrands of
Eq. (2) depends
on $x + x'$. Though each
integration is only over the local regions of space-time dominated by the
inhomogeneities, there are innumerable such averaging volumes spread over the
observable universe -- one for each cosmological coordinate $x$ -- and the
metric within each one depends, of course, not just on the local coordinate
$x'$ but also on the cosmic location of the the averaging volume, given by
the cosmological coordinate $x$. Obviously, the result of the averaging will
depend strongly on the length scale over which it is performed.
In general, that will be larger than, or of the order of, the
characteristic length scale of the inhomogeneities over which we want to
average. \\
This definition is not invariant under a change of
coordinates. It can only be implemented as such in a coordinate system like
Minkowski's in which the cosmological coordinates and the local coordinates
are ``parallel'', or translationally related. Nevertheless, as we shall see,
our averager has some very nice properties. In particular, it is
almost invariant under transformations of the cosmological coordinates, and
under transformations of the complete coordinate $x + x'$ in the weak field
and perturbation cases. \\
We would like to stress that
the quantities that are averaged are always referred to the same scale.
In the process of averaging,
the only observer involved is the one at the local `small
scale', which means that we are treating the problem in a self-consistent way.
In other considerations one might have another observer -- a cosmic
observer -- who does not
see the inhomogeneities. In those cases, we would need to compare
the results of the latter with the averaged ones. \\
There are two motivations for choosing this procedure for averaging. The
first is simply that it implements what we intuitively envision as
averaging over a given length scale, providing
an assignment of an average
value of a quantity to every point $x$ throughout the space-time manifold
taking into consideration the possibly different values of the metric
in different cosmological locations, and specifying unambiguously the
relationship between the large-scale cosmological coordinate system and the
local, small-scale coordinate system, over which the averages are carried out.
One simply takes the averaging volume and shifts it from point to point
throughout the universe, averaging at each
point according to Eq. (2). This is
precisely what is done in making the transition from
microscopic to macroscopic
electromagnetism
\cite{Jackson}, as we have indicated above. Secondly, as we
can explicitly show (see Section IV), the averaging procedure defined in
Eq. (2) yields quantities which are approximately tensorial in character
with respect to the cosmological coordinates $x$, and with respect to the full
coordinates $x + x'$ for weak gravitational fields. That is,
if $Q(x + x')$ is
a tensor with respect to
$x + x'$, $<Q(x)>$ is almost a tensor of the same type with respect to $x$, and
with respect to $x + x'$ for weak fields and perturbed space-times --
deviating from a true tensor by only very small quantities. This is
a very attractive property of our averager. \\
At the same time, there are some other aspects of our averager which, at first
sight, are not
so attractive and require further explanation. First, as we have already
mentioned, the
relationship between the cosmological and the local coordinate systems, as
formally expressed in Eq. (2), can only
be realized in a very retricted class of coordinate systems
-- those in which they can be related to
one another by simple translation, so that the ``complete'' coordinate of
any point can be expressed simply as the sum of the cosmological coordinates
and the local coordinates at that point.
Most coordinate systems will not fulfil this requirement.
This may not be a problem, since, within the demands of coordinate covariance,
we are allowed to choose any coordinate system we want. In particular, any
coordinate system which simplifies the formulation of our problem may be
selected -- as long as the quantities in question are tensors, or nearly
tensors, thus assuring us that they are equivalent to their forms in more
complicated coordinate systems.
For other choices we could also specify the relation
between the cosmological coordinates and the local
coordinates, but the relationship between the two would, in general, be much
more complicated -- not specifiable by a simple sum of the two coordinates,
that is by simple translation. \\
This relation between the cosmological coordinates and the local coordinates
leads another potential difficulty. Although it is clear that
the average we have defined in Eq. (2)
is approximately covariant with
respect to the cosmological coordinates
\cite{Nelson}, as we shall show in Section V,
strictly speaking a transformation of the
cosmological coordinates $x$ in the integrals of
Eq. (2) should be accompanied by a transformation of the
$x'$ coordinates, along with an
induced change in their functional relationship (from a simple sum to something
more complicated), as indicated above. However, even then the
averaging over the local coordinates would still not be invariant under those
changes. In light of the arguments given by Ellis and
Matravers \cite{EllisMatr} and
Ellis, Matravers and Zalaletdinov \cite{EllMaZal} for using preferred coordinate
systems in general relativity which are appropriate to the physical situation
and simplify the problem, even though they break coordinate invariance,
we maintain that this lack of covariance of the averaging procedure
with respect to the local coordinates should not be considered an essential
problem in most cosmological applications. As we shall show, this is certainly
the case for
weak gravitational fields and for perturbations from an exact cosmological
solution (e.g. FLRW) to the field equations. In these cases
the deviations caused by the lack of coordinate covariance of the integrals
with respect to the local coodinates are small. Even for more general cases,
it is likely that these deviations will be small, as long as we are averaging
over volumes for which the space is almost flat -- for which the length
scale of the averaging volumes is smaller than the radius of curvature of the
universe. Thus, in that sense the most important property of our averaging
procedure is that it is covariant relative to transformations of the
cosmological coordinates. \\
Another way of expressing the above objection is that in carrying out our
averaging procedure we are effectively
adding vectors and tensors at different points, which is not a well-defined,
or even an allowed, operation on vectors and tensors
if the space-time is curved. We should really
incorporate bi-vectors in the averaging integrals in order to translate
vectors and tensors to the point designated by the cosmological metric and then
add them, as proposed by Isaacson \cite{Isaac}. The principal reason why we
have not done this is that there is no easy way of defining the needed
bi-vectors without already knowing the cosmological background metric, and
it is precisely this cosmological background metric that we eventually want
to determine
by carrying out the averaging over small scales. We discuss this in more
detail in the next subsection below. (Even though throughout most of this paper
we limit ourselves to cases in which we have a well-defined background -- the
weak-field and perturbed FLRW situations -- we want to be able to use our
averager in more general cases. We shall present one way of doing that
approximately in the last section.) Furthermore, as we have just discussed,
in the case of weak fields
and of almost flat averaging volumes, the errors introduced by neglecting
the bivectors will be small. Zalaletdinov \cite{Zalaletdinov} constructs his
macroscopic theory of gravity with bi-vectors, but these have to be solved
consistently with the other field equations. This is a possible way of
proceeding, but leads to such a complicated theory that solutions in the
simplest cases have yet to be obtained. \\
\subsection{Comparison with other definitions}
There are a number of other definitions of averaging which have been
suggested. We describe them briefly here, and compare them with ours, indicating
its advantages. \\
\begin{itemize}
\item{Noonan's Operator}
The averaging operator given by Noonan \cite{Noonan1,Noonan2} is
\begin{equation}
<Q^\prime>=\frac{{\displaystyle \int \! Q^\prime(x')\, \sqrt{-g(x')} \,d\Omega'}
}{{\displaystyle \int \! \sqrt{-g}\, d\Omega'}},
\end{equation}
where
\begin{itemize}
\item $\sqrt{-g}$ refers to the determinant of a macroscopic metric.
\item The region of integration is finite and its size is
bigger than the usual scale of the
microscopic observer and smaller than that of the macroscopic observer. Using
this fact the denominator may be approximated by $\int \! d\Omega$ in the
cases in which
the variation of $g$ is not abrupt.
\item The dependence of $<Q^\prime>$ on the
space-time coordinates comes from using indefinite integrals. This means
that the variable in the averager is now the boundary of the
integration volume -- which, we believe, is not well-defined for an averager.
\end{itemize}
The three points mentioned above represent the differences from our definition
of the averaging procedure. One possible weakness of Noonan's approach (for
details, see also \cite{Roque}) is the simultaneous introduction of two scales
in his averages: the metric in his definition is one of
large scale, whereas his averaging quantities are over small scales.
\item{Isaccson's Operator}
Isaacson's procedure of averaging \cite{Isaac} directly deals with the
problem that, in general, the result of integrating a tensor field
does not give another tensor, because tensors at different points have
transformation properties which depend on their location, as we have already
indicated. Since one can only add tensors at the same
point, the objective is to carry them to a certain common point and to add
them there. To do that, one has to introduce the {\it bi-vector of parallel
displacement} $j^{\beta}_{\alpha}(x,x^\prime)$ \cite{DeWitt,Synge,Grav}.
This object transforms
as a vector with respect to coordinate transformations at $x$ or at $x^\prime$,
and it has the property that,
given a vector (or tensor, in general)
$A_{\beta}$ at $x^\prime$, then
$A_\alpha(x)=j^{\beta}_{\alpha}(x,x^\prime) A_{\beta}(x^\prime)$
is the unique
vector at $x$ that can be obtained by parallel displacement of
$A_{\beta}$ from $x$ to $x^\prime$ along a geodesic. \footnote{This is
not true if there is more than one geodesic which joins the two
given points $x$ and $x^\prime$, because the transport along each different
geodesic varies. One may avoid such problems arguing that the two points
are close enough to each other as to permit the existence of only one geodesic
joining them.} \\
Given a tensor $T_{\mu\nu}$ Isaacson's averaging operator is defined as:
\begin{equation}
<T_{\mu\nu}(x)>=\int_{all\,space}
j^{\alpha^\prime}_{\mu}(x,x^\prime)
j^{\beta^\prime}_{\nu}(x,x^\prime) T_{\alpha^\prime \beta^\prime}(x^\prime)
f(x,x^\prime) d^4x^\prime
\end{equation}
where $f(x,x^\prime)$ is a weighting function which falls smoothly to zero
when $x$ and $x^\prime$ differ by a distance greater than $d$, its integral is
normalized
to one over all space-time.
This definition carries with it
all the properties of the bi-vector of parallel geodesic displacement,
for example
\begin{equation}
j^{\alpha^\prime}_{\mu}(x,x^\prime)
j^{\beta^\prime}_{\nu}(x,x^\prime)
g_{\alpha^\prime \beta^\prime}(x^\prime)=
g_{\mu \nu}(x).
\end{equation}
This implies that if we select $g_{\alpha^\prime \beta^\prime}(x^\prime)$
to be the small-scale metric, it is impossible to obtain an
averaged metric (corresponding to cosmological scales) different from
the assumed large-scale metric. That is,
\begin{eqnarray}
<g_{\mu\nu}(x)>=\int_{all\,space}
j^{\alpha^\prime}_{\mu}(x,x^\prime)
j^{\beta^\prime}_{\nu}(x,x^\prime) g_{\alpha^\prime \beta^\prime}(x^\prime)
f(x,x^\prime) d^4x^\prime=\nonumber \\
g_{\mu \nu}(x)
\int_{all\,space}
f(x,x^\prime) d^4x^\prime=
g_{\mu \nu}(x).
\end{eqnarray}
This represents a crucial problem for averaging, since, as we have stressed, we
ideally want to obtain metrics on larger scales as averages
of the small-scale metrics, without assuming the particular form of the
large-scale metrics.
\item{ Zotov and Stoeger Averaging}
Zotov and Stoeger first presented a simple procedure in which they
average over an
unbound distribution of stars --in a static background--
and of galaxies --in an expanding background
\cite{Zotov}.
They use an averaging scheme which recalls a three dimensional version of
finding a running mean with the simplest elementary volume: a sphere. Later,
using similar techniques they construct averages over hierarchies of
Swiss-cheese regions in elementary cosmological
cells, which are the smallest volumes which are expanding with the Hubble flow.
Though, they do not provide an averaging procedure which is specifically
defined at each point of cosmological space-time,
but rather construct averages over simple
inhomogeneous configurations which they assign to entire regions, one can
argue that over large volumes this procedure approximately gives an average
at each point. In the examples treated,
the metric (Swiss-cheese)
of the inhomogeneities is considered spherically symmetric and therefore either
Schwarzschild or FLRW. The attempt is to construct the physical large-scale
background metric from averages over inhomogeneities which may be very large
in amplitude compared to the an provisional background. These averages are
obtained by
adding the averages over each inhomogeneity to one another along with the
averages over the provisional ``background'' space in between the
homogeneities. Bi-vectors are not used. We shall come back to discuss this
scheme more fully at the end of the paper, as it can be construed as an
approximate implementation of our averaging procedure in simple non-perturbative
averaging situations.
\item{Buchert and Ehlers averaging of Newtonian Cosmologies}
With a similar approach to that proposed by Noonan, Buchert and Ehlers
\cite{Buchert1,BuchertEhlers} have focused on the averaging problem applied to
the cosmological expansion of the Universe. They
restrict themselves to the Newtonian approach, i.e. the analog of Friedmann's
equation for the motion of self-gravitating presureless fluid.
They propose that the spatial average
of a tensor field $A$ in the domain $D(t)$ should be:
\begin{equation}
<A>_{D} = \frac{1}{V} \int_{D} d^3x A.
\end{equation}
Since the comoving volume $D(t) = a_D^3(t)$, the fluid elements
move on the average according to
\begin{equation}
<\theta>_D = \frac{\dot V}{V} = 3 \frac{\dot{a}_D}{a_D}.
\end{equation}
With this in mind, $a_D(t)$ becomes the new scale factor and is shown to
obey an {\it averaged Raychaudhuri equation}.
Note that there are no spatial dependencies in the averaged quantities, only
time is left as a variable. This means that the averaging scheme yields
the same result at all positions in the synchronous gauge.
They also propose a scheme for general relativity which includes the
determinant of the
spatial metric $g = \det g_{ij}$ in the region of interest:
\begin{equation}
<A>_{D} = \frac{1}{V} \int_{D} d^3x \sqrt{g} A
\end{equation}
with $V = \displaystyle{\int}_D d^3x \sqrt{g}$, which resembles Noonan's
definition even more closely.
This average was then used by Russ et al. \cite{Russ} to compute
the effect of inhomogeneities on the age of a flat Universe. The
sources of the inhomogeneities were taken from the Zel'dovich approximation to
second order, so that their results are valid in
an early stage of the evolution of the flat Universe. They conclude
that the effect on the age of the universe is very small on those scales.
\item{Boersma's Averaging Procedure}
Boersma \cite{Boersma}
derived, from basic assumptions, a generic linearized
spatial averaging operation for
metric perturbations from FLRW, satisfying the
condition that unperturbed FLRW is a stable fixed point of the averaging
(that is, the averaging operation does not introduce spurious
perturbations in the averaged metric).
He specified the correspondence
among points in the real spacetime, the averaged spacetime and the background
spacetime,
by the introduction of a bi-tensor density in the averaging
integrals, which fulfils the same function as the bi-vectors in Isaacson's
approach. He
succeeded in deriving a general form of this bi-tensor density
in terms the background metric and the future-directed unit vectors normal
to the space-like surfaces over which the averages are being performed. With
this formalism, and using Bardeen's \cite{Bardeen} gauge-invariant quantities
Boersma was able
to resolve the gauge problem in his averaging procedure and
apply it to the constraint equations on spacelike hypersurfaces, which are
closely related to the
generalized Friedmann equation.
\end{itemize}
\section{The weak field limit}
\label{WFL}
Using our averager defined in Eq.
(2) we begin to examine its properties
in a very simple, almost trivial way, by considering its application to
gravity in the weak field limit. We do this in order to assure ourselves that
our averaging procedure fulfils the simplest intuitive requirements and to
establish some results which we shall need later in applying it to
perturbed FLRW cosmologies. \\
Let us consider a Minkowskian line element plus small
corrections.
That is,
\begin{equation}
\label{eqdef}
g_{\mu\nu}(x) = \eta_{\mu\nu}(x) + h_{\mu\nu}(x), \qquad |h_{\mu\nu}| \ll 1.
\end{equation}
Applying our averager to this metric we have
\begin{equation}
<g_{\mu\nu}(x)>=\frac{{\displaystyle \int \! d\Omega' \,\sqrt{-g(x + x')}
\,g_{\mu\nu}(x+x')}}
{{\displaystyle \int \! d\Omega' \,
\sqrt{-g(x + x')}}}.
\end{equation}
Since in the weak field limit (see below)
\begin{equation}
\sqrt{-g(x + x')}=1+\frac12 \,\eta^{\mu \nu}h_{\mu \nu}(x + x'),
\end{equation}
and considering that
the product of two or more elements of the matrix of perturbations is
negligible we find, to first (linear) order,
\begin{equation}
<g_{\mu\nu}(x)>=\eta_{\mu\nu}+\frac{\int \! d\Omega' \,
h_{\mu\nu}(x+x')}
{\int \! d\Omega' } + {\mathcal O}(h^{2}).
\end{equation}
Thus, in general, averaging a Minkowski space + perturbations
gives another perturbed Minkowski space -- a result which is obvious, but
reassuring. It is, of course, possible in some circumstances, that the
second term of Eq. (9) will be zero, that is, the
perturbations will average out over larger scales. For instance,
``overdensities'' may be partially or completely compensated
by ``underdensities''. \\
Furthermore, it is clear that
in this weak field case -- and in the case of perturbations to FLRW discussed
in Section V -- the operation of averaging {\it will} commute with
constructing the Einstein field equations, since these are now linear.
Thus, averaging will not introduce any new
terms in the macroscopic field equations themselves, as it does in the exact
case \cite{Ellis,Zotov,Roque}. \\
\label{SecWFL}
Let us now consider a region in space, free of gravitational sources but filled
with gravitational radiation coming from a source situated at infinity. In
these
circumstances we can apply the weak field limit of Einstein's equations
\begin{equation}
\nabla h_{\mu \nu} = 0,
\end{equation}
\begin{equation}
\frac{\partial h^{\mu}_{\nu}}{\partial x^{\mu}}= \frac{1}{2} \frac{\partial
h^{\mu}_{\mu}}{\partial x^{\nu}} .
\end{equation}
Let us now suppose that we have two scales, a macroscopic and a microscopic
scale, both free of sources. Then on both scales
we will see the same behavior, i.e. on each
scale we can write
\begin{equation}
\label{wflimit}
h_{\mu \nu}^{(i)} = e_{\mu \nu}^{(i)} {\rm e}^{ i k^{(i)}_{\lambda}
x^{(i) \lambda}} + C.C.
\end{equation}
for $i=1, 2$, where the $e_{\mu \nu}^{(i)}$ are the polarization tensors in
each region.
The microscopic scale is associated with our definition
and the macroscopic
one ought to be compared with the averaged result.
Recalling the definition of the average in the weak field limit,
\begin{equation}
\label{avh}
<h_{\mu \nu}(x)> = \frac{\int \! d^{4}x' \,\sqrt{-g(x + x')}\,
h_{\mu \nu}(x + x')}
{\int \! d^{4}x'\,\sqrt{-g(x + x')} },
\end{equation}
where $ g = \det {g_{\mu \nu}} = \det(\eta_{\mu \nu} + h_{\mu \nu})$.
Writing Eq.~(\ref{eqdef}) in matrix notation
allows us to compute $g$ using
\begin{equation}
\label{logdet}
\ln \det (1+ A) = tr \ln (1+A).
\end{equation}
by properly defining $A$ in terms of $h_{\mu \nu}$. Thus
\begin{eqnarray}
tr \ln (1+A)& = & \sum_{\alpha} \ln (1+A)_{\alpha \alpha} \nonumber \\
& = & \ln(1 + h_{00}) + \ln(-1 + h_{11}) + \ln(-1 + h_{22}) +
\ln(-1 + h_{33}) \nonumber \\
& = & \ln(1 + h_{00})(-1 + h_{11})(-1 + h_{22})(-1 + h_{33}) \nonumber \\
& \approx & \ln(-1 - h_{00}+ h_{11}+ h_{22}+ h_{33}),
\end{eqnarray}
and
\begin{equation}
-g = 1 + h_{00} - h_{11} - h_{22}- h_{33},
\end{equation}
or
\begin{equation}
\sqrt{-g} \approx 1 + \frac{1}{2}\,\eta^{\mu\nu} h_{\mu\nu}.
\end{equation}
This result is independent of the choice of the gauge, and
the only hypothesis used is that $h_{\mu \nu}$ is small.
Going back to the definition of the average, and using the property just
demonstrated, we find to first order in $h$
\begin{equation}
<h_{\mu \nu}(x)> = \frac{\int \! d^4 x'\, h_{\mu \nu}(x + x')}{\int \! d^4 x'}.
\end{equation}
Therefore, using Eq.~(\ref{wflimit})
\begin{equation}
<h_{\mu \nu}(x)> = \frac{1}{\Omega'}\biggl[e_{\mu\nu} \int \! d^4 x' \,
{\rm e}^{ i k_{\lambda}\,(x^{\lambda} + x'^{\lambda})} + {e}^{*}_{\mu\nu}
\int \! d^4 x' \, {\rm e}^{- i k_{\lambda}\,(x^{\lambda} + x'^{\lambda})}
\biggr],
\end{equation}
or
\begin{equation}
<h_{\mu \nu}> = c_{1} \,e_{\mu \nu} \,{\rm e}^{ i k_{\lambda}\,x^{\lambda}}
+ C.C.
\end{equation}
First of all, note that this averaged quantity
transforms as a
tensor in the weak field limit. Second the polarization
tensors can
now be redefined as they appear multiplied by a constant which contains
information about the microscopic scales of the system. Thirdly,
we have found
that the averaged solution (i.e. the average of the
solution in one scale)
coincides with the solution on another scale.
This was expected, because we have worked in the linear theory. There
Einstein equations
are linearized, and therefore, as we have already indicated, the averaging
procedure does not introduce extra terms in them. This means, of course, that
the averages
are also solutions of the
Einstein weak field equations. \\
We will now extend this analysis to see what happens in the weak field
limit, when considering
a region in which there is a source of gravitational radiation.
In this case, one scale would correspond to that of
the source, while the other could be thought of as the wave zone -- that is, a
scale in which the length scales are much larger than the size of the source.
When averaging, we should compare the average metric $<{h}_{\mu \nu}>$ with
the one obtained for the wave zone.\\
The solution of the weak field Einstein equations in the presence
of a source is the metric \cite{Weinberg}
\begin{equation}
g_{\mu \nu} = \eta_{\mu \nu} + h_{\mu \nu}
\end{equation}
with
\begin{equation}
h_{\mu \nu}(x) = 4 G \int \! d^{3} x'\frac{S_{\mu \nu}(\vec{x}',
t-|\vec{x}-\vec{x}'|)}{|\vec{x}-\vec{x}'|}.
\end{equation}
Combining this
Eq. with Eq. (\ref{avh}) we can write
(to linear order in $h$)
\begin{equation}
\label{averag}
<h_{\mu \nu}(x)> = \frac{4G}{\Omega'} \int \! d^{3} x' d^{4}x''
\frac{S_{\mu \nu}
(\vec{x}',
t + t'' - |\vec{x} + \vec{x}''- \vec{x}'|)}{|\vec{x}+ \vec{x}''-\vec{x}'|}.
\end{equation}
If we now express the energy-momentum tensor as a Fourier integral, we
can analyze each Fourier component separately, and
integrate (or add those components) afterwards.
This means that we can replace in Eq.~(\ref{averag}),
$ S_{\mu \nu}(\vec{x},t) = \hat{S}_{\mu \nu}(\vec{x},\omega)
{\rm e}^{-i \omega t}$, i.e.
\begin{equation}
\label{4}
<h_{\mu \nu}(x)> = \frac{4G}{\Omega'} \int \! d^{3}x'\, d^{4}x'' \,
\hat{S}_{\mu
\nu}(\vec{x}',\omega) \frac{{\rm e}^{- i \omega\,(
t + t'' - |\vec{x} + \vec{x}''- \vec{x}'|)}}{|\vec{x}+ \vec{x}''-\vec{x}'|}
+ C.C.
\end{equation}
\mbox{}Hereafter, we drop the complex conjugate of the quantities to
make the expressions look simpler. However, they should be recalled at the end
of the calculations, as all quantities are real.
So far the only approximation made was to consider a
weak field limit.
The scales in the problem are three:
\begin{enumerate}
\item Scale of the source, given by $x'$.
\item Scale of the averaging procedure (or {\em small scale}), given by
$x''$.
\item Scale of the metric ({\em large scale}),
given mainly by $x$.
\end{enumerate}
This means that scale (1) should be comparable with scale (2), and both much
smaller than scale (3).
Under this assumption we can perform a multipole expansion, which to first
order is
\begin{equation}
|\vec{x} + \vec{x}''- \vec{x}'| \approx |\vec{x}- \vec{x}'| + \vec{x}''
\cdot \frac{(\vec{x}- \vec{x}')}{|\vec{x}- \vec{x}'|}.
\end{equation}
Thus, we can integrate over the {\em small scale} and obtain for the spatial
part
\begin{equation}
I = \int \! d^3 x'' \frac{{\rm e}^{ i \omega |\vec{x} + \vec{x}''- \vec{x}'|}}
{|\vec{x}+ \vec{x}''-\vec{x}'|} =
\frac{{\rm e}^{ i \omega |\vec{x} - \vec{x}'|}}{|\vec{x}-\vec{x}'|} \int \!
d^3 x''
{\rm e}^{ i \omega \vec{x}''
\cdot \frac{(\vec{x}- \vec{x}')}{|\vec{x}- \vec{x}'|}},
\end{equation}
or
\begin{equation}
I = 4 \pi \frac{{\rm e}^{ i \omega |\vec{x} - \vec{x}'|}}
{\omega |\vec{x}-\vec{x}'|}
\left[ \frac{-1}{\omega} R \cos{\omega R} + \frac{1}{w^2}\sin{\omega R}\right],
\end{equation}
where $R$ is the size of the region over which we are averaging. As we
stated before, $R$ should be comparable to the size of the source, and
much smaller than the overall scale of the problem.
Substituting in Eq.~(\ref{4}), we obtain the averaged metric as
\begin{eqnarray}
\label{6}
& <h_{\mu \nu}(x)> & = \frac{16 \pi G}{\Omega'} \int \! d t'' {\rm e}^{-i
\omega \,(t + t'')} \times \nonumber \\
& & \int \! d^{3} x' \,
\frac{{\rm e}^{ i \omega \,|\vec{x} - \vec{x}'|}}{\omega |\vec{x}-\vec{x}'|}\,
\left[ \frac{-1}{\omega} R \cos{\omega R} + \frac{1}{w^2}\sin{\omega R}\,
\right]
\,\hat{S}_{\mu \nu}(\vec{x}',\omega).
\end{eqnarray}
Applying again the multipole expansion for $|\vec{x}'| << |\vec{x}|$,
finally
\begin{equation}
\label{7}
<h_{\mu \nu}(x)> = \frac{16 \pi G}{\Omega'} \,{\rm e}^{-i \omega t}
\,\frac{{\rm e}^{i \omega r}}{r}\,
\frac{1 - {\rm e}^{-i \omega T}}{i \omega}\,
\left[\frac{-1}{\omega} R \cos{\omega R}
+ \frac{1}{w^2}\sin{\omega R}\,\right]\, \int d^3 x'
{\rm e}^{-i \omega \vec{x}' \dot \frac{ \vec{x}}{r}}
\hat{S}_{\mu \nu}(\vec{x}', \omega),
\end{equation}
or,
\begin{equation}
<h_{\mu \nu}(x)> = {\rm e}^{ i k_{\lambda}\, x^{\lambda}}
\,e_{\mu \nu}(\vec{x},\omega) + C. C.
\end{equation}
with
\begin{equation}
e_{\mu \nu}(\vec{x},\omega) = \frac{ 4 G}{r}\frac{4 \pi G}{\Omega'} \frac{1 -
{\rm e}^{-i \omega T}}{i \omega} \,
\left[\frac{-1}{\omega} R \cos{\omega R}
+ \frac{1}{w^2}\sin{\omega R}\,\right] \,\int d^3 x'
{\rm e}^{-i \omega \vec{x}' \dot \frac{ \vec{x}}{r}}
\hat{S}_{\mu \nu}(\vec{x}', \omega).
\end{equation}
Here $e_{\mu \nu}$ is the polarization tensor that an observer in the wave
zone would detect.
Note the contribution from the
integration over the time $t''$, $(1 -
{\rm e}^{-i \omega T})/i \omega $,
with $T$ being the size of the time interval in space-time.
Certainly one expects the choice of the limit of integration $T$ to be
dependent
on the nature of the problem. Let us suppose that only one frequency $\omega$
is being emitted by a source. Therefore, a suitable choice for $T$ will
be $\frac{\pi}{\omega}$. That is, $T$ is basically the characteristic period
of the system.
Note as well that we have only taken into account
characteristic scales of the system which are observed from the microscopic
perspective.
Notice, too, that the averaged metric is, again to first order, a tensor. And
even more, it has the
same form as the solution of Einstein's equations in the weak field limit
considered in the wave zone.
This was to be expected, because we are still dealing
with the linearized equations, and thus, the averaging is a linear procedure.
That is, averaging the equations is still equivalent to averaging the metric,
just because we are working in the weak-field limit.
\section{The approximate tensorial character of the averaged quantities}
In this section we examine whether or not the quantities defined by our
averaging procedure are generally, or approximately, tensors under
certain conditions. \\
We first analyze the tensorial status of our averaging procedure in the
completely general
case. In order to have $<L_{\mu \nu}>$ as a tensor we would need the
transformation of $L_{\mu \nu}$ such that
\begin{equation}
\label{ip}
L^\prime _{\mu \nu} (y + x^\prime) = \frac{\partial y^\alpha}{\partial x^\mu}
\frac{\partial y^\beta}{\partial x^\nu}L_{\alpha \beta} (x+x^\prime)
\end{equation}
under a completely general function $x^\mu=f^\mu(y)$. If Eq.~(\ref{ip})
is valid, we can take the derivatives outside the integral of our averaging
definition and find that the averaged quantities behave as tensors too.
To satisfy Eq.~(\ref{ip}) we need to make a generic
transformation
between $z=x+x^\prime$ and $z^\prime=y+x^\prime$, with a fixed $x^\prime$.
But defining $z^\prime=h(z)$ in this way is equivalent to imposing very
specific constraints
on the original function $f$:
$$z^\prime=y+x^\prime=f(x)+x^\prime=h(x+x^\prime).$$
This is not possible for every conceivable
$f$, and thus the averager is not in general
a tensor. \\
Even when we only transform the cosmological coordinate $x$,
so that the averaging procedure yields almost tensorial quantities with
respect to the cosmological coordinates only (the averaging integrations are
only over $x^\prime$, so that the transformations of the cosmological
coordinates $x$ can be taken through the integrals), \cite{Nelson}
there is an important
problem. Strictly speaking, we cannot transform the
cosmological coordinates without also transforming the local coordinates and
altering the functional relationship connecting them. Furthermore,
the integral over the local coordinates $x'$ is not covariant with respect to
those changes. \\
Can we somehow show that the average is {\it approximately} coordinate
covariant in some cases?
That is indeed true for the
weak-field and the perturbed-FLRW cases. \\
Let $L_{\mu\nu}$ be a tensor. In the weak field
limit our averager applied to $L_{\mu \nu}$ takes the form
\begin{equation}
<L_{\mu\nu}(x)>=\frac{
\int \! d\Omega' \left( 1 + \frac12 \,\eta^{\alpha \beta} h_{\alpha \beta}(x +
x') \right) L_{\mu\nu}
(x+x')}
{\int \! d\Omega' \left( 1+ \frac12 \,\eta^{\alpha \beta} h_{\alpha \beta}(x +
x') \right) }.
\end{equation}
To first order in the perturbation this becomes
\begin{eqnarray}
<L_{\mu\nu}(x)> & = & \frac{1}{\Omega'} \int \! d\Omega' \,L_{\mu\nu}
(x+x') +
\frac{1}{\Omega'} \int \! d\Omega' \,\frac12 \,\eta^{\alpha \beta}
h_{\alpha \beta}(x + x') \, L_{\mu\nu} (x+x') \nonumber \\
& - & \frac{1}{\Omega'^{2}}\int \! d\Omega'\, L_{\mu\nu}(x+x')
\int \! d\Omega''\, \frac12 \,\eta^{\alpha \beta} h_{\alpha \beta}(x + x'').
\end{eqnarray}
where $\Omega' = \int \! d\Omega'$, is the 4-volume and a real number which
only depends on scale of the averaging. \\
We now employ Isaacson's averaging procedure and compare it with our own.
In the Isaacson's case we have the averaging definition given in
Eq. (4),
which yields a {\it bona fide} tensor. We also know that the bi-vectors
$j^{\alpha '}_{\mu}(x, x')$ must satisfy
Eq. (5). In the weak field and
perturbed space-time cases this implies that
\begin{equation}
j^{\alpha '}_{\mu}(x, x') = \delta^{\alpha '}_{\mu} - (1/2)h^{\alpha '}_{\mu}
(x, x').
\end{equation}
Thus,
\begin{equation}
\label{isavg}
<T_{\mu \nu}>_I = \int d^4x' T_{\mu \nu}(x')f(x,x') + O(1),
\end{equation}
where $O(1)$ refers to integrals of first order in $h$. Thus, the first term
in this equation is almost a tensor --differs from a tensor by small terms
of order $h$. Then, our averaging procedure gives an
integral like the first term of
Eq. (\ref{isavg}), except that it
contains a $\sqrt{-g}$ instead of $f(x, x')$ --which makes no difference
in the tensorial character of the integral in the weak field and
perturbed field cases. Thus, our procedure also yields an approximate tensor
in both of these cases, relative to the complete coordinate $x + x'$. \\
\section{The FLRW metric + Perturbations}
\label{SecFRW}
Let us now consider
an FLRW
space-time with perturbations and
apply the averager to the perturbed metric
\begin{equation}
\label{pertflrw}
g_{\mu\nu}(x)=g_{\mu\nu FLRW}(x) + h_{\mu\nu}(x),
\end{equation}
where the FLRW line element in Cartesian coordinate can be written \cite{Bondi}
\begin{equation}
\label{flrwmet}
ds^2=dt^2- \frac{a^2(t)\,[dx^2+dy^2+dz^3]}{\{1 + \frac{1}{4}k[x^2 + y^2 + z^2]
\}^2}.
\end{equation}
Applying the averaging, we find
\begin{equation}
\label{metfrw}
<g_{\mu\nu}(x)>=\frac{{\displaystyle \int \! g_{\mu\nu}(x+x') \,
\sqrt{-g(x + x')} d\Omega'}}
{{\displaystyle \int \! \sqrt{-g(x + x')}\, d\Omega'}}.
\end{equation}
Using Eq.~(\ref{logdet}) the
determinant of the perturbed metric is,
to first order in $h_{\mu\nu}$
\begin{equation}
\label{detFRW}
\det g_{\mu\nu}= \det g_{\mu\nu FLRW}\, ( 1 + h_{\mu \nu}g^{\mu\nu}_{FLRW}),
\end{equation}
and replacing this in Eq.~(\ref{metfrw}) we obtain
\begin{equation}
\label{avFRW}
<g_{\mu\nu}(x)>= \frac{{\displaystyle \int \!\left(g_{\mu\nu FLRW}(x+ x') +
h_{\mu\nu}(x+
x')\right)
\sqrt{-g_{FLRW}(x + x')\left(1 + h_{\mu \nu}g^{\mu\nu}_{FLRW}
\right)} d\Omega'}}
{{\displaystyle \int \!\sqrt{-g_{FLRW}(x + x')\left(1 + h_{\mu \nu}
g^{\mu\nu}_{FLRW}\right)} d\Omega'}}.
\end{equation}
\subsection{What is a perturbation?}
Here we shall define more precisely what we mean by a
perturbation to an
FLRW metric. We will say that $h_{\mu \nu}$ in Eq.~(\ref{pertflrw}) is a
perturbation if, just as in Eq.~(\ref{eqdef}),
\begin{equation}
\label{defpert}
|h_{\mu \nu}| \ll 1.
\end{equation}
Thus, if $h_{\mu \nu}$ is a perturbation to FLRW in
one coordinate system, then in any other coordinate system, in which we can
always write the metric as
\begin{equation}
\label{FRWpert}
g'_{\mu \nu}(y) = g'_{\mu\nu FLRW}(y) + h'_{\mu\nu}(y),
\end{equation}
$h'_{\mu \nu}$ is also a perturbation. Though
$h'_{\mu \nu}$ will naturally
be different from $h_{\mu \nu}$, it will always satisfy Eq.~(\ref{defpert}),
if $h_{\mu \nu}$ itself does so. That is simply because, if we apply a
general coordinate transformation
\begin{equation}
\label{chFRW}
g'^{\alpha \beta}(y) = \frac{\partial y^{\alpha}}{\partial x^{\mu}}
\frac{\partial y^{\beta}}{\partial x^{\nu}}g^{\mu \nu}(x)
\end{equation}
to Eq.~(\ref{pertflrw}), we trivially find that the coordinate transformation
Eq.~(\ref{chFRW}) also relates the $h'_{\mu \nu}$ to the $h_{\mu \nu}$, and
that, therefore, the $h'_{\mu \nu}$ will be small, if the $h_{\mu \nu}$ are.
Thus also the determinate of the transformed metric given in
Eq.~(\ref{FRWpert}) will be given by Eq.~(\ref{detFRW}), with the unprimed
metric variables just replaced by the primed (transformed) metric variables.
\subsection{Averaging FLRW Perturbations }
Let us now obtain a general expression for the average of the perturbed
metric. From Eq.~(\ref{avFRW})
\begin{equation}
<g_{\mu\nu}>=<g_{\mu\nu FLRW} + h_{\mu\nu}> = \frac{N}{V},
\end{equation}
with
\begin{eqnarray}
N & = & \int \! d^4 x' \sqrt{-g_{FLRW}(x + x')} g_{\mu \nu FLRW}(x + x') +
\int \! d^4 x'\sqrt{-g_{FLRW}(x + x')} h_{\mu \nu}(x + x') \nonumber \\
& + & \int \! d^4 x' \sqrt{-g_{FLRW}(x + x')}
\frac{1}{2} g^{\alpha \beta}_{FLRW}(x + x') h_{\alpha \beta}(x + x')
g_{\mu\nu FLRW}(x + x'),
\end{eqnarray}
and
\begin{equation}
V = \int \! d^4 x' \sqrt{-g_{FLRW}(x + x')}(1 + \frac{1}{2}
g^{\alpha \beta}_{FLRW}(x + x') h_{\alpha \beta}(x + x')).
\end{equation}
The denominator can be written as a product, the first term being
the invariant volume for
an FLRW
space-time. Expanding the rest of the terms to
first order in $h_{\mu \nu}$ we get
\begin{eqnarray}
\label{av1FRW}
<g_{\mu\nu}> & = & \frac{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')} g_{\mu\nu FLRW}
(x + x')}{
\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}} +
\frac{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}
h_{\mu \nu}(x + x')}{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}}
\nonumber \\
&+& \frac{\int \! d^4 x'\sqrt{-g_{FLRW}(x+x')}
\frac{1}{2} g^{\alpha \beta}_{FLRW}(x + x') h_{\alpha \beta}(x + x')
g_{\mu\nu FLRW}(x + x')}{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}} \nonumber \\ & - &
\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')} g_{\mu\nu FLRW}(x + x') \times \nonumber
\\ & \times &
\frac{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}\frac{1}{2} g^{\alpha \beta}_{FLRW}(x + x')
h_{\alpha \beta}(x + x')}{\left(\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}\right)^{2}}.
\end{eqnarray} \\
We are now in a position to show that, for our averaging procedure, the average
of
an FLRW metric itself
is, to first order in the local coordinates,
always an FLRW metric, and further
that the average metric of a perturbed
FLRW metric is also
an FLRW metric plus perturbations. These are among the
features
any averaging procedure should possess for it to be considered
minimally adequate. \\
First, since the
length scale represented by the local coordinate $x'$ will
always be much less than that represented by the cosmological coordinates
$x$, we can assume that
the volume of integration, and therefore the domain of $x'$, will
be small with respect to the overall space-time coordinates $x$. Thus,
we can expand the various functions of
$(x + x')$ as multi-variable
(because of the multiple coordinate variables, $x'$, $y'$, $z'$ and $t'$)
Taylor series in these local coordinates, and truncate the series at first
order. We can then write, for instance,
\begin{equation}
\label{0exp}
g_{\mu \nu FLRW}(x+x') \approx g_{\mu \nu FLRW}(x) + \frac{\partial
g_{\mu \nu FLRW}}{\partial x^{\lambda}}(x) x'^{\lambda}.
\end{equation}
Analyzing the first term in Eq.~(\ref{av1FRW}), we find that, in the Cartesian
coordinate form of the metric specified in Eq.~(\ref{flrwmet}),
$<g_{00}^{I}> = 1$. Here and in the discussion which follows, we designate the
four terms in Eq.~(\ref{av1FRW}) by the superscripts
I, II, III, and IV, respectively. Further, using Eq.~(\ref{0exp}) we find also
that
\begin{equation}
<g_{ij}^{I}> = g_{ijFLRW} + \frac{\partial g_{ijFLRW}}{\partial x^{\lambda}}
(x) \frac{\int d^4 x' \sqrt{-g_{FLRW}(x+x')} x'^{\lambda}}
{\int d^4 x' \sqrt{-g_{FLRW}(x+x')}}.
\end{equation}
These results for the first term of Eq.~(\ref{av1FRW}) enable us to determine
what the application of our averager to an exact FLRW metric yields. If we
carry out the integration indicated in the denominator over the averaging
volume given in the local coordinates $x'^{\lambda}$, we find that this term
vanishes, since the integral is odd in $x'^{\lambda}$. Thus, only even terms
of the Taylor series will yield non-zero contributions to the average. Thus,
to first order in the local coordinates, averaging
an FLRW metric yields
an FLRW metric. This result is trivial in the case of averaging a flat FLRW
metric, but is not at first sight
obvious for open and closed FLRW metrics. This
result is somewhat important, because, as Boersma \cite{Boersma} has
emphasized, we do not want the averaging procedure to introduce perturbations
-- we want the average of
an FLRW metric to be an FLRW metric! \\
If we now go back to
Eq.~(\ref{av1FRW}), we see that, since the terms II, III and IV will all be
smaller than the pure FLRW term I, we have the further almost
trivial result that
\begin{equation}
<g_{\mu\nu}> = g_{\mu\nu FLRW} + Perturbations.
\end{equation}
Though these are expected and obvious results, they are very reassuring. Our
averaging procedure is physically and mathematically consistent. If it did
not fulfil these conditions, it would have to be abandoned. \\
Studying terms II, III and IV of Eq.~(\ref{av1FRW}) in more detail, we find
similarly that to first order
\begin{equation}
<g_{\mu\nu}^{II}> = \frac{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}
h_{\mu \nu}(x + x')}{\int \! d^4 x' \sqrt{-g_{FLRW}(x+x')}}
\approx h_{\mu \nu}(x) + \frac{\partial h_{\mu \nu}}{\partial x^{\lambda}}(x)
\frac{\int \! d^4 \sqrt{-g_{FLRW}(x+x')} x' x'^{\lambda}}{\int \! d^4
\sqrt{-g_{FLRW}(x+x')}x'},
\end{equation}
and that
\begin{equation}
<g_{\mu \nu}^{III}> = - <g_{\mu \nu}^{IV}>
\end{equation}
to first order in $h$ and $x'$. \\
Then Eq.~(\ref{av1FRW}) becomes, to first order in the local coordinates,
\begin{eqnarray}
\label{pertreb}
<g_{\mu \nu}(x)> & = & g_{\mu \nu FLRW}(x) +
\frac{\partial g_{\mu \nu FLRW}}{\partial x^{\lambda}} (x)
\frac{\int d^4 x' \sqrt{-g_{FLRW}(x+x')} x'^{\lambda}}
{\int d^4 x' \sqrt{-g_{FLRW}(x+x')}} + \nonumber \\
& & h_{\mu \nu}(x) +
\frac{\partial h_{\mu \nu}}{\partial x^{\lambda}} (x)
\frac{\int d^4 x' \sqrt{-g_{FLRW}(x+x')} x'^{\lambda}}
{\int d^4 x' \sqrt{-g_{FLRW}(x+x')}},
\end{eqnarray}
However, as we just saw above, the integrals in the numerators of the
second and fourth terms of Eq.~(\ref{pertreb}) vanish when taken over the
entire averaging domain, because they are odd in the local coordinates. Thus,
we obtain the somewhat mysterious and apparently trivial result
\begin{equation}
\label{simpres}
<g_{\mu \nu}(x + x')> = g_{\mu \nu FLRW}(x) + h_{\mu \nu} (x),
\end{equation}
to first order in the local coordinates, just what we began with in
Eq.~(\ref{pertflrw}). It appears that the averaging, to first order in the
local coordinates has no effect -- that the $h_{\mu \nu}$ in
Eq.~(\ref{simpres}) is the same as that in Eq.~(\ref{pertflrw}). This requires
some careful comment. \\
Eq.~(\ref{simpres}) is simply the consequence of terminating the Taylor
series in the local coordinates with the linear term. Any further precision
using this approximation resides in the higher-order terms. In particular,
any further correction to the already small $h_{\mu \nu} (x)$ will be smaller
than it
already is (that is, of higher order than either $h_{\mu \nu}(x + x')$
itself or the
value of the local coordinates $x'^{\lambda}$). In general, we should
expect that $|<h_{\mu \nu}(x + x')>| \leq |h_{\mu \nu}(x)|$, but the
possibility of it being $ < |h_{\mu \nu}(x)|$ can only be explored by either
performing the averaging without approximation, or taking the Taylor series
expansion of the metric to higher orders in $x'^{\lambda}$. \\
It does not make any physical sense to go to higher orders to determine the
distorsions averaging introduces in the FLRW background metric itself (in the
closed and open cases -- if we confine ourselves to averaging over
spatial hypersurfaces, it is only in the closed and open cases that higher
order distorsions are introduced by averaging),
since it does not vary significantly over local-
coordinate length scales; it is already smooth on all scales up to those of
the cosmological coordinates. It is intuitively clear that averaging over
an FLRW metric should
give
an FLRW metric. From one point of view, we might
almost say that the FLRW part of the metric should not be averaged -- it has
in a sense already been averaged, or rather is the result of an average having
been already performed, or assumed -- in particular, an average of the density
used in the
field equations. \\
With respect to the perturbed part of the metric
$h_{\mu \nu}(x + x')$, however,
there is a good reason to
carry out the averaging to higher precision. In practice, it may, unlike
the FLRW metric itself, vary a great
deal on local length scales. Averaging over those local variations may
significantly decrease its magnitude on cosmic length scales, even reducing it
to zero. In fact we know that initially there were perturbations on all scales
above that given by photon diffusion (Silk damping). And at our epoch we
see that there are very large density fluctuations (much larger than
perturbations!) on all local and intermediate scales -- density enhancements
as well as voids. In both cases the
perturbations on cosmological scales are really the averages over the metric
fluctuations on smaller scales. \\
Thus, when we write the perturbed FLRW metric
as in Eq.~(\ref{pertflrw}) without further specification, we are in a sense
conflating two very different cases, which must be treated separately. The first
is that in which the metric of the universe on {\it all} scales can be
represented
as
an FLRW metric plus small deviations, which are present on a large
range of scales. In
this case, the expression given in Eq.~(\ref{pertflrw}) is correct on all
scales. But, at the same time it can be restricted to a given scale of
interest -- for instance, the cosmological scale. In doing that, we would
naturally define the perturbed component $h_{\mu \nu}(x)$ as
\begin{equation}
h_{\mu \nu}(x) = <h_{\mu \nu}(x + x')>,
\end{equation}
where the $h_{\mu \nu}(x + x')$ are small on all length scales and where the
averaging procedure is well-defined as applied to an FLRW mertric plus
perturbations: As long as the metric
deviations from the FLRW metric on all scales
are small, the averaging of $h_{\mu \nu}(x + x')$ over a volume of any size
can be performed in a meaningful way, as we have indicated above. That is,
the innumerable small metric fluctuations on local and intermediate scales
can be averaged over to obtain those on cosmic scales. This would be the
situation in the period of the universe up until density perturbations on
one scale or another go nonlinear. It would certainly pertain to the epoch of
recombination when the cosmic microwave background radiation (CMWBR) was last
scattered, and to some hundreds of millions of years afterwards. Thus, it has
important applications to CMWBR anisotropies. \\
The second situation would be as at present, when the metric of the universe
can be be represented by the perturbed FLRW metric in Eq.~(\ref{pertflrw}) only
on cosmic length scales. On local and intermediate scales Eq.~(\ref{pertflrw})
is far from correct -- the metric on these scales is nowhere near FLRW,
deviations from FLRW being very large. Thus, the interpretation of Eq.~
(\ref{pertflrw}) in this case is that on scales much larger than that of
any nonlinear density fluctuation, the universe is close to FLRW, if the
$h_{\mu \nu}(x)$ -- the deviations from FLRW on {\it that} scale -- are small.
There will be lower limit on this length scale, below which this will not be
true. Then this $g_{\mu \nu FLRW}(x) + h_{\mu \nu}(x)$ must be understood as
the average of all
the small-scale metrics -- some of which will be very different from the
cosmological ``background metric -- over a volume of cosmological length
scale. On small and intermediate length scales we will not be able to define a
background, such that the actual metric is always a perturbation (or small
fluctuation) with respect to it.
How is this $g_{\mu \nu FLRW} + h_{\mu \nu}(x)$
to be calculated from the large amplitude small and intermediate scale
fluctuations? And how should we implement our averaging procedure in this case?
This is an important question and will be treated briefly in the next section.
\\
In this second case, the averaging procedure we have outlined here is not
at all adequate. But an adequate one must be constructed, if $h_{\mu \nu}(x)$
-- the deviation of the large-scale metric from FLRW -- is to have any meaning
with respect to the metric fluctuations on smaller scales, many of which
are not at all small. What is usually implicitly assumed in using $h_{\mu \nu}
(x)$ in this context is to consider that we have averaged over the
small and intermediate scale density inhomogeneities to obtain a large-scale
density inhomogeneity which is found to be only marginally different from
the background FLRW density and is therefore the source of a very small large
scale $h_{\mu \nu}(x)$.. However, this uncritically avoids a number of
important mathematical and physical issues which really have to be resolved
before such a procedure is validated -- particularly the effective field
equations which are really operative in averaging over large amplitude
inhomogeneities, including the contribution made by gravitational binding
energy to the average, the background metric which is used in the averaging
process, and the adequacy and validity of any averaging process itself
involving large deviations from either Minkowski or FLRW
\cite{Ellis,Zotov,Zalaletdinov,Roque}. We need to apply our averaging procedure
in these situations and see what it yields. In some simple, idealized cases
an approximate averaging procedure such as that suggested by
by Zotov and Stoeger \cite{Zotov} may, as we point out below, be equivalent
to ours. \\
\mbox{}From this brief discussion, we begin to appreciate how different
these two
cases are, from the point of view of averaging -- and the rather different
problems implicit in each of them. As we have seen, it is only the first case,
involving perturbations on all scales which can be dealt with using the
linearized procedures developed so far, or any procedures limited to
perturbations. \\
It is important to note, furthermore, that, as we have already seen, it is
possible that in either of these two cases,
the average of the perturbations or deviations over a large enough scale will
yield zero,
as in some of the weak-field-limit cases. In fact, strictly speaking, if a
given FLRW background model is a genuine background, there should be a large
enough scale over which averaging yields the exact FLRW model itself --
that is, the average over the perturbations on such a large length scale
gives zero. This means that the
``positive'' and ``negative'' perturbations need to balance on the largest
scales, if the universe is really FLRW above a certain length scale. If this
is {\it not} the case, then we must choose the average
FLRW plus perturbations as the real background. It is also clear, of course,
that generally the average of the perturbations will possess much more
symmetry than the original perturbations themselves. \\
\subsection{Deviations in the Hubble Parameter}
Of course, these issues of averaging will have consequences for the
observational parameters we measure, for example the Hubble parameter $H$.
It can be easily shown from a perturbation treatment of the field equations
that, for a dust equation of state, the deviation of the Hubble parameter from
its value in a best-fit FLRW model is just given by \cite{Padman}
\begin{equation}
\label{Hubble}
\Delta H = - \frac{1}{3} \dot{\delta},
\end{equation}
where $\delta$ is the density contrast, that is $\displaystyle \delta =
\frac{\rho - \rho_b}
{\rho_b}$, where $\rho$ is the mass-energy density and $\rho_b$ is the
background (FLRW) mass-energy density. \\
Now, obviously Eq.~(\ref{Hubble}) depends on how $\delta$ is calculated, and
furthermore we want the time-derivative of the
{\it cosmological $\delta$} -- its
average over the length
scale of the universe. Averages over small or intermediate scales will give
us only the local or intermediate deviations of the Hubble parameter from
its FLRW value. Similar to our discussion of averages over metric
fluctuations in the two general cases in the last subsection, determination
of $\delta$ and therefore $\dot{\delta}$ is relatively easy in the case when
density fluctuations on all scales are perturbations, as long as we have the
required data. Then we can use our procedure to do the averaging. But again, if
we are faced with the second case, where the density inhomogeneities on small
and intermediate scales are large, as at the present time, the correct
averaging procedure of the density fluctuations is unclear. In fact, in
this case, it is not certain that as a consequence of averaging the field
equations at smaller scales over larger scales Eq.~(\ref{Hubble}) holds.
Not only that, but the perturbative treatment from which Eq.~(\ref{Hubble}) is
derived will not be valid in that regime. This issue will be investigated
briefly in the next section, and more
thoroughly in a subsequent paper. Furthermore,
it is obvious again that the scale over which the averaging is done will
determine the result. If we average over intermediate scales -- or determine
$H$
using data sampling intermediate scales, scales at which the Hubble flow can
not yet be recovered, instead of cosmological scales -- then our calculation
or measurement of $H$ will be local, not cosmological.
\section{Averaging in Simple Nonperturbative Cases}
When faced with averaging over large amplitude small and intermediate scale
perturbations we can in principle still apply our averaging procedure. However,
in general it is difficult to see how we could practically implement it
directly. This is true even in simple
idealized cases, where we have, for instance, strong spherically symmetric
inhomogeneities. However,
in such idealized situations, Zotov's and Stoeger's
two-step procedure \cite{Zotov} provides a useful and implementable
approximation to ours. \\
They first average over each individual spherically symmetric inhomogeneity,
which is represented either as a Schwarzschild metric or a local FLRW metric
plus a surrounding underdense annulus. This average is performed over the
whole vacuole
representing the inhomogeneity, without including any region outside it,
and is therefore referred to the center point of the inhomogeneity.\\
Now, very large regions of the universe can be ideally considered to be made
up of collections of these spherically symmetric inhomogeneities, or
spherically symmetric clusters of them, separated by regions which are
either approximately Minkowski -- if the region is not expanding -- or
approximately FLRW, if the region is expanding. These function as a temporary
or intermediate background metric, representing the metric far from the
center of any inhomogeneity or cluster of inhomogeneities. Thus, the picture
is very much like the Swiss-cheese model, except that we envision the
background as temporary or auxilliary -- to enable us to perform the second
step in the averaging. \\
The second averaging is performed over these very large regions -- usually of
cosmological length scale -- consisting of many spherically symmetric
inhomogeneities. It is approximated by:
\begin{equation}
\label{fulavg}
\bar{g}_{\mu \nu} (x) \cong \frac{1}{V_2}[<g_{\mu \nu}>V_1 N + g_{\mu \nu B}
(V_2 - V_1 N)],
\end{equation}
where $<g_{\mu \nu}>$ is the average over a single spherically symmetric
homogeneity, $V_1$ is the volume of each homogeneity, and $N$ is the number
of them in the full cosmological averaging volume $V_2$. $g_{\mu \nu B}$ is the
auxilliary background metric. In this above equation we have idealized all the
inhomogeneities as identical in volume and mass. We can easily generalize to
many different sizes. $x$ is the cosmological coordinate at which the
second averaging is centered. \\
Thus, the second averaging can be conceived as done at each value of the
cosmological coordinate $x$. We simply move the volume $V_2$ around the
universe -- to all different values of $x$ and perform the two-step averaging
procedure. It can be easily seen that this is equivalent to what our
averaging procedure involves, precisely in terms of moving the same volume
around the
universe and performing an average over it. Thus it may be considered as a way
of approximating what our procedure would give
in these situations of strong localized inhomogeneities. As Zotov and
Stoeger point out, $\bar{g}_{\mu \nu}(x)$ will generally not be an FLRW
metric, even if it does not depend on the spatial coordinates -- it will
superficially look like FLRW, but the scale factor will have a different
dependence on time than the FLRW metric. In general, of course, the result of
the averaging will depend on the spatial cosmological coordinates. \\
Only if
the averaging is over a large enough cosmological volume and the universe is
such that averaging over that volume centered at each point in a spacelike
slice of the universe gives exactly the same result, will it yield a scale
factor independent of the spatial coordinates. If the dependence of the average
scale factor on the spatial coordinates is weak and its time dependence
not too much different from FLRW, then the average metric may be represented
by a perturbed FLRW metric
$g_{\mu \nu FLRW}(x) + h_{\mu \nu}$, where now $h_{\mu \nu}
$ is a large length-scale perturbation from a FLRW metric which is defined
via the above averaging procedure. However, strictly speaking, we should
just treat $\bar{g}_{\mu \nu} (x)$ as the average metric over that length
scale, since it will no longer exactly satisfy the Einstein field equations.
\cite{Zotov} It will satisfy field equations which are Einstein's, with an
extra term added, due to the noncommutability of averaging and forming
the Einstein tensor from the metric. \\
Is such averaging approximately tensorial? It seems that it should be, with
respect to the cosmological coordinates. For the prescription for averaging
does not depend on the coordinate system used for the cosmological coordinates
themselves. But do
significant problems arise with regard to the implied averaging over local
coordinates? And is the Zotov-Stoeger averaging over the moving volume really
approximately
equivalent to what our averager would give? We shall investigate these
questions, along with others
in a subsequent paper.
\section{Conclusions}
In this paper we have proposed an intuitively clear averaging procedure for
general relativity and cosmology, which is an extension to that used in
electromagnetism.
We have shown that it gives approximately unique and tensorial results in
weak field and perturbed FLRW cases, and does not lead to any significant
unacceptable results in these cases. Furthermore, it promises to be
easily applicable in cases where fluctuations on all scales are perturbations,
such as up to the epoch in which density perturbations begin to go nonlinear.\\
Finally, we explore the limits of averaging
perturbed FLRW universes, indicating that averaging over very large small
and intermediate scale inhomogeneities, in order to recover the average
perturbation on large scales, requires applying our averaging procedure beyond
the simple FLRW plus perturbations case. We show how that can be done
approximately in
simple idealized cases involving spherically symmetric objects and clusters of
objects in an intermediate background using the Zotov and Stoeger approach.
|
\section{Introduction}
\label{intro}
Doppler surveys of main sequence stars have revealed 15 companions to
main sequence stars that are extrasolar planet candidates. Among
these candidates, 13 have $M \sin i~$ $<$ 5 M$_{\rm JUP}~$. The host stars and
associated descriptions are: 51 Peg (Mayor \& Queloz 1995), 47 UMa
(Butler \& Marcy 1996), 70 Vir (Marcy and Butler 1996), 55 Cnc,
$\upsilon$ And, $\tau$ Boo (Butler et al. 1996), 16 Cygni B (Cochran
et al. 1997), $\rho$ CrB (Noyes et al. 1997), GJ876 (Marcy et
al. 1998, Delfosse et al. 1998), 14 Her (Mayor et al. 1998), HD187123
(Butler et al. 1998), HD195019A \& HD217107 (Fischer et al. 1999),
GJ86 (Queloz et al. 1999) and HD114762 (Latham et al. 1989). The
Doppler measurements reported here suggest the presence of new
planetary candidates around HD210277 and HD168443.
Four main sequence stars harbor Doppler companions that
have $M \sin i~$ = 15--75 M$_{\rm JUP}~$, which may represent the ``brown dwarfs''
(Mayor et al. 1998, Mayor et al. 1997). Indeed the companions to 70
Vir ($M \sin i~$ = 6.8 M$_{\rm JUP}~$) and to HD114762 ($M \sin i~$ = 11 M$_{\rm JUP}~$) may also
represent ``brown dwarfs'' (Marcy \& Butler 1996, Mazeh et al. 1997,
Latham et al. 1989, Boss 1997). The distinction between ``planets''
and ``brown dwarfs'' remains cloudy and rests on two formation
scenarios. Planets form out of the agglomeration of condensible
material in a disk into a rock--ice core (eg. Lissauer 1995). In contrast,
brown dwarfs presumably form by a gravitational instability in gas
(eg. Boss 1998, Burrows et al. 1998). Hybrid formation scenarios
remain viable in which the relative importance of solid core growth
and gas accretion within a disk lead to a continuum in internal structure.
Subsequent collisions may lead to further growth and dynamical
evolution. The current dichotomous taxonomy may describe substellar
physics no more precisely than ``spiral'' and ``elliptical'' summarize
galactic physics.
The first incontrovertible sub--classification within the
substellar regime is revealed in the mass distribution. Companions
having $M \sin i~$ in the decade between 0.5--5 M$_{\rm JUP}~$ outnumber those
between 5--50 M$_{\rm JUP}~$ by a factor of $\sim$3 (eg, Marcy \& Butler 1998,
Mayor et al. 1998). The poor detectability of the lowest--mass companions
implies that the factor of 3 is a lower--limit to the cosmic ratio.
This plentitude of companions having Jovian masses suggests that
qualitatively distinct formation processes predominated, arguably
similar to those associated with the giant planets in our Solar System
(Lin et al. 1998).
The extrasolar planets reveal some peculiarities that may bear on
their formation. The host stars of the extrasolar planet candidates
have higher mean metallicity by a factor of $\sim$2 in abundance compared
with field stars (Gonzalez 1998). Metallicity was not a criterion in
the selection of the target stars for these planet searches. Equally
interesting is that seven planets reside in orbits with a radius less than
0.12 AU, sometimes termed ``51 Peg'' planets (Mayor \& Queloz 1995,
Butler et al. 1998). Precision Doppler surveys are most sensitive to
planets in small orbits, resulting in a selection effect.
Nonetheless, these small orbits challenge us to
explain their existence in a region where both the high temperatures
and the small amount of protostellar material would inhibit formation
{\it in situ}. The 51 Peg planets thus offer support for the prediction
by Goldreich \& Tremaine (1980) and Lin (1986) that Jupiters may
migrate inward from farther out (Lin et al. 1995, Ward 1997, Trilling
et al. 1998).
Perhaps most intriguing about the planet candidates are the orbital
eccentricities. The orbits of the 51 Peg planets may suffer some
tidal circularization (Lin et al. 1998, Terquem et al.1998, Ford et
al. 1998, Marcy et al. 1997), and indeed their orbits are all nearly
circular (see Table 4). In contrast, the planets that orbit farther
than 0.15 AU from their star all reside in non--circular orbits having
$e>$0.1, i.e. more eccentric than for Jupiter ($e$=0.05). Indeed, all
but two have $e>$0.2 . This high occurrence of orbital eccentricity
has lead to a variety of models in which Jupiter--like planets suffer
gravitational interactions with a) other planets (Weidenschilling \&
Marzari 1996, Rasio \& Ford 1996, Lin \& Ida 1996, Levison et
al. 1998), b) the disk (Artymowicz 1993), c) a companion star (Holman
et al. 1996, Mazeh et al. 1997), and d) passing stars in the young
cluster (de la Fuente Marcos and de la Fuente Marcos 1997, Laughlin
and Adams 1998).
The possibility persists that the observed non--circular orbits all
stem from perturbations from a bound companion star, as proposed for
16 Cyg B (Holman et al. 1997, Mazeh et al 1997), rather than from
intrinsic dynamics of planet formation. This paper reports the
detection of two new planetary candidates orbiting at 0.3 and 1.1 AU,
both having large eccentricities. The observations and orbital
solutions are reported in section 2. The search for stellar
companions is discussed in section 3. Section 4 contains a discussion
of the implications for planet formation.
\section{Observations and Orbital Solutions}
\label{obs}
\subsection{Stellar Characteristics of HD210277}
The two stars described here are among 430 G,K, and M-type main
sequence stars currently being monitored at the Keck I telescope for
Doppler variations. HD210277 has an effective temperature of 5570
$\pm$50 K, the average determination from spectral synthesis of
high--resolution spectra (Favata et al. 1997, Fuhrmann 1998, Gonzalez
et al. 1998), which also yields a surface gravity of log $g$ =
4.38 $\pm$ 0.1 (Fuhrmann 1998, Gonzalez et al. 1998). These
surface values imply main--sequence status and correspond to spectral
type G7V (Gray 1997).
The metallicity of HD210277 is measured to be [Fe/H] = +0.24
$\pm$0.02, considerably higher than the average value for field stars,
$<$[Fe/H]$>$ = --0.23, in the Solar neighborhood (Gonzalez et al. 1998, Favata et al. 1997, Fuhrmann 1998). Thus, HD210277
appears to be rich by a factor of 3 in its abundance of heavy
elements, normalized to hydrogen, placing its metallicity within
the upper 5\% of nearby stars. We measure a radial velocity of --21.1
$\pm$ 2 \mbox{km s$^{-1}~$}, which agrees with that of Duquennoy and Mayor (1991),
--21.44 \mbox{km s$^{-1}~$}. Its parallax of 0.047 arcsec (Perryman et al. 1997)
implies an absolute visual magnitude of M$_V$=4.90$\pm$0.05 and a
luminosity, $L$=0.93 L$_{\odot}$. These stellar parameters permit
placement of HD210277 on evolutionary tracks, which yield a mass,
$M$=0.92 $\pm$ 0.02 M$_{\odot}$ and an age of 12 $\pm$2 Gyr (Gonzalez et al. 1998).
One stellar characteristic that bears on the Doppler detectability of
planets is the magnetic field and chromosphere. Spots on a
rotating star can produce spurious Doppler shifts, and chromospheric
emission correlates with spurious Doppler ``noise'' presumably caused
by surface magnetohydrodynamics (Saar et al. 1998). Our spectra
contain the chromospheric H\&K emission lines from which stellar
rotation and stellar age can be estimated (Noyes et al. 1984). This
emission yields the chromospheric index known as the ``Mt Wilson S
Value'' of $S$ = 0.155, implying $R$'(HK)=-5.06, measured from 36
spectra obtained from 1996.5 through 1998.7 (Shirts \& Marcy 1998).
See Baliunas et al. (1998) for a detailed discussion of the $S$ value.
No trend or periodicity are apparent in the S values of HD210277, and
the RMS is 0.006, all of which indicate that HD210277 is
chromospherically quiet. The implied rotation period is, $P_{\rm
Rot}$=40.8 d and the age is 6.9 Gyr. In conjunction with the
aforementioned age of 12 Gyr from tracks, we conclude that HD210277
has an age in the range 7--10 Gyr, but not evolved into the subgiant
regime. Such a chromospherically inactive star may produce spurious
Doppler shifts of no more than $\sim$3 \mbox{m s$^{-1}~$} (Saar et al. 1998, Butler
et al. 1998).
\subsection{Stellar Characteristics of HD168443}
No detailed LTE analysis of HD168443 has been carried out to our
knowledge. A photometric analysis was done by Carney et al. (1994)
giving T$_{\rm eff}$= 5430 and $m/H$=--0.14. The metallicity is
apparently slightly subsolar, similar to the mean for nearby field
stars. Its parallax of 26.4 mas (Perryman et al. 1997) implies an
absolute visual magnitude of M$_V$=4.03$\pm$0.07 and a luminosity,
$L$=2.1 L$_{\odot}$, which places it
$\sim$1.5 mag above the main sequence at its
T$_{\rm eff}$. These stellar parameters suggest a subgiant status and
spectral type, G8IV. Apparently, HD168443 is similar to 70 Vir in
mass, surface characteristics, and metallicity (Marcy \& Butler 1996,
Apps 1998).
Our spectra of HD168443 yield a chromospheric S Value of $S$ = 0.147,
with an RMS of 0.009 during 30 observations from 1996--1998.5,
implying $R$'(HK)=-5.08. No trend or periodicity are apparent in the
S values. The implied rotation period is $P_{\rm Rot}$=37 d, and the
implied age is 7.8 Gyr, from the calibration by Noyes et al. (1984).
In conjunction with its possible subgiant status from above, we
conclude that HD168443 has an age of 7--10 Gyr, slightly evolved
toward subgiant status. We caution that the subgiant status remains
in question, pending spectroscopic assessment of surface gravity.
A mass determination for HD168443 is given by Carney et al. (1994) who
find $M$=0.84 M$_{\odot}$. This mass determination may warrant revision
because it preceded the Hipparcos parallax and because it did not
include revisions to the metallicity dependence of evolutionary tracks
(Bertelli et al. 1994). Based on the Hipparcos data and new tracks,
along with available narrow--band photometry for HD168443, Apps (1998)
estimates a mass of 1.05 $\pm$0.10 M$_{\odot}$ for HD168443. We adopt here
the straight average of the two mass estimates to yield $M$ = 0.945
$\pm$0.1 M$_{\odot}$ .
We measure a radial velocity of --49.0 $\pm$ 2 \mbox{km s$^{-1}~$} (on 1998 Aug),
which along with its high transverse velocity of 44 \mbox{km s$^{-1}~$}, suggests a
kinematic association with the old disk population. Such an old,
chromospherically inactive star may produce spurious Doppler shifts
$\sim$3 \mbox{m s$^{-1}~$} of photospheric origin (Saar et al. 1998, Butler et
al. 1998).
\subsection{Details of the Doppler Measurements}
For both HD210277 and HD168443, spectra were obtained from 1996.5
through 1998.7 with the HIRES echelle spectrometer on the Keck I
telescope (Vogt et al. 1994). We used slit ``B1'' that has a width of
0.57 arcsec and height of 3.5 arcsec. The resolution for these
spectra was $R$=87000, based on the measured FWHM of the spectrometer
instrumental profile. The spectra span wavelengths from 3900--6200
\AA. The wavelength scale and instrumental profile were determined
for each 2--{\AA} chunk of spectrum for each exposure by using iodine
absorption lines superimposed on the stellar spectrum (Butler et
al. 1996). The measured velocities are relative, with an
arbitrary zero--point.
The typical exposure times were $\sim$5 minutes, depending on seeing,
for both stars, yielding a S/N=300 per pixel (1/2 of one resolution
element). Such spectra are expected to carry photon--limited Doppler
precision of 2--3 \mbox{m s$^{-1}~$} (Butler et al. 1996). Indeed, the uncertainty
in the mean velocity of the 400 spectral chunks is typically 2.5 \mbox{m s$^{-1}~$}.
However, our results from 430 stars on the survey reveal a median RMS
velocity of 6 \mbox{m s$^{-1}~$}, which we interpret as the actual scatter that
limits planet detection. Intrinsic photospheric noise of $\sim$3 \mbox{m s$^{-1}~$}
accounts for some of the 6 \mbox{m s$^{-1}~$} scatter (Saar et al. 1998). This
intrinsic stellar effect may be added in quadrature to the
photon--limited errors of 2.5 \mbox{m s$^{-1}~$} to establish an expected Doppler
scatter of 3.9 \mbox{m s$^{-1}~$}. Thus, we infer that unidentified errors of
$\sim$4 \mbox{m s$^{-1}~$} persist in our Doppler results which presumably stem from
inadequacies in our spectral modelling, improvements to which are in
progress.
\subsection{Keplerian Velocities for HD210277}
The 34 measured velocities of HD210277 are listed in Table 1 along
with the JD date. Again, the true uncertainty of each measurement is
$\sim$6 \mbox{m s$^{-1}~$}. A plot of the velocities for HD210277 is shown in
Figure 1. It is apparent that the velocities for HD210277 scatter
with a peak--to--peak variation of $\sim$80 \mbox{m s$^{-1}~$}, and the velocities
are correlated in time. A periodogram analysis revealed no
significant peak because too few cycles have transpired
during the two years of observations and because a Lomb-Scargle
periodogram is not robust for non-sinusoidal variations which result from
eccentric orbits.
A suggestive period of 1.2 yr is evident in the velocities for HD210277, though less than two periods have transpired. The best--fit
Keplerian model yields an orbital period of 437$\pm$25 d, a
semi-amplitude $K$ = 41.0$\pm$5 \mbox{m s$^{-1}~$}, and an eccentricity $e$ =
0.45$\pm$0.08 . The complete set of orbital parameters are given in
Table 3. The RMS to the Keplerian fit is 7.1 \mbox{m s$^{-1}~$}, similar to the
uncertainty and similar to the velocity RMS, 7.6 \mbox{m s$^{-1}~$}, for the orbital
fit to a previously-discovered Keck survey planet HD187123 (Butler et
al. 1998). Thus, the RMS of 7 \mbox{m s$^{-1}~$} for HD210277 implies that a single
companion provides a model that plausibly explains the velocities.
Using the stellar mass of 0.92 M$_{\odot}$, the companion mass is
constrained as, $M \sin i~$ = 1.28 M$_{\rm JUP}~$ , and the semimajor axis is $a$ =
1.10 AU. With periastron and apastron distances of 0.61 and 1.60 AU,
HD210277 is unlikely to harbor additional companions within that
range.
\subsection{Keplerian Velocities for HD168443}
The 30 velocity measurements for HD168443 are listed in Table 2 along
with the JD date. As with HD210277, the true uncertainty of each
measurement is $\sim$6 \mbox{m s$^{-1}~$}. A plot of the velocities for HD168443 is
shown in Figure 2. The velocities for HD168443 scatter with a
peak--to--peak variation of 650 \mbox{m s$^{-1}~$}, with clear temporal correlations
and trends among measurements. A periodogram reveals two dominant
peaks, at $P$=20 d and $P\approx$55 d.
We carried out nonlinear least--squares fits of Keplerian models to
the velocities, starting with trial periods ranging from 3 -- 600 d.
For trial periods near 20 d, the lowest velocity RMS was 69 \mbox{m s$^{-1}~$} which
is clearly inconsistent with expected scatter of 6 \mbox{m s$^{-1}~$}. For trial
periods near 55 d, we found two nearby minima in $\chi^2$,
corresponding to two slightly different orbital periods, $P$=64 d
(RMS=23 \mbox{m s$^{-1}~$}) and $P$=59 d (RMS=36 \mbox{m s$^{-1}~$}). Figure 3 shows the
velocities as a function of orbital phase for the better of those fits
($P$=64 d). That Keplerian fit carries an implied eccentricity of
$e$=0.69, $K$=292 \mbox{m s$^{-1}~$}, and a companion minimum mass of $M \sin i~$ = 4.0
M$_{\rm JUP}~$.
However the scatter to that fit, RMS=23.3 \mbox{m s$^{-1}~$} , clearly exceeds the
expected scatter of 6 \mbox{m s$^{-1}~$} , implying that this fit carries a reduced
$\chi^2$ greater than 4 and hence this model is inadequate. Indeed,
two telltale points located at phase $\sim$0.95 (see Figure 3) were obtained
on consecutive nights. The second velocity was 58 \mbox{m s$^{-1}~$} higher, and yet
according to the Keplerian curve, it should reside lower by 60 \mbox{m s$^{-1}~$}.
We consider this dubious orbital fit to imply that the Keplerian model
fails in some important way.
We modified the model by simply adding a variable linear trend to
Keplerian velocities. Such a model incorporates the possibility of a
long period companion in addition to the shorter--period companion.
This slope introduces only one additional free parameter, as the
``y-intercept'' of the slope is subsumed within the arbitrary
zero--point of the velocities.
The Keplerian--plus--slope model is shown in Figure 4 and
yields a best fit orbital period of,
$P$=57.8 d, $K$=350 \mbox{m s$^{-1}~$}, $e$=0.54, and $M \sin i~$=5.04 M$_{\rm JUP}~$ . The RMS of
the residuals, 12.8 \mbox{m s$^{-1}~$}, is considerably reduced from RMS=23.3 \mbox{m s$^{-1}~$} that
results from a model without a trend. The reduced $\chi^2$ for this
solution is 2.3. All orbital parameters are given in Table 3. Thus,
it appears that the introduction of an {\it ad hoc} slope into the
Keplerian model for HD168443 significantly improves the fit.
However, the RMS of 12.8 \mbox{m s$^{-1}~$} remains larger than the expected scatter
of 6 \mbox{m s$^{-1}~$}, implying that the addition of a velocity slope is too
simple. Introducing an {\it ad hoc} parabolic term in the velocity
trend reduces the RMS to 8 \mbox{m s$^{-1}~$} ($\chi^2$=1.5), superior to that of a
linear trend. However, we feel that introducing this parabolic free
parameter carries only marginal statistical justification. A proper
model that contained a second orbiting companion would require the
introduction of an additional set of Keplerian parameters, for which
we have inadequate constraints. Thus, the only model supported by the
current data is that containing the linear trend.
We tested the predictability of this model with two additional
velocity measurements obtained with the 0.6--m CAT telescope at Lick
Observatory. We obtained spectra on two consecutive nights, centered
on Julian Dates 2451100.644 and 2451101.645 for which the model
containing the Keplerian and linear trend offered a prediction of an
increase in velocity of 52.4 \mbox{m s$^{-1}~$}. On both nights we obtained four
consecutive spectra, each lasting 30 min. Each spectrum was analyzed
separately to derive a Doppler shift. The four velocities were
averaged, to yield the final velocity for each night. The uncertainty
in the mean was computed from the standard deviation of the four
separate measurements, giving an internal error for each night.
These two Lick velocities were --24.2 $\pm$ 6.2 \mbox{m s$^{-1}~$} and +24.1 $\pm$ 3
\mbox{m s$^{-1}~$} on the two nights, respectively, implying that the velocity of HD168443
increased by +48.3 $\pm$ 7 \mbox{m s$^{-1}~$}. This velocity increase agrees
with the prediction of the model (Keplerian plus trend) of +52.4 \mbox{m s$^{-1}~$}.
These Lick velocities were obtained 26 days after the last Keck
measurements were made, on JD=24511074.785, shown in Figure 4. The
alternative model without an imposed velocity trend has a longer
period of $P$=64.3 d, and its predicted change in velocity is --60
\mbox{m s$^{-1}~$}, clearly in conflict (wrong sign) with the observed rise of 48
\mbox{m s$^{-1}~$}. Thus, both the lower velocity RMS and the Lick measurements
favor the model that contains a Keplerian with $P$=57.8 d and a
velocity trend.
The best--fit velocity trend has slope of 89.4 \mbox{m s$^{-1}~$} per yr which could
be caused by a second more distant companion. If so, its minimum
orbital period is $\sim$4 yr. As a benchmark, Jupiter causes a trend
of $\sim$4 \mbox{m s$^{-1}~$} per yr in the Sun during 6 yr. Thus, if the period of
the hypothetical second companion to HD168443 were $\sim$12 yr, its
mass would be at least $\sim$25 M$_{\rm JUP}~$. For the shortest possible
period of 4 yr, the companion mass would be at least 15 M$_{\rm JUP}~$. In
both cases, the companion would be considered a ``brown dwarf'' and
quite possibly a hydrogen--burning star, depending on the actual
period and $\sin i$. Prospective stellar companions are discussed in
section 3.2
\subsection{Velocities for HD114762}
We have obtained 33 velocity measurements for HD114762 since 1994
Nov. They are plotted versus orbital phase in Figure 5. The unseen
companion to this star has been described by Latham et al. (1989),
Mazeh et al. (1996), Cochran et al. (1991), and Hale (1995). Our
velocities offer new measurements of the orbital parameters,
$P$=84.03$\pm$0.1 d, $e$=0.334$\pm$0.02, $K$=618$\pm$6 \mbox{m s$^{-1}~$},
$\omega$=201$\pm$3 deg, and $T_p$=JD2450225.30$\pm$0.6. A revised
mass for HD114762 has been measured by Ng and Bertelli (1998) and
Gonzalez (1998), giving, $M$=0.82$\pm$0.03 M$_{\odot}$, based on its
Hipparcos distance ($d$=40.57 pc, Perryman et al. 1997) and new
stellar evolution models. This stellar mass and the orbital
parameters imply that the companion has $M \sin i~$ = 11.02$\pm$0.5 M$_{\rm JUP}~$.
If the companion mass is truly small compared to the primary star then
the semimajor axis is $a$=0.35 AU. However, Cochran et al. (1991) and
Hale (1995) provide arguments that the companion mass may be large,
possibly stellar. Since that work, several additional considerations
have emerged regarding its status as a candidate planet. HD114762 is
the only planet-candidate found with modest velocity precision, rather
than with high precision of $\sim$10 \mbox{m s$^{-1}~$}. That precision along with
the large survey size (Latham et al. 1989) makes the discovery of an
face-on system more likely. Further, HD114762 has [Fe/H]=-0.6, substantially
more metal--poor than any other planet candidate (Gonzalez 1998). The
standard model of planet formation requires heavy elements to form the
dust which was presumably not abundant in the protoplanetary disk
around HD114762. Finally, the value of $M \sin i~$ (11.02 M$_{\rm JUP}~$) is much
higher than that for all other planet candidates, the highest of which
is $M \sin i~$ = 7.4 M$_{\rm JUP}~$ (70 Vir). Nonetheless, we include HD114762 as a
candidate planetary object in this complete compilation.
\section{Search for Stellar Companions}
\subsection{HD210277}
We examined HD210277 for companion stars as follows. Duquennoy and
Mayor (1991) made 8 radial velocity measurements spanning 6 yr which
exhibited no variation above 220 \mbox{m s$^{-1}~$}. Ground--based astrometry from
1989--1993 revealed no motion at a level of 0.01 arcsec (Heintz 1994).
Lunar occultation measurements revealed no companion to HD210277,
with detection thresholds of $\Delta$ V$_{\rm mag}<$2 within 3 mas
Meyer et al. (1995) . The above measurements, especially those of
Duquennoy and Mayor, jointly rule out stellar companions with masses
as low as 0.1 M$_{\odot}$ within 10 AU. A 0.1 M$_{\odot}$$~$ dwarf orbiting 10
AU from HD210277 would induce velocity variations with semi-amplitude
of 700 \mbox{m s$^{-1}~$} ($\times \sin i$) and a period of 30 yr, detectable as a
trend in velocities of Duquennoy and Mayor (1991), but not observed.
The astrometry of Heintz similarly rules out a stellar companion with
mass down to the substellar limit within 5 AU, which would have induced
astrometric wobbles of 0.02 arcsec.
To search for possible stellar companions beyond 10 AU, we observed
HD210277 on 8 September 1998 UT using the Lawrence Livermore National
Lab adaptive optics system (Max et al. 1997), which is mounted at the
f/17 Cassegrain focus of the Lick Observatory Shane 3-m telescope.
The adaptive optics system performs real-time compensation of
atmospheric seeing using a Shack-Hartmann type wavefront sensor with a
127-actuator deformable mirror. In its current configuration, 61 of
the actuators are actively controlled. For these observations, image
compensation was done with a sampling frequency of 250~Hz using
HD210277 (V=6.54) itself as a wavefront reference, achieving a
closed-loop bandwidth of 20 Hz.
We acquired images using the Lick facility near-IR camera LIRC2
(Gilmore, Rank, \& Temi 1994). The camera has a 256~$\times$~256
pixel HgCdTe NICMOS-3 detector and, when coupled with the AO system, a
plate scale of 0.12 arcsec/pixel. We used both a narrow-band
($\Delta\lambda/\lambda = 0.01$) filter centered on Br$\gamma$ (2.166
\micron) and the broad-band K$^\prime$ filter (1.95-2.35 \micron;
Wainscoat \& Cowie 1992) to span a wide range in radii with good
sensitivity and dynamic range. The star was dithered to 4 positions on
the detector, with total integrations of 240~s in each filter. Images
were reduced in a standard fashion for near-IR images --- bias
subtraction, flat-fielding, and sky subtraction using a master sky
frame constructed from all the images. The angular resolution as
measured by the full-width at half-maximum (FWHM) of the Br$\gamma$
images is 0.18 arcsec, and the images have a mean Strehl ratio of
0.45. The Br$\gamma$ data are most sensitive to companions inside of 0.5
arcsec, and the $K^\prime$-images are more sensitive at larger radii.
Figure 6 presents our 4$\sigma$ upper limits to any stellar companions
to HD210277 combined with $K$-band flux ratios for main sequence
companions derived from Kirkpatrick \& McCarthy (1994). Only the
inner radii are shown for clarity; the deepest portion of the our
images cover 12 arcsec in radius. Our AO data are nearly
diffraction-limited, ruling out any main-sequence dwarf companions
earlier than spectral type M0 from 0.2--12 arcsec (4.2--250 AU) . In
addition, the high Strehl ratio means the images are very sharply
peaked and sensitive to even the lowest mass M dwarfs outside of 0.5
arcsec (11 AU). We rule out any main sequence companion with a
separation of 0.8 arcsec (17 AU) to 12 arcsec (250 AU).
We further rule out any stellar companions out to $\approx1\arcmin$
separations using 2.5 arcsec FWHM $J$ and $K^\prime$ images obtained
from the Lick 3-m with the UCLA two-channel infrared camera known as
"Gemini" (McLean et al. 1994). The $J$ and $K^\prime$ data were taken
simultaneously on 09 October 1998 UT with two 256~$\times$~256
detectors, a Rockwell HgCdTe NICMOS-3 one for $J$ and a Hughes-SBRC
InSb one for $K^\prime$. There are a handful of $K\approx 14-17$
unresolved sources in these images; the majority of these also appear
on the Palomar Sky Survey. Their $J-K$ colors are consistent with
background stars or galaxies, and their numbers are in accord with
field $K$-band galaxy counts (Szokoly et al. 1998). Comparing our
images with those on the Palomar Sky Survey, the only source which
shows noticeable proper motion is HD210277 itself, which exhibits
a magnitude and direction consistent with its nominal proper motion.
\subsection{HD168443}
We have searched for stellar companions to HD168443 in several ways.
A literature search turned up no known companions. We searched for
superimposed spectral lines from a secondary star in the 8400 \AA\
region of our Lick spectra (near the CaII IR triplet). No such lines
were found at a threshold of a few percent of the continuum. This
nondetection rules out any main--sequence companions more massive than
0.5 M$_{\odot}$ within 2.5 arcsec (95 AU) of HD168443, as we use a slit
width of 5 arcsec with the Lick Observatory Coud\'e Auxiliary
Telescope.
Carney et al. (1994) obtained 8 radial velocity velocity measurements
of HD168443 spanning 5 yr and detected no variation above the errors
of 400 \mbox{m s$^{-1}~$}. Any stellar companion more massive than 0.1 M$_{\odot}$$~$
orbiting within 5 AU would have been revealed, except for extreme
values of $\sin i$. The Hipparcos astrometry of HD168443 recorded no
astrometric motion at a level of 2 mas during several years (Perryman
et al. 1997). A stellar companion having 0.1 M$_{\odot}$\ at 5 AU would
induce a (curved) astrometric reflex motion of 16 mas during $\sim$3
yr (1/4 orbital period), as viewed from a distance of 38 pc. Such a
wobble evidently did not occur, thus ruling out stellar companions
within 5 AU, consistent with the velocity data. Hipparcos would not
easily detect stellar companions orbiting beyond 5 AU, as the (more
linear) reflex motion could be absorbed into the assessment of proper
motion. Thus, stellar companions orbiting beyond 5 AU might escape
detection by both the Carney et al. velocities and the Hipparcos
astrometry.
We have not obtained an adaptive optics image of the star, leaving us
little information about stellar companions farther than 5 AU.
However, we were compelled to include a velocity slope of 89 \mbox{m s$^{-1}~$} per
yr in the model of our velocities (Fig 4). This slope could indicate
a stellar companion beyond 5 AU, or a brown dwarf somewhat closer. As
a benchmark, a 0.1 M$_{\odot}$$~$ companion orbiting at 10 AU would induce a
typical velocity slope of 100 \mbox{m s$^{-1}~$} per yr ($\times \sin i$). Our Keck
velocities are consistent with such a stellar companion as well as
more distant and correspondingly more massive ones. The upper mass
limit of 0.5 M$_{\odot}$, imposed by the lack of secondary lines, implies
that the companion must reside within $\sim$30 AU in order to induce
the observed velocity slope of 89 \mbox{m s$^{-1}~$} per yr.
A consistency check on the putative companion is provided by comparing
the absolute velocities obtained by Carney et al (1994) to those found
here. Carney et al. obtained 8 velocity measurements centered at
epoch $\sim$1990 which exhibited an average of --48.9 \mbox{km s$^{-1}~$}, with $\sigma$=0.4
\mbox{km s$^{-1}~$}. Our observation on 1998 Aug 25 gave a velocity of --49.0$\pm$2
\mbox{km s$^{-1}~$} , which agrees with the Carney measurement within the 2 \mbox{km s$^{-1}~$}
uncertainty. This implies an upper limit to the velocity trend of 2
\mbox{km s$^{-1}~$} per 8 yr, which is indeed larger than the trend we actually
detect of 89 \mbox{m s$^{-1}~$} per year.
In summary, any stellar companion must reside beyond 5 AU but not
beyond 30 AU to explain the observed velocity trend, and its mass must
be less than 0.5 M$_{\odot}$ to explain the lack of stellar secondary lines.
A direct search for a stellar companion located 0.15--1 arcsec from HD
168443 seems warranted.
\section{Discussion}
The two extrasolar planet candidates suggested by the data in this paper bring
the total number of such candidates to 17. These candidates all have
$M \sin i~$ $\la$ 5 M$_{\rm JUP}~$ except 70 Vir ($M \sin i~$ = 7.4 M$_{\rm JUP}~$) and HD114762
($M \sin i~$ = 11 M$_{\rm JUP}~$)
which some would place in the ``brown dwarf'' class (Black 1998).
Table 4 lists the basic orbital
parameters and $M \sin i~$ of all 17 known planetary candidates.
A few of the orbital parameters have been slightly modified,
based on our own recent measurements and orbital fits.
The typical uncertainty in the orbital
eccentricity is 0.03, based on Monte Carlo
simulations of the Keplerian fits to data with artificial noise
Table 4 shows that all 9 planet candidates that have a$>$0.2 AU have
eccentricities above 0.1, larger than that for both Jupiter ($e$=0.048) and
Saturn ($e=$0.055) . Figure 7 shows a plot of orbital
eccentricities vs. semimajor axis. All extrasolar planets orbiting
closer than 0.1 AU have small eccentricities. While possibly
primordial, these near--circular orbits for close planets may have
been induced by tidal circularization (cf., Rasio et al. 1996, Marcy
et al. 1997, Terquem et al. 1998).
Apparently, Jupiter--mass companions orbiting from 0.2--2.5 AU, immune
to tides, have large orbital eccentricities.
Apparently, some mechanism commonly produces eccentric orbits
in Jupiter--mass companions that reside from 0.2 -- 2.5 AU in
main--sequence stars. These eccentric planets represent a general
property of 0.3--1.2 M$_{\odot}$\ stars .
Figure 8 shows orbital eccentricities vs. $M \sin i~$. No trend is
apparent at first glance, suggesting that orbital eccentricity is not
correlated with planet mass, within the mass range 0.5--5 M$_{\rm JUP}~$.
However, all 5 planets with $M \sin i~$$<$1.1 M$_{\rm JUP}~$ reside in nearly
circular orbits. This correlation may be a selection effect, as the
lowest--mass planets are more easily detected close to their host
stars in order to induce a detectable Doppler reflex signal. These
close planets are all subject to tidal circularization. Thus, the low
eccentricities among the lowest mass giant planets may not be
considered intrinsic to planet formation.
Figure 9 shows $M \sin i~$ vs. semimajor axis for all 17 planet candidates.
The detectability of planetary companions is shown as the curved line
near the bottom (Cumming et al. 1999).
Apparently, the distribution of planet masses is not a strong function
of semimajor axis from 0.05--2.5 AU for the range of detectable masses, 1--6
M$_{\rm JUP}~$. There is no paucity of either the most or least
massive companions at either extreme of semimajor axis. Of course
there may be some blurring in mass due to $\sin i$. Nonetheless,
we conclude
that if orbital migration within a gaseous disk brings the giant
plants inward, neither that process nor the halting mechanism seems to
depend on planet mass.
Figure 10 shows a histogram of Msini within the range
0$<$15 M$_{\rm JUP}~$ for known companions to main--sequence stars. The
distribution of $M \sin i~$ shows a rapid decline at
roughly 4 M$_{\rm JUP}~$. There are no companions having $M \sin i~$ = 7.5--11 M$_{\rm JUP}~$
and those massive companions would have been easily detected. This
absence seems statistically significant relative to the 14
companions having $M \sin i~$ =0.4--5 M$_{\rm JUP}~$. All selection effects favor
detection of the high--mass companions, and thus the apparent drop in
the $M \sin i~$ histogram from 4 to 7 M$_{\rm JUP}~$ must be real. This drop implies that
the distribution of companion masses, d$N$/d$M$, must indeed
exhibit a decline at $\sim$5 M$_{\rm JUP}~$ with increasing mass, within 2.5 AU.
The highest value of $M \sin i~$ among planet candidates (Fig 10) is for
HD114762 which has $M \sin i~$=11.02 M$_{\rm JUP}~$, which is well above the decline
at $\sim$5 M$_{\rm JUP}~$. Its unknown $\sin i$ leaves an important question
unanswered regarding its true mass and hence any affiliation with
``planets.'' With that possible exception of HD114762, the planetary
mass distribution certainly declines rapidly for masses above 5 M$_{\rm JUP}~$.
The origin of the distribution of semimajor axes and eccentricities
now presents a puzzle. In the standard paradigm, giant planets form
outside 4 AU (Boss 1995, Lissauer 1995). Inward orbital migration
(Lin et al. 1995, Trilling et al. 1998) within the gaseous
protoplanetary disk has been suggested to explain the small orbits
detected to date among extrasolar planets. Such migration makes two
predictions that appear testable. First, orbital migration in a
viscous, gaseous environment is expected to preserve circular orbits
under most circumstances (but see Artymowicz 1993). In contrast, all
9 planet candidates orbiting between 0.2 and 2.5 AU have non--circular
orbits. Second, the orbital migration time scale is proportional to
the orbital period, which leads to rapid orbital decay for
successively smaller orbits. In contrast, the observed orbital
semimajor axes are spread throughout 0.1 to 2.5 AU (though not
necessarily distributed uniformly). No obvious mechanism is known to
halt the migration for these orbits.
Apparently, the orbits with sizes of $\sim$1 AU and large
eccentricities ($e>$0.1) require physical processes that are not explicitly
included within the context of quiescent migration in a dissipative medium.
Scattering of orbits by
other planets, companion stars, or passing stars in the young star
cluster offer mechanisms for producing eccentric orbits (Rasio and
Ford 1995, Lin and Ida 1996, Weidenschilling and Marzari 1996,
Laughlin and Adams 1998). However, these mechanisms do not explicitly
predict small orbits of $\la$1 AU, because significant
energy must be lost from the original orbits of $\sim$5 AU.
One possibility is that planet--scattering continues to occur during
the final era of the remnant gaseous protoplanetary disk. If the disk
remains intact within the inner few AU where the original gas density
was highest, the disk can serve as the reservoir into which the
planet's orbital energy can be deposited, either by dynamical friction
or by tidal interaction between planet and disk. In this scenario,
scattered planets would reside in eccentric orbits subjecting them to
dissipation during periastron passages. Clearly detailed models are
required that include both planet scattering and the dissipative
effects of a weak inner gaseous disk to determine the resulting
planetary orbits.
In any case, we currently have little information about giant planets
that orbit beyond 3 AU. We expect to obtain such information in the coming
years as Doppler programs extend their time baseline. Planets beyond 3 AU
may well reside in predominantly
circular orbits. A population of giant planets that never suffered
significant scattering or migration could comprise these Jupiter
analogs. The lack of main sequence stars having reflex Doppler
periodicities with amplitudes above 30 \mbox{m s$^{-1}~$} already indicates a
paucity of planets having $M>$3 M$_{\rm JUP}~$ within 5 AU (Cumming et
al. 1999). It remains to be determined if the planet mass function
rises rapidly for smaller masses.
\acknowledgements
We thank Kevin Apps for assessment of stellar characteristics, Phil
Shirts for his measurement of Ca II H\&K, and Eric Williams for work
on HD114762. We thank Claire Max, Scot Olivier, Don Gavel and Bruce
Macintosh for the development of the Lick adaptive optics system and
for assistance with the observations. We thank Tom Bida, Tony Misch
and Wayne Earthman for technical help with instrumentation. We
acknowledge support by NASA grant NAGW-3182 and NSF grant AST95-20443
(to GWM), and by NSF grant AST-9619418 and NASA grant NAG5-4445 (to
SSV), and by Sun Microsystems. We thank the NASA and UC telescope
assignment committees for allocations of telescope time.
\clearpage
|
\section{Introduction}
Q2237+0305 comprises a source quasar with a redshift of $z=1.695$ that is gravitationally lensed by a foreground galaxy at $z=0.0394$ producing 4 images with separations of $\sim 1''$. Each of the 4 images are observed through the bulge of a galaxy which has an optical depth in stars that is of order unity (eg. Kent \& Falco 1988; Schneider et al. 1988; Schmidt, Webster \& Lewis 1998). In addition, the proximity of the lensing galaxy means that the effective transverse velocity may be high. The combination of these facts make Q2237+0305 the ideal object from which to study microlensing. Indeed, Q2237+0305 is the only object in which cosmological microlensing has been confirmed (Irwin et al. 1989; Corrigan et al. 1991).
Numerical microlensing simulations (eg. Wambsganss, Paczynski \& Schneider 1990; Witt, Kayser \& Refsdal 1993) have shown that the statistics of microlensed high magnification events (HMEs) obtained from long term monitoring may provide information on properties of the lens such as the stellar mass function and the percentage of mass in stars for the bulge. However, the monitoring period required is greater than 100 years (Wambsganss, Paczynski \& Schneider 1990). There are several other unknown quantities in the problem. These include the magnitude and direction of any transverse motion, as well as the source size. In particular, the magnitude and direction of the transverse motion are degenerate with the density of caustics (a function of the mean compact object mass) for a given set of microlensing statistics.
Microlensed fluctuation in the quasars continuum results from motion due to both a galactic transverse velocity and to the random proper motion or stream motion of stars and compact objects. Wyithe, Webster \& Turner (1999a) (hereafter WWTa) define the equivalent transverse velocity as the transverse velocity in a model containing stars with static positions, that produces a rate of microlensing most closely resembling that of a model containing stellar proper motions. Foltz et al. (1992) measured the central velocity dispersion of Q2237+0305 to be $\sim$ 215$km\,sec^{-1}$, and theoretical models (Schmidt, Webster \& Lewis 1998) predict a value of $\sim$ 165$km\,sec^{-1}$. If the dispersion is isotropic then (as discussed in Wyithe, Webster \& turner (1999b) (hereafter WWTb)) the equivalent transverse velocity calculated from the overall microlensing rate (all 4 images) of Q2237+0305 is larger than this value. The microlensing rate that results from the line-of-sight velocity dispersion (of an isotropic distribution) is therefore comparable to that of the likely transverse velocity. This means that the often made assumption that random proper motions provide a negligible contribution to microlensing in Q2237+0305 is incorrect.
While the microlensing rate resulting from random stellar motions is dependent on the mean microlens mass, the equivalent transverse velocity (WWTa; WWTb) that describes the microlensing rate is not. We use this fact to break the degeneracy between microlensing rate and mean microlens mass, and hence obtain useful limits on the mass function along the line-of-sight through the bulge of the lensing galaxy.
This paper is presented in 5 parts. Sec.~\ref{mass} contains a general discussion of microlensing and the mass function. Sec.~\ref{models} discusses the numerical methods used to model microlensing in Q2237+0305 and Sec.~\ref{mass_limits} describes a method to place limits on the mass function through consideration of microlensing due to both a transverse velocity and a stellar velocity dispersion.
\section{Microlensing and the stellar mass function}
\label{mass}
\begin{table*}
\begin{center}
\caption{\label{ml}A collection of results from microlensing data for the average mass of compact objects in galactic halos and/or bulges of the Milky Way and Quasar lensing galaxies.}
\begin{tabular}{|c|c|c|}
\hline
Milky Way halo & Results for mean masses of dark halo objects toward LMC. & $\langle m\rangle=0.13^{+.08}_{-.05}M_{\odot}$\\
(Alcock et al. 1997a) & The ranges include statistical ($1\sigma$) as well as systematic &\hspace{7mm} -$0.55^{+.38}_{-.21}M_{\odot}$ \\
& uncertainties due to choice of halo model. & \vspace{3mm}\\
Milky Way bulge & Microlensing event time-scales are consistent with a mean & $0.1M_{\odot}\la\langle m\rangle\la1.0M_{\odot} $ \\
(Alcock et al. 1997b) & mass of compact objects and stars in the Milky Way bulge.\vspace{3mm}\\
0957+561 halo & The limit has 99\% and 95\% significance levels for assumed & \\
(Schmidt \& Wambsganss 1998) & source sizes of $4\times10^{14}cm$ and $4\times10^{15}cm$, and scales as & $\langle m\rangle \ga 0.001\left(\frac{v_{t}}{600}\right)^{2}$ \\
& the square of the assumed transverse velocity $v_{t}$. & \vspace{3mm}\\
2237+0305 bulge and halo & The microlensing rate of current light curves is consistent & \\
(Lewis \& Irwin 1996) & with a mean mass of stars and compact objects. The value & $0.1M_{\odot}\la\langle m\rangle\left(\frac{v_{t}}{600}\right)^{2} \la10M_{\odot} $\\
& scales as the square of the assumed transverse velocity $v_{t}$. & \vspace{3mm}\\
2237+0305 bulge and halo & The proper motions of microlenses provide a minimum micro- & $\langle m\rangle = 0.010^{+?}_{-.005}M_{\odot}$ \\
This paper. & lensing rate. The limits given for the mean mass have a 99\% & \hspace{7mm}$-0.29^{+?}_{-.18}M_{\odot}$ \\
& significance level, and depend on the bulge dynamics assumed. & \\\hline
\end{tabular}
\end{center}
\end{table*}
Stellar mass functions have traditionally been measured through the combination of an observed luminosity function and an empirical mass-luminosity relationship. However, because of the difficulties inherent in the observations of faint stars, these determinations become uncertain as the hydrogen burning limit ($\sim 0.08M_{\odot}$) is approached and crossed. In contrast to this approach, microlensing uses the mass of a lens to magnify the observed flux of a background source through gravitational deflection of the light bundle. The statistics obtained are therefore free of any bias introduced by the hydrogen burning limit, and so microlensing is a powerful tool for determining the contribution to the mass function of low mass stars and dark compact objects.
Gravitational microlensing is observed in two very different regimes. Paczynski (1986) suggested surveying millions of stars in the Magellanic Clouds for microlensing induced flux variation as a way of detecting solar mass compact objects in the halo of our own galaxy. Several groups have undertaken such searches (eg. Alcock et al. 1997a; Renault et al. 1998). In a similar vein, Paczynski (1991) and Griest et al. (1991) suggested microlensing experiments towards the Galactic bulge as a method to probe the masses of stars along the line-of-sight in the Galactic disc. Microlensing searches towards the Galactic bulge have since found a higher rate of microlensing events than are found towards the LMC (eg. Alcock et al. 1997b; Udalski et al. 1994). In a very different microlensing regime, the observed flux of a back-ground quasar can be altered by gravitational lensing by stars or halo objects in a foreground galaxy. Q2237+0305 is an example of a quasar that lies very close to a galactic line-of-sight. In such cases multiple imaging occurs, allowing microlensed variation to be easily separated from intrinsic fluctuations which are observed in all images.
The analyses of microlensing in the Galactic and cosmological regimes are very different. While there are only a few sources (quasar images) in the cosmological case rather than the millions available in Galactic microlensing searches, the optical depth (or equivalently the probability of lensing) is $\sim 10^{6}$ greater. The transverse velocity is a critical parameter for the determination of a mass or masses responsible for a microlensing event, regardless of the microlensing scenario. Unfortunately it is unknown in both cases. In Galactic microlensing calculations the velocities are inferred from an assumed distribution. However in the cosmological case the transverse velocities of the lensing objects are not independent. Rather, they are equal except for the contribution of the individual stellar proper motions. Moreover, cosmological microlensing results from the gravitational contribution of a large ensemble of many hundreds of masses rather than on a single microlens. Microlensed, multiply-imaged quasars are an excellent probe of the mass function of compact objects along the quasar image line-of-sight because they interact with a large number of microlenses.
Several authors have placed limits on the masses of microlenses responsible for cosmological microlensing. Schmidt \& Wambsganss (1998) use the lack of observed variation in Q0957+561 to place a lower limit on the mass of microlenses in the halo of the lensing galaxy. They rule out a mean mass of halo objects $\langle m\rangle\ll 10^{-2}M_{\odot}$. However their determination is dependent on the source size, and the fraction of smooth matter employed in their calculations. A transverse velocity of $v_{t}=600\,km\,sec\,^{-1}$ is assumed. The uncertainty in this assumed value is the most serious problem for this kind of analysis because the values of mass obtained are $\propto v_{t}^{2}$. Lewis \& Irwin (1996) compared the monitoring data of Q2237+0305 with simulations using a structure function to analyse variability. They too assume a transverse velocity of $v_{t}=600\,km\,sec\,^{-1}$ and conclude that the mean mass of objects in Q2237+0305 is $0.1M_{\odot}\la\langle m\rangle \la 10M_{\odot}$. This result is also proportional to the square of the unknown transverse velocity.
While they do not measure precisely the same quantity, the above results are supportive of those of the MACHO experiment (eg Alcock et al. 1997a,b). From microlensing observations towards the LMC by the Galactic halo, Alcock et al. (1997a) conclude that the average MACHO masses are $0.08-0.93M_{\odot}$. The quoted range includes both statistical uncertainties and systematic effects introduced by the assumption of halo model. In addition, they find evidence for an excess of events over those predicted from stellar lensing alone. Alcock et al. (1997b) find that the range of time-scales towards the Galactic bulge are consistent with microlens masses of $0.1M_{\odot}\la\langle m\rangle \la 1.0M_{\odot}$. Tab.~\ref{ml} summarises the results described.
\section{The microlensing models}
\label{models}
\subsection{Microlensing models for Q2237+0305}
\begin{table}
\begin{center}
\caption{\label{params}Values of the total optical depth and the magnitude of the shear at the position of each of the 4 images of Q2237+0305. The quoted values are those of Schmidt, Webster \& Lewis (1998). $\kappa_{*}$ and $\kappa_{c}$ are the optical depths in stars and in smoothly distributed matter respectively.}
\begin{tabular}{|c|c|c|}
\hline
Image & $\kappa=\kappa_{*}+\kappa_{c}$ & $|\gamma|$ \\ \hline
A & 0.36 & 0.40 \\
B & 0.36 & 0.40 \\
C & 0.69 & 0.71 \\
D & 0.59 & 0.61 \\ \hline
\end{tabular}
\end{center}
\end{table}
Throughout the paper, standard notation for gravitational lensing is used. The Einstein radius of a 1$M_{\odot} $ star in the source plane is denoted by $\eta_{0} $. The normalised shear is denoted by $\gamma$, and the convergence or optical depth by $\kappa$. The model for gravitational microlensing consists of a very large sheet of point masses that simulates the section of galaxy along the image line-of-sight, together with a shear term that includes the perturbing effect of the mass distribution of the lensing galaxy as a whole. The normalised lens equation for a field of point masses with an applied shear in terms of these quantities is
\begin{equation}
\vec{y}= \left( \begin{array}{cc}
1-\kappa_{c}-\gamma & 0 \\
0 & 1-\kappa_{c}+\gamma
\end{array} \right)\vec{x} + \sum_{j=0}^{N_{*}}m^{j}\frac{(\vec{x}^{j}-\vec{x})}{|\vec{x}^{j}-\vec{x}|^{2}}
\label{lens_map}
\end{equation}
Here $\vec{x}$ and $\vec{y}$ are the normalised image and source positions respectively, and the $\vec{x}_{i}^{j}$ and $m^{j}$ are the normalised positions and masses of the microlenses. $\kappa_{c}$ is the optical depth in smoothly distributed matter. Eqn. \ref{lens_map} is solved for the macroimage magnification at many points along a predefined source trajectory through the inversion technique of Lewis et al. (1993) and Witt (1993). The region of the lens plane in which image solutions need to be found to ensure that $99\%$ of the total macro-image flux is recovered from a source point was described by Katz, Balbus \& Paczynski (1986). The union of areas of in the lens plane corresponding to the flux collection area of each point on the source line is known as the shooting region, the method for determining the dimensions of which is described in Lewis \& Irwin (1995), and Wyithe \& Webster (1999). The radius of the disc of point masses is chosen to be 1.2 times that required to cover this shooting region.
We assume the macro-parameters for Q2237+0305 calculated by Schmidt, Webster \& Lewis (1998) (the values are shown in Tab.~\ref{params}). Two models are considered for the mass distribution, one with no continuously distributed matter and one where smooth matter contributes 50\% of the surface mass density. Both the microlensing rate due to a transverse velocity (Witt, Kayser \& Refsdal 1993; Lewis \& Irwin 1996; WWTb), as well as the corresponding rate due to proper motions (WWTa) are not functions of the detail of the microlens mass distribution, but depend only on the mean microlens mass.
The mean of the microlens mass function can therefore be determined independently from its form. On the other hand, information on the form of the mass function is unobtainable through consideration of the microlensing rate. We limit our attention to models in which all the point masses have identical mass.
Each of our models was computed for a source track of length 10$\eta_{o}$ (corresponding to $\sim70$ years for an effective transverse velocity of $600\,km\,sec^{-1}$). We set 500 to be the minimum number of stars in a model. Two orientations were chosen for the transverse velocity with respect to the galaxy, with the source trajectory being parallel to the A$-$B or C$-$D axes. At each orientation, 100 simulations were made for each of the 4 images of Q2237+0305 in combination with each of the model mass distributions. The two orientations bracket the range of possibilities, and because the images are positioned approximately orthogonally with respect to the galactic centre correspond to shear values of $\gamma_{A},\gamma_{B}<0,\gamma_{C},\gamma_{D}>0$ and $\gamma_{A},\gamma_{B}<0,\gamma_{C},\gamma_{D}>0$ respectively. Tab.~\ref{param} shows the number of stars used for each of these models along with the mean magnification $\langle \mu \rangle$, and the theoretical magnification $\langle \mu_{th}\rangle$ for comparison. The average magnification was found from the combination of the two sets of models, and the error calculated from the standard deviation of a subset of 6 simulations (3 per model orientation). Fig.~\ref{amp_dist} shows the magnification distributions for each image in each model. Simulations having $\gamma<0$ and $\gamma>0$ are represented by light and dark lines respectively. The distributions are quantitatively comparable to those presented for similar values of $\kappa$ and $\gamma$ in Lewis \& Irwin (1995), and demonstrate that the magnification distribution is independent of the source direction as required.
\begin{table}
\begin{center}
\caption{\label{param}Values for the number of stars in the microlensing models (left column $\gamma>0$; right column $\gamma<0$), the average model magnifications and the theoretical values for comparison.}
0\% smooth matter
\begin{tabular}{|c|c|c|c|c|}
\hline
Image & \multicolumn{2}{c}{No.\ Stars} & $\langle \mu \rangle$ & $\langle \mu_{th} \rangle$ \\ \hline\hline
A/B & 763 & 500 & 3.97$\pm$.28 & 4.01 \\
C & 1022 & 500 & 2.39$\pm$.11 & 2.45 \\
D & 2887 & 1301 & 4.79$\pm$.14 & 4.90 \\ \hline \\\\
\end{tabular}
50\% smooth matter
\begin{tabular}{|c|c|c|c|c|}
\hline
Image & \multicolumn{2}{c}{No.\ Stars} & $\langle \mu \rangle$ & $\langle \mu_{th} \rangle$ \\ \hline\hline
A/B & 500 & 500 & 4.01$\pm$.19 & 4.01 \\
C & 500 & 500 & 2.44$\pm$.09 & 2.45 \\
D & 1080 & 500 & 4.84$\pm$.20 & 4.90 \\ \hline
\end{tabular}
\end{center}
\end{table}
\begin{figure}
\vspace*{85mm}
\special{psfile=fig1a.epsi hscale=40 vscale=40 hoffset=-70 voffset=250 angle=270}
\special{psfile=fig1b.epsi hscale=40 vscale=40 hoffset=-70 voffset=130 angle=270}
\caption{\label{amp_dist} The magnification distributions for images A/B (left), C (centre) and D (right). The top and bottom rows show the distributions for models that have no smooth matter component, and a 50\% smooth matter component respectively. The light and dark lines represent distributions that are sampled parallel to and at right angles to the shear.}
\end{figure}
We assume that microlensing is produced through the combination of a galactic transverse velocity with each of two classes of proper motion for individual stars with respect to the galaxy: an isotropic velocity dispersion and a circular stream motion. Moreover, we assume that the magnitude of the dispersion or stream motions are the same for each of the four images. We take the theoretical value of $\sigma_{*}\sim 165\,km\,sec^{-1}$ for the line-of-sight velocity dispersion of the stars in the galactic bulge. This is lower than the value observed by Foltz et al. (1992), so lower limits placed on the mass are more conservative.
\subsection{The effective transverse velocity}
\begin{figure}
\vspace*{80mm}
\special{psfile=fig2.epsi hscale=95 vscale=95 hoffset=-245 voffset=250 angle=270}
\caption{\label{vel_lim} Plots of the cumulative probability ($P_v$) for the effective transverse velocity. The solid and dot-dashed lines correspond to the models with trajectory directions that are aligned with the C-D and A-B image axes respectively. The dark and light lines represent results from models containing 0\% and 50\% continuously distributed matter. The upper and lower plots correspond to the cases where the photometric error was assumed to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D. The model assumed $1M_{\odot}$ microlenses.}
\end{figure}
We define the effective transverse velocity as the transverse velocity that produces a microlensing rate from a static field model equal to that of the observed light-curve. The effective transverse velocity therefore describes the microlensing rate due to the combination of the effects of a galactic transverse velocity and random microlens proper motion.
Our calculation of effective transverse velocity utilises the cumulative histogram of derivatives calculated from the 6 difference light-curves (A-B, A-C, A-D, B-C, B-D, C-D). Through computation of the difference light-curves, intrinsic flux variation, as well as the systematic component of the observational uncertainty are removed from the data. WWTb describes a procedure for determining the probability that the effective transverse velocity is less than an assumed value. That paper provides correct upper limits, but does not obtain the cumulative probability function. A modified procedure to determine the cumulative probability for effective transverse velocity from ensembles of mock data-sets is described below.
Many mock observations of the 6 difference light-curves were produced from model light-curves with a sampling rate and period that are identical to the published monitoring data (Irwin et al. 1989, Corrigan et al. 1991, $\O$stensen et al. 1996). Observational errors were also simulated by applying a random fluctuation to each point on the light curve, distributed according to a Gaussian with a half-width $\sigma$ describing the published photometric error.
The simulations used two different estimates for the uncertainty in the photometric magnitudes. In the first case a small error was assumed (SE). For images A and B, $\sigma_{SE}$=0.01 mag, and for images C and D $\sigma_{SE}$=0.02 mag. In the second case, a larger error was assumed (LE). For images A and B, $\sigma_{LE}$=0.02 mag and for images C and D $\sigma_{LE}$=0.04 mag. The observational error in Irwin et al. (1989) was 0.02 mag. We note that the adoption of a larger error leads to a smaller bound on the measured upper limit of effective transverse velocity, and subsequently a larger lower bound on the mean mass.
The simulations assume a point source since the value of effective transverse velocity measured with a model is not a sensitive function of the source size (in the range of interest) assumed (WWTb). The insensitivity arises from the fact that the measurement is derived predominantly from points that are not part of a HME.
We define a set of effective transverse velocities ($V_{eff}$) and determine the probability, based on light-curve derivatives that each of these correctly describes the true value for Q2237+0305 ($V_{gal}$). For each model, 5000 mock observations were produced at each effective transverse velocity from the four sets (one per image) of 100 10$\eta_{o}$ simulated light-curves. Due to the finite sampling period, histograms of microlensed light-curve derivatives produced from the individual mock observations describe typical but not average behaviour. Therefore an average histogram was also computed from each set of mock observations. For each mock observation at each pre-defined effective transverse velocity ($V_{eff}$), a mock measurement of effective transverse velocity $v_{eff}$ was made by minimising the KS difference between the average histogram for $v_{eff}$ and the histogram of the mock observation. Thus we calculate the function representing the likelihood $p_{lhood}(v_{eff}|V_{eff})$ for observing $v_{eff}$ given an assumption for the true value ($V_{eff}$).
Similarly, from the observed histogram of microlensed light-curve derivatives we find the effective transverse velocity ($v_{obs}$) that best describes the observed microlensing rate. Using Bayes' theorem we calculate the posterior probability that the effective galactic transverse velocity $V_{gal}$ is less than an assumed value $V_{eff}$:
\begin{eqnarray}
\nonumber P_{v}(V_{gal}<V_{eff}|v_{obs})&=&\\
&&\hspace{-25mm}\int_{0}^{V_{eff}}p_{lhood}(v_{obs}|V_{eff}')\,p_{prior}(V_{eff}')\,dV_{eff}',
\end{eqnarray}
where $p_{prior}(V_{eff})$ is the prior probability for $V_{eff}$.
We have assumed two different and physically un-motivated priors. These are respectively flat:
\begin{eqnarray}
\nonumber p_{prior}(V_{eff})&\propto& dV_{eff}\hspace{10mm} 0<V_{eff}<2000\sqrt{\langle m\rangle}\\
p_{prior}(V_{eff})&\equiv&0\hspace{16mm} otherwise,
\end{eqnarray}
and logarithmic:
\begin{eqnarray}
\nonumber p_{prior}(V_{eff})&\propto&\frac{dV_{eff}}{V_{eff}}\hspace{10mm} 0<V_{eff}<2000\sqrt{\langle m\rangle}\\
p_{prior}(V_{eff})&\equiv&0\hspace{16mm} otherwise.
\end{eqnarray}
These priors bracket the range and spread of physically plausible priors such as a Gaussian based on observed peculiar velocities (eg Mould et al. 1993) or those computed from numerical studies. Our results are insensitive to the choice of these priors, and we conclude that our estimate of probability is dominated by the microlensing observations rather than our assumed prior for $V_{eff}$.
$P_v(v_{gal}<V_{eff}|v_{obs})$ is plotted in Fig.~\ref{vel_lim} for the models discussed in this paper ($\langle m\rangle = 1M_{\odot}$). In these plots the solid and dot-dashed lines correspond to source trajectories along the C-D image axis and A-B image axis, and the dark and light lines correspond to models with 0\% and 50\% smoothly distributed matter. The upper and lower plots correspond to the cases where the photometric error was assumed to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D.
\subsection{The contribution to microlensing of stellar proper motions}
The average histogram of light-curve derivatives can be computed for the case where microlensing results from a random velocity dispersion rather than a transverse velocity. We define the equivalent transverse velocity as the transverse velocity that in combination with a static microlensing model produces a microlensing rate closest to the rate produced by a random velocity dispersion of the point masses. The microlensing rates are considered equivalent at the transverse velocity that produces a cumulative histogram of light-curve derivatives closest (has the minimum possible KS difference) to the corresponding proper motion histogram.
The formalism required for the computation of the cumulative distribution of derivatives resulting from proper motion of stars was developed in WWTa. The upper panel of Tab.~\ref{stream_tab} shows the values of equivalent transverse velocity of a Gaussian 1-d velocity dispersion with a half-width of $\sigma_{*}=165\,km\,sec^{-1}$. These values were computed from the difference light-curves for models of Q2237+0305 (the quoted error describes the range of values obtained over three separate sets of simulations). The two cases shown are for a trajectory that is parallel to the shear in images A and B ($\gamma_{A},\gamma_{B}>0$), and for one that is parallel to the shear in images C and D ($\gamma_{C},\gamma_{D}>0$). In all cases the effective transverse velocity is larger than the 1-d velocity dispersion.
\section{The mass of compact objects in Q2237+0305}
\label{mass_limits}
\subsection{Placing limits on the mass using effective transverse velocity}
\begin{table*}
\begin{center}
\caption{\label{stream_tab}Table showing the equivalent transverse velocities and the 99\%, 95\% and 90\% lower limits for the mean microlens mass as well as the most likely value. Values for the mass limits are shown corresponding to the cases where the error was $\sigma_{SE}=0.01$ mags in images A/B, $\sigma_{SE}=0.02$ mags in images C/D, and where the error was $\sigma_{LE}=0.02$ mags in images A/B, $\sigma_{LE}=0.04$ mags in images C/D (in parentheses). Values are shown that correspond to an isotropic velocity dispersion of $\sigma{*}=165\,km\,sec^{-1}$ (top table), a circular stream velocity of $v_{stream}=233\,km\,sec^{-1}$ (bottom table).}
\begin{center}
\vspace{5mm}
$\sigma{*}=165\,km\,sec^{-1}$
\begin{tabular}{|c|c|c|c|c|c|c|c|}
\hline
Trajectory & Smooth & Equiv. & Bayesian & $m_{low}(99\%)$ & $m_{low}(95\%)$ & $m_{low}(90\%)$ & mode \\
Orientation & Matter & Vel. ($km\,sec^{-1}$) & prior & ($M_{\odot}$) & ($M_{\odot}$) & ($M_{\odot}$) & ($M_{\odot}$) \\ \hline\hline
Shear: & 0\% & 280 $\pm$15 & log & 0.050 (0.111) & 0.111 (0.277) & 0.176 (0.484) & 0.139 (0.288) \\
$\gamma_{A},\gamma_{B}>0,\:\:\gamma_{C},\gamma_{D}<0 $ & 50\% & 199 $\pm$12 & log & 0.019 (0.039) & 0.042 (0.095) & 0.066 (0.161) & 0.054 (0.087) \vspace{3mm}
\\
Shear: & 0\% & 270 $\pm$10 & log & 0.053 (0.111) & 0.111 (0.270) & 0.173 (0.463) & 0.139 (0.281) \\
$\gamma_{A},\gamma_{B}<0,\:\:\gamma_{C},\gamma_{D}>0 $ & 50\% & 181 $\pm$4 & log & 0.018 (0.036) & 0.039 (0.084) & 0.061 (0.143) & 0.042 (0.087) \\ \hline
Shear: & 0\% & 280 $\pm$15 & flat & 0.098 (0.230) & 0.270 (0.739) & 0.505 (1.496) & 0.222 (0.450) \\
$\gamma_{A},\gamma_{B}>0,\:\:\gamma_{C},\gamma_{D}<0 $ & 50\% & 199 $\pm$12 & flat & 0.042 (0.090) & 0.121 (0.301) & 0.240 (0.643) & 0.069 (0.139) \vspace{3mm}
\\
Shear: & 0\% & 270 $\pm$10 & flat & 0.107 (0.244) & 0.293 (0.792) & 0.553 (1.638) & 0.222 (0.450) \\
$\gamma_{A},\gamma_{B}<0,\:\:\gamma_{C},\gamma_{D}>0 $ & 50\% & 181 $\pm$4 & flat & 0.044 (0.093) & 0.135 (0.330) & 0.278 (0.729) & 0.069 (0.239) \\ \hline
\end{tabular}
\end{center}
\vspace{5mm}
$v_{stream}=233\,km\,sec^{-1}$
\begin{tabular}{|c|c|c|c|c|c|c|c|}
\hline
Trajectory & Smooth & Equiv. & Bayesian & $m_{low}(99\%)$ & $m_{low}(95\%)$ & $m_{low}(90\%)$ & mode \\
Orientation & Matter & Vel. $(km\,sec^{-1})$ & prior & ($M_{\odot}$) & ($M_{\odot}$) & ($M_{\odot}$) & ($M_{\odot}$) \\ \hline\hline
Shear: & 0\% & 129 $\pm$7 & log & 0.010 (0.023) & 0.024 (0.061) & 0.039 (0.113) & 0.027 (0.043) \\
$\gamma_{A},\gamma_{B}>0,\:\:\gamma_{C},\gamma_{D}<0 $ & 50\% & 99 $\pm$5 & log & 0.006 (0.011) & 0.013 (0.030) & 0.022 (0.056) & 0.013 (0.021) \vspace{3mm}
\\
Shear: & 0\% & 120 $\pm$2 & log & (0.011 (0.023) & 0.025 (0.064) & 0.042 (0.120) & 0.021 (0.043) \\
$\gamma_{A},\gamma_{B}<0,\:\:\gamma_{C},\gamma_{D}>0 $ & 50\% & 85 $\pm$3 & log & 0.005 (0.009) & 0.011 (0.024) & 0.018 (0.044) & 0.010 (0.021) \\ \hline
Shear: & 0\% & 129 $\pm$7 & flat & 0.028 (0.071) & 0.105 (0.319) & 0.251 (0.823) & 0.034 (0.069) \\
$\gamma_{A},\gamma_{B}>0,\:\:\gamma_{C},\gamma_{D}<0 $ & 50\% & 99 $\pm$5 & flat & 0.014 (0.031) & 0.053 (0.144) & 0.136 (0.395) & 0.017 (0.027) \vspace{3mm}
\\
Shear: & 0\% & 120 $\pm$2 & flat & 0.027 (0.069) & 0.108 (0.33) & 0.273 (0.897) & 0.034 (0.687) \\
$\gamma_{A},\gamma_{B}<0,\:\:\gamma_{C},\gamma_{D}>0 $ & 50\% & 85 $\pm$3 & flat & 0.014 (0.031) & 0.060 (0.159) & 0.162 (0.454) & 0.017 (0.027) \\ \hline
\end{tabular}
\vspace{5mm}
\end{center}
\end{table*}
In this section we describe a method for determining the lower limit for the average compact object mass. The microlensing rate has a minimum possible value determined by the size of the stellar velocity dispersion, thus there is a minimum mass which can explain the observed rate. This is quantified by noting that the measured effective transverse velocity must be greater than the equivalent transverse velocity.
There is a simple scaling that relates models of the same optical depth but consisting of model stars with a different mass. The dimensionless Einstein radius of a point mass is $\sqrt{m}$. Therefore, reducing all microlens masses by a factor $a$ reduces all physical distances (in both the source and lens planes) between dimensionless coordinates by a factor $\sqrt{a}$ (eg. Witt, Kayser \& Refsdal 1993). For a given transverse velocity this results in an increase by a factor of $1/\sqrt{a}$ in the gradient of the light-curve at all points. Similarly, the rate of change of magnification at a static source point due to stellar proper motions increases by a factor $1/\sqrt{a}$. Since it results from the proper motions the minimum rate is larger if a smaller mean microlens mass is assumed. However the minimum rate cannot statistically exceed the observed rate. The ratio of equivalent transverse velocity to velocity dispersion is independent of the mean microlens mass, however the measurement of effective transverse velocity is proportional to $\sqrt{a}$. This allows limits to be placed on the value of $a$ and therefore the stellar mass from the observed microlensing rate.
\begin{figure}
\vspace*{155mm}
\special{psfile=fig3.epsi hscale=95 vscale=95 hoffset=-485 voffset=465 angle=270}
\caption{\label{mass_lim} The differential distributions for the mean microlens mass. The solid and dotted curves represent the resulting functions when the photometric error was assumed to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D. The light and dark lines correspond to the assumptions of logarithmic and flat priors for effective transverse velocity referred to in the text.}
\end{figure}
\begin{figure*}
\vspace*{130mm}
\special{psfile=fig4.epsi hscale=95 vscale=95 hoffset=-245 voffset=390 angle=
270}
\caption{\label{contour}Contours of percentage peak height of the function $p_{m,v}(\langle m\rangle,V_{tran})$. The contours shown are the 0.1\%, 1.0\%, 3.6\%, 14\%, 26\% and 61\% levels. The solid and dotted curves represent the resulting functions when the photometric error was assumed to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D.}
\end{figure*}
\subsection{\label{propmot}Mass limits from random proper motions}
For a model where $\gamma_{A},\gamma_{B}>0$, and the simulated errors are assumed to be $\sigma_{SE}=0.01$ and $\sigma_{SE}=0.02$ mags in images A/B and C/D respectively, the effective transverse velocity is less than $1346\sqrt{\langle m\rangle} \,km\,sec^{-1}$ at the 99\% level, where $\langle m\rangle$ is the microlens mass assumed.
In the absence of a galactic transverse velocity, the minimum level of microlensing is described by the equivalent transverse velocity of the stellar velocity dispersion $V_{equiv}(DISP)=280\,km\,sec^{-1}$ (see Tab.~\ref{stream_tab}). The value of $\langle m\rangle$ at which $V_{upper}(99\%)$ drops below $V_{equiv}(DISP)$ is $m_{low}(99\%)$:
\begin{equation}
\label{mass_low}
m_{low}(99\%)=\left(\frac{V_{equiv}(DISP)}{V_{upper}(99\%)}\right)^{2}\sim 0.04.
\end{equation}
\noindent The minimum mass of stars in this model is therefore $\sim 0.04M_{\odot}$. This calculation assumes a value for the one dimensional velocity dispersion of $165\,km\,sec^{-1}$. However we note that all lower limits obtained in this section are proportional to the square of this value. The procedure also assumes that the measured effective transverse velocity is independent of the assumed source size. WWTb show that this is approximately true for source sizes smaller than $\approx 4\times 10^{16}\sqrt{m}\,cm$. This value compares favourably with the findings of Wambsganss, Paczynski \& Schneider (1990) that a source smaller than $\sim 2\times 10^{15}\sqrt{\langle m\rangle /.225}\,cm$ is required to typically reproduce the observed HME amplitude.
The curves shown in Fig.~\ref{vel_lim} represent the cumulative probability $P_v(V_{gal}<V_{eff}\sqrt{m})$ that the effective transverse velocity $V_{gal}$ is less than $V_{eff}\sqrt{m}$.
It follows that the cumulative probability that the mean microlens mass is less than the value
\begin{equation}
\nonumber M=\left(\frac{V_{equiv}(DISP)}{V_{eff}}\right)^{2}
\end{equation}
is
\begin{equation}
\nonumber U_m(\langle m\rangle<M|V_{equiv})=1-P_v(V_{gal}<V_{eff}\sqrt{M}).
\end{equation}
The calculation of $m_{low}(99\%)$ performed above assumes that the physical transverse velocity of the galaxy with respect to the observer source line of sight $V_{tran}=0$. However since the galaxy may have a component of transverse motion, $V_{equiv}$, which describes the minimum microlensing rate is a function of $V_{tran}$. The dependence of $V_{equiv}$ on $V_{tran}$ is determined according to the relationships described in WWTb. Since the galactic transverse velocity is not necessarily zero, $U_m(\langle m\rangle<M|V_{equiv})$ must be determined at a series of transverse velocities and combined with prior probabilities for $V_{tran}$ to obtain estimates for the probability of the mean microlens mass. We have assumed the flat and logarithmic priors employed for the calculation of $P_v(V_{gal}<V_{eff})$:
\begin{eqnarray}
\nonumber U_m(\langle m\rangle<M)&=&\\
&&\hspace{-25mm}\int U_m(\langle m\rangle<M|V_{tran})p_{prior}(V_{tran})dV_{tran}
\end{eqnarray}
The probability density $p_{m}(\langle m\rangle)$ for the mean microlens mass is obtained by taking the derivative $\frac{d\,U_m}{d\,\langle m\rangle}$. Note that this distribution does not represent the mass function.
Fig.~\ref{mass_lim} shows the functions $p_{m}(\langle m\rangle)$ for the 0\% smooth matter and 50\% smooth matter models for both directions considered.
The solid and dotted curves represent functions assuming the photometric error to be $\sigma_{SE}$=0.01 mag in images A/B and $\sigma_{SE}$=0.02 in images C/D, and $\sigma_{LE}$=0.02 mag in images A/B and $\sigma_{LE}$=0.04 in images C/D.
The light and dark lines refer to the logarithmic and flat priors for transverse velocity. Results for $m_{low}(99\%)$, $m_{low}(95\%)$, $m_{low}(90\%)$ and the mode of the distribution are shown in the top panel of Tab.~\ref{stream_tab}. In each column, the values correspond to the two assumptions for the simulated error referred to above. $m_{low}(99\%)$ is $>0.02M_{\odot}$ in all models considered. The mode of the distribution ranges between 0.04$M_{\odot}$ and 0.45$M_{\odot}$. It is interesting to note that these models produce distribution modes that are similar to the mean of a Salpeter distribution with a lower cutoff at the hydrogen burning limit. We also note that the distribution is highly asymmetric, with the mode being of the same order as the lower limit.
\subsection{The effect of systematic assumptions}
In contrast to the calculation of $P_v(V_{gal}<V_{eff}\sqrt{\langle m\rangle})$ the assumed prior for $V_{eff}$ makes a significant difference to the microlens mass limits obtained. In particular, low mass microlenses are less likely if large transverse velocities are assumed. This dependence on the prior illustrates the limitation of the light curve data for breaking the degeneracy between galactic transverse velocity and microlens mass in the high mass-velocity regime.
The assumption of a larger photometric uncertainty yields a lower estimate of the transverse velocity (WWTb). This in turn means that a larger estimate is made for the mean microlens mass. The effect is readily apparent in Fig.~\ref{mass_lim} as well as from Tab.~\ref{stream_tab}.
We find lower microlens mass limits if a non-zero fraction of smooth matter is assumed. This results from the combination of a larger determination of effective transverse velocity and a smaller relative contribution to microlensing of the stellar proper motions resulting in a smaller minimum for microlensing rate. This is expected since no microlensing will be observed due to a galaxy composed entirely of continuously distributed matter. The plots in Fig.~\ref{mass_lim} demonstrate the level of dependence on smooth matter component.
$p_{m}(\langle m\rangle|V_{tran})$ is extremely sensitive to the level of photometric error assumed at large $\langle m\rangle$. In addition, microlenses of arbitrarily high mass can produce the required rate in combination with a correspondingly large transverse velocity. Therefore at large $\langle m\rangle$, $p_{m}(\langle m\rangle)$ is a sensitive function of the prior assumed. For these reasons our method does not place useful upper limits on the mean microlens mass.
\subsection{Simultaneous limits on mean microlens mass and transverse velocity}
\label{simul}
To explore the co-dependence of the determined mean microlens mass and galactic transverse velocity we have computed the two dimensional distribution
\begin{equation}
p_{m,v}(\langle m\rangle,V_{tran})=p_{m}(\langle m\rangle|V_{tran})\,p_{v}(V_{tran}|\langle m\rangle).
\end{equation}
Figure~\ref{contour} shows contour plots of this distribution for the assumed source orientations, smooth matter fraction and photometric errors. All possible combinations of the two values of each parameter are included, so that eight models are shown. The contours are at 0.1, 1.0, 3.6, 14, 26, 61 percent of the peak height, and so the extrema of the inner 4 contours represent $4\sigma$, $3 \sigma$, $2\sigma$, and $1\sigma$ limits on the single variables. The most important benefit of combining the distributions of $V_{tran}$ and $\langle m \rangle$ is that upper limits can be placed on the mean microlens mass and galactic transverse velocity. The limits on $\langle m \rangle$ and $V_{tran}$ are subject to both random and systematic uncertainties. Assuming that the chosen values for these parameters bracket the likely models, upper and lower limits for $\langle m \rangle$, and an upper limit for $V_{tran}$ can be found by using the most extreme values of the contour limits. This yields the 95 per cent result that $0.01 M_{\odot} \la \langle m \rangle \la 1 M_{\odot}$, and $V_{tran}\la 300\,km\,sec^{-1}$ (95\%). The upper limit is lower than that obtained by Lewis \& Irwin (1996), although sufficiently high that it does not conflict with any popular models of galactic halos or bulges. The elongation of the outer contours with a power law slope of $\frac{1}{2}$ illustrates the degeneracy between microlens mass and transverse velocity at large mass.
\subsection{The effect of trajectory direction}
\begin{figure}
\vspace*{85mm}
\special{psfile=FIG5.ps hscale=45 vscale=45 hoffset=-18 voffset=-50 angle=0}
\caption{\label{magmaps} The magnification maps for images A (top) and C (bottom), in the cases of 0\% smooth matter (left) and 50\% smooth matter (right). The point masses each had a mass of 1$M_{\odot}$}
\end{figure}
Fig.~\ref{magmaps} shows the magnification maps for images A and C in the cases of 0\% and 50\% continuously distributed matter. The caustic networks in both images are significantly stretched so that a source could travel a significant distance in the direction of caustic clustering and be subject to little microlensed flux variation.
The measured effective transverse velocity is dependent on trajectory direction, with a larger measurement resulting from cases where the source trajectory is perpendicular to the shear in images C and D ($\gamma_{C},\gamma_{D}<0$). The larger measurement quantifies the greater stretching of the caustic network in these images, creating a larger difference between the rate of microlensing for sources moving with and against the shear. The variation with orientation of the measurements of effective transverse velocity is slightly larger when the model contains a component of smooth matter due to additional stretching of the caustic network.
A similar argument applies to the determination of the equivalent transverse velocity of a stellar velocity dispersion. The rate of microlensing due to stellar proper motions at a series of fixed source points is independent of the direction in which those points are sampled, regardless of the contributing fraction of smoothly distributed matter. However microlensing that results from a transverse velocity has a rate that varies more with direction in images C and D. The equivalent transverse velocity is therefore a function of direction. The effect is more pronounced in models that include a smoothly distributed mass component.
Thus there are two separate effects. Firstly, if the source trajectory is aligned with the image C-D axis the measured effective transverse velocity increases. This effect is magnified by the inclusion of a component of smooth matter. In addition, the equivalent transverse velocity of the stellar proper motions is also increased under the same circumstances. Interestingly, during our calculation of lower mass limit these effects tend to have a cancelling effect, rendering the limits reasonably insensitive to the source trajectory direction assumed.
\subsection{Mass limits from stellar stream motions}
The calculations in Sec \ref{propmot} and Sec \ref{simul} assumed that microlensing results from the combination of stellar proper motions and galactic transverse velocity. However the bulge dynamics may be dominated by rotational motion resulting in stream motions of the starfield. For an isothermal sphere model of the bulge, the rotational velocity is $\sqrt{2} \sigma_{*}=233\,km\,sec^{-1}$.
Kundic, Witt \& Chang (1993) obtained the expression for the caustic velocity $v_{caust}$ resulting from a stream motion $\vec{v}_{stream} = (v_{stream},0)$:
\begin{equation}
\label{stream}
v_{caust}=(1-\gamma-\kappa_{c}) |\vec{v}_{stream}|.
\end{equation}
If the rotational motion is assumed to be circular then the stream motions are perpendicular to the image-galactic centre axis, and therefore parallel to the shear vector at each image. We choose the stream motions to be in the $x_{1}$ direction making the shear positive for all 4 images. An equivalent transverse velocity was defined for the stream motion and determined in analogy to that for the stellar proper motions. For a circular stream motion of $v_{stream}=233\,km\,sec^{-1}$ equivalent transverse velocities ($V_{equiv}(STREAM)$) were calculated for the two transverse directions discussed previously. These values are presented in Tab.~\ref{stream_tab}. When combined with the equivalent velocities for an isotropic velocity dispersion, the values describe the contribution to the microlensing rate of the extreme cases for stellar motions in the galactic bulge.
The lower panel in Tab.~\ref{stream_tab} shows results for models with stream motions. Values are shown for $m_{low}(99\%)$, $m_{low}(95\%)$, $m_{low}(90\%)$ and the mode for all models considered. $m_{low}(99\%)$ is $> 0.005M_{\odot}$ in all models.
The equivalent transverse velocities presented in Tab.~\ref{stream_tab} show that an orbital stream motion produces a much lower microlensing rate than a transverse velocity of equal magnitude (independent of its direction). The reason for this behaviour is apparent from Eqn. \ref{stream}; a positive shear reduces the caustic velocity resulting from stream motions. In addition, the caustic velocity and hence microlensing rate are also decreased when there is a component of smooth matter. The lower microlensing rate translates to an equivalent transverse velocity that is smaller than the one obtained from a stellar velocity dispersion. This in turn means a lower estimate of $m_{low}(P)$. In the case discussed, the mass limit obtained is only $\sim 0.2$ times that where the proper motions are assumed to be isotropic.
The last row of Tab.~\ref{ml} summarises the range of values for the mean microlens mass obtained by this study. These compare favourably with the results of related studies, none of which find that Jupiter mass compact objects are the dominant component of mass in the bulges and or halos of the spiral galaxies studied.
\section{Conclusion}
Calculations of the transverse velocity from published monitoring data of Q2237+0305, as well as complimentary characterisations of caustic clustering from previous studies show that the caustic networks produced by populations of microlenses have a structure independent of the mass spectrum, but a scale length proportional to the square root of the mean microlens mass. Previous determinations of average mass from microlensing data have therefore been proportional to the square of the unknown transverse velocity.
However a minimum rate of microlensing is produced by the proper motions of microlenses in the bulge of the lensing galaxy. We have used limits on the effective transverse velocity obtained from published monitoring data for Q2237+0305 in combination with calculations of the microlensing effect of isotropic stellar proper motions to calculate the probability for the mean microlens mass. We find that the most likely value for the mean microlens mass is 0.04-0.45$M_{\odot}$. The lower limits on the mean microlens mass responsible for the observed microlensing are $\langle m\rangle \,> 0.02-0.24M_{\odot}$ (99\% level). A similar calculation assuming an orbital stream motion of stars rather than a velocity dispersion produces a most likely value of 0.01-0.05$M_{\odot}$ and a lower limit of $\langle m\rangle \,>0.005-0.071M_{\odot}$ (99\% level). Through joint consideration of the probability for mean microlens mass and transverse velocity under the assumption of an isotropic velocity dispersion we obtain a mass between $0.01M_{\odot}$ and $1.0M_{\odot}$ (95\%), and a galactic transverse velocity less than 300$\,km\,sec^{-1}$ (95\%).
Our result that very low mass objects do not comprise a significant mass fraction of microlenses in Q2237+0305 supports the results of Schmidt \& Wambsganss (1998) who rule out microlenses in Q0957+561 having $\langle m\rangle~\ll~0.01\left(\frac{v_t}{600\,km\,sec^{-1}}\right)^2$. The most likely mean microlens mass of $\sim 0.2M_{\odot}$ supports both the conclusions of Lewis \& Irwin (1996) who found that microlenses in Q2237+0305 have $0.1M_{\odot}\la\langle m\rangle\left(\frac{v_t}{600\,km\,sec^{-1}}\right)^2\la10M_{\odot}$, and of the MACHO collaboration (Alcock et al. 1997a,b) who find from microlensing in the Milky Way that $\langle m\rangle\ga 0.05M_{\odot}$ in the halo and $0.1M_{\odot}\la\langle m\rangle\la1.0M_{\odot}$ towards the bulge.
While our limits depend on the correctness of the published macrolensing parameters, as well as on the fraction of smoothly distributed matter assumed, they do not depend on other unknowns present in the problem including the source size, source intensity profile, effective transverse velocity and direction of the source trajectory. We have investigated models having smooth matter fractions of 0 and 0.5, and find that smaller masses can explain the observed microlensing rate if a non-zero smooth matter fraction is assumed. Our results depend on the Bayesian prior chosen for galactic transverse velocity. However we have chosen priors to bracket physically reasonable choices and find that the contribution of the prior to the systematic uncertainty is smaller than that of the assumption for smooth matter fraction.
Our results unambiguously rule out a significant contribution of Jupiter mass compact objects to the mass distribution of the bulge whether the microlenses move predominantly in orbital or random motions. However if the microlens proper motions are distributed according to an isotropic velocity dispersion then the lower limit obtained is of the same order as the hydrogen burning limit.
\section*{Acknowledgements}
The authors would like to thank Chris Fluke and Daniel Mortlock for helpful discussions. This work was supported in part by NSF grant AST98-02802. JSBW acknowledges the support of an Australian Postgraduate award and a Melbourne University Overseas Research Experience Award.
|
\section{Introduction}
\placetable{tbl-1}
Seyfert galaxies are the nearest and most numerous class of active
galaxy and, although they are radio-weak, the nuclei of many Seyferts
contain compact non-thermal radio sources which often account for a
large fraction of the total radio luminosity of the galaxy. Aperture
synthesis with instruments such as the VLA and MERLIN has resolved
these radio sources, in many cases revealing linear, highly-collimated
structures ({\it eg} Ulvestad \& Wilson 1984a,b, Pedlar et al. 1993)
and there is now little doubt that at least some Seyfert galaxies are
capable of producing radio jets which are qualitatively similar to
those found in radio-loud quasars and radio galaxies, albeit several
orders of magnitude smaller and less powerful.
Because of their proximity, Seyfert nuclei present us with the
opportunity to study jet structure on the smallest possible spatial
scales, with a level of detail which is impossible for more luminous,
but also more distant, quasars and radio galaxies. In this paper we
report on very long baseline interferometer (VLBI) observations of
four Markarian Seyfert galaxies, Mrk~1, Mrk~3, Mrk~231 and Mrk~463E
(see Table~1). Preliminary results of this project were presented by
Ghosh et al. (1994).
We assume $H_{0} = 75$~kms$^{-1}$Mpc$^{-1}$ throughout this paper. The
spectral index, $\alpha$, is defined according to the relation
$S\propto \nu^{\alpha}$, where $S$ is the flux density at frequency
$\nu$.
\subsection{Radio emission and the standard model of Seyfert nuclei}
Several studies have attempted to use radio observations of large
samples of objects to test models of Seyfert activity. In particular,
the radio properties are regarded as ideal for testing schemes which
attempt to unify Seyfert 1s and 2s via viewing angle
effects. According to this scenario all Seyfert nuclei are surrounded
by an optically thick torus which lies in the plane perpendicular to
the axis of the radio jets. When our line of sight to the nucleus
intersects the torus, the central engine and broad line region (BLR)
are blocked from view and we classify the object as a Seyfert 2. If
our view of the nucleus is unobstructed, we call the object a Seyfert
1. Assuming that the radio emission is isotropic and unaffected by
dust obscuration, the model predicts that the average radio
luminosities of Types 1 and 2 should be the same and also that the
radio jets in Seyfert 1s should be foreshortened on average relative
to those in Seyfert 2s due to projection effects.
To date, radio surveys of Seyferts have produced conflicting results,
some of which can be ascribed to selection effects in the samples
themselves ({\it eg} due to the greater difficulty in identifying
Seyfert 2s; on average those that are detected tend to be more
luminous than Seyfert 1s). However, even when such effects have been
removed, or compensated for as far as is possible, confusion still
remains. Kukula et al. (1995) found that the radio structures of
Seyfert 1s in the CfA sample (Huchra \& Burg 1992) were on average too
compact to be be foreshortened versions of the Type 2 objects in the
same sample, indicating that orientation is not the only effect at
work in this particular case. Indeed, only 1/26 Seyfert~1s in the CfA
sample showed any sign of extended radio structure down to the
0.24$''$ resolution of the VLA. Meanwhile, Roy et al. (1994), in a
single-baseline VLBI study of several Seyfert samples, found that
Type~1 objects were {\it less} likely than Type~2s to contain compact
nuclear radio components, contrary to what would be expected from the
simple unified model. Finally, Thean et al. (1998) report that Seyfert
1s and 2s in the Extended 12$\mu$m AGN Sample (Rush, Malkan \&
Spinoglio 1993) are {\it equally likely} to contain extended radio
sources, with no significant difference between the size distributions
of the two types.
In practice the effects of the various selection criteria and
observational biases operating in these samples are very difficult to
quantify without a better understanding of the physical processes
which are responsible for the radio emission. Detailed studies of
individual Seyferts offer perhaps the best means of achieving this
goal, and VLBI - though an inefficient tool for large surveys - is
ideally suited to this kind of work.
\subsection{VLBI observations of Seyfert nuclei}
VLBI provides the highest angular resolution of any imaging technique
but in the past sensitivity constraints were extremely tight,
effectively limiting VLBI studies of active galaxies to the small
fraction of objects which are radio loud. However, technological
advances made over the last decade have resulted in large increases in
sensitivity, and imaging of the brightest radio-quiet objects is now
well within the capability of modern instruments.
Several recent VLBI studies of Seyfert nuclei have provided detailed
images of structures close to the central engine itself (eg Halkides,
Ulvestad \& Roy 1997; Mundell et al. 1997; Oosterloo et al. 1998;
Ulvestad, Wrobel \& Carilli 1998; Roy et al. 1998). Often the radio
structure on milliarcsecond (parsec) scales is misaligned with the
large-scale ($\sim 100$~pc) radio jets familiar from low-resolution
maps. These discoveries pose challenges for traditional unified models
of Seyferts because they imply that the aligned arcsecond-scale radio
continuum and optical line emission cannot necessarily be used to
trace the symmetry axis of the central engine.
However, in some cases the physical properties of the VLBI radio
source suggest an entirely different origin to the emission on larger
scales. In the nucleus of NGC~1068, the archetypal Seyfert~2 galaxy,
Gallimore, Baum \& O'Dea (1997) found a linear radio structure $\sim
1$~pc across, lying almost perpendicular to the arscecond-scale
synchrotron jet. They suggest that this feature represents free-free
emission from the inner edge of a molecular torus surrounding the AGN
- the first direct observational evidence for such a structure.
Ulvestad, Roy, Colbert \& Wilson (1998), consider a similar
explanation for the misaligned sub-parsec radio structure in NGC~4151.
\section{Observations and data reduction}
Observations were made in Autumn 1990 (see Table~1 for dates) with the
European VLBI Network (EVN), a radio interferometer spanning Western
Europe and at this time incorporating telescopes in Germany
(Effelsberg), Sweden (Onsala), Italy (Medicina) and the U.K. (Jodrell
Bank). The observing frequency was 1655~MHz (18~cm), giving an angular
resolution in the final maps of $\sim20$~mas - a factor of 2.5 higher
than that of the restored Hubble Space Telescope. The data recording
was in MK{\sc iii} mode~B with a bandwidth of 28~MHz. Two calibration
sources, OQ208 and DA193, were interleaved with the four target
sources during the 4 days of the observing run, with approximately 10
hours spent on each target source. The amplitude calibration was
carried out using the system temperatures measured before and after
each 13-minute scan and the gain curves supplied by each
station. Global fringe-fitting was performed on the calibration
sources using point source models, and the resulting single-band and
multi-band delays were interpolated to the entire data-set. The data
for each source were then fringe-fitted again and delays and rates
were adjusted before frequency and time averaging were performed. For
Mrk~3, the previously made 18-cm MERLIN map (see below) was used for
the model, both for global fringe fitting and for the first round of
self calibration. For the other three sources, point source models
were used and hence the absolute position information has not been
retained. The calibration sources were also processed through several
cycles of phase-self-calibration and mapping. The resulting flux
densities for the two sources at this epoch of observation were
$S_{OQ208} = 0.8$~Jy and $S_{DA193} = 1.8$~Jy. We estimate an
uncertainty of $\sim 10\%$ in the flux densities. For the final maps,
a uniform weighting scheme was used for the Fourier transform in order
to maximize the angular resolution. The RMS noise levels were
typically 0.15 - 0.2~mJy.
Eight hours of additional 18-cm (1658~MHz) observations were made of
one object, Mrk~3, on April 26$^{\rm th}$ 1993 with the Multi-Element
Radio Linked Interferometer Network (MERLIN), operated from Jodrell
Bank. The MERLIN and EVN data were calibrated separately and the
visibilities then combined, with appropriate weighting, to give a
single large dataset. The MERLIN data helps to fill the `hole' in the
center of the $uv$ plane caused by the EVN's lack of short
baselines. After the Fourier transform and deconvolution, the
resulting map retained both the high angular resolution of the EVN and
MERLIN's greater sensitivity to extended radio emission.
\section{Results}
Compact radio components were detected in all four Seyferts, with
three objects displaying prominent linear radio structures. The
measured positions, flux densities, sizes and various derived
properties of the individual radio components are given in Table~2
(Mrk 1, 231 \& 463) and Table~3 (Mrk~3). Flux densities and positions
were measured from the maps using the {\sc aips} task {\sc
tvstat}. Deconvolved angular sizes for the components were obtained
from two-dimensional Gaussian fits to the data using the task {\sc
jmfit} or, for the more irregular features, were estimated from the
maps.
Estimates of the magnetic field density in the radio-emitting regions
derived from minimum energy arguments ({\it eg} Miley 1980), give
typical values of $\sim 10^{-7}$T (Tables~2 \& 3). The corresponding
lifetime for synchrotron-emitting electrons in a field of this
strength is $\sim10^{4}$~years. Brightness temperatures, $T_{B}$, of
the compact radio features are typically $10^{6-7}$K, similar to the
values measured in other Seyfert nuclei with the VLBA (Gallimore et
al. 1997: Ulvestad et al. 1998; Carilli et al. 1998). However, the
spatial resolution of the current maps is insufficient to determine
whether parsec- or subparsec-scale free-free emission from a molecular
torus might be present in any of the objects in our sample.
\placetable{tbl-2}
\placetable{tbl-3}
\subsection{Markarian~1}
\placefigure{f1}
This Seyfert 2 galaxy is perhaps best known for the prominent water
maser in its nucleus (Braatz, Wilson \& Henkel 1994). Keel (1996)
suggests that the galaxy is interacting with the nearby object
NGC~451. No evidence for broad lines - indicative of a `hidden'
Seyfert~1 nucleus - has been found in Mrk~1, either in polarized light
(Kay 1994) or in the infrared (Veilleux, Goodrich \& Hill 1997).
Mulchaey \& Wilson (1995) report that the [O{\sc iii}]$\lambda5007$
emission in Mrk~1 is extended along PA83$^{\circ}$, whilst
Ulvestad, Wilson \& Sramek (1981), using the VLA at 6~cm, found the
radio structure to consist of a barely resolved nuclear source with
weaker emission extending $\sim400$~mas ($\sim 120$~pc) to the
south.
In the EVN map (Figure~1) we find a compact radio core (deconvolved
size $\sim10$~mas~$= 3$~pc) surrounded by a halo of emission
approximately 100~mas ($\sim 30$~pc) across. This extended emission
appears to be elongated in approximately the same direction as the
optical emission line region, although we note that the restoring beam
also has a similar orientation. There is also some evidence for weak
emission extending to the south, perhaps leading into the larger
structure reported by Ulvestad, Wilson \& Sramek. We find no evidence
for linear radio structure in this object on any scale larger than a
few parsecs.
\subsection{Markarian~3}
\placefigure{f2}
Mrk~3 is an early-type (S0 or elliptical) galaxy with a Seyfert 2
nucleus, although the presence of broad emission lines in its
polarized spectrum argues strongly for a hidden Seyfert~1 nucleus in
this object (Schmidt \& Miller 1985; Miller \& Goodrich 1990, Tran
1995), as does the photon budget calculated by Capetti et al. (1995a).
It is the nearest object in the current study; at a distance of only
55~Mpc we obtain a spatial resolution of 5~pc with the EVN. In an
earlier study of this object with MERLIN at 6~cm, we found a
highly-collimated `S'-shaped radio structure with a bright radio
`hotspot' at its western end (Kukula et al. 1993). HST imaging of the
Narrow Line Region shows that the jet is surrounded by a sheath of
line-emitting gas (Capetti et al. 1995a), but there is no direct
one-to-one correspondence between the individual radio components and
brightness peaks in the line emission.
By concatenating 18-cm data from the EVN and MERLIN, with suitable
weighting of the data, we retain both the high angular resolution of
the former and the sensitivity to more extended emission provided by
the shorter baselines of the latter. The result, shown in Figure~2,
reveals a wealth of detail, with both the jets and the western
hotspot containing complicated substructure. In Figure~2, we adopt the
same naming scheme for the radio components as used to describe our 6-cm
MERLIN map, with the addition of component 1a, at the eastern end of
the structure (which was only marginally detected in the MERLIN map),
and with the refinement of letters to denote the most prominent
small-scale features.
The jets consist of a string of compact knots encased in a streamer of
more diffuse emission with a marked `S'-shaped curvature. There
appears to be no systematic increase or decrease in the brightness of
the knots along the length of the jet. The jet is resolved in the
transverse direction and its width is fairly constant ($\sim 100$~mas)
along its length. However there are several regions in which there
appears to be little or no radio emission: between the weak, diffuse
components 1a \& 1; on either side of the unresolved component 4; and
between the jet and the bright western `hotspot'.
\subsubsection{Spectral index variations along the jet}
\placefigure{f3}
In order to study the variation in spectral index across the radio
source we convolved our 18-cm EVN map with a $0.09''\times0.07''$ beam
to make it directly comparable to our previous 6-cm MERLIN map (Kukula et
al. 1993). Using the {\sc aips} task {\sc comb} to compare the two
datasets resulted in the spectral index map reproduced in
Figure~3. Spectral indices for the individual components are listed in
Table~3.
The map shows the jet to have a steep spectrum, with a typical
spectral index of $\alpha \sim -1$. The compact features generally
have a slightly flatter spectrum than the diffuse emission,
although the only knot with a genuinely flat spectrum is the unresolved
component~4.
The proximity of component~4 to the peak of the optical continuum
emission (as measured by Clements 1981), and its position at the
center of symmetry of the `S', first led us to suggest that it might
be the radio core of Mrk~3 (Kukula et al. 1993). The newly-derived
flat spectrum for this component supports this view since it is
characteristic of synchrotron self-absorption (SSA), a property which
is often observed in the core components of other AGN. If this is the
case then the jet in Mrk~3 is two-sided and we see no emission on
either side of the nucleus for the first $\sim 30$~pc.
If SSA is indeed responsible for the flat radio spectrum of this
component then its brightness temperature must be in excess of
$10^{9}$K, although at present we can only place a lower limit of
$3.5\times10^{6}$K on $T_{B}$ (Table~3). In order for the core to be
synchrotron self-absorbed it must have a maximum size of $\sim 2$~mas
($\sim 0.5$~pc). This prediction will be tested by forthcoming VLBA
observations.
\subsubsection{Interpretation of the radio structures in Mrk~3}
The radio source in Mrk~3 is one of the nearest known examples of an
AGN containing a pair of highly-collimated radio jets. The details
revealed in our new map of Mrk~3 therefore show the structure of such
systems on a very small spatial scale. Here we discuss the nature of
the jets in Mrk~3 based on this small-scale evidence.
{\bf General structure:} As noted previously, the radio emission
follows an overall `S'-shaped distribution. This could be due either
to a change in the ejection axis of the jets with time, or else to the
interaction of the jets with a rotating interstellar medium. As
Figure~4 shows, the brightest optical line emission tends to be
associated with the leading (convex) edge of the `S' (Capetti et
al. 1995a) - an observation which could favor either scenario.
{\bf The flat-spectrum core:} In radio-loud sources the innermost
radio components observed with VLBI are usually assumed to mark the
first recollimation shocks within the jet. The identification of an
unresolved flat-spectrum component in Mrk~3 therefore provides us with
the best estimate of the position of the central engine. The
deconvolved size of this source is $<10$~pc, suggesting that the first
shocks in the jet must occur within 5~pc of the central
engine. Clearly an immediate goal for future VLBI would be to study
this region in more detail and hopefully to resolve the core component
into two recollimation shocks. Flux density comparisons might then
give some clues as to the presence of relativistic velocities in the
inner jet regions and the orientation of the two jets with respect to
our line of sight.
{\bf The western hotspot:} The hotspot itself appears to be
edge-brightened, but with a complex of bright knots in the center. The
overall appearance is suggestive of a shell or bowshock where the jet
is interacting violently with the external medium. The knots could be
the result of radio emission over a relatively large 3D interaction
surface at the head of the jet which is seen in
projection. Alternatively they could mark a series of increasingly
violent collisions as the jet is diverted within the hotspot before
finally impacting on the external medium.
\placefigure{f4}
{\bf A comparison of the western and eastern jets:} Apart from their
participation in the overall `S'-shaped morphology of the source, the
two radio jets display a number of obvious differences. The
most striking of these is the absence at the end of the eastern jet
of strong radio emission and of any structure analogous to the bright
western hotspot. Another obvious difference is that, unlike its western
counterpart, the eastern jet contains two bright radio components
within 100~pc of the nucleus.
It is tempting to suggest that these two circumstances are related.
If, as suggested above, the radio knots mark the sites of shocks
where oscillations in the jet direction have caused it to strike the
walls of the channel, then the presence of two strong collisions early
on in the course of the eastern jet might cause much of the bulk
kinetic energy of the jet material to be thermalised, leading to a
large drop in the jet's Mach number.
Radio maps at lower resolution (eg Unger et al. 1986; Kukula et
al. 1993) show that beyond the structure visible in the current maps
the eastern jet emission curves to the south and continues to fade.
Interestingly, the HST study by Capetti et al. (1995a) shows that the
line emission in this region {\it does not} fade but remains bright
(Figure~4). However, the ionisation state here, as traced by the
[O{\sc iii}]/H$\alpha$ ratio, is two to three times higher than in the
rest of the NLR, indicating lower gas densities. By contrast, the
diffuse line emission associated with the western hotspot is
relatively faint and has a low ionisation, which Capetti et
al. interpret as evidence for a high-density shell of shocked material
surrounding the radio lobe.
The differences in the properties of the two jets recall those between
the two main types of structure found in classical radio galaxies:
FRII sources, in which supersonic jets end in sharp-edged hotspots
confined by shocks; and FRIs, where the jets appear to become subsonic
before they terminate (see, for instance, Leahy 1991). Thus the
western jet in Mrk~3 is analogous to an FRII jet, maintaining a high
Mach number along its length and terminating in a violent shock where
it impinges on the external medium. In contrast, the eastern jet,
once it has passed through the two bright knots, behaves more like an
FRI source, dissipating into its surroundings without producing strong
shocks. Thus, the radio source in Mrk~3, although roughly an order of
magnitude less luminous than classical radio sources and confined to
the nuclear regions of its host, might offer insights into both types
of structure displayed by large, powerful radio galaxies. Further
high-resolution work on this object, such as our own forthcoming VLBA
observations, will be necessary in order to determine the spectral
indices of the compact radio features more accurately, and to search
for proper motions and other changes within the jet.
\subsection{Markarian~231}
\placefigure{f5}
Mrk~231 is an ultra-luminous infrared galaxy (Sanders et al. 1988),
whose Seyfert 1 spectrum also contains multiple blueshifted
absorption-line systems (Boksenberg et al. 1977). Boksenberg et
al. also estimate 2 magnitudes of extinction towards the optical
nucleus of Mrk~231, implying that the object is technically a
radio-quiet quasar rather than a Seyfert galaxy, according to the
luminosity criterion of Schmidt \& Green (1981).
Mrk~231 possesses complex radio structure on a variety of physical
scales, consisting of both compact, high-surface-brightness features
and regions of more diffuse emission. At the lowest angular
resolutions radio images of Mrk~231 show a bright, nuclear source with
a large region of faint emission extending $\sim 30$~arcsec ($\sim
2.4$~kpc) to the south (Hutchings \& Neff 1987; Carilli, Wrobel \&
Ulvestad 1998). The nuclear radio source remains point-like at VLA
resolutions ({\it eg} Kukula et al. 1995) but with VLBI the component
is resolved into a small, high-surface-brightness central
component embedded in a diffuse, roughly elliptical region of emission
$\sim 200$~mas ($\sim 160$~pc) across (Carilli et al. 1998). Carilli
et al. interpret this diffuse structure as synchrotron emission from
the inner part of the molecular gas disc detected in CO by Bryant \&
Scoville (1996).
The compact radio source at the center of the `disc' has a convex
radio spectrum, and a high brightness temperature (McCutcheon \&
Gregory 1978; Ulvestad, Wilson \& Sramek 1981; Neff \& Ulvestad 1988).
Neff \& Ulvestad (1988), using three-station VLBI with the EVN in the
early 1980s, were able to demonstrate that this component is elongated
in a north-south direction, perpendicular to the major axis of the
molecular disc. More recent VLBA observations have further resolved
the source into a core-lobe structure (Ulvestad, Wrobel \& Carilli
1998; Carilli et al. 1998) with a bright central component and
southern lobe separated by $\sim 30$~mas ($\sim 20$~pc), and a weaker
lobe 20~mas to the north. At high frequencies the `core' itself is
resolved into a 2-milliarcsec (1.5-pc) triple source which is
misaligned with the `lobes' to the north and south (Ulvestad, Wrobel
\& Carilli 1998), reminiscent of the parsec-scale radio structures of
NGC~1068 (Gallimore et al. 1997) and NGC~4151 (Ulvestad et al. 1998).
The lack of short baselines in the EVN means that it is most sensitive
to radio structures which are intermediate in size between the
small-scale ($\sim 30$~mas) core-lobe source and the larger {$\sim
200$~mas) `disc' detected with the VLBA by Carilli et al. (1998). In
the current EVN map of Mrk~231 (Figure~5) the diffuse `disc' emission
is resolved out, leaving only the linear nuclear source. The core-lobe
structure from the VLBA maps is only partially resolved by the EVN and
forms the brightest part of a somewhat larger ($\sim 100$~pc) jet-like
structure aligned in PA$\sim20^{\circ}$. In addition to the VLBA
source, this `jet' contains two more radio components $\sim 70$~mas
(50~pc) and 130~mas (90~pc) to the southwest.
The total flux density of the VLBA core-lobe source does not appear to
have varied significantly during the six years separating the EVN and
VLBA observations ($94\pm 10$~mJy at 1.6~GHz with the EVN in September
1990, and $100\pm 10$~mJy at 1.4~GHz with the VLBA in December
1996). Our measured fluxes are also in good agreement with those of
Lonsdale, Smith \& Lonsdale (1993), made in September 1991 at
1.5~GHz. At higher frequencies, where the emission from the extended
steep-spectrum components is less dominant, the radio source is known
to be variable (McCutcheon \& Gregory 1978; Ulvestad, Wilson \& Sramek
1981; Neff \& Ulvestad 1988).
Clearly, our understanding of this object is far from complete. The
presence of structures on a variety of scales and in different
orientations - the $\sim160$-pc `disc' identified by Carilli et
al. (1998), the $\sim50$-pc `jet' seen in the current study, and the
misaligned $1.5$-pc nuclear triple source detected with the VLBA by
Ulvestad, Wrobel \& Carilli (1998) - suggest that several processes
are contributing to the radio continuum emission in Mrk~231.
Ulvestad, Wrobel \& Carilli (1999) provide a more detailed discussion
of the various radio components in this object and the relationships
between them.
\subsection{Markarian~463E}
\placefigure{f6}
\placefigure{f7}
This object forms the eastern component of an interacting pair of
galaxies. Its Seyfert 2 spectrum contains polarized broad H$\alpha$
(Miller \& Goodrich 1990, Tran 1995) and, in the near infrared, broad
Pa$\beta$ (Veilleux, Goodrich \& Hill 1997), indicating the presence of
a hidden Seyfert 1 nucleus. Mazzarella et al. (1991) infer large
amounts of visual extinction towards the nucleus and speculate that,
but for this, Mrk~463E would be classified as a quasar. Ground-based
studies of the [O{\sc iii}] emission show that the nuclear region is
elongated in PA$\sim180^{\circ}$ and that fainter emission fills a
roughly biconical volume extending $\sim20$~arcsec ($\sim$18~kpc) to
the north and south of the nucleus (Hutchings \& Neff 1989).
The object possesses radio structure on several different scales, all -
like the [O{\sc iii}] emission - aligned in a roughly north-south
direction. Mazzarella et al. (1991) report radio components 4~arcsec
(3.6~kpc) north and 18~arcsec (16.2~kpc) south of the nucleus. Strong
radio emission is also associated with the nucleus itself: existing
radio images (Unger et al. 1986, Neff \& Ulvestad 1988, Mazzarella et
al. 1991) show a bright, slightly extended radio source coincident
with the optical brightness peak of the galaxy, with a second radio
component lying $1.2$~arcsec (1~kpc) to the south. In terms of both
luminosity and overall size the radio source associated with Mrk~463E
is intermediate between the sources typical of the majority of Seyfert
galaxies and those found in radio galaxies and quasars.
Optical continuum imaging with the F517N and F547N filters of the HST
(Uomoto et al. 1993) resolves the nuclear region, revealing a bright
component $\sim 500$~mas across, with an optical `jet' extending
southwards for 820~mas (740~pc) and stopping just short of the
southern radio component. Uomoto et al. attribute this radiation to a
mixture of emission lines produced {\it in situ} and scattered light
(both continuum and emission-line) from the hidden Seyfert~1 nucleus,
rather than optical synchrotron emission.
At a distance of $\sim 200$~Mpc, the 20-mas EVN beam corresponds to a
physical size of 18~pc. In our new map (Figure~6) the brightest radio
source is resolved into four discrete components, all aligned in
roughly the same direction as the large-scale structure. Three of the
sources form a closely spaced triplet, whilst the fourth lies some
300~mas (270~pc) further south. The radio emission to the south of
the optical `jet' is also resolved, and is elongated in the same
direction as the overall radio and optical structures. The radio
components become progressively weaker as one moves southwards.
\subsubsection{A new interpretation of the nucleus of Mrk~463E}
The optical brightness peak of Mrk~463E has a Type~2 (narrow line)
spectrum and has traditionally been assumed to mark the site of the
supermassive black hole responsible for the activity in this
object. Since the brightest radio component in Mrk~463E is roughly
coincident with the optical peak it followed that this was the radio
source associated with the obscured central engine. However, in other
Seyferts with prominent linear radio structures this is not always the
case; in both Mrk~3 (Kukula et al. 1993; this paper) and Mrk~6 (Kukula
et al. 1996) the brightest radio component occurs at the end of the
jet furthest from the nucleus, is resolved, edge-brightened, and has a
steep radio spectrum - all the hallmarks of a `hotspot' where the jet
material is ploughing into the ISM of the host galaxy. In both cases,
the radio emission is associated with enhanced narrow-line
emission, which appears to surround the radio structure like a shell
or halo. In Mrk~3 the hidden central engine is associated with an
unresolved, flat-spectrum radio component. In Mrk~6, the Seyfert~1
nucleus does not appear to coincide with any of the detected radio
components. Roy et al. (1994) invoked free-free absorption by ionized
gas in the narrow line region to explain the lack of compact nuclear
radio sources in their sample of Seyfert~1s (in Seyfert~2s we observe
the object `side on' and so the depth of ionized gas between the
observer and the central engine is much reduced). The same mechanism
might also account for the absence of a radio component at the
position of the Type~1 nucleus in Mrk~6.
Tremonti et al. (1996) used HST imaging polarimetry to locate the
hidden Seyfert~1 nucleus in Mrk~463E. They report that the magnetic
polarization vectors converge on the southern tip of the optical
`jet', indicating that this, and not the bright narrow-line region
north of the `jet', is the true site of the central engine in this
object.
The detailed radio structure of Mrk~463E revealed by our new EVN
observations, when considered in conjunction with the optical
structures reported by Uomoto et al. (1993), lends strong support to
this finding. In Figure~7 we show the current 18-cm radio map
(greyscale) superimposed on the HST continuum image by Uomoto et
al. (contours). There is some question as to the positional accuracy
of the two images. The radio data has undergone self-calibration,
causing the absolute positional information to be lost. More
seriously, HST pointing is generally only accurate to within
$\sim1$~arcsec, and the true pixel scale of the HST Planetary Camera
image may be slightly different from the canonical value of
43.9~mas~pixel$^{-1}$. In performing the registration of the two
images we have forced the brightness peaks to coincide. Several
aspects of the argument set out below rely heavily on this assumption,
so it is clearly imperative that more accurate radio and,
particularly, optical astrometry be carried out on this object.
The overall outline of the optical emission in the HST image is
roughly wedge-shaped, with its apex at or near the southern radio
component, 1.2~arcsec from the presumed `Seyfert~2 nucleus'. The
enhanced optical emission forming the `jet' begins immediately north
of this radio component and continues north for almost a kiloparsec,
incorporating three bright knots, before terminating abruptly. A gap
of $\sim 0.1$~arcsec (90~pc) separates the `jet' from the bright,
resolved `nucleus' to the north and a second radio component appears
to lie in the gap.
Our suggested interpretation of Mrk~463E is that the `Seyfert~2
nucleus' at the northern end of the jet is in fact a region of
enhanced narrow-line emission associated with a radio hotspot. In
accordance with the data of Tremonti et al. (1996) we suggest that the
central engine itself lies at the southern end of the optical jet,
although for the reasons outlined below we feel that it is unlikely to
be exactly coincident with the southern radio component. The overall
wedge-shape of the extended optical emission can be accounted for by
shadowing of the radiation field of the AGN by an optically-thick
torus surrounding the central engine - the favored model for several
other Seyferts with similar optical morphologies ({\it eg} Wilson \&
Tsvetanov 1994).
We interpret the linear optical feature seen in the HST image as an
expanding cylindrical sheath or halo of material surrounding the
channel of the radio jet. Such a model was suggested by Capetti and
co-workers to explain the very similar structures seen in emission
lines in several Seyferts, including Mrk~3 \& 6 (Capetti et al. 1995a,
\& b, 1996). The material in the halo has been pushed aside by the
passage of the radio plasma, and is subsequently ionized by UV photons
from the central engine. The high density of the halo relative to the
surrounding gas also makes it a more effective site for scattering of
the nuclear continuum. Uomoto et al. (1993) claim that the F547M to
F517N flux ratio of the `jet' indicates a mixture of optical continuum
and emission lines. This can be explained in our model by
contributions from line emission produced {\it in situ} by the
swept-up, ionized gas, as well as polarized continuum and broad
permitted line emission (already detected by Miller \& Goodrich 1990)
from the obscured nucleus which have been scattered into our line of
sight by electrons in the halo. Double-peaked line profiles would be
indicative of an expanding cylindrical sheath of material. Free-free
continuum emission from shocked gas in the halo might also be
present. Optical spectroscopy and spectropolarimetry with high spatial
resolution will be required in order to determine the precise
contributions from each mechanism.
The estimated lifetime for $\lambda$~18cm synchrotron-emitting
electrons in this source (derived from the minimum energy magnetic
field values listed in Table~2) is $\sim10^{4}$~years. Hence, unless
the bulk velocity of the jet is an appreciable fraction of the speed
of light ($\geq 0.01c$) the electrons in the plasma will radiate most
of their energy before traveling more than a few tens of parsecs from
the origin of the jet. At larger distances a fresh injection of energy
will be required in order to produce appreciable amounts of
synchrotron emission. We would therefore expect to detect strong
synchrotron emission at sites where the jet plasma is interacting
violently with its surroundings, leading to the re-acceleration of
electrons, but very little where the jet is flowing freely.
Thus, in the region of the optical `jet' the plasma has already
drilled out a channel for itself, through which it flows relatively
unimpeded, and produces little or no synchrotron emission. The radio
emission immediately north of the linear optical feature might
indicate that the jet is colliding with an obstacle at this point. It is
not clear why the optical emission should simultaneously fall in
brightness, but Taylor, Dyson \& Axon (1992) point out that the shocks
produced by the collision of a jet with an external medium can heat
the swept-up material to temperatures $\gg 10^{4}$K, too large for the
emission of forbidden lines to occur.
Following this encounter, which appears to alter the course of the
jet, the plasma feeds into the bright radio hotspot. This is the site
at which the jet is actively carving a channel into the local ISM. The
observed radio structure of the hotspot consists of three successively
brighter components, the first two of which are compact and a third
which is resolved in a direction roughly orthogonal to that of the
jet. We suggest that these features are analogous to the internal
radio structure observed in the western hotspot of Mrk~3 (Figure~2)
and that the bright transverse feature is associated with a bowshock
at the head of the jet. Mazzarella et al. (1991) detected $\sim
320$~mJy of emission in this region with the VLA at 20~cm, implying
that over 100~mJy of extended flux has been `resolved out' by the
small beam of the EVN. The VLA map by Mazzarella et al. also contains
a weak radio component 4~arcsec (3.6~kpc) north of the nucleus. This
could indicate either that the jet is not entirely disrupted in the
structure which we have dubbed the hotspot, or else that the hotspot
was formed when dense material drifted into the path of the jet,
cutting off the radio structure further to the north.
Unger et al. (1986) found that the radio emission in Mrk~463E has a
steep spectrum between 408 and 1666~MHz. Although the angular
resolution of their maps was not sufficient to resolve the individual
knots which make up the northern component, a steep radio spectrum in
this region lends credence to our suggestion that this is a radio
hotspot rather than the central engine itself. The radio component to
the south of the optical `jet' also has a steep spectrum and the
emitting region is resolved in our EVN map. This makes it unlikely to
be directly associated with the central engine, since the nuclear
components observed in other Seyferts tend to be compact, with flat
spectra.
However, further evidence that the true nucleus is located in this
vicinity is the mention by Neff \& Ulvestad (1988) of a region of
extremely blue optical continuum emission apparently coincident with
the southern radio component (in our EVN map). We suggest that this
blue emission might be a signature of the hidden Seyfert~1 nucleus.
New studies of the broad-line emission from Mrk~463E (polarized
H$\alpha$ in the optical and broad Pa$\beta$ in the near infrared)
with improved spatial resolution would also assist in defining the
scattering geometry and locating the central engine.
Mazzarella et al. (1991) also found a knot of radio emission 18~arcsec
(16.2~kpc) south of the nucleus in their 20~cm VLA image, which could
be construed as a radio hotspot at the end of a southern counterpart
of the jet in the north (although, of course, it might equally well be
entirely unrelated to the AGN). The relative weakness of the radio
emission in this component, and the lack of any detectable emission
from the counterjet itself, might be explained in a number of ways.
If the counterjet is directed away from us, into the plane of the sky,
and the bulk velocity is relativistic then Doppler dimming could lead
to a reduction in the observed radio emission. A difference in the
environment to the south of the nucleus might also lead to less radio
emission: if the counterjet encounters little resistance from the ISM
then there will be few opportunities for the re-acceleration of
electrons and a corresponding absence of synchrotron radiation. A
lower density in the southern ISM would also account for the greater
length of the counterjet (16~kpc) relative to the maximum extent of
the northern radio emission (3.6~kpc). Certainly, in view of the
disturbed nature of the Mrk~463 system, we would not necessarily
expect the galaxy's density contours to be symmetrical about the
nucleus of Mrk~463E.
A final possibility is that the counterjet is currently dormant or
inactive and the southern radio component is therefore the fading remnant
of a more luminous hotspot.
The ground-based [O{\sc iii}] image by Hutchings \& Neff (1989) shows
a fan-shaped region of emission extending $\sim 20$~arcsec to the
north of the nucleus. However, to the south the [O{\sc iii}] emission
is concentrated in a bright, resolved knot at a distance of $\sim
10$~arcsec from the nucleus, with little extended emission. This could
indicate either larger amounts of obscuration between the Earth and
the ionized gas to the south (perhaps because the southern cone of
ionising photons is directed away from us) or else, once again, a lack
of gas in this region.
However, at this stage we cannot rule out the possibility that much of
the faint [O{\sc iii}] and radio emission on arsecond scales is
associated with starforming regions rather than the central engine of
Mrk~463E.
To summarize: we suggest that the optical and radio brightness peak in
Mrk~463E has been mistakenly identified as the location of the active
nucleus in this object, when in fact it represents the working surface
of a jet, which originates $\sim 1$~arcsec to the south. Since our
understanding of this object depends crucially on the ability to match
up high-resolution images made in many different regions of the
spectrum, more accurate astrometry at all wavelengths is urgently
required.
\section{Discussion}
Three of the objects in our study (Mrk~3, 231 and 463E) were
observed previously with the EVN in the early 1980s at 1.6~GHz (Neff \&
Hutchings 1988). Our maps agree with the previous data in terms of
the overall morphology of the sources but advances in instrumentation
and software have resulted in significant improvements in image
quality and the new maps therefore contain much more detail.
Most of the compact radio emission in all four Seyferts has a steep radio
spectrum (component~4 in Mrk 3 has a flat spectrum)
and is clearly non-thermal in origin. Neff \& Ulvestad (1988)
point out that unrealistically large supernova rates would be required
in order to reproduce the radio luminosities observed in the compact
radio components of Mrk~3, 231 and 463; based on our current
measurements the same appears to be the case for Mrk~1. Moreover, the
combination of high ($\geq 10^{5}$K) brightness temperatures and steep
radio spectra appears to rule out supernovae as the primary source of
the synchrotron emission since, in a starburst complex, brightness
temperatures would be limited to $\leq 10^{5}$K and one would expect
the radio spectrum to flatten for $T_{B}>10^{4}$K (see Condon 1992).
An AGN is therefore the most likely source of the VLBI radio
emission, and in three of the objects we have mapped, the radio
morphology also strongly suggests the presence of less powerful forms
of the jets found in radio-loud sources. However, there is evidence
for galaxy interactions/mergers in all four cases, and the more
diffuse and amorphous radio emission (for example, that detected on
arcsecond scales in Mrk~231 by Hutchings \& Neff 1987, or the radio
halo in Mrk~1) may well be the result of starburst activity rather
than a direct product of the AGN.
What can be learned from these four objects? We cannot hope to draw
generalized conclusions about Seyferts as a class from such a small
sample. Also, optical imaging and kinematical studies indicate that
all four objects in our study are currently interacting with
neighboring galaxies (Mrk~1, 3 \& 463E) and/or have recently
undergone mergers (Mrk~3 \&~231). Though such processes may well play
a r\^{o}le in triggering and fueling the activity in all active
galaxies ({\it eg} Phinney 1994), the prominent disturbances in these
four galaxies might account for some of the complexity in their radio
structures. However, the current observations do highlight a number
of important points which should be taken into account in any
consideration of Seyfert radio properties.
With very high angular resolution, jet-like structures {\it can} be
found even in compact Type~1 objects such as Mrk~231, in which the
axis of the central engine is probably aligned quite closely with the
line of sight. The presence of large-scale ($>100$~pc) collimated
radio jets in Mrk~3 and 463E also serves to reinforce the message that
Seyfert~1s are capable of producing such structures since, although
these two objects have Type~2 spectra as observed from our own vantage
point, spectropolarimetry and infrared spectroscopy indicate that both
contain a hidden Type~1 nucleus.
On the other hand, there are Seyferts such as Mrk~1 which, although
they possess respectably luminous radio sources, show no signs of
jet-like structure on scales as small as a few parsecs. Mrk~1 is a
particularly good example of this class because, as a Type~2 Seyfert,
the unified model predicts that the ejection axis for a putative radio
jet should be close to the plane of the sky, thus minimizing any
projection effects. It is interesting to note that, in the present
sample, both of the Type~2 Seyferts which show evidence for a hidden
Type~1 nucleus also contain radio jets, whilst Mrk~1, in which all
attempts to find broad lines have so far failed, does not.
It appears that whilst some Seyfert galaxies of both types are
certainly capable of producing linear, jet-like radio sources, the
radio structures in Seyferts can range in size from extremely small
($<10$~pc) nuclear sources through to jets on scales of tens, hundreds
and even thousands of parsecs. This large distribution of sizes is
intrinsic to the Seyfert population and is quite separate from viewing
angle effects. Possibly the radio jets in Seyferts are relatively
short-lived compared to the lifetime of the active nucleus, so that
when we observe a sample of objects, we see jets in various stages of
development, as well as objects in which jets are not currently
active. Alternatively, there may be some Seyferts of both spectral
types which are simply incapable of producing radio jets, perhaps due
to environmental factors such as the gas density/kinematics in the
nuclear region of the host galaxy and the orientation of the central
engine relative to the plane of the host.
The factors at work in the host galaxy (dust and gas content,
interaction and star-formation history) and the active nucleus
(immediate environment and orientation of the central engine, the
size, age and morphology of the radio source), as well as the biases
and omissions introduced by observational constraints, combine to make
the observed properties of a particular Seyfert galaxy distinctive and
unique. This point is illustrated by the cases of Mrk~3 and 463E, in
which jet and counterjet have very different appearances due, perhaps,
to the different conditions which they encounter {\it en route}. One
should not forget that Seyfert galaxies are complicated systems
composed of many elements, whose properties can change and evolve on
relatively short timescales. Consequently one should not be too
surprised if simplistic models fail on some level when applied
globally to large samples of objects.
Further studies of the relationship between the AGN and its host need
to be carried out before we can be confident about our interpretation
of Seyfert radio surveys. High-resolution imaging at many wavelengths
will clearly play a major r\^{o}le in such studies, but, as the present
work demonstrates, accurate astrometry will be crucial in order to
match up the different datasets.
\section{Summary}
The maps presented here clearly demonstrate that VLBI can be a highly
effective tool for investigating the radio structures of Seyfert
nuclei, even though the radio emission from these objects is
relatively weak when compared to radio-loud sources. Since Seyferts
are in general much closer to us than other types of AGN, the spatial
resolution achieved by VLBI is unprecedented, allowing us to study the
structure of the radio jets on scales of a few parsecs and to
investigate in detail the complex interactions between the jets and
their environment.
Radio jets exist in both broad- and narrow-line Seyferts, although the
scale of such structures varies from a few tens of parsecs to several
kiloparsecs. The range in sizes appears to be intrinsic to the Seyfert
population and is independent of any additional projection effects
caused by the orientation of the central engine. However, there are
also objects which possess luminous nuclear radio sources yet which
show no evidence for jets, even on scales of $\leq10$~pc.
In Mrk~3 and 463E we find evidence for jets containing internal
shocks and terminating in luminous, edge-brightened hotspots
resembling the structures seen in large-scale FRII radio
sources. However, in general the observed radio properties of an
individual Seyfert nucleus are likely to be a complex function of the
local environment as well as of the central engine itself, making
each object unique. This has important consequences for the
interpretation of radio surveys of Seyferts.
\acknowledgements
The authors would like to thank Bill Junor for help in reducing the
data, Ant Holloway for producing Figure~4, Christina Tremonti and Alan
Uomoto for useful discussions on the subject of Mrk~463, and Ger de
Bruyn, George Miley and Dave Graham for their invaluable contribution
to the early stages of this project. We are also grateful to James
Ulvestad, the referee, whose comments resulted in significant
improvements to this paper. MJK acknowledges PPARC support and STScI
funding (grant number O0573). The EVN is a large-scale facility of the
European Union and is administered by the European Consortium for
VLBI. MERLIN is a national facility operated by the University of
Manchester on behalf of PPARC. This research has made use of the
NASA/IPAC Extragalactic Database (NED) which is operated by the Jet
Propulsion Laboratory, California Institute of Technology, under
contract with the National Aeronautics and Space Administration.
|
\section{INTRODUCTION}
Prior to the pioneering study of ``flashing galaxies'' by Searle, Sargent , \&
Bagnuolo (1973), galaxy evolution was thought to be a smooth, exponentially
decaying process lasting several billion years. The discovery of a significant
population of galaxies actively
forming stars dramatically changed this picture (Harris \& van den
Berg 1981; Seiden \& Gerola 1982; Schweizer 1986; Burstein 1986;
Kumai, Basu , \& Fujimoto 1993). Many galaxies are now known to experience
a series of episodes of enhanced star formation during their lifetime,
especially through merging events (Schweizer 1986; Zepf, Geisler, \&
Ashman 1994), but also following the formation of a bar (Phillips 1996;
Martin \& Friedli 1997).
Although known to be very important in the evolution of galaxies, the star
formation process is usually poorly accounted for in galactic evolution
models. ``Recipes'' (e.g. Kennicutt 1998) are provided to account for the
global characteristics of the star formation rate versus the interstellar
medium (ISM); those are derived from the global
or azimuthally averaged properties of observed
galaxies while star formation is a local phenomenon. Little is known about
the local aspects of the star formation process.
The way the star formation propagates, the dynamics of the star formation,
and its relation to the physical properties of the ISM are usually
parameterized as a percolation phenomenon (Mueller \& Arnett 1976;
Gerola \& Seiden 1978; Seiden, Schulman , \& Gerola 1979; Seiden \& Gerola
1982) or regulated by feedback or porosity
(Navarro \& Steimetz 1997; Silk 1997).
All are ad-hoc parameterizations of a very complicated process.
One of the most difficult tasks in the quantification of the star formation
process is to assign precise ages to star forming regions.
Starburst galaxies are ideal objects to study the local properties of
various star forming regions and their effects on the properties of a galaxy.
Starbursts form many compact ($\sim$ 2-5 pc), luminous and blue
($-$11 $>$ M$_{UV}$ $>$ $-$18) super star clusters (SSC; Meurer {et~al}. 1995)
that radiate tremendous amounts of light in all wavebands. Also, a
reasonable fraction of this light can easily ionize many atoms
to the first or second stage of ionization, while the high supernovae
rate associated with starbursting regions
excites and ionizes elements to even higher stages.
These characteristics of starbursting regions provide the opportunity
to study the physical properties of the gas. The knowledge of the gas
properties is essential if one wants to constrain the parameters
involved in the quantification of the star formation process. Several
parameters of a stellar population other than age, can cause a
variation of the integrated magnitude and colors of a SSC. Extinction,
metallicity and the mass of stars formed at present or at any given
time by the cluster are all factors that can affect its observed
characteristics. It is a difficult task to disentangle which parameter
or combinations of parameters cause the differences in the observed
integrated magnitudes and colors of stellar clusters. For example, a
young cluster embedded in dust can be mistaken for an old, unreddened
cluster.
Nevertheless, with appropriate tools, the effect of these various
parameters can be separated. In this paper, I will show
that the extinction degeneracy can be lifted by a reasonable amount by studying
the SSC integrated light with the (B$-$H) vs. (H$-$K) two color diagram
(BHK diagram). I discuss in \S 2 the main physical
arguments that support this choice of colors and I propose an
application of the method to the interacting system Arp 299 in \S 3.
\section{CHRONOMETRY WITH THE BHK DIAGRAM}
\subsection{MODELING}
Figure \ref{fig1} shows the theoretical behavior of the BHK colors
for an instantaneous coeval burst
of star formation forming 10$^6$ M$_\odot$ of stars ranging from
1 to 120 M$_\odot$ according to a Salpeter initial mass function
using the models of Leitherer et al. (1999). These models are
implemented with the new set of stellar evolution models of the Geneva
group. For masses of 12~--~25~M$_\odot$ (depending on metallicity)
and above, the Leitherer et al. models are using evolutionary
tracks with enhanced mass loss of Meynet et al. (1994). The tracks
of Schaller et al. (1992), Schaerer et al. (1993a, 1993b), and
Charbonnel et al. (1993) with standard mass loss are used between
12 and 0.8~M$_\odot$. Atmosphere models are from Lejeune, Buser, \& Cuisinier
(1997) for stars which are not in a Wolf-Rayet phase. Wolf-Rayet stars
are modeled with the atmospheres of Schmutz, Leitherer, \& Gruenwald (1992).
A discussion of the uncertainties related to the modeling technique can
be found in Leitherer et al. (1999).
A modification to the models was necessary since stellar evolution
models fail to reproduce the red supergiants (RSG) features at low
metallicities. As the study of Origlia et al. (1998) pointed out, the
low metallicity models does not reproduce the CO 1.62$\mu$ and CO
2.29$\mu$ indices as well as the J$-$K colors of a selected sample of
LMC young clusters. However, they pointed out that if the fraction of
time spent as a RSG during the core-helium phase is forced to at least
50 \% and if the RSG temperature is maintained to less than 4000 K,
the models agrees well with the observations. Our modeling technique
was modified according to this prescription.
The calculated paths on the BHK diagram of Figure \ref{fig1} are for
metallicities of 0.05 Z$_{\odot}$ and Z$_{\odot}$ and no extinction is
shown. In the next three sections, I discuss how different
fundamental parameters like the metallicity, extinction, dust emission
and nebular emission can affect the locus of the observed points on
the BHK diagram and why the B, H and K colors are the best choice for
this type of study.
\subsection{THE EFFECT OF METALLICITY}
For a single metallicity, the colors are distributed in a u-shaped
form (see Figure~\ref{fig1}), where two different branches can be
easily seen. The ``young branch'' runs almost diagonally from the
right to the left of the diagram. The light originating within a young
starbursting region may be dominated by massive O stars ($\sim$80 \%) and
nebular emission ($\sim$20 \%) in the B band, while in the H and K bands,
the nebular emission may contribute up to as much as $\sim$70 to 85 \% of the total flux.
As the nebular emission becomes negligible with the massive O stars evolving
to become red supergiants, a turnoff occurs in the integrated colors. The color
path then takes the ``old branch'', a path that mainly runs along the
(B$-$H) axis. After 10 Myr, the color path turns around and follows the
old branch. This behavior causes a degeneracy between regions aged between
6 and 10 Myr and those aged between 10 and 50 Myr. As we will see in section
\S 3.2, this degeneracy can easily be lifted with spectroscopic data.
After 50 Myr (not shown in Figure \ref{fig1}), the colors do not change
significantly and the evolutionnary path remain in the same region
than the 50 Myr point.
Models in Figure \ref{fig1} also show that an increase in the
metallicity of the SSC produces a different behavior of the colors
(Figure \ref{fig1}). This behavior can be mistaken for an age change.
However, known abundance properties in galaxies
help; the metal distribution in small starburst galaxies
shows no or very shallow abundances gradients
(Kobulnicky \& Skillman 1997; Devost, Roy, \& Drissen 1997; Kobulnicky 1998).
The two metallicity tracks shown in Figure~\ref{fig1} would correspond to
an extreme gradient of several dex kpc$^{-1}$; such a change in
metallicity between two close SSC belonging to the same galaxy is
most unlikely. Thus, the tracks of a group of SSC from the same
galaxy should show parts falling on the same track on the
BHK color diagram, in the absence of extinction.
\subsection{THE EFFECT OF EXTINCTION AND DUST EMISSION}
The extinction vector on the BHK diagram is more or less parallel to
the old branch since E(H-K) $\sim$ 0.1 E(B-H). This alignment is the
key point to extinction correction. Extinction will
move any points in the old or in the young branch along a vector
that is parallel to the old branch. This has two important advantages.
First, the structure of the old branch will still be recognizable since
its alignment with the extinction vector will
prevent its spread by local extinction differences. Global
dereddening is then possible by fitting together the theoretical
old branch path and the observed old branch structure. The second
benefit is that extinction of points belonging to the young branch
will not be moved to the old branch by extinction. An example
of such a degeneracy is given in Figure \ref{fig2}.
This two color diagram plots the behavior of the (B$-$J) vs. (J$-$H)
with the same model parameters as in Figure \ref{fig1} for solar
metallicity stars. We can see that the young and old branches are
still there but extinction of points belonging to the young branch
will cross the path of the old branch, making the distinction
between an age and an extinction effect almost impossible. Also,
the misalignment of the any of the two branches with the extinction
vector causes a spreading of the points by local extinction
differences and it is very unlikely that the structure of
any of the two branches will be conserved.
Warm dust emission will also modify the behavior of the observed infrared
colors of SSCs. Satyapal {et~al}. (1995) showed that
warm and hot dust emission
contaminates only the K band by adding at most 20\% to its
total flux. Thus, hot dust will make the observed points redder
by a factor of at most 0.2 mag along the
(H$-$K) axis. This value is not high enough to produce a degeneracy
between the young and old branches but will move the points belonging
to the young branch further to the red thus enhancing the difference
between the two branches.
\subsection{THE EFFECT OF NEBULAR LINES}
Nebular emission is also a factor that can produce degeneracies. In
the infrared bands, the main contribution comes from the Br$\gamma$
emission line which contaminates the K band. The models of Leitherer
et al. (1999) show that for a continuous burst of star formation, the
equivalent width of the Br$\gamma$ line decreases from $\sim$ 500 \AA\
when the burst is younger than 3 Myr to less than 100 \AA\ when the
burst is older than 4 Myr. This means that the Br$\gamma$ nebular
line will add $\sim$ 0.12 mag to the K band light when the cluster is
younger than 3 Myr and less than 0.02 mag when the cluster is older
than 4 Myr. The implications for the BHK diagram is that the
Br$\gamma$ emission line pushes the points belonging to the young
branch further to the red and thus, once again, enhancing the
difference between the old and the young branch.
The nebular component in the optical part of the spectrum is much more
important. The hydrogen and the oxygen lines can make an important
contribution to the measured flux and produce unwanted
degeneracies. The H$\alpha$ line contaminates the R band filter while
the H$\beta$ and [O {\sc iii}]($\lambda\lambda$5007,4959) can make
significant contribution to the V band flux. However, the B band is
affected to a lesser degree than the V band filter since the
transmission of the B filter at the wavelength of these lines is about
15\% less than the transmission of the V filter. Also, even moderate
redshifts will shift these lines further into the V waveband. So for
our purpose, the B band filter is minimizing the contamination by the
nebular lines.
\subsection{TIME RESOLUTION}
The considerations outlined in the latter sections restrict the choice
of colors to B, H and K. The alignment of the old branch with the
extinction vector as well as the effects of dust emission and nebular
line emission are strong arguments in favor of this choice. The
achievable time resolution of this method is mainly limited by the
extinction differences from one cluster to the other. However, even
with no data on extinction, discrimination of regions younger and
older than 4 Myr is still feasible. The age determination in this case
is solely based on the color differences between the old and young
branches. Nevertheless, if the exciting stars are not completely
embedded in dust and are themselves visible in the optical domain,
nebular lines ratios measured from line-of-sight spectra to the
stellar knots allow a first-order correction. In such a situation,
extinction correction using the Balmer lines decrement helps to lift a
good part of the (B$-$H) degeneracy, then allowing age determinations
with a time resolution limited by the assumed dust and gas properties.
Uncertainties in the dust modeling mainly comes from the assumptions
made on its spatial distribution. Dust extinction can be modeled
assuming a foreground uniform screen or a clumpy layer in front the
extincted stellar population, or assuming that the dust is mixed
within the stellar population (Natta \& Panagia 1984, Calzetti, Kinney
\& Storchi-Bergmann 1994). Observations of starburst galaxies seem to
rule out dust and stars mixed internally. The intrinsic nature of a
starburst, e.g. the high stellar winds and the frequent supernovae
explosions, may in fact favor the foreground screen geometry (Calzetti
1998). However, very young clusters can be mixed with dust and suffer
very high extinction (Calzetti et al. 1997). This type of regions
maybe missed by the BHK diagram since they will not be detected in the
broad band images.
Nevertheless, for the regions detected in the broad band images, the
worst case scenario would be to make a correction for extinction in
the BHK diagram for a clumpy medium using the foreground screen
approximation. In that case, for a value of A$_V$ $\sim$ 2, the
difference between assuming a foreground screen and a clumpy dust
distribution produces a color difference $\Delta$E(B$-$V) of 0.35. This
produces some uncertainty on the dating of the burst, the magnitude of
which depends on where the burst is along its evolutionary track in
Figure~\ref{fig1}. A small change in (B-H) produces a larger time
error for a 1 Myr old burst than it does for a 6 Myr old burst.
The uncertainty on the dating of the burst will also be affected by
the model used to correct for extinction. There is a fair amount of
evidence that the extinction derived from the stellar continuum is
systematically less than that derived from the nebular gas
(e.g. H$\alpha$/H$\beta$ ratios). This seems to indicate that the dust
is mostly mixed with the gas, and the stars are mostly seen through
regions of lower optical depth. However, a higher degree of accuracy
can be reached if the ratio of the extinction derived from the stellar
continuum to the value derived from the nebular gas is known. This
ratio is believed to be between 2 and 3 for starburst galaxies (Meurer
et al. 1995, Calzetti, Kinney \& Storchi-Bergmann 1994). Using the
extinction law of Kinney et al. (1994) to scale between
$\Delta$E(B-V) and $\Delta$(B-H) and a conversion factor between the
continuum extinction and the nebular gas extinction of 2.5, the error
produced by the assumed spatial geometry translates in the BHK diagram
to values of $\sim$ 1.4 Myr along the old branch and $\sim$ 2.0 Myr
along the young branch. Changing the conversion factor between the
continuum extinction and the Balmer extinction does not change
significantly the error estimate on the age determination.
For ages ranging from 10 to 50 Myr, the time resolution has a value
worse than 11 Myr. The type of resolution derived for these ages
or greater, is incompatible with the purpose of dating star forming
regions to infer the properties of the star formation process. This
means that the BHK diagram can only be useful to trace star formation
to a reasonable resolution for regions that are younger than 10 Myr.
\subsection{A DOUBLE BURST MODEL}
I explored the behavior in the BHK diagram of
a model that forms star clusters in two bursts.
The super star cluster A in NGC 1569 maybe the result of
such a burst. Gonz\'alez Delgado et al. (1997) found that in this
cluster, red supergiants features as well as Wolf-Rayet features were
present and suggested a double burst scenario to explain the
coexistence of these features within the same cluster. However, the
detection of red supergiants and Wolf-Rayet features within the same
spatial location may also be caused by the alignment of two clusters
along the same line of sight (De Marchi et al. 1997).
I modeled the BHK colors of two identical unresolved bursts occurring 1
Myr (double burst model 1) and 3 Myr (double burst model 2) apart. The
integrated colors of the double burst were derived using:
\begin{center}
$({\rm B-H})_{{\rm DB}}=({\rm B-H})_{\rm O} - 2.5log(\frac{1+{\rm A}}{1+{\rm B}})$
\end{center}
where,
\begin{center}
${\rm A}=10^{0.4({\rm M_{B_O}-M_{B_S}})}$ \quad\mbox{and}\quad ${\rm B}=10^{0.4({\rm M_{H_O}-M_{H_S}})}$
\end{center}
Subscript DB means double burst, subscript O relates to the
original burst and subscript S relates to the second burst.
M$_{{\rm B_O}}$, M$_{{\rm B_S}}$, M$_{{\rm H_O}}$ and M$_{{\rm H_S}}$
are the absolute B and H magnitudes of the original and second burst
respectively. The same algorithm was used for the (H$-$K) color.
The physical parameters of the model are the same as the solar
metallicity burst of Figure~\ref{fig1} and no chemical evolution
occurs between the two bursts.
The results are compared to the single burst model in
Figure~\ref{fig3}. The paths of the single burst and the double burst
model 1 are almost identical but differ significantly between the
single burst and the double burst model 2. When both bursts are at
young age (t $<$ 4 Myr) the age estimate is not affected. At
intermediate age, a shift occurs in the (B$-$H) color so that a 5 Myr
old single burst can be mistaken for an 6 Myr old double burst (model
2; Figure \ref{fig3}). This means that if a single burst approximation
is made when the studied cluster formed in two bursts separated by 3
Myr, the time estimate can be altered at ages between 6 and 10 Myr by
a value of 1 to 2 Myr. However, if the two bursts are resolved and
occurred at different spatial locations, each stellar populations will
be given an age by the BHK diagram that will not be affected by the
double burst scenario.
\section{A TEST CASE: Arp 299}
I have tested the BHK diagram method of chronometry by exploring the
SSCs in the interacting system Arp 299. This system includes two
galaxies; NGC 3690 and IC694. The assumed distance to this object is
42 Mpc (Nordgen, et al. 1997; H$_0=$75 km s$^{-1}$ Mpc$^{-1}$).
Several UV bright young SSCs (Meurer et al. 1995) are present as well
as bright infrared sources. With their K band image, Wynn-Williams et
al. (1991) identified four bright infrared sources belonging to the
whole system, three of which are associated with NGC~3690 and one
associated with IC 694. More recently, Satyapal et al. (1997)
established that the light coming from Arp 299 can be entirely
explained by a starburst model. I will show how an age determination
of the various sources with the BHK diagram gives insight into the
star formation history of Arp 299.
\subsection{OBSERVATIONS AND DATA REDUCTION}
Observations of Arp 299 were obtained during three different observing
runs. First, wide band BVRI images were obtained at the 1.6 meter
telescope of the Observatoire du mont M\'egantic (OMM) in March
1996. The detector was the Thompson THX 1024 $\times$ 1024 CCD with
pixel size of 19 $\mu$ and a plate scale of 0.31\arcsec\
pix$^{-1}$. The JHK band images were taken with the
Canada-France-Hawaii Telescope (CFHT) in January 1997 using the
infrared camera MONICA (Nadeau et al. 1994). Spectroscopy of Arp 299
was also performed at the CFHT with MOS-ARGUS, a fiber optics system
that allows 2-D spectroscopy with a $\sim$ 12\arcsec $\times$
10\arcsec\ field of view and sampling fiber apertures of
0.4\arcsec. The detector used was the STIS2 2K $\times$ 2K thin CCD
with 21$\mu$ pixels.
\subsubsection{Wide band images}
For optical imaging analysis, bias, flatfield and dark images were
taken during the same night the observations were performed. The
corresponding correction was made for each image of Arp 299. Flux
calibration was achieved with Feige 56. Exposure times and other
relevant information about the observing run are listed in Table
\ref{tbl-1}. Bad pixels were removed by interpolation. Images in one
filter were combined using a combination algorithm to eliminate the
signature of the cosmic rays and to increase the signal-to-noise (S/N)
ratio. The mean FWHM of the point-spread function as measured from
stars in the field of the four combined images is $\sim$ 1\arcsec.
The sky conditions the night of the observations allowed reliable
photometry. The images were then aligned together using field
stars. All the successive image reduction steps were performed with
IRAF.
The second set of observations was done at the CFHT in the J, H, and K
bands with the IR camera MONICA (Nadeau et al. 1994) at the f/8 focus
on January 20 1997 (Table \ref{tbl-1}). The images were corrected for
flatfield and dark current. The flux calibration was made with the IR
UKIRT standard stars FS2, FS21, and FS25. Bad pixels were corrected
with a mask. In each band, four images were combined using a rejection
routine to raise the S/N ratio and remove the cosmic ray
signature. The images were aligned using point sources common to the
three wavebands. During the observations, the seeing was 0.7\arcsec.
Once again, the sky was clear and allowed for reliable photometric
calibration. Most of the data reduction steps were performed using
software developed by the Montreal group (D. Nadeau). Steps not
requiring this software were done with IRAF.
Both set of images were calibrated by deriving the zeropoints of the
magnitude system from the flux of their respective standard
stars. With this type of calibration, there are two main sources of
uncertainty in the measured magnitudes. One comes from the uncertainty
on the value of the zeropoints and the other one comes from photon noise.
However, as we will see in \S 3.2, only the bright points were used
for analysis so the photon noise is negligible. In this case, the
error on the derived magnitudes only comes from the zeropoints
uncertainty which is of the order of 0.1 in B and H and 0.2 in K which
translates to an error of 0.15 in (B$-$H) and 0.22 in (H$-$K) (see the
error bars of Figure \ref{fig5}).
The optical and infrared images were then scaled together using the I
band image of the OMM and the J band image of the CFHT. Sources common
to both images were used as reference points.
\subsubsection{2-D Spectroscopy}
Observations with the 2-D spectrograph MOS-ARGUS were done during the
nights of February 27 to March 1 1998 with the B600 grism and the STIS
II CCD at the f/8 focus of the CFHT. The spectral range is from 3600
to 7000 \AA\ and the dispersion is 2.2~\AA\ pix$^{-1}$ (Table
\ref{tbl-1}). MOS-ARGUS is a fiber optic spectrograph that puts 594
fibers arranged in an hexagon shaped bundle that allows for the
collection of that many spectra per exposure. The fibers are grouped
in 25 rows that contain a number of fibers increasing from 18 to 30,
giving a maximum field of view of 12\arcsec\ $\times$ 10\arcsec. Each
fiber sampled an aperture of 0.4\arcsec\ which corresponds to a
scale of 80 pc at the assumed distance of 42 Mpc. Data reduction was
done with IRAF procedures that I specially developed for reduction of
the MOS-ARGUS data. A monochromatic image reconstruction software was
built using the software package IDL. All spectra were bias and
flatfield corrected. Special care was taken to correct for fiber
transmission and wavelength response.
\subsection{ANALYSIS AND RESULTS}
Figure \ref{fig4} shows the K image of Arp 299 labeled with the
conventional symbols of Wynn-Williams {et~al}. (1991). The data of
Mazzarella \& Boroson (1993) shows surprisingly uniform oxygen
abundances at about a third solar all over the galaxy. The
logarithmic extinction differences at H$\beta$\ between the various
regions they studied are of the order of $\Delta c =$ 0.2 $-$ 0.3
which corresponds to a $\Delta$(B$-$H) of 0.6.
Figure~\ref{fig5} compares the models with the observed color of the
0.7\arcsec\ $\times$ 0.7\arcsec\ (150 $\times$ 150 pc) aperture points
in the binned images; I chose a threshold in absolute M$_B$ and M$_K$
magnitude to have only points brighter than $-$10 and $-$14,
respectively. Plotting the observed points instead of the integrated
magnitudes ensures that most of the spatial information is extracted
from the observations since a single resolution element covers 150
pc. Ideally, one would want to be able to resolve each individual
cluster and derive the integrated colors. This would minimizes the
probability of having more than one burst within one aperture and give
the most reliable spatial information. However, such resolution are
unreachable from the ground for Arp 299.
The parameters of the model in Figure~\ref{fig5} are the same as those
shown in Figure \ref{fig1} with the exception that Figure \ref{fig5}
was made using the stellar evolutionary tracks of metallicity 0.40
solar. Also, dust emission corresponding to 0.2 mag in K was added to
the original evolutionary track in order to compare the data with
more realistic models. The track with dust emission is then shifted of
a value of 0.2 magnitude to the red along the x axis. The best global
fit of the theoretical tracks to the data gives an extinction
correction E(B$-$V) = 0.16, which is in agreement with the value
derived by Meurer {et~al}. (1995) and with the value inferred from the
data of Mazzarella \& Boroson (1993).
The ages were determined in the following way. All the points that
fall between the two old branches (see Figure \ref{fig5}) but with
(B$-$H) $<$ 1.2 are considered to be aged between 4 and 7 million
years. All the points with (B$-$H) $>$ 1.2 are given an age of 7 to
10 Myr. Notice that the points belonging to this part of the diagram
could be aged between 10 and 50 Myr (see \S 2.2) but the nebular
analysis (next paragraph) shows that this is not the case. The value
of (B$-$H) = 1.2 is set by the uncertainty on the color that is caused
by the measured extinction differences since in this case, extinction
is the major source of uncertainty. All the points that lie outside
the two tracks are labeled to be younger than 4 Myr. These points
belong to the young branch and have been shifted to this part of the
diagram by extinction, dust emission and nebular line emission. The
derived chronological map is shown on Figure \ref{fig6}. This map
shows that: {\em i}) objects C, B1 and A are the youngest of the
system; {\em ii}) a region to the south west of object A and one to
the south of object C, are approximately 3 Myr older and {\em iii})
object B2 and a region located to the south west of object A are the
oldest regions of Arp 299.
A further test of the validity and of the limitations of this approach
can be made with the 2-D spectroscopy of MOS-ARGUS. The equivalent
width of H$\alpha$ (EW(H$\alpha$)) is known to be a very strong
function of age, because it traces very young star forming regions
(Leitherer et al. 1999). However, it cannot be used give precise ages
due to geometrical considerations (Devost, Roy, \& Drissen 1997).
Nevertheless, a young region should show up as a high and peaked
region in EW(H$\alpha$). Figure \ref{fig7} shows a contour map of
EW(H$\alpha$) for NGC 3690, derived from MOS-ARGUS spectroscopic
observations superposed to the wide R band image. The contours
indicate a direct correlation between object C and the highest
EW(H$\alpha$) region located to the north of the image. This spatial
correlation between the stellar light and the nebular light is very
distinctive of young regions. The region with high EW(H$\alpha$)
located to the west of B2 is also most interesting. This region has a
high value of EW(H$\alpha$) but has a very low underlying continuum.
Careful examination of the HST UV image of Meurer et al. (1995) shows
that this region doesn't contain any UV bright sources and no sources
are detected in our B or in our K band image. This lack of continuum
light means that this region is probably very young and still buried
in dust. The continuum light is heavily extincted so this region is
not detected in the broad band images. Such a region is then missed by
the BHK diagram and EW(H$\alpha$) is needed to locate them.
Region B2, also seen in the wide R band image, is an older object
according to EW(H$\alpha$) since it is more extended and has a much
smaller value of EW(H$\alpha$). These results also confirm those obtained
previously with the BHK diagram. Nevertheless, the BHK diagram
analysis alone cannot assign a precise age to region B2 because the
evolutionary paths of Figure \ref{fig5} are degenerate for ages
between 7 to 10 Myr and 10 to 50 Myr. The EW(H$\alpha$) is necessary
to lift this degeneracy. The models of Leitherer et al. 1999 shows that
the {\em integrated} EW(H$\alpha$) has a value of $\sim$ 1 \AA\ at 14 Myr
and rapidly becomes undetectable as the stellar population becomes
older. The value of the {\em surface} EW(H$\alpha$) derived in Figure
\ref{fig7} ($\sim$ 5 \AA) around object B2 is too high for this object to
be in the 10 - 50 Myr age range so this region must be in the 7 -10
Myr age range.
\section{DISCUSSION AND CONCLUSION}
The clear distinction between the two branches seen in the BHK diagram
allows for reliable discrimination between regions younger than 4 Myr
and those older than this age, even if no data on extinction is
available. The BHK diagram clearly categorizes three (A, B1, C) of
the four infrared sources of Arp 299 as being very young ($<$ 4
Myr). Also, with the help of the EW(H$\alpha$), region B2 is
identified as being between 7 and 10 Myr. The correlation between the
highest EW(H$\alpha$) region and region C, labeled young from the BHK
diagram analysis, supports the validity and usefulness of this diagram
as a tool for determining the relative ages of young stellar
populations in starbursting regions. Also, a region located to the
west of object B2 has also a high value of EW(H$\alpha$) and is missed
by the BHK diagram. Clearly, very young regions still buried in dust
are very likely to be missed by the diagram since they are not detected
in broad band images.
The nebular analysis is found to be a useful complementary tool to the
BHK diagram. In addition to the complementary use of the
EW(H$\alpha$), data on extinction with any of the Balmer lines allows
to acheive a better time resolution. This is essential to derive more
precise star formation properties since extinction is the main
contributor to the uncertainty on the time labeling. In the case of
Arp 299, the extinction differences inferred from the data of
Mazzarella \& Boroson (1993) limits our resolution to $\sim$ 3 Myr.
Also, their longslit data only sampled a small part of the galaxy and
the extinction from the unsampled regions is unknown. The use of 2-D
spectroscopy will reduce considerably the uncertainty on
extinction. Corrections using the H$\alpha$/H$\beta$\ line ratio will
lower the time resolution down to the limit imposed by the uncertainty
of the assumed dust model for extinction which I have shown to be of
the order of 1.5 to 2 Myr.
\acknowledgments
I thank Gilles Joncas and Ren\'e Doyon who kindly obtained the infrared
images of Arp 299 at CFHT. Discussions with Jean-Ren\'e Roy, Carmelle
Robert and Claus Leitherer were most helpful. The comments of D. Calzetti
were appreciated. I would also like to thank the
anonymous referee whose suggestions and comments significantly
helped to improve the paper. This research was supported by the
Natural Sciences and Engineering Research Council of Canada, by FCAR
of the Government of Qu\'ebec and by the STScI DDRF. I am thankful to
the STScI, and in particular to Claus Leitherer, for hospitality
during my stay when part of this work was accomplished.
\clearpage
|
\section{Introduction}
An immersion of a smooth manifold $M$ into a smooth manifold $W$ is
a smooth map with everywhere injective differential. Two immersions
are regularly homotopic if they can be connected by a continuous
1-parameter family of immersions.
An immersion is generic if all its self-intersections are
transversal. In the space ${\mathcal{F}}$ of immersions $M\to W$, generic
immersions form an open dense subspace. Its complement is
{\em the discriminant hypersurface}, $\Sigma\subset{\mathcal{F}}$.
Two generic immersions belong to the same path component of ${\mathcal{F}}-\Sigma$
if they can be connected by a regular homotopy, which at
each instance is a generic immersion. We shall consider
the classification of generic immersions up to regular homotopy
through generic immersions. It is similar to the
classification of embeddings up to diffeotopy (knot theory).
In both cases, all topological properties of equivalent maps are the
same.
An {\em invariant of generic immersions} is a function on ${\mathcal{F}}-\Sigma$
which is locally constant. The value of such a function along a
path in ${\mathcal{F}}$ jumps at intersections with $\Sigma$. Invariants may be
classified according to the complexity of their jumps. The most basic
invariants in this classification are called
{\em first order invariants} (see Section ~\ref{mainres}).
In \cite{A}, Arnold studies generic regular plane curves (i.e. generic
immersions $S^1\to{\mathbb{R}}^2$). He finds three
first order invariants $J^+$, $J^-$, and
$\operatorname{St}$. In \cite{E2}, the author considers the case $S^3\to{\mathbb{R}}^5$. Two
first order
invariants $J$ and $L$ are found. In both these cases, the only
self-intersections of generic immersions are transversal double points
and in generic 1-parameter families there appear isolated
instances of self-tangencies and triple points. The values of the
invariants $J^{\pm}$ and $J$ change at instances of self-tangency and
remain constant at instances of triple points. The invariants $\operatorname{St}$
and $L$ change at triple points and remain constant at
self-tangencies.
In this paper we shall consider high-dimensional analogs of these
invariants. The most straightforward generalizations arise for
immersions $M^{2n-1}\to W^{3n-1}$, where self-tangencies and
triple points are the only degeneracies in generic 1-parameter families.
The above range of dimensions is included in a
2-parameter family, $M^{nm-1}\to W^{n(m+1)-1}$, $m,n>1$, where
generic immersions do not have $k$-fold self-intersection points if
$k>m$ and in generic 1-parameter families there appear isolated
instances of $(m+1)$-fold self-intersection. Under these circumstances,
we find first order invariants of generic immersions.
(For precise statements, see Theorem
~\ref{thmL} and Theorem ~\ref{thmJ} in Section ~\ref{mainres}, where
the main results of this paper are formulated.)
In particular, if $n$ is
even and $M^{nm-1}$ is orientable and satisfies a certain homology
condition (see Theorem ~\ref{thmL} (b)) then there exists an
integer-valued invariant $L$ of generic immersions
$M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ which is an analog of Arnold's $\operatorname{St}$: It
changes under instances of $(m+1)$-fold self-intersection, does not
change under other degeneracies which appear in generic 1-parameter
families, and is additive under connected sum. The value of $L$
at a generic immersion $f$ is the linking number of a
copy of the set of $m$-fold self-intersection points of $f$, shifted in
a special way, with $f(M)$ in ${\mathbb{R}}^{n(m+1)-1}$. (See Definition
~\ref{dfnL}.)
For immersions of odd-dimensional spheres in codimension two there
appear restrictions on the possible values of $L$ (see Theorem
~\ref{thmdivL}). This phenomenon is especially interesting in the case
of immersions $S^{4j-1}\to{\mathbb{R}}^{4j+1}$. Using a result of Hughes and
Melvin \cite{HM} we prove that the range of $L$ is
related to Bernoulli numbers: For any
generic immersion $f\colon S^{4j-1}\to{\mathbb{R}}^{4j+1}$, $L(f)$ is divisible
by $p^r$, where $p$ is a prime and $r$ is an integer such that
$p^{r+k}$ divides $2j+1$ and $p^{k+1}$ does not divide $\mu_j$ for
some integer $k$. Here $\mu_j$ is the denominator of $\frac{B_j}{4j}$,
where $B_j$ is the $j^{\rm th}$ Bernoulli number.
In general, it is not known if the above result gives all restrictions
on the range of $L$. However, in special cases it does. For example,
for immersions $S^3\to{\mathbb{R}}^5$ the range of $L$ is ${\mathbb{Z}}$ (see
~\cite{E2} and also Remark ~\ref{lkrmk}) and for immersions
$S^{67}\to{\mathbb{R}}^{69}$ it is $35{\mathbb{Z}}$ (see Section ~\ref{rmksonl}).
In ~\cite{E1}, the author gives a complete classification of generic
immersions
$M^{k}\to{\mathbb{R}}^{2k-r}$, $r=0,1,2$, $k\ge 2r+4$ up to regular homotopy
through generic immersions (under some conditions on the lower
homotopy groups of $M$).
Here, the class of a generic immersion is determined by its
self-intersection with induced natural additional
structures (e.g. spin structures). The existence of invariants such as
$L$ mentioned above implies that the corresponding classification in
other dimensions is more involved (see Remark ~\ref{nonloc}).
The invariants of generic immersions in Theorems
~\ref{thmL} and ~\ref{thmJ} give rise to
invariants of regular homotopy which take values in finite cyclic
groups (Section ~\ref{rhinv}). We construct examples showing that,
depending on the source and target manifolds, the regular homotopy
invariants may or may not be trivial (Section ~\ref{expbm}). It would
be interesting to relate these invariants to invariants arising from
the cobordism theory of immersions (see for example Eccles ~\cite{Ec}).
\section{Statements of the main results}\label{mainres}
Before stating the main results we define the notions of invariants of
orders zero and one. They are similar to knot invariants of finite type
introduced by Vassiliev in ~\cite{V}.
As in the Introduction, let ${\mathcal{F}}$ denote the space of immersions and
let $\Sigma\subset{\mathcal{F}}$ denote the discriminant hypersurface.
The set $\Sigma^1\subset\Sigma$ of non-generic immersions which appear at isolated
instances in generic 1-parameter families (see Lemmas
~\ref{vdef1sm} and ~\ref{vdef1big} for descriptions of such immersions
$M^{nm-1}\to W^{n(m+1)-1}$) is a
smooth submanifold of ${\mathcal{F}}$ of codimension one.
(${\mathcal{F}}$ is an open subspace of the space of smooth maps, thus an
infinite dimensional manifold and the notion of codimension in ${\mathcal{F}}$
makes sense.)
If $f_0\in\Sigma^1$ then there is a neighborhood $U(f_0)$ of $f_0$ in
${\mathcal{F}}$ cut in two parts by
$\Sigma^1$. A coherent choice of a positive and negative part of $U(f_0)$
for each $f_0\in\Sigma^1$ is a {\em coorientation} of $\Sigma$. (The
discriminant hypersurface in the space of immersions
$M^{nm-1}\to W^{n(m+1)-1}$ will be cooriented in Section ~\ref{codiscr}.)
Let $a$ be an invariant of generic immersions and let $f_0\in\Sigma^1$.
Define the jump $\nabla a$ of $a$ as
$$
\nabla a(f_0)=a(f_+)-a(f_-),
$$
where $f_+$ and $f_-$ are generic immersions in the positive
respectively negative part of $U(f_0)$. Then $\nabla a$ is a locally
constant function on $\Sigma^1$.
An invariant $a$ of generic immersions is a {\em zero order
invariant} if $\nabla a\equiv0$.
Let $\Sigma^2$ be the set of all immersions which appear in generic
2-parameter families but can be avoided in generic 1-parameter
families (see Lemma ~\ref{vdef2} for the case
$M^{nm-1}\to W^{n(m+1)-1}$). Then $\Sigma^2\subset\Sigma$ is a smooth
codimension two submanifold of ${\mathcal{F}}$.
An invariant $a$ of generic immersions is a {\em first order
invariant} if $\nabla a(f_0)=\nabla a(f_1)$ for any immersions
$f_0,f_1\in\Sigma^1$ which can be joined by a path in
$\Sigma^1\cup\Sigma^2$ such that at intersections with $\Sigma^2$ its tangent
vector is transversal to the tangent space of $\Sigma^2$.
\begin{thm}\label{thmL}
Let $m>1$ and $n>1$ be integers and let $M^{nm-1}$ be a closed
manifold.
\begin{itemize}
\item[{\rm (a)}] If $m$ is even and
$H_{n-1}(M;{\mathbb{Z}}_2)=0=H_{n}(M;{\mathbb{Z}}_2)$ then there exists a unique (up to
addition of zero order invariants) first order ${\mathbb{Z}}_2$-valued
invariant $\Lambda$ of generic
immersions $M^{nm-1}\to {\mathbb{R}}^{n(m+1)-1}$ satisfying the following
conditions: It jumps by $1$ on the part of $\Sigma^1$ which consists
of immersions with one $(m+1)$-fold self-intersection point and does
not jump on other parts of $\Sigma^1$.
\item[{\rm (b)}] If $n$ is even, $M$ is orientable, and
$H_{n-1}(M;{\mathbb{Z}})=0=H_{n}(M;{\mathbb{Z}})$ then there exists a unique (up to
addition of zero order invariants) first order integer-valued
invariant $L$ of generic
immersions $M^{nm-1}\to {\mathbb{R}}^{n(m+1)-1}$ satisfying the following
conditions: It jumps by $m+1$ on the part of $\Sigma^1$ which consists
of immersions with one $(m+1)$-fold self-intersection point and does
not jump on other parts of $\Sigma^1$.
\end{itemize}
\end{thm}
Theorem ~\ref{thmL} is proved in Section ~\ref{pfthmL}. The invariants
$\Lambda$ and $L$ are defined in Definition ~\ref{dfnLambda} and
Definition ~\ref{dfnL}, respectively. If appropriately normalized,
$\Lambda$ and $L$ are additive under connected summation of generic
immersions (see Section ~\ref{csgi}).
\begin{thm}\label{thmdivL}
Let $f\colon S^{2m-1}\to{\mathbb{R}}^{2m+1}$ be a generic immersion.
\begin{itemize}
\item[{\rm (a)}] If $m=4j+1$ then $2j+1$ divides $L(f)$.
\item[{\rm (b)}] If $m=4j+3$ then $4j+4$ divides $L(f)$.
\item[{\rm (c)}] If $m=2j$ then $p^r$ divides $L(f)$ for every prime
$p$ and integer $r$ such that $p^{r+k}$ divides
$2j+1$ and $p^{k+1}$ does not divide $\mu_j$ for
some integer $k$. Here $\mu_j$ is the denominator of
$\frac{B_j}{4j}$, where $B_j$ is the $j^{\rm th}$
Bernoulli number.
\end{itemize}
\end{thm}
We prove Theorem ~\ref{thmdivL} in Section ~\ref{pfthmdivL}.
\begin{thm}\label{thmJ}
Let $m>1$ and $n>1$ be integers and let $M^{nm-1}$ be a closed
manifold and let $W^{n(m+1)-1}$ be a manifold.
\begin{itemize}
\item[{\rm (a)}] If $n$ is odd then, for each integer $2\le r\le m$
such that $m-r$ is even, there exist a unique (up to addition of zero
order invariants)
first order integer-valued invariant $J_r$, of generic immersions
$M^{nm-1}\to W^{n(m+1)-1}$ satisfying the following conditions:
The invariant jumps by $2$ on the part of $\Sigma^1$ which consists
of immersions with one degenerate $r$-fold self-intersection point and
does not jump on other parts of $\Sigma^1$.
\item[{\rm (b)}] If $n=2$ then there exists a unique (up to
addition of zero order invariants) first order integer-valued
invariant $J$, of
generic immersions $M^{2m-1}\to W^{2m+1}$ satisfying the following
conditions:
The invariant jumps by $1$ on the part of $\Sigma^1$ which consists
of immersions with one degenerate $m$-fold self-intersection point and
does not jump on other parts of $\Sigma^1$.
\end{itemize}
\end{thm}
Theorem ~\ref{thmJ} is proved in Section ~\ref{pfthmJ}.
If appropriately normalized, $J_r$ and $J$ are additive under
connected summation of generic immersions (see Section ~\ref{csgi}).
The value of $J_r$
on a generic immersion is the Euler characteristic of its resolved
$r$-fold self-intersection manifold, the value of $J$ is the number of
components in its $m$-fold self-intersection (which is a closed
1-dimensional manifold).
\section{Generic immersions and generic regular deformations}
In this section we define generic immersions and describe the
immersions corresponding to non-generic instances in generic 1- and
2-parameter families. We also describe the versal deformations of
these non-generic immersions.
\subsection{Generic immersions}
Let $M$ and $W$ be smooth manifolds and let
$f\colon M\to W$ be an immersion. A point $q\in W$ is a {\em $k$-fold
self-intersection point of $f$} if $f^{-1}(q)$ consists of exactly $k$
points. Let
$\Gamma_k(f)\subset W$ denote the set of $k$-fold self-intersection
points of $f$ and let $\widetilde {\Gamma}_k(f)=f^{-1}(\Gamma_k(f))\subset M$
denote its preimage. Note that $f|{\widetilde{\Gamma}_k(f)}\to\Gamma_k(f)$
is a $k$-fold covering.
\begin{dfn}\label{dfngi}
Let $m>1$ and $n>1$. An immersion
$f\colon M^{nm-1}\to W^{n(m+1)-1}$ is {\em generic} if
\begin{itemize}
\item[{\bf g1}] $\Gamma_k(f)$ is empty for $k>m$, and
\item[{\bf g2}] if $q=f(p_1)=\dots=f(p_r)\in\Gamma_r(f)$ $(r\le m)$
then for any $i$, $1\le i\le r$
$$
df\, T_{p_i}M+\cap_{s\ne i}df\, T_{p_s}M=T_q W.
$$
\end{itemize}
\end{dfn}
A standard application of the jet-transversality theorem shows that
the set of generic immersions is open and dense in the space of all
immersions ${\mathcal{F}}$.
We remark that the set of $k$-fold self-intersection points
$\Gamma_k(f)$ of a generic immersion
$f\colon M^{nm-1}\to W^{n(m+1)-1}$ is a smooth submanifold of
$W$ of dimension $n(m-k+1)-1$. Moreover, the deepest
self-intersection $\Gamma_m(f)$ is a closed manifold.
Lemma ~\ref{giloc} below gives a
local coordinate description of a generic immersion close to a
$k$-fold self-intersection point. To state it we
introduce some notation: Let $x_i\in{\mathbb{R}}^{nm-1}$, we
write $x_i=(x_i^0,x_i^1,\dots,x_i^{k-1})$, where
$x_i^0\in{\mathbb{R}}^{n(m-k+1)-1}$ and $x_i^r\in{\mathbb{R}}^{n}$, for $1\le r\le k-1$.
Similarly, we write $y\in{\mathbb{R}}^{n(m+1)-1}$ as $y=(y^0,y^1,\dots,y^k)$, where
$y^0\in{\mathbb{R}}^{n(m-k+1)-1}$ and $y^r\in{\mathbb{R}}^{n}$, for $1\le r\le k$.
\begin{lma}\label{giloc}
Let $f\colon M^{nm-1}\to W^{n(m+1)-1}$ be a generic immersion and let
$q=f(p_1)=\dots=f(p_k)$ be a $k$-fold self intersection point of
$f$. Then
there are coordinates $y$ in $V\subset W^{n(m+1)-1}$ centered at $q$
and coordinates $x_i$ on $U_i\subset M^{nm-1}$ centered at $p_i$,
$1\le i\le k$ such that $f$ is given by
\begin{align*}
f(x_1)&=(x_1^0,x_1^1,\dots,x_1^{k-2},x_1^{k-1},0),\\
f(x_2)&=(x_2^0,x_2^1,\dots,x_2^{k-2},0,x_1^{k-1}),\\
{}&\vdots{}\\
f(x_k)&=(x_k^0,0,x_k^1,x_k^2,\dots,x_k^{k-1}).
\end{align*}
(That is, if $y=f(x_i)$ then $y^0(x_i)=x_i^0$, $y^{r}(x_i)=x_i^r$ for
$1\le r\le k-i$, $y^{k-i+1}(x_i)=0$, and $y^{r}(x_i)=x_i^{r-1}$ for
$k-i+2\le r\le k$.)
\end{lma}
\begin{pf}
The proof is straightforward.
\end{pf}
When the source and target of a generic immersion are oriented and the
codimension is even then there are induced orientations on the
self intersection manifolds:
\begin{prp}
Let $n=2j>1$, $m>1$, $2\le k\le m$, and let
$f\colon M^{2jm-1}\to W^{2j(m+1)-1}$ be a generic immersion.
Orientations on $M^{2jm-1}$ and $W^{2j(m+1)-1}$ induce an orientation on
$\Gamma_k(f)$.
\end{prp}
\begin{pf}
Let $N$ denote the normal bundle of the immersion. The
decomposition
$$
f^*TW^{2j(m+1)-1}=TM^{2jm-1}\oplus N
$$
induces an orientation on $N$.
If $q\in\Gamma_k(f)$, $q=f(p_1)=\dots=f(p_k)$ then
$$
T_qW^{2j(m+1)-1}=T_q\Gamma_k(f)\oplus_{i=1}^k N_{p_i}.
$$
The orientation on $N$ induces an orientation on $\oplus_{i=1}^k
N_{p_i}$. Since the dimension of the bundle $N$ is $2j$ which is even,
the orientation on the sum is independent on the ordering of the
summands. Hence, the decomposition above induces a well-defined
orientation on $T\Gamma_k(f)$.
\end{pf}
\subsection{The codimension one part of the discriminant hypersurface}
Our next result describes (the 1-jets of) immersions in $\Sigma^1$ (see
Section ~\ref{mainres}).
\begin{lma}\label{1jetSi1}
If $f_0\colon M^{nm-1}\to W^{n(m+1)-1}$ is an immersion in $\Sigma^1$ then
{\bf g1} and {\bf g2} of Definition \ref{dfngi} holds, except at one
$k$-fold
$(2\le k\le m+1)$ self-intersection point $q=f_0(p_1)=\dots=f_0(p_k)$,
where,
\begin{itemize}
\item[{\rm (a)}] if $k=2$,
$$
\dim(df_0\, T_{p_1}M+df_0\,
T_{p_2}M)=n(m+1)-2,
$$
or
\item[{\rm (b)}] if $2<k\le m+1$, for $i\ne l$
$$
\dim(df_0\, T_{p_i}M +\cap_{r\ne i,r\ne l}df_0\,
T_{p_r}M)=n(m+1)-1
$$
and
$$
\dim(df_0\, T_{p_i}M+\cap_{r\ne i}df_0\,
T_{p_r}M)=n(m+1)-2.
$$
\end{itemize}
\end{lma}
\begin{pf}
We have to show that degeneracies as above appears at
isolated parameter values in generic 1-parameter families, and that
further degeneracies (of the 1-jet) may be avoided . This follows
from the jet-transversality theorem applied to maps
$M^{nm-1}\times [-\delta,\delta]\to W^{n(m+1)-1}$.
\end{pf}
A $k$-fold self-intersection point $q$ of an immersion
$f_0\colon M^{nm-1}\to W^{n(m+1)-1}$ where {\bf g1} and {\bf g2} of
Definition ~\ref{dfngi} does not hold will be called a
{\em degenerate self-intersection point of $f_0$}. If $2\le k\le m$ we
say that $q$ is a
{\em $k$-fold self-tangency point of $f_0$} if $k=m+1$ we say that $q$
is a {\em $(m+1)$-fold self-intersection point of $f_0$}.
Recall that a deformation $F$ of a map $f$ is called {\em versal} if
any deformation of $f$ is equivalent (up to left-right action of
diffeomorphisms) to one induced from $F$.
Let $f_0$ be an immersion in $\Sigma^1$. Then its versal deformation
$f_t$ is a 1-parameter
deformation. In other words, it is a path $\lambda(t)=f_t$ in ${\mathcal{F}}$
which intersects $\Sigma^1$ transversally at $f_0$ and thus,
$f_t$ are generic immersions for small $t\ne 0$.
Next, we shall describe immersions $f_0\in\Sigma^1$ in local coordinates
close to their degenerate self intersection point.
Self-tangency points and $(m+1)$-fold self-intersection points are
treated in Lemma ~\ref{vdef1sm} and Lemma ~\ref{vdef1big},
respectively.
To accomplish this we need a more detailed description of coordinates
than that given in Lemma ~\ref{giloc}, we use coordinates as there
with one more ingredient:
We write (when necessary) $x_i^r\in{\mathbb{R}}^n$ and $y_i^r\in{\mathbb{R}}^n$, $r> 1$
as $x_i^r=(\xi_i^r,u_i^r)$ and
$y^r=(\eta^r,v^r)$, where $\xi_i^r,\eta^r\in{\mathbb{R}}$ and
$u_i^r,v_i^r\in{\mathbb{R}}^{n-1}$. (Greek letters for scalars and Roman for
vectors).
\begin{lma} \label{vdef1sm}
Let $f_0\colon M^{nm-1}\to W^{n(m+1)-1}$ be an immersion in $\Sigma^1$
and let $q=f_0(p_1)=\dots=f_0(p_k)$ be a point of degenerate $k$-fold
self intersection $2\le k \le m$. Then there are coordinates $y$ on
$V\subset W$
centered at $q$ and coordinates $x_i$ on $U_i\subset M$ centered at
$p_i$, $1\le i\le k$ such that in these coordinates the versal
deformation $f_t$, $-\delta<t<\delta$ of $f_0$ is constant
outside of $\cup_i U_i$ and in neighborhoods of $p_i\in U_i$
it is given by
\begin{itemize}
\item[{\rm (a)}]
\begin{align*}
f_t(x_1)&=(x_1^0,x_1^1,\dots,x_1^{k-2},x_1^{k-1},0),\\
f_t(x_2)&=(x_2^0,x_2^1,\dots,x_2^{k-2},0,x_1^{k-1}),\\
{}&\vdots{}\\
f_t(x_{k-1})&=(x_{k-1}^0,x_{k-1}^1,0,x_{k-1}^2,\dots,x_{k-1}^{k-1}),
\end{align*}
(That is, if $y=f_t(x_i)$ then {\em for $1\le i\le k-1$},
$y^0(x_i)=x_i^0$, $y^1(x_i)=x_i^1$, $y^{r}(x_i)=x_i^r$ for
$1\le r\le k-i$, $y^{k-i+1}(x_i)=0$, and $y^{r}(x_i)=x_i^{r-1}$ for
$k-i+2\le r\le k$.)
\item[{\rm (b)}]
\begin{align*}
f_t(x_k)=&
\left(\begin{matrix}
x_k^0, & (\xi_k^1,0), & (\xi_k^2,u_k^1), & \dots,
& (\xi_k^{k-1},u_k^{k-2}), & (-\xi_k^2-\dots-\xi_k^{k-1},u_k^{k-1})
\end{matrix}\right)\\
+&
\left(\begin{matrix}
0, & (0,0), & (0,0), &\dots, & (0,0), & (Q(x_k^0,\xi_k^1)+t,0)
\end{matrix}\right)
\end{align*}
where $Q$ is a nondegenerate quadratic form in the $n(m-k+1)$
variables $(x_k^0,\xi_k^1)$.
\end{itemize}
\end{lma}
\begin{pf}
It is straightforward to see that
we can find coordinates $x_i$ on $U_i$, $i=1,\dots,k$, so that up to
first order of approximation $f_0|U_i$ is given by the expressions in
(a) and the first term in (b) above.
We must consider also second order terms: Let $N$ be a linear subspace
in the coordinates $y$ transversal to $Tf_0(U_1)\cap\dots\cap
Tf_0(U_{k-1})$. Let $\phi\colon f(U_k)\to N$ be orthogonal projection
onto $N$. Then
$\operatorname{ker}(d\phi)=Tf_0(U_1)\cap\dots\cap Tf_0(U_{k})$
and $\operatorname{coker}(d\phi)$ is
1-dimensional. The second derivative
$d^2\phi\colon\operatorname{ker}(\phi)\to\operatorname{coker}(\phi)$
must be a non-degenerate quadratic form, otherwise, we can avoid $f_0$ in
generic 1-parameter families. (This is a consequence of the jet
transversality theorem.) The Morse lemma then implies that, after
possibly adjusting the coordinates in $U_k$ by adding quadratic
expressions in $(x_k^0,\xi_k^1)$ to $\xi_k^j$, $j>1$, there exists
coordinates for $f_0$ as stated.
Finally, we must prove that the deformation $f_t$ as given above is
versal. A standard result in singularity theory says that it is enough
to prove that the deformation is infinitesimally versal. This is
straightforward.
\end{pf}
In Lemma ~\ref{vdef1big} below we will use coordinates as at generic
$m$-fold self intersection points. That is,
$x=(x^0,x^1,\dots,x^{m-1})\in{\mathbb{R}}^{nm-1}$ and
$y=(y^0,y^1,\dots, y^m)\in{\mathbb{R}}^{n(m+1)-1}$, where $x^0,y^0\in{\mathbb{R}}^{n-1}$
and $x_k,y_k\in{\mathbb{R}}^n$ for $k\ne 0$.
\begin{lma}\label{vdef1big}
Let $f_0\colon M^{nm-1}\to W^{n(m+1)-1}$ be an immersion in $\Sigma^1$
and let $q=f_0(p_1)=\dots=f_0(p_{m+1})$ be a point of $(m+1)$-fold
self-intersection. Then there are coordinates $y$ on $V\subset W$
centered at $q$ and coordinates $x_i$ on $U_i\subset M$ centered at
$p_i$, $1\le i\le k$ such that in these coordinates the versal
deformation $f_t$, $-\delta<t<\delta$ of $f_0$ is constant
outside of $\cup_i U_i$ and in neighborhoods of $p_i\in U_i$
it is given by
\begin{itemize}
\item[{\rm (a)}]
\begin{align*}
f_t(x_1)&=(x_1^0,x_1^1,\dots,x_1^{m-2},x_1^{m-1},0),\\
f_t(x_2)&=(x_2^0,x_2^1,\dots,x_2^{m-2},0,x_1^{m-1}),\\
{}&\vdots{}\\
f_t(x_{m})&=(x_{m}^0,0,x_{m}^1,x_{m}^2,\dots,x_{m}^{m-1}),
\end{align*}
(That is, if $y=f_t(x_i)$ then {\em for $1\le i\le m$},
$y^0(x_i)=x_i^0$, $y^{r}(x_i)=x_i^r$ for
$1\le r\le m-i$, $y^{m-i+1}(x_i)=0$, and $y^{r}(x_i)=x_i^{r-1}$ for
$m-i+2\le r\le m$.)
\item[{\rm (b)}]
\begin{align*}
&f_t(x_k)=\\
&\left(\begin{matrix}
0, & (\xi_k^1,u^0_k), & (\xi_k^2,u_k^1), & \dots,
& (\xi_k^{m-1},u_k^{m-2}), & (-\xi_k^1-\dots-\xi_k^{m-1}+t,u_k^{m-1})
\end{matrix}\right)
\end{align*}
\end{itemize}
\end{lma}
\begin{pf}
The proof is similar to the proof of Lemma ~\ref{vdef1sm}, but easier.
\end{pf}
\subsection{The codimension two part of the discriminant hypersurface}
\begin{lma}\label{vdef2}
Let $f_{0,0}\colon M^{nm-1}\to W^{n(m+1)-1}$ be an immersion in
$\Sigma^2$. Then either {\rm (a)} or {\rm (b)} below holds.
\begin{itemize}
\item[{\rm (a)}] $f_{0,0}$ has two distinct degenerate
self-intersection points $q_1$ and $q_2$. Locally around $q_i$,
$i=1,2$
$f_{0,0}$ is as in Lemma ~\ref{vdef1sm} or Lemma ~\ref{vdef1big}.
The versal deformation of $f_{0,0}$ is a product of the corresponding
1-parameter versal deformations.
\item[{\rm (b)}] $f_{0,0}$ has one degenerate $k$-fold $2\le k\le m$
self-intersection point $q$. There are coordinates $y$ centered at $q$
and coordinates $x_i$ centered at $p_i$, $1\le i\le k$ such that in
these coordinates the versal deformation $f_{s,t}$, $-\delta<t,s<\delta$ of
$f_{0,0}$ is constant outside of $\cup_i U_i$ and in a neighborhood of
$p_i\in U_i$ it is given by
\begin{itemize}
\item[{\rm (b1)}]
\begin{align*}
f_{s,t}(x_1)&=(x_1^0,x_1^1,\dots,x_1^{k-2},x_1^{k-1},0),\\
f_{s,t}(x_2)&=(x_2^0,x_2^1,\dots,x_2^{k-2},0,x_1^{k-1}),\\
{}&\vdots{}\\
f_{s,t}(x_{k-1})&=(x_{k-1}^0,x_{k-1}^1,0,x_{k-1}^2,\dots,x_{k-1}^{k-1}),
\end{align*}
That is, if $y=f_{s,t}(x_i)$ then {\em for $1\le i\le k-1$},
$y^0(x_i)=x_i^0$, $y^1(x_i)=x_i^1$, $y^{r}(x_i)=x_i^r$ for
$1\le r\le k-i$, $y^{k-i+1}(x_i)=0$, and $y^{r}(x_i)=x_i^{r-1}$ for
$k-i+2\le r\le k$.
\item[{\rm (b2)}]
\begin{align*}
&f_{s,t}(x_k)=\\
&\left(\begin{matrix}
x_k^0, & (\xi_k^1,0), & (\xi_k^2,u_k^1), & \dots,
& (\xi_k^{k-1},u_k^{k-2}), & (-\xi_k^2-\dots-\xi_k^{k-1},u_k^{k-1})
\end{matrix}\right)\\
+&
\left(\begin{matrix}
0, & (0,0), & (0,0), &\dots, & (0,0), & (Q(x_k^0)+\xi_k^1((\xi_k^1)^2+s)+t,0)
\end{matrix}\right)
\end{align*}
\end{itemize}
where $Q$ is a nondegenerate quadratic form in the $n(m-k+1)-1$
variables $x_k^0$.
\end{itemize}
\end{lma}
\begin{pf}
The jet-transversality theorem applied to maps of
$M^{2jm-1}\times [-\delta,\delta]^2$ into $W^{2j(m+1)-1}$ shows that
immersions with points of $k$-fold self-intersection $k\ge m+2$, as
well as immersions with points of $k$-fold self-intersection points
$2\le k\le m+1$ at which the 1-jet has further
degenerations than the 1-jets of Lemma ~\ref{1jetSi1} can be avoided in
generic 2-parameter families.
Assume that $f_{0,0}$ has a degenerate $k$-fold self-intersection
point. If $k=m+1$ it is easy to see that $f_{0,0}$ has the same local
form as the map in Lemma ~\ref{vdef1big}. So, immersions with $(m+1)$-fold
self-intersection points appears along 1-parameter subfamilies in
generic 2-parameter families. If $k<m+1$ we proceed as in the proof of
Lemma ~\ref{vdef1sm} and
construct the map $\phi$. Applying the jet transversality theorem once
again we see that if the rank of $d^2\phi$ is smaller than
$n(m-k+1)-1$ then $f_{0,0}$ can be avoided in generic 2-parameter
families. If the rank of $d^2\phi$ equals $n(m-k+1)-1$ then the third
derivative of $\phi$ in the direction of the null-space of $d^2\phi$
must be non-zero otherwise $f_{0,0}$ can be avoided in generic
2-parameter families.
With this information at hand we can find coordinates as
claimed. Finally, as in the proof of Lemma ~\ref{vdef1sm}, we must
check that the deformation given is infinitesimally versal. This is
straightforward.
\end{pf}
\begin{figure}[htbp]
\begin{center}
\includegraphics[angle=0, width=6cm]{cod2.eps}
\end{center}
\caption{The discriminant intersected with a small generic
2-disk.}\label{figcod2}
\end{figure}
Pictures (a) and (b) in Figure \ref{figcod2} correspond to cases (a) and
(b) in Theorem \ref{vdef2}. The codimension two parts of the
discriminant are represented by points.
In (a) two branches of the discriminant, which consist of immersions
with one degenerate $r$-fold and one degenerate $k$-fold
self-intersection point respectively, intersect in $\Sigma^2$. In (b)
the smooth points of
the semi-cubical cusp represents immersions with one degenerate
$k$-fold self-intersection point. The singular point represents
$\Sigma^2$. If an immersion in $\Sigma$ above the singular
point ($\Sigma^2$) is moved below it then the index of the quadratic form
$Q$ in local coordinates close to the degenerate self-intersection
point (see Lemma \ref{vdef1sm}) changes.
\section{Definition of the invariants}
In this section we define the invariants $J_r$, $J$, $L$ and
$\Lambda$. To this end, we describe resolutions of
self-intersections (for the definitions of $J_r$ and $J$) and
we compute homology of image-complements of and
of normal bundles (for the definitions of $L$ and $\Lambda$).
\subsection{Resolution of the self intersection}
For generic immersions $f\colon M^{nm-1}\to W^{n(m+1)-1}$ let
$
\Gamma^j(f)=\Gamma_j(f)\cup\Gamma_{j+1}(f)\cup\dots\cup\Gamma_m(f)
$, for $2\le j\le m$ and let $\widetilde{\Gamma}^j(f)=f^{-1}(\Gamma^j(f))$.
Resolving $\Gamma^j(f)$ we obtain a smooth manifold $\Delta^j(f)$:
\begin{lma}\label{resol}
Let $m>1$ and $n>1$ and $2\le j\le m$ be integers. Let
$f\colon M^{nm-1}\to W^{n(m+1)-1}$ be a generic immersion. Then there exists
closed $(n(m-j+1)-1)$-manifolds $\widetilde{\Delta}^j(f)$ and $\Delta^j(f)$,
unique up to diffeomorphisms, and immersions
$\sigma\colon\widetilde{\Delta}^j(f)\to M$ and $\tau\colon\Delta^j(f)\to W$
such that the diagram
$$
\begin{CD}
\widetilde{\Delta}^j(f) & @>{\sigma}>> & f^{-1}(\Gamma^j(f))\subset M\\
@VpVV & & @VVfV\\
\Delta^j(f) & @>{\tau}>> & \Gamma^j(f)\subset W
\end{CD},
$$
commutes. The maps $\sigma$ and $\tau$ are surjective, have multiple
points only along $\widetilde{\Gamma}^{j+1}(f)$ and $\Gamma^{j+1}(f)$
respectively, and $p$ is a $j$-fold cover.
\end{lma}
\begin{pf}
This is immediate from Lemma ~\ref{giloc}: Close to a $k$-fold self
intersection point $\Gamma_k(f)$ looks like the intersection of $k$
$(nm-1)$-planes in general position in ${\mathbb{R}}^{n(m+1)-1}$.
\end{pf}
\subsection{Definition of the invariants $J_r$ and $J$}
\begin{dfn}\label{dfnJ_r}
Let $m>1$ and $n>1$ be integers and assume that $n$ is odd. For a
generic immersion $f\colon M^{nm-1}\to W^{n(m+1)-1}$ and an integer
$2\le r\le m$ such that $m-r$ is even, define
$$
J_r(f)=\chi(\Delta^r(f)),
$$
where $\chi$ denotes Euler characteristic.
\end{dfn}
\begin{dfn}\label{dfnJ}
Let $m>1$ be an integer. For a
generic immersion $f\colon M^{2m-1}\to W^{2m+1}$, define
$$
J_r(f)=\text{The number of components of $\Gamma_m(f)$.}
$$
\end{dfn}
\begin{lma}\label{wdJ}
The functions $J_r$ and $J$ are invariants of generic immersions.
\end{lma}
\begin{pf}
Let $f_t$, $0\le t\le 1$ be a regular homotopy through generic
immersions. If $F\colon M\times [0,1]\to W\times[0,1]$ is the
immersion $F(x,t)=(f_t(x),t)$ then
$\Gamma^j(F)\cong\Gamma^j(f_0)\times I$. It follows that
$\Delta^j(f_0)$ is diffeomorphic to $\Delta^j(f_1)$.
\end{pf}
\subsection{Homology of complements of images}\label{imcom}
\begin{lma}\label{Hcom}
Let $f\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be a generic
immersion. Assume that $M$ is closed. Then
\begin{itemize}
\item[{\rm (a)}]
$$
H_{n-1}({\mathbb{R}}^{n(m+1)-1}-f(M);{\mathbb{Z}}_2)\cong H^{nm-1}(M;{\mathbb{Z}}_2)\cong{\mathbb{Z}}_2,
$$
and
\item[{\rm (a)}] if $M$ is oriented then
$$
H_{n-1}({\mathbb{R}}^{n(m+1)-1}-f(M);{\mathbb{Z}})\cong H^{nm-1}(M;{\mathbb{Z}})\cong{\mathbb{Z}}.
$$
\end{itemize}
\end{lma}
\begin{pf}
Alexander duality implies that
$H_{n-1}({\mathbb{R}}^{n(m+1)-1}-f(M))\cong H^{nm-1}(f(M))$. By
Lemma \ref{giloc}, we can choose
triangulations of $M$ and $f(M)$ such that the map $f\colon M\to f(M)$
is a bijection on the $(nm-k)$-skeleta for $1\le k\le n$.
\end{pf}
\begin{rmk}\label{idHcom}
In case (a) of Lemma \ref{Hcom} we will use the unique
${\mathbb{Z}}_2$-orientation of
$M$ and the duality isomorphism to identify
$H_{n-1}({\mathbb{R}}^{n(m+1)-1}-f(M);{\mathbb{Z}}_2)$ with ${\mathbb{Z}}_2$. In case (b) of
Lemma \ref{Hcom}, a ${\mathbb{Z}}$-orientation of $M$
determines an isomorphism $H^{nm-1}(M;{\mathbb{Z}})\to{\mathbb{Z}}$.
We shall use this and the duality isomorphism to
identify $H_{n-1}({\mathbb{R}}^{n(m+1)-1}-f(M);{\mathbb{Z}})$ with ${\mathbb{Z}}$.
A small ($n-1$)-dimensional sphere going around a fiber in the normal
bundle of $f(M)-\Gamma^2(f)$ generates these groups.
\end{rmk}
\subsection{Homology of normal bundles of immersions}
Consider an immersion $f\colon M^{nm-1}\to W^{n(m+1)-1}$.
Let $N$ denote its normal bundle. Then $N$ is a
vector bundle of dimension $n$ over $M$. Choose a Riemannian metric
on $N$ and consider the associated bundle $\partial N$ of unit vectors
in $N$. This is an ($n-1$)-sphere bundle over $M$.
Let $\partial F\cong S^{n-1}$ denote a fiber of $\partial N$.
\begin{lma}\label{Hfib}
Let $f\colon M^{nm-1}\to W^{n(m+1)-1}$ be an immersion. Let
$i\colon \partial F\to\partial N$ denote the inclusion of the
fiber.
\begin{itemize}
\item[{\rm (a)}] If $H_{n-1}(M;{\mathbb{Z}}_2)=0=H_n(M;{\mathbb{Z}}_2)$ then
$$
i_\ast\colon H_{n-1}(\partial F;{\mathbb{Z}}_2)\to H_{n-1}(\partial N;{\mathbb{Z}}_2),
$$
is an isomorphism.
\item[{\rm (b)}] If $M$ and $W$ are oriented and
$H_{n-1}(M;{\mathbb{Z}})=0=H_n(M;{\mathbb{Z}})$ then
$$
i_\ast\colon H_{n-1}(\partial F;{\mathbb{Z}})\to H_{n-1}(\partial N;{\mathbb{Z}}),
$$
is an isomorphism.
\end{itemize}
\end{lma}
\begin{pf}
This follows from the Leray-Serre spectral sequence.
\end{pf}
\begin{rmk}\label{idHfib}
In case (a) of Lemma \ref{Hfib} we use the isomorphism $i_\ast$ and a
canonical
generator of $H_{n-1}(\partial F;{\mathbb{Z}}_2)$ to identify
$H_{n-1}(\partial N;{\mathbb{Z}}_2)$ with ${\mathbb{Z}}_2$.
In case (b) of Lemma \ref{Hfib}, orientations of $M$ and $W$ induce an
orientation of the fiber sphere $\partial F$. We use this orientation
and $i_\ast$ in Lemma \ref{Hfib} to identify $H_{n-1}(\partial
N;{\mathbb{Z}})$ with ${\mathbb{Z}}$.
\end{rmk}
\subsection{Shifting the $m$-fold self-intersection}
Let $f\colon M^{nm-1}\to W^{n(m+1)-1}$ be a generic immersion.
Recall that the set of $m$-fold self-intersection points of $f$ is a
closed submanifold $\Gamma_m(f)$ of $W^{n(m+1)-1}$ of dimension
$n-1$ and its preimage $\widetilde{\Gamma}_m(f)$ is a closed submanifold of the
same dimension in $M^{nm-1}$.
\begin{lma}\label{sctn}
If $f\colon M^{nm-1}\to W^{n(m+1)-1}$ is an immersion then
there exists a smooth section $s\colon\widetilde{\Gamma}_m(f)\to\partial N$,
where $\partial N$ is the unit sphere bundle of the normal bundle of
$f$.
\end{lma}
\begin{pf}
This is immediate: The base space of the $n$-dimensional
vector bundle $N|\widetilde{\Gamma}_m(f)$ has dimension $n-1$.
\end{pf}
Let $\phi$ denote the canonical bundle map over
$f\colon M^{nm-1}\to W^{n(m+1)-1}$,
$$
\phi\colon N\to TW^{n(m+1)-1},
$$
and let $s$ be as in Lemma \ref{sctn}.
We define a vector field $v\colon \Gamma_m(f)\to TW^{n(m+1)-1}$
along $\Gamma_m(f)$: For $q=f(p_1)=\dots=f(p_m)$ let
$$
v(q)=\phi(s(p_1))+\dots+\phi(s(p_m))).
$$
It is easy to see that $v$ is well-defined and smooth.
We now restrict attention to the case where the target manifold is
Euclidean space.
\begin{dfn}\label{shif}
For a generic immersion $f\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$,
let $\Gamma_m'(f,v,\epsilon)$ be the submanifold of ${\mathbb{R}}^{n(m+1)-1}$
obtained by shifting $\Gamma_m(f)$ a small distance $\epsilon>0$ along $v$.
\end{dfn}
\begin{rmk}\label{skeps}
Note that for $\epsilon>0$ small enough $\Gamma_m'(f,v,\epsilon)\cap
f(M)=\emptyset$. Moreover, if $a>0$ is such that for all $0<\epsilon<a$
the corresponding $\Gamma_m'(f,v,\epsilon)$ is disjoint from $f(M)$ then
$\Gamma_m'(f,v,\epsilon)$ and $\Gamma_m'(f,v,a)$ are homotopic in
${\mathbb{R}}^{n(m+1)-1}-f(M)$.
\end{rmk}
\subsection{Definition of the invariants $L$ and $\Lambda$}
\begin{dfn}\label{dfnLambda}
Let $f\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be a generic immersion.
Assume that $M$ satisfies
\begin{equation*}
H_{n-1}(M;{\mathbb{Z}}_2)=0=H_n(M;{\mathbb{Z}}_2).
\tag{C$\Lambda$}
\end{equation*}
Define $\Lambda(f)\in{\mathbb{Z}}_2$ as
$$
\Lambda(f)=[\Gamma_m'(f,v,\epsilon)]-s_\ast[\widetilde{\Gamma}_m(f)]\in{\mathbb{Z}}_2,
$$
where $\epsilon>0$ is very small and
$
[\Gamma_m'(f,v,\epsilon)]\in H_{n-1}({\mathbb{R}}^{n(m+1)-1}-f(M);{\mathbb{Z}})\cong{\mathbb{Z}}_2
$
(see Remark \ref{idHcom})
and
$s_\ast[\widetilde{\Gamma}_m]\in H_{n-1}(\partial N;{\mathbb{Z}})\cong{\mathbb{Z}}_2$ (see Remark
\ref{idHfib}).
\end{dfn}
\begin{dfn}\label{dfnL}
Let $f\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be a generic immersion. Assume
that $n$ is even and that $M$ is oriented and satisfies
\begin{equation*}
H_{n-1}(M;{\mathbb{Z}})=0=H_n(M;{\mathbb{Z}}).
\tag{C$L$}
\end{equation*}
Define
$L(f)\in{\mathbb{Z}}$ as
$$
L(f)=[\Gamma_m'(f,v,\epsilon)]-s_\ast[\widetilde{\Gamma}_m(f)]\in{\mathbb{Z}},
$$
where $\epsilon>0$ is very small and
$[\Gamma_m'(f,v,\epsilon)]\in
H_{n-1}({\mathbb{R}}^{2m(j+1)-1}-f(M);{\mathbb{Z}})\cong{\mathbb{Z}}$
(see Remark ~\ref{idHcom})
and
$s_\ast[\widetilde{\Gamma}_m]\in H_{n-1}(\partial N;{\mathbb{Z}})\cong{\mathbb{Z}}$ (see Remark
\ref{idHfib}).
\end{dfn}
\begin{rmk}
If $\operatorname{Tor}(H_{n-2}(M;Z),{\mathbb{Z}}_2)=0$
and
$M$ satisfies condition $({\rm C}L)$ in Definition ~\ref{dfnL}
then by the universal coefficient
theorem $M$ also satisfies condition $({\rm C}\Lambda)$ in Definition
~\ref{dfnLambda}. It follows from Remarks ~\ref{idHcom}, \ref{idHfib} that
in this case $\Lambda=L\mod{2}$.
\end{rmk}
\begin{lma}\label{wdL}
$\Lambda$ and $L$ are well-defined. That is, they do
neither depend on the choice of $s$, nor on the choice of
$\epsilon>0$.
\end{lma}
\begin{rmk}
We shall often drop the awkward notation
$\Gamma_m'(f,v,\epsilon)$ and simply write $\Gamma'(f)$. This is
justified by Lemma ~\ref{wdL}.
\end{rmk}
\begin{pf}
The independence of $\epsilon>0$ follows immediately from Remark
~\ref{skeps}. We will therefore not write all $\epsilon$'s out in the
sequel of this proof.
Let $s_0$ and $s_1$ be two homotopic sections of
$\partial N|\widetilde{\Gamma}_m(f)$. Let $v_0$ and $v_1$ be the
corresponding normal vector fields along $\Gamma_m(f)$.
A homotopy $s_t$ between $s_0$ and $s_1$ induces a homotopy $v_t$ between
$v_0$ and $v_1$ and hence between $\Gamma_m'(f,v_0)$ and
$\Gamma_m'(f,v_1)$ in ${\mathbb{R}}^{n(m+1)-1}-f(M)$. This shows that $\Lambda$
and $L$ only depend on the homotopy class of $s$.
To see that $\Lambda$ and $L$ are independent of the vector field we
introduce the notion of adding a local twist: Let
$s\colon \widetilde{\Gamma}_m(f)\to\partial N$ be
a section and let $p\in\widetilde{\Gamma}_m(f)$. Choose a neighborhood $U$ of $p$ in
$\widetilde{\Gamma}_m(f)$ and a trivialization $\partial N|U\cong U\times
S^{n-1}$ such that $s(u)\equiv w$, where $w$ is a point in
$S^{n-1}$. Let $D^{n-1}$ be disk inside $U$ and let $\sigma\colon D\to
S^{n-1}$ be a smooth map of degree $\pm 1$ such that $\sigma\equiv w$ in
a neighborhood of $\partial D$. Let $s^{\rm tw}$ be the section which is $s$
outside of $D$ and $\sigma$ in $D$. We say that $s^{\rm tw}$ is the result of
adding a local twist to $s$. (In the case when $M$ is
oriented and $n$ is even, the local twist is said to be positive if
the degree of
$\sigma$ is $+1$ and negative if the degree of $\sigma$ is $-1$.)
Let $s^{\rm tw}$ be the vector field obtained by adding a twist to $s$. Let
$v^{\rm tw}$ and $v$ be the vector fields along $\Gamma_m(f)$ obtained
from $s^{\rm tw}$ and $s$ respectively. Clearly,
$$
{s^{\rm tw}}_\ast[\Gamma_m(f)]=s_\ast[\Gamma_m(f)]\pm 1
\quad\text{ and }\quad
[\Gamma_m'(f,v^{\rm tw})]=[\Gamma_m'(f,v)]\pm 1.
$$
Hence, $\Lambda$ and $L$ are invariant under adding local twists.
A standard obstruction theory argument shows that
if $s$ and $s'$ are two sections of $\partial N|\Gamma_m$ then by
adding local twists to $s'$ we can obtain a section $s''$ which is
homotopic to $s$. Hence, $\Lambda$ and $L$ are independent of the
choice of section.
\end{pf}
\begin{lma}
$\Lambda$ and $L$ are invariants of generic immersions.
\end{lma}
\begin{pf}
Let $f_t$, $0\le t\le 1$ be a regular
homotopy through generic immersions. Consider the induced map
$$
F\colon M\times I\to{\mathbb{R}}^{n(m+1)-1}\times I,\quad F(x,t)=(f_t(x),t),
$$
where $I$ is the unit interval $[0,1]$. Shifting
$\Gamma_m(F)\cong\Gamma_m(f_0)\times I$ off $F(M\times I)$ using a
suitable
vector field in $N|{\widetilde \Gamma}_m(F)$ it is easy to see that
$L(f_0)=L(f_1)$.
\end{pf}
\section{Additivity properties, connected sum, and reversing orientation}
In this section we study how our
invariants behave under two natural operations on generic immersions:
connected sum and reversing orientation.
\subsection{Connected summation of generic immersions}\label{csgi}
For (oriented) manifolds $M$ and $V$ of the same dimension, let
$M\operatorname{\ensuremath{\sharp}} V$ denote the (oriented) connected sum of $M$ and $V$.
Let $f\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ and $g\colon
V^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be two generic immersions.
We shall define the connected sum $f\operatorname{\ensuremath{\sharp}} g$ of these. It will be a
generic immersion $M\operatorname{\ensuremath{\sharp}} V\to{\mathbb{R}}^{n(m+1)-1}$.
Let $(u,x)$,
$u\in{\mathbb{R}}$ and $x\in{\mathbb{R}}^{n(m+1)-2}$ be coordinates on
${\mathbb{R}}^{n(m+1)-1}$. Composing the immersions $f$ and $g$ with
translations we may assume that $f(M)\subset\{u\le-1\}$ and
$g(V)\subset\{u\ge 1\}$. Choose a point
$p\in M$ and a point $q\in V$ such that there is only one point in
$f^{-1}(f(p))$ and in $g^{-1}(g(q))$. Pick an arc $\alpha$ in
${\mathbb{R}}^{n(m+1)-1}$ connecting
$f(p)$ to $g(q)$ and such that $\alpha\cap(f(M)\cup
g(V))=\{f(p),g(q)\}$. Moreover, assume that $\alpha$ meets $f(M)$ and
$g(V)$ transversally at its endpoints. Let $N$ be the normal bundle of
$\alpha$. Pick a (oriented) basis of $T_{f(p)}f(M)$ and an
(anti-oriented) one
of $T_{g(q)}(g(V))$. These give rise to $nm-1$ vectors over $\partial a$
in $N$. Extend these vectors to $nm-1$ independent normal vector
fields along $\alpha$. Using a suitable map of $N$ into a tubular
neighborhood of $\alpha$ these vector fields give rise to an embedding
$\phi\colon\alpha\times D\to{\mathbb{R}}^{n(m+1)-1}$, where $D$ denotes
a disk of dimension $nm-1$, such that
$\phi|f(p)\times D$ is an (orientation preserving) embedding
into $f(M)$ and $\phi|g(q)\times D$ is an (orientation
reversing) embedding into $g(V)$. The tube
$\phi(\alpha\times\partial D)$ now joins
$f(M)-\phi(f(p)\times\operatorname{int}(D))$ to
$g(V)-\phi(g(q)\times\operatorname{int}(D))$. Smoothing the corners we get a
generic immersion $f\operatorname{\ensuremath{\sharp}} g\colon M\operatorname{\ensuremath{\sharp}} V\to{\mathbb{R}}^{n(m+1)-1}$.
\begin{lma}
Let $f,g\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be generic immersions.
The connected sum $f\operatorname{\ensuremath{\sharp}} g$ is independent
up to regular homotopy through generic immersions of both the choices
of points $f(p)$ and $g(q)$ and the choice of the path $\alpha$ used
to connect them.
\end{lma}
\begin{pf}
This is straightforward.
(Note that the preimages of self-intersections has codimension $n>1$.)
\end{pf}
\begin{lma}
If $f,f'\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ and
$g,g'\colon V^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$, $m,n>1$ are regularly homotopic
through generic immersions then $f\operatorname{\ensuremath{\sharp}} g$ and $f'\operatorname{\ensuremath{\sharp}} g'$ are
regularly homotopic through generic immersions.
\end{lma}
\begin{pf}
This is straightforward
\end{pf}
\begin{prp}\label{csJ}
The invariants $J_r$ and $J$ are additive under connected summation.
\end{prp}
\begin{pf}
$\Delta^j(f\operatorname{\ensuremath{\sharp}} g)=\Delta^j(f)\sqcup\Delta^j(g)$.
\end{pf}
Note that if $M^{nm-1}$ and $V^{nm-1}$ are manifolds which both
satisfy condition $({\rm C}\Lambda)$ in Definition ~\ref{dfnLambda} or
condition $({\rm C}L)$ in Definition ~\ref{dfnL} then so does $M\operatorname{\ensuremath{\sharp}} V$.
\begin{prp}\label{csL}
Let $f\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ and
$g\colon V^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be generic immersions.
\begin{itemize}
\item[{\rm (a)}] If $M$ and $V$ both satisfy condition
$({\rm C}\Lambda)$ then
$$
\Lambda(f\operatorname{\ensuremath{\sharp}} g)=\Lambda(f)+\Lambda(g).
$$
\item[{\rm (b)}] If $n$ is even and $M$ and $V$ are both oriented and
both satisfy condition $({\rm C}L)$ then
$$
L(f\operatorname{\ensuremath{\sharp}} g)=L(f)+L(g).
$$
\end{itemize}
\end{prp}
\begin{pf}
Note that $\Gamma_m(f\operatorname{\ensuremath{\sharp}} g)=\Gamma_m(f)\sqcup\Gamma_m(g)$.
Consider case (b). Choose
$2j$-chains $D$ and $E$ in $\{u<-1\}$ and $\{u>1\}$ bounding
$\Gamma_m(f)$ and $\Gamma_m(g)$ respectively and disjoint from the arc
$\alpha$, used in the construction of $f\operatorname{\ensuremath{\sharp}} g$. Then
$$
L(f\operatorname{\ensuremath{\sharp}} g)=(D\cup E)\cdot f\operatorname{\ensuremath{\sharp}} g(M\operatorname{\ensuremath{\sharp}} V)=
D\cdot f(M)+E\cdot g(V)=L(f)+L(g).
$$
Case (a) is proved in exactly the same way.
\end{pf}
\subsection{Changing orientation}
The invariants $J_r$, $J$, and $\Lambda$ are clearly orientation
independent. In contrast to this, the invariant $L$ is orientation
sensitive.
To have $L$ defined, let $n=2j$ and consider an oriented closed manifold
$M^{2jm-1}$ which satisfies condition (C$L$).
\begin{prp}
Assume that there exists an orientation reversing diffeomorphism
$r\colon M\to M$. Let $f\colon
M^{2jm-1}\to{\mathbb{R}}^{2j(m+1)-1}$ be a generic immersion. Then $f\circ r$
is a generic immersion and
$$
L(f\circ r)=(-1)^{m+1}L(f)
$$
\end{prp}
\begin{pf}
Note that $\Gamma_m(f\circ r)=\Gamma_m(f)=\Gamma_m$. The orientation of
$\Gamma_m$ is induced from the decomposition
$$
T_q{\mathbb{R}}^{2j(m+1)-1}=T_q\Gamma_m(f)\oplus N_{1}\oplus\dots\oplus N_{m}.
$$
The immersions $f$ and $f\circ r$ induces opposite orientations on
each $N_i$ hence the orientations induced on $\Gamma_m$ agrees if $m$
is even and does not agree if $m$ is odd.
Let $D$ be a $2j$-chain bounding $\Gamma_m$, with its orientation
induced from $f$. If $m$ is even then
$$
L(f\circ r)=D\cdot f(r(M))=D\cdot -f(M)=-L(f).
$$
If $m$ is odd then
$$
L(f\circ r)=-D\cdot f(r(M))=-D\cdot -f(M)=L(f).
$$
\end{pf}
\begin{prp}
Let $R\colon {\mathbb{R}}^{2j(m+1)-1}\to{\mathbb{R}}^{2j(m+1)-1}$ be reflection in a
hyperplane. Let $f\colon M^{2jm-1}\to{\mathbb{R}}^{2j(m+1)-1}$ be a generic
immersion. Then $R\circ f$ is a generic immersion and
$$
L(R\circ f)=(-1)^mL(f)
$$
\end{prp}
\begin{pf}
Note that the oriented normal bundle of $Rf$ is $-RN$.
So the correctly oriented $2j$-chain bounding $\Gamma_{m}(R\circ f)$
is $(-1)^{m+1}RD$. Thus,
$$
L(R\circ f)=(-1)^{m+1} (RD\cdot Rf(M))=(-1)^m(D\cdot f(M))=(-1)^mL(f).
$$
\end{pf}
\section{Coorientations and proofs of Theorems ~\ref{thmL} and
~\ref{thmJ}}
In this section we prove Theorems ~\ref{thmL} and ~\ref{thmJ}. To
do that we need a coorientation
of the discriminant hypersurface in the space of immersions.
\subsection{Coorienting the discriminant}\label{codiscr}
\begin{rmk}\label{nabla0}
Let $a$ be an invariant of generic immersions. If $a$ is
${\mathbb{Z}}_2$-valued then $\nabla a$ (see Section ~\ref{mainres}) is
well-defined without reference to any coorientation of
$\Sigma$. Moreover, if $a$ is integer-valued and $\Delta$ is a union of
path components of $\Sigma^1$ then the notion $\nabla a|\Delta\equiv0$
is well-defined without reference to a coorientation of $\Sigma$.
\end{rmk}
We shall coorient the relevant parts (see Remark ~\ref{nabla0}) of the
discriminant hypersurface. That is, we shall find coorientations of the
parts of the discriminant hypersurface in the space of generic
immersions where our invariants have non-zero jumps. These
coorientations will be {\em continuous} (see ~\cite{E3}, Section
7). That is, the intersection number of any generic small loop in
${\mathcal{F}}$ and $\Sigma^1$ vanishes and the coorientation extends continuously
over $\Sigma^2$.
\begin{dfn}\label{coJr}
Let $m>1$ and $n>1$. Assume that $n$ is odd. Let $2\le
r\le m$ and assume that $m-r$ is even. Let $f_0\colon M^{nm-1}\to
W^{n(m+1)-1}$ be a generic immersion in $\Sigma^1$ with one degenerate
$r$-fold self-intersection point. Let $f_t$ be a versal deformation of
$f_0$. We say that $f_{\delta}$ is on the positive side of $\Sigma^1$ and
$f_{-\delta}$ on the negative side if
$\chi(\Delta^r(f_\delta))>\chi(\Delta^r(f_{-\delta}))$.
\end{dfn}
\begin{rmk}
Note that by Lemma ~\ref{vdef1sm} $\chi(\Delta^r(f_\delta))$ is
obtained from
$\chi(\Delta^r(f_{-\delta}))$ by a Morse modification. Since the
dimension of $\chi(\Delta^r(f_\delta))$ is $n(m-r+1)-1$ is even the
Euler characteristic changes under such modifications.
\end{rmk}
\begin{dfn}\label{coJ}
Let $m>1$. Let $f_0\colon M^{2m-1}\to
W^{2m+1}$ be a generic immersion in $\Sigma^1$ with one degenerate
$m$-fold self-intersection point. Let $f_t$ be a versal deformation of
$f_0$. We say that $f_{\delta}$ is on the positive side of $\Sigma^1$ and
$f_{-\delta}$ on the negative side if the number of components in
$\Gamma_m(f_\delta)$ is larger than the number of components in
$\Gamma_m(f_{-\delta})$.
\end{dfn}
\begin{rmk}
Note that by Lemma ~\ref{vdef1sm} $\Gamma_m(f_\delta)$ is
obtained from
$\Gamma_m(f_{-\delta})$ by a Morse modification. Since these are
1-manifolds the number of components change.
\end{rmk}
The construction of the relevant coorientation for $L$ is more
involved. It is based on a high-dimensional counterpart of the
notion of over- and under-crossings in classical knot theory.
Let $f_0\colon M^{2j(m+1)-1}\to W^{2j(m+1)-1}$ be an immersion of
oriented manifolds. Assume that $f_0$ has an ($m+1$)-fold
self-intersection point,
$q=f_0(p_1)=\dots=f_0(p_{m+1})$. Let $f_t$ be a versal deformation of
$f_0$. Let $U_i$ be small neighborhoods of $p_i$
and let $S_i^t$ denote the oriented sheet $f_t(U_i)$. Note that
$S_i^0\cap\dots\cap S^0_{m+1}=q$ and that
$S^t_1\cap\dots\cap S^t_{m+1}=\emptyset$ if $t\ne 0$.
Let $D^t_i=\cap_{j\ne i}S_j^t$. Let $w_i$ be a line transversal to
$TS_i^0+TD^0_i$ at $q$. For small $t\ne 0$ both $S_i^t$ and
$D_i^t$ intersects $w_i$. Orienting the line from $S_i^t$ to
$D_i^t$ gives a local orientation $(D_i^t,S_i^t,\vec{w}_i)$
of ${\mathbb{R}}^{2j(m+1)-1}$. Comparing it with the standard orientation of
${\mathbb{R}}^{2j(m+1)-1}$ we get a sign
$\sigma_t(i)=\operatorname{Or}(D_i^t,S_i^t,\vec{w}_i)$, where $\operatorname{Or}$ denotes the
sign of the orientation. Note that, if we orient
the line from $D_i^t$ to $S_i^t$ we get the opposite orientation
$-\vec{w}_i$ of $w$ and
$$
\operatorname{Or}(D_i^t,S_i^t,\vec{w}_i)=\operatorname{Or}(S_i^t,D_i^t,-\vec{w}_i),
$$
since $D_i^t$ and $S_i^t$ are both odd-dimensional. Note also that
$\sigma_t(i)=-\sigma_{-t}(i)$ (see Lemma \ref{vdef1big}).
We next demonstrate that $\sigma_t(i)=\sigma_t(j)$ for all $i,j$:
Let $N_i^t$ be the oriented normal bundle of $S_i^t$. Let $w$ be a vector
from $D_1^t$ to $D_2^t$, transversal to both
$TS_1^0+TD^0_1$ and $TS_2^0+TD^0_2$. Orient it from
$D_1^t$ to $D_2^t$. Then
$$
\sigma_t(1)=\operatorname{Or}(S_1^t,D^t_1,\vec w),
$$
and hence $TD^t_1+w$ gives a normal bundle $N_1^t$ and by the
convention used to orient the normal bundle
$\operatorname{Or}(N_1^t)=\sigma_t(1)\operatorname{Or}(D_1^t,\vec w)$. In a similar way it follows that
$\operatorname{Or}(N_2^t)=\sigma_t(2)\operatorname{Or}(D_2^t,-\vec w)$.
Now, by definition
\begin{align*}
1= & \operatorname{Or}(D_1^t,N_2^t,\dots,N_m^t)
=\sigma_t(2)\operatorname{Or}(D_1^t,(D_2^t,-\vec w),N_3^t,\dots,
N_m^t)=\\
= & \text{[$\dim(D_2^t)$ is odd]}=
-\sigma_t(2)\operatorname{Or}(D_1^t,-\vec w,D_2^t,N_3^t,\dots, N_m^t)=\\
= & \sigma_t(2)\sigma_t(1)\operatorname{Or}(N_1^t,D_2^t,N_3^t,\dots,N_m^t)
=\sigma_t(1)\sigma_t(2).
\end{align*}
Hence, $\sigma_t(1)\sigma_t(2)=1$ as claimed.
\begin{dfn}\label{coL}
Let $f_0\colon M^{2j(m+1)-1}\to W^{2j(m+1)-1}$ be an immersion of
oriented manifolds. Assume that $f_0$ has an $(m+1)$-fold
self-intersection point. Let $f_t$ be a versal deformation of
$f_0$. We say
that $f_{\delta}$ is on the positive side of $\Sigma^1$ at $f_0$ if
$$
\sigma_\delta(1)=\dots=\sigma_\delta(m+1)=+1.
$$
\end{dfn}
\subsection{First order invariants}
The following obvious lemma will be used below.
\begin{lma}\label{jump}
Any first order invariant of generic immersions is uniquely (up to
addition of zero order invariants) determined by its jump.\qed
\end{lma}
\subsection{Proof of Theorem ~\ref{thmJ}}\label{pfthmJ}
We know from Lemma ~\ref{wdJ} that $J$ and $J_r$ are invariants of
generic immersions. We must calculate their jumps.
We start in case (a): Assume that $n$ is odd.
Let $f_t\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$, $t\in [-\delta,\delta]$ be
a versal deformation of $f_0\in\Sigma^1$ and fix $r$, $2\le r\le m$ such
that $m-r$ is even.
If
$f_0$ has a degenerate $k$-fold intersection point $2\le k\le m+1$ and
$k\ne r$ then $\Delta^r(f_{-\delta})$ is diffeomorphic to
$\Delta^r(f_{\delta})$: If $k<r$ then $\Delta^r(f_{-\delta})$ is
not affected at all by the versal deformation. If $k>r$ then the
immersed submanifold $\tau^{-1}(\Gamma_k(f_{-\delta}))$ of
$\Delta^r(f_{-\delta})$ is changed by surgery (or is
deformed by regular homotopy if $k=m+1$) under the
versal deformation. This does not affect the diffeomorphism class of
$\Delta^r(f_{-\delta})$.
If $f_0$ has a degenerate $r$-fold self-intersection point then
$\Delta^r{f_{\delta}}$ is obtained from $\Delta^r{f_{-\delta}}$ by a
surgery (see Lemma ~\ref{vdef1sm}). Since the dimension of
$\Delta^r(f_{-\delta})$ is $n(m-r+1)-1$ which is even this changes the
Euler characteristic by $\pm2$.
According to our coorientation conventions $\nabla J_r(f_0)=2$ if
$f_0$ has a degenerate $r$-fold self-intersection point. Also,
$\nabla J_r(f_0)=0$ if $f_0$ has any other degeneracy.
The jump of $J$ in case (b) is $1$ by the same argument.
It is evident from Lemma ~\ref{vdef2} (see Figure
\ref{figcod2}) that $J$ and $J^r$ are first order invariants. The
theorem now follows from Lemma ~\ref{jump}.\qed
\subsection{Proof of Theorem \ref{thmL}}\label{pfthmL}
We start with (b):
Let $f_t\colon M^{2jm-1}\to{\mathbb{R}}^{2j(m+1)-1}$, $t\in [-\delta,\delta]$
be a generic one-parameter family intersecting $\Sigma^1$ in $f_0$. If
the degenerate intersection point of $f_0$ is a $k$-fold intersection
point with $2\le k\le m-1$ then clearly
$L(f_{-\delta})=L(f_{\delta})$, since the $m$-fold self-intersection
is not affected under such deformations.
Assume that $f_0$ has a degenerate $m$-fold self-intersection
point. Without loss of generality we may assume that $f_t$ is a
deformation of the form in Lemma \ref{vdef1sm} (we use coordinates as
there):
Let $F(x,t)=(f_t(x),t)$. Shifting $((\{Q(y^0,v^1)=t\},0\dots,0),t)$ we
obtain a $2j$-chain bounded by
$\Gamma_m'(f_\delta)\times\delta-\Gamma_m'(f_{-\delta})\times {-\delta}$
in
${\mathbb{R}}^{2m(j+1)-1}\times[-\delta,\delta]-F(M\times[-\delta,\delta])$.
It follows that $L(f_\delta)=L(f_{-\delta})$.
If $f_0$ has an $(m+1)$-fold self-intersection then
the discussion preceding Definition ~\ref{coL} shows that
$L(f_{\delta})=L(f_{-\delta})+(m+1)$: At an $(m+1)$-fold self
intersection point $m+1$ crossings are turned into crossings of
opposite sign.
Hence, $\nabla L(f_0)=m+1$ if $f_0$ has an $(m+1)$-fold
self-intersection point and $\nabla L(f_0)=0$ if $f_0$ has any other
degeneracy.
The calculation of $\nabla \Lambda$ in (a) is analogous. Let us
just make a remark about the parity of $m$: At an $(m+1)$-fold
self-intersection point $m+1$ crossings are
changed. Hence, at instances of $(m+1)$-fold self-intersection the
invariant $\Lambda$ changes by $1\in{\mathbb{Z}}_2$ if $m+1$ is odd and does not
change if $m+1$ is even.
It is
immediate from Lemma ~\ref{vdef2}, see Figure \ref{figcod2} that
$\Lambda$ and $L$ are first order invariants.
The theorem now follows from Lemma ~\ref{jump}.
\qed
\section{Invariants of regular homotopy}\label{rhinv}
A function of immersions $M\to W$ which is constant on path components
of ${\mathcal{F}}$ will be called an {\em invariant of regular homotopy}. Our
geometrically defined invariants of generic immersion give rise to
torsion invariants of regular homotopy.
\begin{dfn}
Let $n$ be odd. Let $f\colon M^{nm-1}\to W^{n(m+1)-1}$ be an
immersion. Let $f'$ be a generic immersion regularly homotopic to $f$.
For $2\le r\le m$ such that $m-r$ is even, define $j_r(f)\in{\mathbb{Z}}_2$ as
$$
j_r(f)=J_r(f')\mod{2}.
$$
\end{dfn}
\begin{prp}\label{rhJ}
The function $j_r$ is an invariant of regular homotopy.
\end{prp}
\begin{pf}
Clearly it is enough to show that for any two regularly homotopic
generic
immersions $f_0$ and $f_1$, $j_r(f_0)=j_r(f_1)$. Let $f_t$ be a
generic regular homotopy from $f_0$ to $f_1$. Then $f_t$ intersects
$\Sigma$ transversally in a finite number of points in $\Sigma^1$. It
follows from Theorem ~\ref{thmJ} that $j_r$ remains unchanged at such
intersections. Hence, $j_r(f_0)=j_r(f_1)$.
\end{pf}
\begin{dfn}
Let $n$ be even. Assume that $M^{nm-1}$ is a manifold which satisfy
condition $({\rm C}L)$. Let $f\colon M^{nm-1}\to {\mathbb{R}}^{n(m+1)-1}$ be an
immersion. Let $f'$ be a generic immersion regularly homotopic to $f$.
Define $l(f)\in{\mathbb{Z}}_{m+1}$ as
$$
l(f)=L(f')\mod{(m+1)}.
$$
\end{dfn}
\begin{prp}\label{rhL}
The function $l$ is an invariant of regular homotopy.
\end{prp}
\begin{pf}
The proof is identical to the proof of Proposition ~\ref{rhJ}.
\end{pf}
\begin{dfn}
Let $m>1$ be odd. Assume that $M^{nm-1}$ is a manifold which satisfy
condition $({\rm C}\Lambda)$. Let $f\colon M^{nm-1}\to {\mathbb{R}}^{n(m+1)-1}$ be an
immersion. Let $f'$ be a generic immersion regularly homotopic to $f$.
Define $\lambda(f)\in{\mathbb{Z}}_2$ as
$$
\lambda(f)=\Lambda(f').
$$
\end{dfn}
\begin{prp}
The function $\lambda$ is an invariant of regular homotopy.
\end{prp}
\begin{pf}
This follows from the proof of Theorem ~\ref{thmL}, where it is noted that
when $m$ is odd $\Lambda$ remains constant when a regular homotopy
intersects $\Sigma^1$.
\end{pf}
\section{Sphere-immersions in codimension two}
In this section Theorem \ref{thmdivL} will be proved. To do that we
will first discuss the classifications of sphere-immersions and
sphere-embeddings up to regular homotopy.
\subsection{The Smale invariant}
In Smale's classical work \cite{S} it is proved that there is a
bijection between the set of regular homotopy classes ${\mathbf {Imm}}(k,n)$ of
immersions $S^k\to{\mathbb{R}}^{k+n}$ and the
elements of the group $\pi_k(V_{{k+n},k})$, the $k^{\rm th}$ homotopy
group of the Stiefel manifold of $k$-frames in $(k+n)$-space.
If $f\colon S^k\to{\mathbb{R}}^{k+n}$ is an immersion we let
$\Omega(f)\in\pi_k(V_{{k+n},k})$ denote its Smale invariant.
Via $\Omega$ we can view ${\mathbf {Imm}}(k,n)$ as an Abelian group.
If the codimension is two the groups appearing in Smale's
classification are easily computed: The exact homotopy
sequence of the fibration
$$
SO(2)\hookrightarrow SO(2m+1)\to V_{2m+1,2m-1}
$$
implies that
$\pi_{2m-1}(V_{2m+1,2m-1})\cong\pi_{2m-1}(SO(2m+1))$. Bott-periodicity
then gives:
$$
\pi_{2m-1}(V_{2m+1,2m-1})=\begin{cases}
{\mathbb{Z}} &\text{if } m=2j,\\
{\mathbb{Z}}_2 &\text{if } m=4j+1,\\
0 &\text{if } m=4j+3.
\end{cases}
$$
\begin{rmk}\label{geomope}
It is possible to identify the group operations in ${\mathbf {Imm}}(k,2)$ geometrically.
Kervaire \cite{K} proves that the Smale invariant $\Omega$ is
additive under connected sum of immersions. This gives the geometric
counterpart of addition.
If $f\colon S^n\to{\mathbb{R}}^{n+2}$ is an immersion, $r\colon S^n\to S^n$
is an orientation reversing diffeomorphism, and $n\ne 2$ then
$\Omega(f)=-\Omega(f\circ r)$. (See ~\cite{E2} for the case
$S^3\to{\mathbb{R}}^5$, the other cases are analogous). This gives
the geometric counterpart of the inverse operation in ${\mathbf {Imm}}(n,2)$ for
$n\ne 2$.
It is interesting to note that for immersions $f\colon S^2\to{\mathbb{R}}^4$,
$\Omega(f)=\Omega(f\circ r)$: The Smale invariant can in this
case be computed as the algebraic number of self-intersection points.
This number is clearly invariant under reversing orientation. (The same is
true also for immersions $S^{2k}\to{\mathbb{R}}^{4k}$, $k\ge 1$.)
\end{rmk}
\begin{lma}\label{lhom}
The regular homotopy invariant $l$ induces a homomorphism
$$
{\mathbf {Imm}}(2m-1,2)\to{\mathbb{Z}}_{m+1}.
$$
\end{lma}
\begin{pf}
Note that the dimensions are such that $L$ is defined and
spheres certainly satisfy condition $({\rm C}L)$. The invariant $L$ is
additive under
connected summation and changes sign if an immersion is composed on
the left with an orientation reversing diffeomorphism. Hence, $l$
induces a homomorphism.
\end{pf}
Let ${\mathbf {Emb}}(n,2)\subset{\mathbf {Imm}}(n,2)$ be the set of regular homotopy classes
which contain embeddings. By Remark ~\ref{geomope}
${\mathbf {Emb}}(n,2)$ is a subgroup of ${\mathbf {Imm}}(n,2)$. A result of Hughes and
Melvin ~\cite{HM} states that
${\mathbf {Emb}}(4j-1,2)\subset{\mathbf {Imm}}(4j-1,2)\cong{\mathbb{Z}}$ is a subgroup of index
$\mu_j$. Here $\frac{B_j}{4j}=\frac{\nu_j}{\mu_j}$, where $B_j$ is the
$j^{\rm th}$ Bernoulli number and $\nu_j$ and $\mu_j$ are coprime
integers.
\subsection{Proof of Theorem ~\ref{thmdivL}}\label{pfthmdivL}
By Lemma \ref{lhom},
$l\colon{\mathbf {Imm}}(2m-1,2)\to{\mathbb{Z}}/(m+1){\mathbb{Z}}$ is a homomorphism and clearly,
${\mathbf {Emb}}(2m-1,2)\subset\operatorname{ker}(l)$.
In case (b) ${\mathbf {Imm}}(8j+5,2)=0$ and hence the image of $l$ is zero,
which proves that $L(f)$ is always divisible by $m+1=4j+4$.
In case (a) ${\mathbf {Imm}}(8j+1,2)\cong{\mathbb{Z}}/2{\mathbb{Z}}$. Hence, for any immersion $f$,
$l(f\operatorname{\ensuremath{\sharp}} f)=0$. Thus, $L(f\operatorname{\ensuremath{\sharp}} f)=2L(f)$ is divisible by
$m+1=4j+2$, which implies that $L(f)$ is divisible by $2j+1$.
In case (c) ${\mathbf {Emb}}(4j-1,2)\cong \mu_j{\mathbb{Z}}$ as a subgroup of
${\mathbf {Imm}}(4j-1,2)\cong{\mathbb{Z}}$. Hence, for any immersion $f$, $m+1=2j+1$
divides
$$
L(f\operatorname{\ensuremath{\sharp}}\dots[\mu_j \text{ summands}]\dots\operatorname{\ensuremath{\sharp}} f)=\mu_jL(f).
$$
Thus, if if $p$ is a prime and $r,k$ are integers such that $p^{r+k}$
divides $2j+1$ and $p^{k+1}$ does not divide $\mu_j$ then $p^{r}$
divides $L(f)$.
\qed
\section{Examples and problems}\label{expbm}
In this section we construct examples of 1-parameter families of
immersions which shows that all the first order invariants we have
defined are non-trivial. We also discuss some problems in connection
with the regular homotopy invariants defined in Section ~\ref{rhinv}.
\subsection{Examples}
Using our local coordinate description of immersions with one
degenerate self-intersection point we can construct examples showing
that the invariants $J$, $J_r$, $\Lambda$ and $L$ are non-trivial. We
start with $\Lambda$ and $L$:
Choose $m$ standard spheres $S_1,\dots,S_m$ of dimension $(nm-1)$
intersecting in general position in ${\mathbb{R}}^{n(m+1)-1}$ so that
$S_1\cap\dots\cap S_m\cong S^{n-1}$. Pick a point
$p\in S_1\cap\dots\cap S_m$ and let
$S_{m+1}$ be another standard $(nm-1)$-sphere intersecting
$S_1\cap\dots\cap S_m$ at $p$ so that in a neighborhood of $p$ the
embeddings are given by the expressions in Lemma ~\ref{vdef1big} and
so that $p$ is the only degenerate intersection point. Let
$f_0\colon S^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be the immersions which is the
connected sum of $S_1,\dots,S_{m+1}$. Let $f_t$ be a versal
deformation of $f$. Then, after possibly reversing the direction of
the versal deformation we have $\Lambda(f_{\delta})=1\in{\mathbb{Z}}_2$, if $n$
is odd or $L(f_{\delta})=\pm(m+1)$ if $n$ is even.
In the latter case we would get the other sign of $L(f_{\delta})$ if
the
orientation of $S_{m+1}$ in the construction above is reversed. To
distinguish these two immersions denote one by $h^+$ and the other by
$h^-$, so that $L(h^+)=m+1=-L(h^-)$. Then it is an immediate
consequence of Theorem ~\ref{thmL} that:
{\em Any generic regular homotopy from $h^+$ to $h^-$ has at least two
instances of $(m+1)$-fold self-intersection.}
Theorem ~\ref{thmL} has many corollaries as the one just mentioned.
To see that $L$ and $\Lambda$ are nontrivial for other source
manifolds. We use connected sum with the immersions
$S^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ just constructed and Lemma ~\ref{csL}.
The proof that the invariants $J$ and $J_r$ are nontrivial are along
the same lines: Construct a sphere-immersion into Euclidean space with
one degenerate self- intersection point as in the local models in Lemma
~\ref{vdef1sm}. Then embed this Euclidean space with immersed sphere
into any target manifold and use connected sum together with Lemma
~\ref{csJ}.
\begin{rmk}\label{Lrange}
Applying connected sum to the sphere-immersion constructed above in
the case when it is an immersion $S^{2m-1}\to{\mathbb{R}}^{2m+1}$ shows that
for any integer $k$ there exists a generic immersions
$f\colon S^{2m-1}\to{\mathbb{R}}^{2m+1}$ such that $L(f)=(m+1)k$.
\end{rmk}
\begin{rmk}\label{nonloc}
Let $f_0\colon M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ be an immersion in $\Sigma^1$
with an $(m+1)$-fold self-intersection point and let $f_t$, $-\delta<
t<\delta$ be a versal deformation of $f_0$.
Let $\Gamma_{\pm}=\cup_{i=1}^m\Gamma_{i}(f_{\pm\delta})$ and
${\widetilde \Gamma}_{\pm}=f_{\pm\delta}^{-1}(\Gamma_{\pm})$. Then if
$U_{\pm}$ are small regular neighborhoods of $\Gamma_{\pm}$ and
${\widetilde U}_{\pm}=f_{\pm\delta}^{-1}(U_{\pm})$ then there exist
diffeomorphisms $\phi$ and $\psi$ such that the following diagram
commutes
$$
\begin{CD}
({\widetilde U}_-,{\widetilde \Gamma}_-) & @>\psi>> & ({\widetilde U}_+,{\widetilde \Gamma}_+)\\
@V{f_{-\delta}}VV & {} & @VV{f_{+\delta}}V\\
(U_-,\Gamma_-) & @>\phi>> & (U_+,\Gamma_+)
\end{CD}.
$$
That is, the local properties of a generic immersion close to its
self-intersection does not change at instances of $(m+1)$-fold
self-intersection points in generic 1-parameter families.
Assume that $M$ fulfills the requirements of Theorem \ref{thmL}.
Then $\Lambda$ or $L$ is defined and
$\Lambda(f_{-\delta})\ne \Lambda(f_{\delta})$ or
$L(f_{-\delta})\ne L(f_{\delta})$, respectively. This implies that
$f_{-\delta}$ and $f_{\delta}$ are not regularly homotopic through
generic immersions. Hence, knowledge of the local properties of a
generic immersion $M^{nm-1}\to{\mathbb{R}}^{n(m+1)-1}$ close to its
self-intersection is not enough to
determine it up to regular homotopy through generic immersions.
\end{rmk}
\subsection{Problems}\label{pbm}
{\em Are the regular homotopy invariants $l$, $\lambda$ and $j_r$
nontrivial?}
A negative answer to this question implies restrictions on
self-intersection manifold. A positive answer gives non-trivial
geometrically defined regular homotopy invariants.
\subsection{Remarks on the invariant $l$}\label{rmksonl}
Theorem \ref{thmdivL} gives information on the possible range of $l$
for sphere-immersions in codimension two. We consider the most
interesting cases of immersions
$S^{4j-1}\to{\mathbb{R}}^{4j+1}$, $j\ge 1$ (Theorem \ref{thmdivL} (c)).
In the first case $S^3\to{\mathbb{R}}^5$, $l$ is non-trivial. This was shown in
\cite{E2}. Hence, for any integer $b$ there are generic immersions
$f\colon S^3\to{\mathbb{R}}^5$ such that $L(f)=b$.
\begin{rmk}\label{lkrmk}
The invariant $l$ is called $\lambda$ in \cite{E2} and is the
$\mod{3}$ reduction of an integer-valued invariant called $\operatorname{lk}$.
The definition of $\operatorname{lk}$ given in ~\cite{E2} differs slightly from the
definition of $L$ given here. There is an easy indirect way to see that,
nonetheless, the two invariants are the same:
Due to Theorem ~\ref{thmL} and Theorem 2
in \cite{E2}, $L-\operatorname{lk}$ is an invariant of regular homotopy which is $0$
on embeddings and additive under connected summation. Assume that
$L(f)-L'(f)=a$ for some immersion. Then, since $\mu_1=24$, the
connected sum of $24$ copies of any immersion is regularly homotopic
to an embedding. Hence, $24a=0$ and therefore $a=0$. Thus, $L\equiv\operatorname{lk}$.
\end{rmk}
If $2j+1$ is prime then Theorem \ref{thmdivL} does not impose any
restrictions on $l$ since, in this case, $2j+1$ divides $\mu_j$ (see
Milnor and Stasheff \cite{MS}).
On the other hand, if none of the prime factors of $2j+1$ divides
$\mu_j$, Theorem \ref{thmdivL} implies that $l$ is
trivial and in such cases Remark \ref{Lrange} allows us to determine
the range of $L$ which is $(2j+1){\mathbb{Z}}$. The first two cases where this
happens are $S^{67}\to{\mathbb{R}}^{69}$ and $S^{107}\to{\mathbb{R}}^{109}$, where $2j+1$ equals
$35=5\cdot 7$ and $55=5\cdot 11$ and $\mu_j$ equals $24=2^3\cdot 3$ and
$86184=2^3\cdot 3^4\cdot 7\cdot 19$, respectively.
\subsection{Remarks on the invariant $j_2$}
The first case in which to consider the invariant $j_2$ are
immersions of a $5$-manifold into an $8$-manifold.
Theorem 7.30 in ~\cite{E1} shows that
the self-intersection surface of a generic immersions of
$S^5\to{\mathbb{R}}^8$ must have {\em even} Euler characteristic (even though it may
be non-orientable). Thus, in this case $j_2$ is trivial. (Note that
$\pi_5(V_{8,5})\cong{\mathbb{Z}}_2$ so that there are two regular homotopy
classes of immersions $S^5\to{\mathbb{R}}^8$. Hence, it is non-trivial to see
that $l$ is trivial in this case.)
In contrast, the invariant $j_2$ is non-trivial for immersions
$S^5\to{\mathbb{R}} P^8$: Clearly, $S^5$ embeds in ${\mathbb{R}} P^8$. Thus, there is
an immersion $f$ with $j_2(f)=0$.
Consider three hyperplanes $H_1$, $H_2$, $H_3$ in general position in
${\mathbb{R}} P^8$. Fix a point $q\in {\mathbb{R}} P^8-(H_1\cup H_2\cup H_3)$. Note that
$X={\mathbb{R}} P^8-\{p\}$ is a non-orientable line bundle over $H_i$,
$i=1,2,3$. Choose sections $v_i$ of $X$ over $H_i$ such
that $\{v_i=0\}\cong{\mathbb{R}} P^6$ meets $H_1\cap H_2\cap H_3$ transversally
in $H_i$ for $i=1,2,3$. Now, $H_1\cap H_2\cap H_3\cong{\mathbb{R}} P^5$.
Let
$g\colon {\mathbb{R}} P^5\to{\mathbb{R}} P^8$ be the corresponding embedding.
Restricting $v_i$ to ${\mathbb{R}} P^5$ we get three normal vector fields
$v_1,v_2,v_3$ along ${\mathbb{R}} P^5\subset {\mathbb{R}} P^8$ and
$\{v_1=v_2=v_3=0\}\cong{\mathbb{R}} P^2$.
Let $p\colon S^5\to{\mathbb{R}} P^5$ be the universal cover.
Then $h=g\circ p\colon S^5\to{\mathbb{R}} P^8$ is an immersion. Let
$K_i=h^{-1}(\{v_i=0\})\cong S^4$.
Choose Morse functions $\phi_i\colon S^5\to{\mathbb{R}}$ such that
$\{\phi_i=0\}=K_i$. Let $\epsilon>0$ be
small. Then $f\colon S^5\to{\mathbb{R}} P^8$, given by
$$
f(x)=h(x)+\epsilon\sum_{i=1}^3\phi_i(x)v_i(h(x))\quad\text{for $x\in S^5$},
$$
is a generic immersion with $\Gamma_2(f)\cong{\mathbb{R}} P^2$. Hence, $j_2(f)=1$.
|
\section{Introduction} \setcounter{equation}{0}
To gain a better understanding of the structure of the lattice theory
and to interprete correctly the numbers obtained in Monte Carlo simulations,
it is very instructive to compare gauge variant quantities such
as gauge and fermion field propagators with the corresponding analytical
perturbative results. In this respect, compact $~U(1)~$ pure gauge theory
within the Coulomb phase serves as a very useful `test ground', because in
the weak coupling limit this theory is supposed to describe noninteracting
photons. However, previous lattice studies \cite{nak1,bmmp,clas,zmod,bddm}
have revealed some rather nontrivial effects. It has been shown that the
standard Lorentz (or Landau) gauge fixing procedure leads to a
$~\tau$--dependence of the transverse gauge field correlator
$~\Gamma_T(\tau;{\vec p})~$ being inconsistent with the expected zero-mass
behavior \cite{nak1}. Numerical \cite{bmmp,bddm,for} and analytical
\cite{clas} studies have shown that there is a connection between `bad'
gauge (or Gribov) copies and the appearance of periodically closed double
Dirac sheets (DDS). The removal of DDS restores the correct perturbative
behavior of the transverse photon correlator $~\Gamma_T(\tau;{\vec p})~$ with
momentum $~{\vec p}\ne 0~$. However, it does not resolve the Gribov ambiguity
problem \cite{grib} completely.
Other Gribov copies connected with zero--momentum modes
of the gauge fields still appear, which can `damage' such observables as
the zero--momentum gauge field correlator $~\Gamma(\tau;0)~$ or the
fermion propagator $~\Gamma_{\psi}(\tau)~$ \cite{zmod,bddm,zmlf}.
After having understood, why gauge variant lattice correlators behave
unexpectedly from the perturbation theory point of view, one can search
for the `true' gauge fixing procedure. This constitutes the main goal of
this note. We propose a {\it zero--momentum Lorentz gauge} (ZML), which
permits to get rid of the lattice artifacts and provides correct values for
various correlation functions.
We are going to compare also the standard Lorentz gauge fixing procedure (LG)
\cite{mo1} with the axial Lorentz gauge (ALG) proposed in \cite{nak2}. We
show that ALG produces just the same problems as LG does and,
therefore, cannot resolve the Gribov ambiguity problem.
\section{Gauge fixing : Lorentz gauge and axial Lorentz gauge}
\setcounter{equation}{0}
The standard Wilson action with $~U(1)~$ gauge group is \cite{wil}
\begin{equation}
S(U ) = \beta \sum_x \sum_{\mu > \nu }
\Bigl( 1 - \cos \theta_{x,\mu\nu} \Bigr) ~,
\label{action_u1}
\end{equation}
\noindent where the link variables are
$~U_{x\mu} =\exp (i\theta_{x\mu}) \in U(1)~$ and
$~\theta_{x \mu} \in (-\pi, \pi]~$. The plaquette angles are given by
$~\theta_{x;\mu\nu} = \theta_{x;\mu} + \theta_{x+{\hat \mu};\nu}
- \theta_{x+{\hat \nu};\mu} - \theta_{x\nu}~$.
This action makes the part of the full QED action $S_{QED}$, which is
supposed to be compact if we consider QED as arising from a subgroup of
a non--abelian (e.g., grand unified) gauge theory \cite{pol2}.
The plaquette angle $~\theta_P \equiv \theta_{x;\, \mu \nu}~$
can be split up: $~\theta_P = [\theta_P] + 2\pi n_P$, where
$~[\theta_{P}] \in (-\pi ;\pi ]$ and $n_P = 0, \pm 1, \pm 2$.
The plaquettes with $~n_P \neq 0~$ are called Dirac plaquettes.
The dual integer valued plaquettes $~m_{x,\mu \nu} =
\frac {1}{2} \varepsilon_{\mu \nu \rho \sigma} n_{x,\rho \sigma}~$
form Dirac sheets \cite{dgt}.
In our calculations we monitored the total number of Dirac plaquettes
$N_{DP}^{(\mu\nu)}$ for every plane $(\mu;\nu)=(1;1),\ldots ,(3;4)$
and $~ N_{DP} \equiv \max_{(\mu\nu)} N_{DP}^{(\mu\nu)}$.
The appearance of periodically closed double Dirac sheet means
that at least for one out of six planes $(\mu\nu)$ the number of Dirac
plaquettes $N_{DP}^{(\mu\nu)}$ should be
\begin{equation}
N_{DP}^{(\mu\nu)} \ge 2\frac{V_4}{N_{\mu}N_{\nu}}~.
\end{equation}
\noindent For example, on the lattice $~12\times 6^3~$ the appearance of DDS
means $~N_{DP} \ge 72$.
In lattice calculations the usual choice of the Lorentz (or Landau)
gauge is
\begin{equation}
\sum_{\mu=1}^4 {\bar \partial}_{\mu} \sin \theta_{x\mu} = 0,
\label{gf1}
\end{equation}
\noindent which is equivalent to finding an extremum of the
functional $F(\theta)$
\begin{equation}
F(\theta) = \frac{1}{V_4} \sum_{x} F_{x}(\theta )~;
\qquad F_{x}(\theta) = \frac{1}{8}\sum_{\mu=1}^4
\Bigl[ \cos \theta_{x\mu}+ \cos \theta_{x-{\hat \mu};\mu}\Bigr]
\label{gf2}
\end{equation}
\noindent with respect to the (local) gauge transformations
\begin{equation}
U_{x\mu} \longrightarrow \Lambda_{x} U_{x \mu} \Lambda_{x+{\hat \mu}}^{\ast}~;
~~\Lambda_x=\exp\{i\Omega_x\} \in U(1)~.
\label{gauge_tr}
\end{equation}
The standard gauge fixing procedure (referred in what follows as LG)
consists of the maximization of the value $~F_x(\theta)~$ at some site
$~x~$ under the local gauge transformations $~\Lambda_x~$, then at
another site and so on. After a certain number of gauge fixing sweeps a
local maximum $~F_{max}~$ of the functional $~F(\theta^{\Lambda})~$
is reached.
In order to improve the convergence of the iterative gauge fixing procedure
one can use the optimized overrelaxation procedure \cite{mo3}
with some parameter $\alpha$
which depends on the volume and $~\beta~$.
For convergence criteria we use
$$ \max \left |\sum_{\mu} {\bar \partial}_\mu\sin \theta_{x\mu} \right |< 10^{-5}
\qquad \mbox{and} \qquad
\frac{1}{V} \sum_x \left|\sum_\mu {\bar \partial}_\mu\sin \theta_{x\mu}
\right |< 10^{-6}~.$$
\noindent In Figure \ref{fig:fmax_12x06_b01p10_LG}a we show
a typical Monte Carlo time history of $~F_{max}~$
obtained using the standard Lorentz gauge fixing procedure.
Partially, the comparatively big dispersion of $~F_{max}~$
is due to periodically closed double Dirac sheets \cite{bmmp,clas}.
In Figure \ref{fig:fmax_12x06_b01p10_LG}b we show the corresponding
time history of $~ N_{DP} = \max_{(\mu\nu)} N_{DP}^{(\mu\nu)}~$.
For the given lattice size and $\beta$--value $~\sim 20\%~$
of configurations turn out to possess DDS. Exactly because of the
configurations containing DDS the transverse photon correlator
\begin{equation}
\Gamma_T(\tau;{\vec p})~=~\langle \Phi(\tau;{\vec p}) ~\Phi^{\ast}(0;{\vec p}) \rangle ~,
\qquad \Phi(\tau;{\vec p})~=~
\sum_{{\vec x}} \exp(i {\vec p} {\vec x} +\frac{i}{2}p_{\mu}) ~\sin \theta_{\tau {\vec x}, \mu}
\end{equation}
\noindent $(~\mu=1,3, ~~{\vec p}=(0,p,0)~) $ exhibits an unphysical `tachion--like'
behavior \cite{bmmp,clas,bddm}. In Figure
\ref{fig:pcr_12x06_b01p10_3gauge} we show the
normalized photon correlator $~\Gamma_T(\tau;{\vec p}) / \Gamma_T(0;{\vec p})~$
for lowest non-vanishing momentum and LG together with results obtained with
other Lorentz gauge fixing procedures to be discussed lateron. We
clearly see for LG the deviation from the expected zero-mass behavior.
All the observations described above have not been seen changing, when
$~\beta~$ and/or the lattice size were increased considerably
\cite{bddm}.
It is a long--standing believe \cite{zwan} that the
`true' gauge copy corresponds to the {\it absolute} maximum of $~F(\theta)$.
Thus it would be highly desirable to find a Lorentz gauge fixing
procedure which produces a (unique) gauge copy of the given configuration
with the absolute maximum of $~F(\theta)~$.
However, if one repeatedly subjects a configuration
$~\bigl\{\theta_{x\mu}\bigr\}~$ to a random gauge transformation as in
Eq.(\ref{gauge_tr}) and then subsequently applies to it the LG
procedure, one usually obtains gauge (Gribov) copies with different
values of $~F_{max}$.
An attempt to resolve this problem has been made in \cite{nak2} where
a modified Lorentz gauge fixing procedure has been proposed. This
procedure (which we refer to as the axial Lorentz gauge fixing or ALG)
consists of the two steps:
\begin{itemize}
\item[i)] first transform every configuration to satisfy a maximal tree
temporal gauge condition (`axial' gauge) \cite{creu};
\item[ii)] then apply the Lorentz gauge
fixing procedure.\footnote{In \cite{nak2} an analytically known
local solution of the Lorentz gauge condition has been applied, instead
of an iteratively obtained one. But, as we have convinced ourselves, this
solution does not resolve the problems we are discussing here.}
\end{itemize}
\noindent
An axial gauge with a chosen maximal tree is unique by definition.
In practice, this is easily checked by random gauge
transformations applied first. Consecutive gauge fixing steps
-- e.g. the Lorentz gauge iterations --
will lead always to the same result as long as we do not change the
detailed prescription for these steps. The question is, whether this
'unique' Lorentz gauge obtainable for each gauge field configuration
resolves the problems mentioned above.
The answer is 'no'. We do not find the absolute maximum of the gauge
functional in the majority of the cases. There is a quite high percentage
of Gribov copies left containing DDS (around $~10 \%~$ for $~\beta = 1.1~$
and a $~12\times 6^3~$ lattice). As a consequence the transverse non-zero
momentum photon propagator does not come out correctly again. The
corresponding data are shown in Figure \ref{fig:pcr_12x06_b01p10_3gauge}, too.
Doubling of the linear lattice size and enlarging $~\beta~$ (we checked
$~\beta=2.0~$) do not improve the behavior.
In paper \cite{bmmp} a Lorentz gauge fixing prescription
with a preconditioning step based on a non-periodic gauge transformation
was proposed. The latter has been chosen in such a way that
the spatial Polyakov loop averages were transformed into real numbers
as a first step. We convinced ourselves that the Lorentz gauge fixed
configurations with very high probability did not contain DDS.
As a consequence, the photon correlator becomes correct.
However, this gauge fixing procedure does not provide the absolute
maximum of the Lorentz gauge functional, too.
Thus, the lattice Gribov problem for QED in the Coulomb phase has not
been solved.
\section{Zero--momentum Lorentz gauge (ZML)}
\setcounter{equation}{0}
As is well known, apart from the local symmetry Wilson action $~S~$ has an
additional (global) symmetry with respect to non--periodic transformations
\begin{equation}
U_{x\mu} \longrightarrow {\bar \Lambda}_x U_{x\mu} {\bar\Lambda}^{\ast}_{x+{\hat \mu}}
= U_{x\mu} \cdot e^{-ic_{\mu}}~;
\qquad
{\bar\Lambda}_x = e^{i\sum_{\mu}c_{\mu}x_{\mu}}~,
\label{global1}
\end{equation}
\noindent or equivalently
\begin{equation}
\theta_{x\mu} \longrightarrow \theta_{x\mu} -c_{\mu}~.
\label{global2}
\end{equation}
\noindent Note that these transformations do not spoil the periodicity of the
gauge fields $~U_{x\mu}~$ and $~\theta_{x\mu}~$,
respectively. The transformation in
Eq.(\ref{global2}) changes the zero--momentum mode $~\phi_{\mu}~$ of the
link angle $~\theta_{x\mu} \equiv \phi_{\mu} + \delta\theta_{x\mu}~$, where
$~\sum_x \delta\theta_{x\mu}=0~$, and a proper choice of the parameters
$~c_{\mu}~$ can make $~\phi_{\mu}~$ equal to zero.
It is rather evident that for the infinitesimal fluctuations
$~\delta\theta_{x\mu}~$ about the zero--momentum mode $~\phi_{\mu}~$
the absolute maximum of the functional $~F(\theta)~$ corresponds to
the case $~\phi_{\mu}=0~$.
Our main statement (to be proved in this section) is that in most of
the cases ($\mbox{}_{\textstyle \sim}^{\textstyle > } 99.99\%$) gauge copies of the given configuration
$~\theta_{x\mu}~$ are due to $~a)$ double Dirac sheets; $~b)$
the zero--momentum modes of this field.
It is the exclusion of DDS and of the zero--momentum modes that
permits to obtain a gauge copy of the given configuration with the
{\it absolute} maximum of the functional $~F(\theta)~$.
We define our gauge fixing prescription (which we refer to as the
zero--momentum Lorentz gauge or ZML gauge) as follows. Every iteration
consists of one sweep with (global) transformations as in
Eq.(\ref{global2}) and one sweep with local
gauge transformations as in the standard Lorentz gauge fixing procedure.
In Figures \ref{fig:fmax_12x06_b01p10_ZML}a,b we show time histories of
$~F_{max}~$ and of $~N_{DP}~$ for the gauge fixing
procedure with the minimization of the zero--momentum modes of the gauge
field. Both the time histories should be compared with the corresponding
ones for the standard Lorentz gauge shown in Figures
\ref{fig:fmax_12x06_b01p10_LG}a,b. One can see that after
suppression of the zero--momentum modes the average value of
$~F_{max}~$ became essentially larger as compared
with the LG case. At the same time the number of DDS (to be precise,
the number of configurations with $~N_{DP} \ge 72$) has drastically decreased
($\sim 1\%$). Typically, those configurations with DDS have smaller values
of $~F_{max}~$ than configurations without DDS have.
It is a very easy task to remove the remaining DDS. If a DDS in a ZML gauge
has really appeared, then perform a random gauge transformation to the same
field configuration and repeat the ZML procedure again. As a result
gauge field configurations do not contain DDS.
To convince ourselves that the ZML gauge fixing procedure provides an
absolute maximum of the functional $~F(\theta)~$ we generated
many random gauge copies for every thermalized configuration
$~\{\theta_{x\mu}\}~$. The number of these random gauge copies
$~N_{RC}~$ varied between $~10~$ and $~1000~$ for different
$~\beta$'s and lattices.
Let $~\Bigl\{\theta^{(j)}_{x\mu}\Bigr\}~$ be the $~j^{th}~$ gauge copy
of the configuration $~\Bigl\{ \theta_{x\mu} \Bigr\}~$ obtained with
the random gauge transformation as in Eq.(\ref{gauge_tr}), $~j=1,\ldots ,
N_{RC}~$.
For any configuration $~\Bigl\{ \theta_{x\mu} \Bigr\}~$ we define a
`variance' $~\delta F_{max}(\theta)~$ of $~F_{max}(\theta)~$
\begin{equation}
\delta F_{max}(\theta) =
\max_{ij}\Bigl( F_{max}(\theta^{(i)}) - F_{max}(\theta^{(j)}) \Bigr)~;
~~ i,j =1,\ldots , N_{RC}~.
\end{equation}
\noindent Different gauge copies can, in principle, have different values
of $~F_{max}(\theta)~$. Therefore, its `variance'
$~\delta F_{max}(\theta)~$ can be non--zero.
For example, the standard gauge fixing procedure (LG) described above
gives typical values $~\delta F_{max}(\theta)\sim 0.06\div 0.07~$
for $~N_{RC}~=~10~$.
As an example we show in Figure \ref{fig:var_12x06_b01p10}a a time
history of the `variance' $~\delta F_{max}(\theta)~$ for the standard
LG and for ZML. In the case of ZML this `variance' is {\it zero}
\footnote{In rather rare cases ($\mbox{}_{\textstyle \sim}^{\textstyle < } 0.01\%$) in ZML gauge $~\delta
F_{max}(\theta)\ne 0~$. So far we have no explanation for it.}.
In Figure \ref{fig:var_12x06_b01p10}b we compare values of $~F_{max}~$
for these two gauges. The lower broken line for the LG corresponds to
the first gauge copy which is just a thermalized configuration produced
by our updating subroutine. Applying random gauge transformations we
produced different random gauge copies ($N_{RC}=10$ in this case)
and then chose the gauge copy with the maximal value of $~F_{max}~$. The
corresponding values are represented by the solid LG line in Figure
\ref{fig:var_12x06_b01p10}b. One can see that even this is far below the
solid line which corresponds to the ZML gauge. Increasing the number
$N_{RC}$ up to $1000$ does not change this conclusion.
One reason for the non--zero value of the `variance' is the appearance of
gauge copies containing double Dirac sheets \cite{bmmp,clas,bddm}.
Gauge copies with double Dirac sheets have typically much lower values
of $~F_{max}(\theta)~$ as compared to $~F_{max}~$ for gauge copies
without DDS. Moreover, a contribution of configurations with DDS
`spoils' the photon correlator $~\Gamma_T(\tau;{\vec p})~$ with $~{\vec
p}\ne 0~$ which leads to a wrong dispersion relation inconsistent with
the dispersion relation for the massless photon. Double Dirac sheets
represent a spectacular example of lattice artifacts which can lead to
a misleading interpretation of the results of numerical calculations
(see also \cite{neu}).
But, as we have seen, the mere exclusion of gauge copies with DDS in the
standard Lorentz gauge fixing procedure does {\it not} yet provide a
zero value of the `variance' $~\delta F_{max}(\theta)~$.
Different gauge copies (without DDS) still have different values
of $~F_{max}(\theta)~$. The exclusion of the zero-momentum modes turns
out to be sufficient to remove the ambiguity.
\section{Conclusions} \setcounter{equation}{0}
Now let us summarize our findings.
\vspace{0.5cm}
Our main result is that $~\sim 99.99\%~$ (if not all) gauge copies for the
given gauge field configuration are due to two reasons :
\begin{itemize}
\item[a)] periodically closed double Dirac sheets;
\item[2)]
the zero--momentum mode of the gauge field
$~\theta_{x\mu}~$.
\end{itemize}
\noindent We didn't find any other reason for the appearance of the gauge
copies.
\noindent The minimization of the zero--momentum mode can be performed
sweep by sweep using a global transformation as in Eq.(\ref{global2}).
We proposed a modified gauge fixing procedure consisting of local
gauge transformations in Eq.(\ref{gauge_tr}) and global transformations
in Eq.(\ref{global2}) (ZML gauge).
The exclusion of the DDS (which appear rather rarely in the ZML gauge) can
be easily performed on the algorithmic level as described in the text.
The application of the ZML gauge fixing procedure provides us with
{\it absolute} maximum of the functional $~F(\theta)~$.
We have shown that the gauge fixing procedure with axial gauge
preconditioning (ALG) cannot solve the problem of the Gribov ambiguity
in this theory. The axial gauge preconditioning cannot exclude the
appearance of DDS as well as of nonzero values of the zero--momentum
mode.
In this paper we present our results only for the bosonic sector
of the theory. However, we believe that this study solves ultimatively
the Gribov ambiguity puzzle in the case of quenched compact QED within
the gauge as well as the fermionic sector.
Work on the fermion case is in progress \cite{zmlf}.
The inclusion of dynamical fermions changes somewhat the symmetry group
of the full action, i.e. the parameters $~c_{\mu}~$ in Eq.'s
~(\ref{global1}),~(\ref{global2}) can have only discrete values.
This case needs some additional study.
\section*{Acknowledgements} \setcounter{equation}{0}
Financial support from grant INTAS-96-370, RFBR grant 99-01-01230
and the JINR Heisenberg-Landau program is kindly acknowledged.
\vspace{0.5cm}
|
\section*{1. Introduction}
\setcounter{section}{1}
\setcounter{equation}{0}
\noindent
The generalized H\'{e}non-Heiles Hamiltonian is
\begin{equation}
H=\frac{1}{2}\left(\dot x^2+\dot y^2+Ax^2+By^2\right)+
Dx^2y^{m-2}-\frac{C}{m}y^m,
\end{equation}
where $\dot x=\frac{dx}{dt}$, $\dot y=\frac{dy}{dt}$
and $A$, $B$, $C$
and $D$ are nonzero parameters. It corresponds to the original form
given by H\'{e}non and Heiles [1] in their studies of astronomical
problems when $A=B=C=D=1$ and $m=3$. Here, we consider the cubic form
($m=3$) of (1.1), with the corresponding equations of motion
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\ddot x+Ax+2Dxy=0,
\vspace{1mm}\\
\ddot y+By+Dx^2-Cy^2=0.
\end{array}
\end{equation}
Several studies have been made regarding the integrability properties
and singularity structures of various versions of (1.2); for example,
Chang, Tabor and Weiss~[2] have analy\-sed~(1.2) with $A=B=1$, whereas
Chang, Greene, Tabor and Weiss [3] have considered the same system
for general~$m$. In keeping with their notation, we rescale the
variables,
\begin{equation}
x\rightarrow \frac{x}{C},\qquad y\rightarrow \frac{y}{C},\qquad
\lambda=\frac{D}{C},
\end{equation}
to express (1.2) as
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\ddot x+Ax+2\lambda xy=0,
\vspace{1mm}\\
\ddot y+By+\lambda x^2-y^2=0.
\end{array}
\end{equation}
The leading-order behaviours of solutions about an arbitrary
singularity, as well as their resonance structures (in the sense of
the ARS algorithm, Ablowitz, Ramani and Segur [4, 5]), have been
derived (see [2, 3]; Chang, Tabor, Weiss and Corliss [6]; Bountis,
Segur and Vivaldi [7]; Weiss [8]; Tabor [9, p. 337]).
We summarize the results brief\/ly, in order to f\/ix notation and to
single out the present objectives. Setting $\tau=t-t_0$, where $t_0$
is an arbitrary constant (the location of the singularity), and
\begin{equation}
x\sim a\tau^{\alpha},\qquad y\sim b\tau^{\beta},\qquad \tau\rightarrow 0,
\end{equation}
two types of leading-order singular behaviours are found:
\begin{equation}
\hspace*{-5pt}\begin{array}{ll}
\mbox{(i) }& \alpha=\beta=-2,
\vspace{1mm}\\
\mbox{(ii) }& \beta=-2, \quad Re(\alpha)>\beta.
\end{array}
\end{equation}
Carrying the analysis further by writing
\begin{equation}
x\sim a\tau^{\alpha}+p\tau^{\alpha+k_r},\qquad
y\sim b\tau^{\alpha}+q\tau^{\beta+k_r},
\end{equation}
and determining those values of $k_r$ for which $p$ or $q$ is
undetermined, one f\/inds the corresponding resonances. It turns out
that
\begin{equation}
\hspace*{-5pt}\begin{array}{ll}
\mbox{(i)}& \alpha=\beta=-2,\quad k_r=-1, 6, r, \bar{r},
\vspace{2mm}\\
&\displaystyle \mbox{where}\quad r,\bar{r}=\frac{5}{2}\pm\frac{1}{2}
\sqrt{1-24(\frac{1}{\lambda}+1)};
\vspace{3mm}\\
\mbox{(ii)} & \displaystyle Re(\alpha)>-2,\quad \beta=-2,\quad \alpha,
\bar{\alpha}=\frac{1}{2}\pm\frac{1}{2}\sqrt{1-48\lambda},\quad k_r=-1, 6, 0, r,
\bar{r},
\vspace{3mm}\\
&\mbox{where}\quad r, \bar{r}=\mp\sqrt{1-48\lambda}.
\end{array}
\end{equation}
In case (ii), $\alpha$ and $\bar{\alpha}$ are two possible leading orders,
with corresponding resonances $r$ and~$\bar{r}$. This case occurs
only for $\lambda>-\frac{1}{2}$, since $Re(\alpha)>-2$.
The nature of the resonances for each value of $\lambda$ is
summarized as follows:
\medskip
\noindent
{\bf Case(i)}
\[
\hspace*{-5pt}\begin{array}{lcl}
\displaystyle \lambda<-\frac{24}{23} & : & r \quad \mbox{and} \quad
\bar{r} \quad \mbox{are complex}\vspace{3mm} \\
\displaystyle \lambda=-\frac{24}{23} & : & \displaystyle r=\bar{r}=\frac{5}{2}\vspace{3mm}\\
\displaystyle -\frac{24}{23}<\lambda<-\frac{1}{2} &:& r>0,\quad \bar{r}>0,\quad
r\neq\bar{r}\vspace{3mm}\\
\displaystyle \lambda=-\frac{1}{2}&:& r=5,\quad \bar{r}=0\vspace{3mm}\\
\displaystyle -\frac{1}{2}<\lambda<0 &:& r>0,\quad \bar{r}<0\vspace{3mm}\\
\lambda>0 &:& r \quad \mbox{and} \quad \bar{r} \quad \mbox{are complex}
\end{array}
\]
{\bf Case (ii)}
\[
\hspace*{-5pt}\begin{array}{lcl}
\displaystyle -\frac{1}{2}<\lambda<0& :& r<0,\quad \bar{r}>0\vspace{3mm}\\
\displaystyle 0<\lambda<\frac{1}{48} &:& r<0,\quad \bar{r}>0\vspace{3mm}\\
\displaystyle \lambda=\frac{1}{48} &:& r= \bar{r}=0\vspace{3mm}\\
\displaystyle \lambda>\frac{1}{48}&:& r \quad \mbox{and} \quad \bar{r} \quad
\mbox{are pure imaginary}
\end{array}
\]
Those cases where a negative resonance is present (in addition to $-1$)
likely correspond to singular, rather than general, solutions (see,
e.g., [9, p.~339]). The signif\/icance of repeated resonances (case~(i),
$\lambda=-\frac{24}{23}$, and case (ii), $\lambda=\frac{1}{48}$) is,
at present, unknown. The case $\lambda=-\frac{1}{2}$ is somewhat
anomalous [9, p.~340] and we shall comment upon it later. In those
cases where there exist four distinct resonances, the usual $-1$ and
three others with nonnegative real parts, general solutions may be
written down in series expansions about the arbitrary singularity
$t_0$. These may be Laurent series, if the system passes the
Painlev\'{e} test~[5] or, otherwise, psi-series (Hille~[10, p.~249]).
It has been found [2] that the system is integrable when $\lambda=-1$
and $A=B$, when $\lambda=-\frac{1}{6}$, and when $\lambda=-\frac{1}{16}$ and
$B=16A$. In the latter case, although the system is integrable, it
passes only the ``weak'' Painlev\'{e} test in that it admits a
rational movable branch point. Outside of these three cases, the
system possesses solutions with movable branch points (logarithmic,
irrational or complex), and a psi-series expansion is required to
represent the general solution about an arbitrary branch point. Such
expansions constitute formal general solutions containing four
arbitrary constants, the remaining coef\/f\/icients being well def\/ined by
a self-consistent recursion relation.
In this paper, we prove the absolute convergence of such series on
real intervals of the form $0<\tau<R$, $R>0$, thus establishing that
the formal solutions are actual solutions. Similar results have been
obtained by Melkonian and Zypchen~[11] for the Lorenz system (see,
e.g.,~[9, p.~344] and Sparrow~[12]), where only logarithmic
psi-series occur. A study of the convergence problem for a class of
second- and third-order ordinary dif\/ferential equations has been made
by Hemmi and Melkonian~[13]. Some of the key ideas within the proofs
may be extracted from a paper by Hille~[14] regarding second-order
quadratic systems. Studies of general systems of partial dif\/ferential
equations which admit WTC (Weiss, Tabor and Carnevale~[15])
expansions containing logarithms have been made by Kichenassamy and
Srinivan [16] and Kichenassamy and Littman~[17], who have proved
convergence by entirely dif\/ferent methods. Logarithmic psi-series
occur for the Cubic H\'{e}non Heiles system~(1.4) when $\lambda=-1$ and
$A\neq B$, or when $\lambda=-\frac{1}{16}$ and $B\neq 16A$, and these may
be dealt with by the methods discussed in [11, 13, 16, 17].
Here, we restrict our attention to the non-logarithmic cases.
Sections~2 and~3 deal with the case(i)-leading orders, involving a
series with complex exponents (Section~2) and one with irrational
exponents (Section 3). The necessary Lemmas are relegated to Appendix~A.
The series which we employ in Sections~2 and~3 are not the same as
the ones given, e.g., in~[3], and we conf\/irm the validity
(self-consistency) of our version in Appendix~B. Section~4 concerns
the case(ii)-leading orders where, once again, we employ a new
series, the validity of which is conf\/irmed in Appendix~C. Concluding
remarks are made in Section~5, including a discussion of the case
$\lambda=-\frac{1}{2}$.
\section*{2. Case(i)-leading orders:
{\mathversion{bold}
$\alpha=\beta=-2$,\\[2mm]
\hspace*{20pt}$k_r=-1, 6,
\frac{5}{2}\pm\frac{1}{2}\sqrt{1-24\left(\frac{1}{\lambda}+1\right)}$}
complex}
\setcounter{section}{2}\setcounter{equation}{0}
\noindent
Here, the resonances $r,
\bar{r}=\frac{5}{2}\pm\frac{1}{2}\sqrt{1-24(\frac{1}{\lambda}+1)}$ are complex,
with $\lambda<-\frac{24}{23}$ or $\lambda>0$.
The general solution may be taken in the form
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\sum_{m=0}^{\infty}
a_{klm}\tau^{k-2+rl+\bar{r}m},\vspace{3mm}\\
\displaystyle y=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\sum_{m=0}^{\infty}b_{klm}
\tau^{k-2+rl+\bar{r}m}.
\end{array}
\end{equation}
The self-consistency of (2.1), as well as the presence of four
arbitrary constants, is shown in Appendix~B. It is also shown therein
that (2.1) is equivalent to
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\left[a_{kl}\tau^{k-2+rl}+\bar{a}_{kl}\tau^{k-2+\bar{r}l}\right],
\vspace{3mm}\\
\displaystyle y=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\left[b_{kl}\tau^{k-2+rl}+\bar{b}_{kl}\tau^{k-2+\bar{r}l}\right],
\end{array}
\end{equation}
and that (2.2) is equivalent to the solution given in [3]. Either
(2.1) or (2.2) may be employed to prove convergence. However, the
advantage of starting with (2.1) is that it is much easier to deal
with a single, triply-indexed series such as (2.1) than a sum of two,
doubly-indexed ones such as (2.2), vis-\`{a}-vis the conf\/irmation of
its validity.
The convergence proof consists of resumming (2.2) into more tractable
forms, and showing that the latter are majorized by series which
converge on an interval, by the ratio test. Thus, let
$\varepsilon=\frac{1}{2}\sqrt{24(\frac{1}{\lambda}+1)-1}>0$, note that $r-1,
\bar{r} -1=\frac{3}{2}\pm i\varepsilon$, and write
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\left[a_{kl}\tau^{k-2+l+\left(\frac{3}{2}+i\varepsilon\right)l}
+\bar{a}_{kl}\tau^{k-2+l+\left(\frac{3}{2}-i\varepsilon\right)l}\right]
\vspace{3mm}\\
\displaystyle \qquad
=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\left[(a_{kl}+
\bar{a}_{kl})\cos(\varepsilon l z)+i(a_{kl}-\bar{a}_{kl})\sin(\varepsilon
lz)\right]e^{\frac{3}{2}lz}\tau^{k-2+l},
\end{array}
\end{equation}
where $z=\ln(\tau)$, $\tau>0$. It follows that
\begin{equation}
x=\sum_{\gamma=0}^{\infty}f_{\gamma}(z)\tau^{\gamma-2},
\end{equation}
where
\begin{equation}
f_{\gamma}(z)=\sum_{k+l=\gamma}\left[(a_{kl}+\bar{a}_{kl})\cos(\varepsilon
lz)+i(a_{kl}-\bar{a}_{kl})\sin(\varepsilon lz)\right]e^{\frac{3}{2}lz}
\end{equation}
is a polynomial in $e^{\frac{3}{2}z}$ of degree $\gamma$, with
coef\/f\/icients that are bounded functions of $z$. Similarly,
\begin{equation}
y=\sum_{\gamma=0}^{\infty}h_{\gamma}(z)\tau^{\gamma-2},
\end{equation}
where
\begin{equation}
h_{\gamma}(z)=\sum_{k+l=\gamma}\left[(b_{kl}+\bar{b}_{kl})\cos(\varepsilon
lz)+i(b_{kl}-\bar{b}_{kl})\sin(\varepsilon lz)\right]e^{\frac{3}{2}lz}
\end{equation}
is a polynomial of the same type as $f_{\gamma}(z)$.
Let $u(t)=\dot{x}(t)\equiv\frac{dx}{dt}$ and
$v(t)=\dot{y}(t)\equiv\frac{dy}{dt}$, then
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle u=\dot x=\sum_{\gamma=0}^{\infty}g_{\gamma}(z)
\tau^{\gamma-3}=\sum_{\gamma=0}^{\infty}\left[(\gamma-2)f_{\gamma}
+f_{\gamma}'\right]\tau^{\gamma-3},\vspace{3mm}\\
\displaystyle v=\dot
y=\sum_{\gamma=0}^{\infty}k_{\gamma}(z)\tau^{\gamma-3}=
\sum_{\gamma=0}^{\infty}\left[(\gamma-2)h_{\gamma}+h_{\gamma}'\right]\tau^{\gamma-3},
\end{array}
\end{equation}
where $f'_{\gamma}\equiv\frac{d f_{\gamma}}{dz}, h'_{\gamma}\equiv\frac{d
h_{\gamma}}{dz}$, and $g_{\gamma}$ and $k_{\gamma}$ are also polynomials of the
same type as $f_{\gamma}$ and~$h_{\gamma}$.
Express (1.4) as a system of four f\/irst-order equations in $x, u, y,
v$ to obtain
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\dot x=u,
\vspace{1mm}\\
\dot u=-Ax-2\lambda xy,
\vspace{1mm}\\
\dot y=v,
\vspace{1mm}\\
\dot v=-By-\lambda x^2+y^2.
\end{array}
\end{equation}
Substitution of (2.4), (2.6) and (2.8) into (2.9) gives, for
$\gamma\geq 0$,
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle f_{\gamma}'+(\gamma-2)f_{\gamma}-g_{\gamma}=0,
\vspace{3mm}\\
\displaystyle g_{\gamma}'+(\gamma-3)g_{\gamma}+
Af_{\gamma-2}+2\lambda\sum_{\mu=0}^{\gamma}f_{\gamma-\mu}h_{\mu}=0,
\vspace{3mm}\\
h_{\gamma}'+(\gamma-2)h_{\gamma}-k_{\gamma}=0,
\vspace{3mm}\\
\displaystyle k_{\gamma}'+(\gamma-3)k_{\gamma}+Bh_{\gamma-2}+\lambda\sum_{\mu=0}^{\gamma}f_{\gamma-\mu}f_{\mu}
-\sum_{\mu=0}^{\gamma}h_{\gamma-\mu}h_{\mu}=0.
\end{array}
\end{equation}
For $\gamma=0$, we f\/ind
\begin{equation}
f_0=\pm\frac{3}{\lambda}\sqrt{\frac{1}{\lambda}+2},\qquad g_0=-2f_0,
\qquad h_0=-\frac{3}{\lambda},\qquad k_0=\frac{6}{\lambda}.
\end{equation}
For $\gamma>0$, (2.10) is expressible as
\begin{equation}
\vec{f}_{\gamma}'+\mbox{\mathversion {bold}$ A$}_{\gamma}\vec{f}_{\gamma}=\vec{F}_{\gamma},
\end{equation}
where
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle\vec{f}_{\gamma}=\left(\begin{array}{c}f_{\gamma}\\
g_{\gamma}\\h_{\gamma}\\k_{\gamma}\end{array}\right),
\qquad
\mbox{\mathversion {bold}$ A$}_{\gamma}=\left(\begin{array}{cccc}\gamma-2 & -1 &0&0\vspace{2mm}\\
-6&\gamma-3&2\lambda f_0&0\vspace{2mm}\\
0&0&\gamma-2&-1\vspace{2mm}\\
2\lambda f_0&0&\displaystyle \frac{6}{\lambda}&\gamma-3\end{array}\right),
\vspace{4mm}\\
\vec{F}_{\gamma}=\left(\begin{array}{c}
0\vspace{3mm}\\
\displaystyle -Af_{\gamma-2}-2\lambda\sum\limits_{\mu=1}^{\gamma-1}f_{\gamma-\mu}h_{\mu}
\vspace{3mm}\\
0\vspace{3mm}\\
\displaystyle -Bh_{\gamma-2}-\lambda\sum\limits_{\mu=1}^{\gamma-1}f_{\gamma-\mu}f_{\mu}
+\sum\limits_{\mu=1}^{\gamma-1}h_{\gamma-\mu}h_{\mu}\end{array}\right)
\equiv\left(\begin{array}{c}F_{\gamma}\\
G_{\gamma}\\H_{\gamma}\\K_{\gamma}\end{array}\right).
\end{array}
\end{equation}
The eigenvalues of $\mbox{\mathversion {bold}$ A$}_{\gamma}$ are $\gamma+1, \gamma-6$ and
$\gamma-\left(\frac{5}{2}\pm i\varepsilon\right)$, precisely $\gamma-k_r$ where $k_r$
is a resonance. The matrices $\mbox{\mathversion {bold}$ A$}_{\gamma}$ are simultaneously
diagonalizable by a matrix $\mbox{\mathversion {bold}$ P$}$ independent of $\gamma$. $\mbox{\mathversion {bold}$ P$}$ is
the matrix of eigenvectors,
\begin{equation}
\mbox{\mathversion {bold}$ P$}=\left(\begin{array}{cccc}
-\lambda f_0&-\lambda f_0&\lambda f_0&\lambda f_0
\vspace{3mm}\\
3\lambda f_0&-4\lambda f_0&\displaystyle \lambda f_0\left(\frac{1}{2}+i\varepsilon\right)& \displaystyle \lambda
f_0\left(\frac{1}{2}-i\varepsilon\right)
\vspace{3mm}\\
3&3&\displaystyle 3\left(2+\frac{1}{\lambda}\right)&\displaystyle 3\left(2+\frac{1}{\lambda}\right)
\vspace{3mm}\\
-9&12& \displaystyle 3\left(2+\frac{1}{\lambda}\right)\left(\frac{1}{2}
+i\varepsilon\right)&\displaystyle 3\left(2+\frac{1}{\lambda}\right)\left(\frac{1}{2}
i\varepsilon\right)\end{array}\right),
\end{equation}
and $\mbox{\mathversion {bold}$ P$}^{-1}\mbox{\mathversion {bold}$ A$}_{\gamma}\mbox{\mathversion {bold}$ P$}=\mbox{\mathversion {bold}$ D$}_{\gamma}$, where
\begin{equation}
\mbox{\mathversion {bold}$ D$}_{\gamma}=\left(\begin{array}{cccc}\gamma+1&0&0&0\vspace{2mm}\\
0&\gamma-6&0&0\vspace{2mm}\\
0&0&\displaystyle \gamma-\left(\frac{5}{2}+i\varepsilon\right)&0\vspace{2mm}\\
0&0&0&\displaystyle \gamma-\left(\frac{5}{2}-i\varepsilon\right)
\end{array}\right).
\end{equation}
For $\gamma>6$ (in which case all eigenvalues of $\mbox{\mathversion {bold}$ A$}_{\gamma}$ have positive
real parts), the solution of the matrix dif\/ferential equation (2.12)
may be expressed as
\begin{equation}
\vec{f}_{\gamma}(z)=\int_{-\infty}^z e^{\mbox{\mathversion {bold}$ A$}_{\gamma} (x-z)}\vec{F}_{\gamma}(x)dx
=\int_{-\infty}^z\mbox{\mathversion {bold}$ P$} e^{\mbox{\mathversion {bold}$ D$}_{\gamma} (x-z)}\mbox{\mathversion {bold}$ P$}^{-1}\vec{F}_{\gamma}(x)dx,
\end{equation}
where the exponential within the integrals denotes the exponential of
a matrix.
Given an $m\times n$ matrix $\mbox{\mathversion {bold}$ A$}=(a_{ij})$, def\/ine its norm as
\begin{equation}
\|\mbox{\mathversion {bold}$ A$}\|=\max_{1\leq i\leq m}\left(\sum_{j=1}^n|a_{ij}|\right).
\end{equation}
It follows from (2.16) that
\begin{equation}
\|\vec{f}_{\gamma}(z)\|\leq \|\mbox{\mathversion {bold}$ P$}\|\|\mbox{\mathversion {bold}$ P$}^{-1}\|\int_{-\infty}^{z}e^{(\gamma-6)(x-z)}\|
\vec{F}_{\gamma}(x)\|dx,
\end{equation}
since $\|e^{\mbox{\mathversion {bold}$ D$}_{\gamma}(x-z)}\|=e^{(\gamma-6)(x-z)}$ for $x\leq z$. $\mbox{\mathversion {bold}$ P$}$
depends upon $\lambda$, so that $\|\mbox{\mathversion {bold}$ P$}\|\|\mbox{\mathversion {bold}$ P$}^{-1}\|\leq M(\lambda)$, a
constant (independent of $\gamma$). Thus,
\begin{equation}
\|\vec{f}_{\gamma}(z)\|\leq M(\lambda)
\int_{-\infty}^{z}e^{(\gamma-6)(x-z)}\|\vec{F}_{\gamma}(x)\|dx.
\end{equation}
\noindent
{\bf Theorem 2.1.} {\it There exists $K>0$ such that for all $z<0$ and
$\gamma\geq 1$,}
\begin{equation}
\|\vec{f}_{\gamma}(z)\|\leq\frac{(2K)^{\gamma}}{\sqrt{\gamma+1}}.
\end{equation}
\noindent
{\bf Proof.} Since $f_{\gamma}(z)$ is a polynomial in
$X=e^{\frac{3}{2}z}$ of degree $\gamma$, with coef\/f\/icients that are
bounded functions of $z$ (linear combinations of $\cos(\varepsilon lz)$ and
$\sin(\varepsilon lz)$),
\begin{equation}
|f_{\gamma}(z) | \leq
\sum_{m=0}^{\gamma}a_{m}^{(\gamma)}e^{\frac{3}{2}mz}\equiv P_{\gamma}(X),
\end{equation}
with $a_m^{(\gamma)}\geq 0$ for $0\leq m\leq \gamma$. By Lemma A1 of
Appendix A (with $q=1,\ n_{\gamma}=\gamma,\ M=1$ and $p=1$), given $N>1$,
there exists $K_1>0$ such that
\begin{equation}
|f_{\gamma}(z)|\leq P_{\gamma}(X)\leq
\frac{(K_1+KX)^{\gamma}}{\sqrt{\gamma+1}}
\qquad\mbox{for}\quad 1\leq\gamma\leq N-1.
\end{equation}
Similarly, there exist positive constants $K_2$, $K_3$ and $K_4$
corresponding to the polynomials $g_{\gamma}$, $h_{\gamma}$ and
$k_{\gamma}$, repectively. With $K=\max\limits_{1\leq i\leq 4}\{K_i\}$,
we obtain
\begin{equation}
\|\vec{f}_{\gamma}(z)\|\leq \frac{(K+Ke^{\frac{3}{2}z})^{\gamma}}
{\sqrt{\gamma+1}}\leq\frac{(2K)^{\gamma}}{\sqrt{\gamma+1}}
\qquad\mbox{for}\quad 1\leq\gamma\leq N-1,\ z<0,
\end{equation}
since $0<e^{\frac{3}{2}z}<1$. The inequality (2.23) constitutes our
inductive hypothesis, and we shall show that (2.20) holds for
$\gamma=N$, thus establishing that it holds for all $N>1$.
By (2.13) and the inductive hypothesis (2.23), we f\/ind that
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle |G_N|\leq
|A||f_{N-2}|+2|\lambda|\sum_{\mu=1}^{N-1}|f_{N-\mu}||h_{\mu}|
\vspace{3mm}\\
\displaystyle \phantom{G_N}\leq
|A|\frac{(2K)^{N-2}}{\sqrt{N-1}}+2|\lambda|
\sum_{\mu=1}^{N-1}\frac{(2K)^N}{\sqrt{N-\mu+1}\,\sqrt{\mu+1}}
\vspace{3mm}\\
\displaystyle \phantom{G_N}\leq \left
\{\frac{|A|(2K)^{-2}}{\sqrt{N-1}}+2|\lambda|\pi\right\}(2K)^N
\end{array}
\end{equation}
by Lemma A2 of Appendix A. Similarly,
\begin{equation}
|K_N|\leq\left\{\frac{|B|(2K)^{-1}}{\sqrt{N-1}}+(|\lambda|+1)\pi
\right\}(2K)^N.
\end{equation}
Thus,
\begin{equation}
\|\vec{F}_N\|\leq E(2K)^N,
\end{equation}
where
\begin{equation}
E\leq\max\{|A|+2|\lambda|\pi, |B|+(|\lambda|+1)\pi\}.
\end{equation}
Thus, from (2.19) and (2.26), we obtain
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle \|\vec{f}_{N}(z)\| \leq
M(\lambda)E\int_{-\infty}^z e^{(N-6)(x-z)}(2K)^Ndx
\vspace{3mm}\\
\displaystyle \phantom{\|\vec{f}_{N}(z)\|} =\left\{\frac{M(\lambda)E\sqrt{N+1}}{N-6}
\right\}\frac{(2K)^N}{\sqrt{N+1}}\qquad
\mbox{for}\quad N>6.
\end{array}
\end{equation}
For all suf\/f\/iciently large $N$, $\frac{M(\lambda)E\sqrt{N+1}}{N-6}<1$, so
beginning with such an $N>6$, we obtain
\begin{equation}
\|\vec{f}_N(z)\|\leq \frac{(2K)^N}{\sqrt{N+1}},
\end{equation}
which completes the proof.
By (2.4),
\begin{equation}
x=\sum_{\gamma=0}^{\infty}f_{\gamma}(z)\tau^{\gamma-2},
\end{equation}
and by (2.20),
\begin{equation}
|f_{\gamma}(z)\tau^{\gamma-2}|\leq
\frac{(2K)^{\gamma}}{\sqrt{\gamma+1}}\tau^{\gamma-2},
\end{equation}
and the series
\begin{equation}
\sum_{\gamma=0}^{\infty}\frac{(2K)^{\gamma}}{\sqrt{\gamma+1}}\tau^{\gamma-2}
\end{equation}
converges absolutely by the ratio test for $\tau<\frac{1}{2K}$. Thus,
the series (2.30) for $x$ converges absolutely for $0<\tau<R$, where
$R$ is at least $\frac{1}{2K}$. Similarly, the series for $\dot x,\ y$
and $\dot y$ converge absolutely for $0<\tau<R$.
\section*{3. Case(i)-leading orders: {\mathversion{bold}$\alpha=\beta=-2$,\\[2mm]
\hspace*{20pt} $k_r=-1, 6, \frac{5}{2}\pm\frac{1}{2}\sqrt{1-24\left(\frac{1}
{\lambda}+1\right)}$} irrational}
\setcounter{section}{3}\setcounter{equation}{0}
\noindent
In the present case, the resonances $r,
\bar{r}=\frac{5}{2}\pm\frac{1}{2}\sqrt{1-24\left(\frac{1}{\lambda}+1\right)}$
are irrational, positive and distinct,
with $-\frac{24}{23}<\lambda<-\frac{1}{2}$. The general solution is given
by (2.2) (or (2.1)) as in the complex-resonance case of Section 2,
but the details of the resummation are slightly dif\/ferent since,
in the present case, $r<1$ and/or $\bar{r}<1$ is possible.
Let $n$ be the least positive integer such that $\mu_1=r-\frac{1}{n}>0$
and $\mu_2=\bar r -\frac{1}{n}>0$. (Any positive integer satisfying the
latter two conditions will also do.) Then the solution $x$ in (2.2)
may be expressed as
\begin{equation}
x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\left[a_{kl}\tau^{k-2+\frac{1}{n}l+\mu_1l}
+\bar{a}_{kl}\tau^{k-2+\frac{1}{n}l+\mu_2l}\right].
\end{equation}
Let $w=\tau^{1/n}$ and $z=\ln(\tau)$, $\tau>0$, then
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\left[a_{kl}w^{nk-2n+l}e^{\mu_1
lz}+\bar{a}_{kl}w^{nk-2n+l}e^{\mu_2 lz}\right]
\vspace{3mm}\\
\displaystyle \qquad =\sum_{\gamma=0}^{\infty}\sum_{nk+l=\gamma}\left(a_{kl}e^{\mu_1
lz}+\bar{a}_{kl}e^{\mu_2 lz}\right)w^{\gamma-2n}
=\sum_{\gamma=0}^{\infty}f_{\gamma}(z)w^{\gamma-2n},
\end{array}
\end{equation}
where
\begin{equation}
f_{\gamma}(z)=\sum_{nk+l=\gamma}
\left(a_{kl}e^{\mu_1 lz}+\bar{a}_{kl}e^{\mu_2 lz}\right)
\end{equation}
is a polynomial of degree $\gamma$ in the two variables $e^{\mu_1 z}$
and $e^{\mu_2 z}$. Similarly,
\begin{equation}
y=\sum_{\gamma=0}^{\infty}h_{\gamma}(z)w^{\gamma-2n},
\end{equation}
where
\begin{equation}
h_{\gamma}(z)=\sum_{nk+l=\gamma}\left(b_{kl}e^{\mu_1 lz}
+\bar{b}_{kl}e^{\mu_2 lz}\right)
\end{equation}
is a polynomial of the same type as $f_{\gamma}(z)$.
Let
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle u=\dot x=\sum_{\gamma=0}^{\infty}g_{\gamma}(z)w^{\gamma-3n}
=\sum_{\gamma=0}^{\infty}\left[\left(\frac{\gamma}{n}-2\right)f_{\gamma}+
f_{\gamma}'\right]w^{\gamma-3n},\vspace{3mm}\\
\displaystyle v=\dot
y=\sum_{\gamma=0}^{\infty}k_{\gamma}(z)w^{\gamma-3n}=\sum_{\gamma=0}^{\infty}
\left[\left(\frac{\gamma}{n}-2\right)h_{\gamma}+h_{\gamma}'\right]w^{\gamma-3n},
\end{array}
\end{equation}
where, as before, $\dot x=\frac{dx}{dt},\
f_{\gamma}'=\frac{df_{\gamma}}{dz}$, etc., and $g_{\gamma}$ and $k_{\gamma}$
are polynomials of the same type as $f_{\gamma}$ and $h_{\gamma}$.
Substitution of (3.2), (3.4) and (3.6) into the system (2.9) gives,
for $\gamma=0$,
\begin{equation}
f_0=\pm\frac{3}{\lambda}\sqrt{\frac{1}{\lambda}+2},\qquad
g_0=-2f_0,\qquad h_0=-\frac{3}{\lambda},\qquad k_0=\frac{6}{\lambda},
\end{equation}
(as (2.11)) and, for $\gamma>0$,
\begin{equation}
\vec{f}_{\gamma}'+\mbox{\mathversion {bold}$ A$}_{\gamma}\vec{f}_{\gamma}=\vec{F}_{\gamma}
\end{equation}
(as (2.12)), where $\vec{F}_{\gamma}$ is again given by (2.13), but now,
\begin{equation}
\mbox{\mathversion {bold}$ A$}_{\gamma}=\left(\begin{array}{cccc}
\displaystyle \frac{\gamma}{n}-2&-1&0&0\vspace{2mm}\\
-6&\displaystyle \frac{\gamma}{n}-3&2\lambda f_0&0\vspace{2mm}\\
0&0&\displaystyle \frac{\gamma}{n}-2&-1\vspace{2mm}\\
2\lambda f_0&0&\displaystyle \frac{6}{\lambda}&\displaystyle \frac{\gamma}{n}-3\end{array}\right),
\end{equation}
and coincides with the $\mbox{\mathversion {bold}$ A$}_{\gamma}$ of (2.13) if $n=1$. The eigenvalues
of $\mbox{\mathversion {bold}$ A$}_{\gamma}$ in the present case are
$\frac{\gamma}{n}-k_r$, where $k_r=-1,
6, r, \bar{r}$ are the resonances.
The matrix $\mbox{\mathversion {bold}$ P$}$ which diagonalizes all of the $\mbox{\mathversion {bold}$ A$}_{\gamma}$ is the same
as (2.14) which, for the present situation, is expressed as
\begin{equation}
\mbox{\mathversion {bold}$ P$}=\left(\begin{array}{cccc}
-\lambda f_0 &-\lambda f_0& \lambda f_0& \lambda f_0
\vspace{2mm}\\
3\lambda f_0 &-4\lambda f_0 & \lambda f_0(r-2)&\lambda f_0(\bar{r}-2)
\vspace{2mm}\\
3&3&\displaystyle 3\left(2+\frac{1}{\lambda}\right)&\displaystyle 3\left(2+\frac{1}{\lambda}\right)\vspace{2mm}\\
-9&12&\displaystyle 3\left(2+\frac{1}{\lambda}\right)(r-2)&\displaystyle 3\left(2+\frac{1}{\lambda}\right)
(\bar r-2)\end{array}\right),
\end{equation}
with $\mbox{\mathversion {bold}$ P$}^{-1}\mbox{\mathversion {bold}$ A$}_{\gamma}\mbox{\mathversion {bold}$ P$}=\mbox{\mathversion {bold}$ D$}_{\gamma}$ and
\begin{equation}
\mbox{\mathversion {bold}$ D$}_{\gamma}=\left(\begin{array}{cccc}
\displaystyle \frac{\gamma}{n}+1&0&0&0\vspace{2mm}\\
0&\displaystyle \frac{\gamma}{n}-6&0&0
\vspace{2mm}\\
0&0&\displaystyle \frac{\gamma}{n}-r&0\vspace{2mm}\\
0&0&0&\displaystyle \frac{\gamma}{n}-\bar r \end{array}\right).
\end{equation}
As before, $\|\mbox{\mathversion {bold}$ P$}\|\,\|\mbox{\mathversion {bold}$ P$}^{-1}\|$ is bounded by a constant $M(\lambda)$,
the solution of the system (3.8) is expressible as
\begin{equation}
\vec{f}_{\gamma}(z)=\int_{-\infty}^z \mbox{\mathversion {bold}$ P$} e^{\mbox{\mathversion {bold}$ D$}_{\gamma}(x-z)}\mbox{\mathversion {bold}$ P$}^{-1} \vec{F}_{\gamma}(x)dx
\end{equation}
for $\frac{\gamma}{n}>6$ (noting that within the range of $\lambda$ presently
under consideration, $0<r$, $\bar r<6$, so that all eigenvalues of
$\mbox{\mathversion {bold}$ A$}_{\gamma}$ are positive for $\frac{\gamma}{n}>6$), and
\begin{equation}
\|\vec{f}_{\gamma}(z)\|\leq M(\lambda)\int_{-\infty}^z e^{\left(\frac{\gamma}{n}-6\right)
(x-z)}\|\vec{F}_{\gamma}(x)\|dx.
\end{equation}
\noindent
{\bf Theorem 3.1.} {\it There exists $K>0$ such that for all $z<0$
and $\gamma\geq 1$,}
\begin{equation}
\|\vfgz\|\leq\frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}.
\end{equation}
\noindent
{\bf Proof.} Since $f_{\gamma}(z)$ is a polynomial of degree $\gamma$ in
$X=e^{\mu_1z}$ and $Y=e^{\mu_2z}$, Lemma A3 of Appendix A (with $M=1$
and $p=1$) implies that given $N>1$, there exists $K_1>0$ such that
\begin{equation}
|f_{\gamma}(z)|\leq
\frac{(K_1+K_1e^{\mu_1z}+K_1e^{\mu_2z})^{\gamma}}{\sqrt{\gamma+1}}
\qquad\mbox{for}\quad 1\leq\gamma\leq N-1.
\end{equation}
Similarly, there exist positive constants $K_2$, $K_3$ and $K_4$
corresponding to the polynomials $g_{\gamma}$, $h_{\gamma}$ and $k_{\gamma}$,
respectively. With $K=\max\limits_{1\leq i\leq 4}\{K_i\}$, we obtain
\begin{equation}
\|\vfgz\|\leq \frac{(K+Ke^{\mu_1z}+Ke^{\mu_2z})^{\gamma}}{\sqrt{\gamma+1}}
\qquad\mbox{for}\quad 1\leq\gamma\leq N-1.
\end{equation}
Since $0<e^{\mu_i z}<1$ for $i=1, 2$ and for all $z<0$, we have
\begin{equation}
\|\vfgz\|\leq\frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}\qquad\mbox{for}\quad
1\leq\gamma\leq N-1.
\end{equation}
The inequality (3.17) constitutes our inductive hypothesis, and we
shall show that (3.14) holds for $\gamma=N$, thereby establishing that it
holds for all $N>1$.
As in Section 2, we f\/ind that
\begin{equation}
\|\vec{F}_N\|\leq E(3K)^N,
\end{equation}
where $E$ is bounded as in (2.27). Thus, from (3.13) and (3.18), we
obtain
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle \|\vec{f}_N(z)\|\leq
M(\lambda)E\int_{-\infty}^ze^{\left(\frac{N}{n}-6\right)(x-z)}(3K)^Ndx
\vspace{3mm}\\
\displaystyle \phantom{\|\vec{f}_N(z)\|}=\left\{\frac{M(\lambda)E\sqrt{N+1}}
{\frac{N}{n}-6}\right\}\frac{(3K)^N}{\sqrt{N+1}}
\qquad\mbox{for}\quad \frac{N}{n}>6.
\end{array}
\end{equation}
For all suf\/f\/iciently large $N$ (and any $n\geq 1$),
$\frac{M(\lambda)E\sqrt{N+1}}{\frac{N}{n}-6}<1$, so beginning with such an
$N>6n$, we obtain
\begin{equation}
\|\vec{f}_N(z)\|\leq \frac{(3K)^N}{\sqrt{N+1}},
\end{equation}
which completes the proof.
By (3.2),
\begin{equation}
x=\sum_{\gamma=0}^{\infty}f_{\gamma}(z)w^{\gamma-2n},
\end{equation}
and by (3.14),
\begin{equation}
|f_{\gamma}(z)w^{\gamma-2n}|\leq\frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}w^{\gamma-2n},
\end{equation}
and the series
\begin{equation}
\sum_{\gamma=0}^{\infty}\frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}w^{\gamma-2n}
\end{equation}
converges absolutely by the ratio test for $w<\frac{1}{3K}$, i.e., for
$\tau<\frac{1}{(3K)^n}$. Thus, the series (3.21) for $x$ converges
absolutely for $0<\tau<R$, where $R$ is at least $\frac{1}{(3K)^n}$.
Similarly, the series for $\dot x$, $y$ and $\dot y$ converge
absolutely for $0<\tau<R$.
\section*{4. Case(ii)-leading orders:
{\mathversion{bold} $Re(\alpha)>-2$, $\beta=-2$}}
\setcounter{section}{4}\setcounter{equation}{0}
\noindent
As mentioned in Section 1, if $Re(\alpha)>-2$ and $\lambda>-\frac{1}{2}$, two
types of leading-order and resonance structures are possible, namely,
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle \mbox{(a) Leading order}\quad
\alpha=\frac{1}{2}+\frac{1}{2}\sqrt{1-48\lambda}\,,\ \mbox{resonances}\quad
k_r=-1, 6, 0, r;\vspace{3mm}\\
\displaystyle \mbox{(b) Leading order}\quad
\bar\alpha=\frac{1}{2}-\frac{1}{2}\sqrt{1-48\lambda}\,,\ \mbox{resonances}\quad
k_r=-1, 6, 0, \bar r,
\end{array}
\end{equation}
where $r, \bar r=\mp\sqrt{1-48\lambda}$. For $\lambda<\frac{1}{48}$, $r<0$, so
that (a) does not correspond to a general solution, and will
therefore not be considered. (See Conte, Fordy and Pickering [18] for
a discussion of negative resonances and the appropriate series
expansions.) In case (b), $\bar r>0$ for $-\frac{1}{2}<\lambda<0$ and
$0<\lambda<\frac{1}{48}$, and $\bar r$ is pure imaginary for
$\lambda>\frac{1}{48}$. The general solution takes the form
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\sum_{m=0}^{\infty}
a_{klm}\tau^{k+\bar\alpha +(2+\bar\alpha)m+\bar r l},
\vspace{3mm}\\
\displaystyle y=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\sum_{m=0}^{\infty}
b_{klm}\tau^{k-2+(2+\bar\alpha)m+\bar r l},
\end{array}
\end{equation}
discussion of the validity of which, as well as other relevant
issues, is relegated to Appendix~C. If $\bar r$ is pure imaginary
($\lambda>\frac{1}{48}$), then the methods employed earlier do not apply to
(4.2) (see Section~5), and we have not determined whether the series
converge or not. For $\bar r>0$ ($\lambda<\frac{1}{48}$), we proceed as in
the preceding sections.
Let $n$ be the least positive integer such that $\bar\beta=\bar
r-\frac{1}{n}>0$, let $w=\tau^{1/n}$ and $z=\ln(\tau),\ \tau>0$, in
order to express $x$ in (4.2) as
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\sum_{m=0}^{\infty}a_{klm}\tau^{k+2m+\frac{1}{n}l+\bar\alpha+\bar\alpha
m+\bar\beta l}
\vspace{3mm}\\
\displaystyle \qquad =\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\sum_{m=0}^{\infty}
a_{klm}w^{nk+2nm+l+\bar\alpha n}\tau^{\bar\alpha m+\bar\beta l}
=\sum_{\gamma=0}^{\infty}f_{\gamma}(z)w^{\gamma+\bar\alpha n},
\end{array}
\end{equation}
where
\begin{equation}
f_{\gamma}(z)=\sum_{nk+2nm+l=\gamma}a_{klm}e^{\bar\alpha mz+\bar\beta lz}
\end{equation}
is a polynomial of degree $\gamma$ in the two variables $e^{\bar\alpha z}$
and $e^{\bar\beta z}$. Similarly,
\begin{equation}
y=\sum_{\gamma=0}^{\infty}h_{\gamma}(z)w^{\gamma-2n},
\end{equation}
where
\begin{equation}
h_{\gamma}(z)=\sum_{nk+2nm+l=\gamma}b_{klm}e^{\bar\alpha mz+\bar\beta lz}
\end{equation}
is a polynomial of the same type as $f_{\gamma}(z)$.
Let
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle u=\dot x=\sum_{\gamma=0}^{\infty}g_{\gamma}(z)w^{\gamma+\bar\alpha
n-n}=\sum_{\gamma=0}^{\infty}\left[\left(\frac{\gamma}{n}+\bar
\alpha\right)f_{\gamma}+f_{\gamma}'\right]w^{\gamma+\bar\alpha n-n},
\vspace{3mm}\\
\displaystyle v=\dot y=\sum_{\gamma=0}^{\infty}k_{\gamma}(z)w^{\gamma-3n}
=\sum_{\gamma=0}^{\infty}\left[\left(\frac{\gamma}{n}-2\right)h_{\gamma}+h_{\gamma}'
\right]w^{\gamma-3n},
\end{array}
\end{equation}
and substitute (4.3), (4.5) and (4.7) into the system (2.9) to
obtain, for $\gamma=0$,
\begin{equation}
f_0\ \mbox{ arbitrary},\qquad g_0=\bar\alpha f_0,\qquad h_0=6,
\qquad k_0=-12,
\end{equation}
and for $\gamma>0$, the system
\begin{equation}
\vec{f}_{\gamma}'+\mbox{\mathversion {bold}$ A$}_{\gamma}\vec{f}_{\gamma}=\vec{F}_{\gamma},
\end{equation}
where, as before,
\begin{equation}
\vec{f}_{\gamma}=\left(\begin{array}{c} f_{\gamma}\\
g_{\gamma}\\h_{\gamma}\\k_{\gamma}\end{array}\right),\qquad
\vec{F}_{\gamma}=\left(\begin{array}{c} F_{\gamma}\\
G_{\gamma}\\H_{\gamma}\\K_{\gamma}\end{array}\right),
\qquad F_{\gamma}=H_{\gamma}=0,
\end{equation}
and here,
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle G_{\gamma}=-Af_{\gamma-2n}-2\lambda\sum_{\mu=1}^{\gamma-1}f_{\gamma-\mu}h_{\mu},
\vspace{3mm}\\
\displaystyle K_{\gamma}=-Bh_{\gamma-2n}-\lambda\sum_{\mu=0}^{\gamma-4n}e^{2\bar\alpha
z}f_{\gamma-4n-\mu}f_{\mu}+\sum_{\mu=1}^{\gamma-1}h_{\gamma-\mu}h_{\mu},
\end{array}
\end{equation}
and
\begin{equation}
\mbox{\mathversion {bold}$ A$}_{\gamma}=\left(\begin{array}{cccc}
\displaystyle \frac{\gamma}{n}+\bar\alpha & -1 & 0&0\vspace{2mm}\\
12\lambda & \displaystyle \frac{\gamma}{n}+\bar\alpha-1&2\lambda f_0&0\vspace{2mm}\\
0&0&\displaystyle \frac{\gamma}{n}-2&-1\vspace{2mm}\\
0&0&-12&\displaystyle \frac{\gamma}{n}-3\end{array}\right).
\end{equation}
As expected, the eigenvalues of $\mbox{\mathversion {bold}$ A$}_{\gamma}$ are $\frac{\gamma}{n}-k_r$, where
$k_r=-1, 0, 6, \bar r$ are the resonances. The diagonalizing matrix
for $\mbox{\mathversion {bold}$ A$}_{\gamma}$ is
\begin{equation}
\mbox{\mathversion {bold}$ P$}=\left(\begin{array}{cccc}
\displaystyle -\frac{\bar \alpha}{12}f_0&1&-\lambda f_0&1\vspace{2mm}\\
\lambda f_0&\bar\alpha&-\lambda f_0(\bar\alpha+6)&1-\bar\alpha\vspace{2mm}\\
1&0&3(2\bar\alpha+5)&0\vspace{2mm}\\
-3&0&12(2\bar\alpha+5)&0\end{array}\right),
\end{equation}
with $\mbox{\mathversion {bold}$ D$}_{\gamma}=\mbox{\mathversion {bold}$ P$}^{-1}\mbox{\mathversion {bold}$ A$}_{\gamma}\mbox{\mathversion {bold}$ P$}$, and $\|\mbox{\mathversion {bold}$ P$}\|\,\|\mbox{\mathversion {bold}$ P$}^{-1}\|\leq
M(\lambda,f_0)$, a constant independent of $\gamma$, but depending upon $\lambda$
and the arbitrary constant $f_0$. Since $-\frac{1}{2}<\lambda<0$ or
$0<\lambda<\frac{1}{48}$, we have $0<\bar r<5<6$, so that $k_r=6$ is again
the largest resonance, and all eigenvalues of $\mbox{\mathversion {bold}$ A$}_{\gamma}$ are positive
for $\frac{\gamma}{n}>6$. Thus, for $\frac{\gamma}{n}>6$, the solution of (4.9)
may be expressed as
\begin{equation}
\vec{f}_{\gamma}(z)=\int_{-\infty}^z \mbox{\mathversion {bold}$ P$} e^{\mbox{\mathversion {bold}$ D$}_{\gamma}(x-z)}\mbox{\mathversion {bold}$ P$}^{-1}\vec{F}_{\gamma}(x)dx,
\end{equation}
with
\begin{equation}
\|\vfgz\|\leq M(\lambda,f_0)\int_{-\infty}^z e^{\left(\frac{\gamma}{n}-6\right)
(x-z)}\|\vec{F}_{\gamma}(x)\|dx.
\end{equation}
\noindent
{\bf Theorem 4.1.} {\it There exists $K>0$ such that for all $z<0$
and $\gamma\geq 1$},
\begin{equation}
\|\vfgz\|\leq \frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}.
\end{equation}
\noindent
As the proof is practically identical to the one of Theorem 3.1, we
omit the details.
By (4.3),
\begin{equation}
x=\sum_{\gamma=0}^{\infty}f_{\gamma}(z)w^{\gamma+\bar\alpha n}
\end{equation}
and by (4.16),
\begin{equation}
\left|f_{\gamma}(z)w^{\gamma+\bar\alpha n}\right|\leq
\frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}w^{\gamma+\bar\alpha n},
\end{equation}
and the series
\begin{equation}
\sum_{\gamma=0}^{\infty}\frac{(3K)^{\gamma}}{\sqrt{\gamma+1}}w^{\gamma+\bar\alpha n}
\end{equation}
converges absolutely by the ratio test for $w<\frac{1}{3K}$, i.e., for
$\tau<\frac{1}{(3K)^n}$. Thus, the series (4.17) for $x$ and,
similarly, the series (4.5) for $y$ and (4.7) for $\dot x$ and $\dot
y$ converge absolutely for $0<\tau<R$, where $R$ is at least
$\frac{1}{(3K)^n}$.
\section*{5. Concluding Remarks}
\setcounter{section}{5}\setcounter{equation}{0}
\noindent
Three types of psi-series solutions of the cubic H\'{e}non-Heiles
system have been shown to be absolutely convergent on intervals of
the form $0<\tau<R$, $R>0$. It is clear that the series for $x$, $\dot x$,
$y$ and $\dot y$ converge uniformly on compact subintervals of
$(0,R)$, thereby justifying the termwise dif\/ferentiation of the
series.
The resummations that have been performed in order to proceed with
the convergence proofs amount to the replacement of multiply-indexed
series with constant coef\/f\/icients by single series with polynomial
coef\/f\/icients. This procedure makes use of the fact that the
resonances appearing as exponents within the psi-series have positive
real parts. For example, to obtain (2.3) and (2.6) from (2.2), it is
essential that $Re(r)>0$ and $Re(\bar{r})>0$. Otherwise, as occurs in
(4.2) when $\bar{r}$ is pure imaginary, a resummation would give rise
to series containing $f_{\gamma}(z)$ which are not polynomials, but
inf\/inite series.
The situation when $\lambda=-\frac{1}{2}$ is very interesting. Here,
$\alpha=-2$ (the leading order for $x$), $r=5$ and $\bar r=0$. But the
leading coef\/f\/icient for $x$ is $\pm\frac{3}{\lambda}\sqrt{2+\frac{1}{\lambda}}=0$,
contradicting the fact that $x=O(\tau^{-2})$. The correct
leading-order behaviours have been given in [7] and~[9, p.~340]. In
our notation,
\[
\hspace*{-5pt}\begin{array}{l}
\displaystyle x\,\sim\,(-30)^{1/2}\tau^{-2}(\ln(\tau))^{-\frac{1}{2}},
\vspace{3mm}\\
\displaystyle y\,\sim\,6\tau^{-2}+\frac{5}{2}\tau^{-2}(\ln(\tau))^{-1}.
\end{array}
\]
A full series expansion of the form
\[
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}a_{kl}\tau^{k-2}
(\ln(\tau))^{-l-\frac{1}{2}},\vspace{3mm}\\
\displaystyle y=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}b_{kl}\tau^{k-2}(\ln(\tau))^{-l},
\end{array}
\]
fails, due to an incompatible resonance at $(k,l)=(0,2)$. We are not
aware of the correct series in this case.
\subsection*{Acknowledgments}
\noindent
This work was supported in part by the Natural Sciences and
Engineering Research Council of Canada. I wish to thank the referee
for helpful comments and references.
\renewcommand{\theequation}{\Alph{section}\arabic{equation}}
\section*{Appendix A}
\setcounter{equation}{0}
\noindent
This section contains the Lemmas necessary for the convergence
proofs.
\medskip
\noindent
{\bf Lemma A1.} {\it Let $X\neq 0$ have constant sign, let $q\geq 1$
be an integer, and for $\gamma\geq1$, let
\[
P_{\gamma}(X)=\sum_{m=0}^{n_{\gamma}}c_m^{(\gamma)}X^m
\]
be a sequence of
polynomials of degree $n_{\gamma}=[\frac{\gamma}{q}]$. Given an integer $N>q$,
$0<M\leq 1$ and $p>0$, there exists $K$ with}
$\mbox{sign}(K)=\mbox{sign}(X)$ {\it and} $p|K|>1$ {\it such that}
\[
|P_{\gamma}(X)|\leq M\frac{(p|K|+KX)^{\gamma/q}}{\sqrt{\gamma+1}}\qquad
\mbox{\it for}\quad 1\leq \gamma\leq N-1.
\]
This is Lemma 4.4 of [13], to which we
refer the reader for a proof.
\medskip
\noindent
{\bf Lemma A2.}
\[
\lim_{\gamma \rightarrow\infty}\sum_{\mu=0}^{\gamma-1}\frac{1}{\sqrt{\mu+1}\,
\sqrt{\gamma-\mu}}=\pi.
\]
\noindent
The result follows easily by the proof of the integral test for
convergence of series. For the details, we refer to Hemmi [19, p.~64].
\medskip
\noindent
{\bf Lemma A3.} {\it Let $X, Y>0$ and for $\gamma\geq 1$, let
$P_{\gamma}(X,Y)$ be a sequence of polynomials of degree $\gamma$ in $X$ and
$Y$, i.e.,
\[
P_{\gamma}(X,Y)=\sum_{\beta=0}^{\gamma}\sum_{\mu=0}^{\beta}c_{\mu\beta}^{(\gamma)}X^{\mu}
Y^{\beta-\mu}.
\]
Given an integer $N>1$, $0<M\leq 1$ and $p>0$, there exists $K>0$
such that
\[
|P_{\gamma}(X,Y)|\leq M\frac{(pK+KX+KY)^{\gamma}}{\sqrt{\gamma+1}}
\qquad\mbox{for}\quad 1\leq \gamma\leq N-1.
\] }
\noindent
{\bf Proof.} Let
\[
K=\max_{\mbox{\scriptsize $\begin{array}{c}1\leq\gamma\leq
N-1\\0\leq\beta\leq\gamma\\0\leq\mu\leq\beta\end{array}$}}
\left\{\frac{|c_{\mu\beta}^{(\gamma)}|
\sqrt{\gamma+1}}{\left(\begin{array}{c}\gamma\\ \beta\end{array}\right)\left(
\begin{array}{c}\beta\\ \mu\end{array}\right)Mp^{\gamma-\beta}}\right\}^{1/\gamma},
\]
then for $1\leq\gamma\leq N-1$, $0\leq\beta\leq\gamma$,
$0\leq\mu\leq\beta$, we have
\[
|c_{\mu\beta}^{(\gamma)}|\leq\frac{M}{\sqrt{\gamma+1}}\left(\begin{array}{c}\gamma
\\ \beta\end{array}\right)\left(\begin{array}{c}\beta \\ \mu\end{array}\right)K^{\gamma}
p^{\gamma-\beta},
\]
which implies that
\[
\hspace*{-5pt}\begin{array}{l}
\displaystyle |P_{\gamma}(X,Y)|\leq\sum_{\beta=0}^{\gamma}\sum_{\mu=0}^{\beta}
|c_{\mu\beta}^{(\gamma)}|X^{\mu}Y^{\beta-\mu}
\vspace{3mm}\\
\displaystyle \phantom{|P_{\gamma}(X,Y)|} \leq\frac{M}{\sqrt{\gamma+1}}\sum_{\beta=0}^{\gamma}
\sum_{\mu=0}^{\beta} \left(\begin{array}{c}\gamma \\ \beta\end{array}\right)
\left(\begin{array}{c}\beta \\ \mu\end{array}\right)K^{\gamma}p^{\gamma-\beta}X^{\mu}
Y^{\beta-\mu}\vspace{3mm}\\
\displaystyle \phantom{|P_{\gamma}(X,Y)|}=
\frac{M}{\sqrt{\gamma+1}}\sum_{\beta=0}^{\gamma}\sum_{\mu=0}^{\beta}
\left(\begin{array}{c}\gamma \\ \beta\end{array}\right)\left(\begin{array}{c}
\beta \\ \mu\end{array}\right) (pK)^{\gamma-\beta}(KX)^{\mu}(KY)^{\beta-\mu}
\vspace{3mm}\\
\displaystyle \phantom{|P_{\gamma}(X,Y)|}=M\frac{(pK+KX+KY)^{\gamma}}{\sqrt{\gamma+1}},
\end{array}
\]
by the binomial theorem.
\section*{Appendix B}
\setcounter{section}{2}
\setcounter{equation}{0}
\noindent
First, we show that the solution (2.1), namely,
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\sum_{m=0}^{\infty}a_{klm}\tau^{k-2+rl+\bar r m},
\vspace{3mm}\\
\displaystyle y=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\sum_{m=0}^{\infty}b_{klm}\tau^{k-2+rl+\bar r m},
\end{array}
\end{equation}
of the system (1.4), corresponding to the case(i)-leading orders and
non-integral resonances, is self-consistent and contains four
arbitrary constants.
Substitution of (B1) into (1.4) gives
\begin{equation}
a_{000}=\pm\frac{3}{\lambda}\sqrt{2+\frac{1}{\lambda}},
\qquad b_{000}=-\frac{3}{\lambda},
\end{equation}
and for $(k,l,m)\neq (0,0,0)$, the coef\/f\/icient recursion relations
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle (k+rl+\bar r m)(k+rl+\bar rm -5)a_{klm}+2\lambda a_{000}b_{klm}
\vspace{3mm}\\
\displaystyle \qquad=-Aa_{k-2,l,m}-2\lambda\sum_{(0,0,0)\prec(p,q,s)\prec(k,l,m)}
a_{pqs}b_{k-p,l-q,m-s},
\vspace{3mm}\\
\displaystyle 2\lambda a_{000}a_{klm}+\left[(k+rl+\bar r m -2)(k+rl+\bar r
m-3)+\frac{6}{\lambda}\right]b_{klm}
\vspace{3mm}\\
\displaystyle \qquad=-B b_{k-2,l,m}+
\hspace{-0.7cm}\sum_{(0,0,0)\prec(p,q,s)\prec(k,l,m)}
\hspace{-0.7cm}[-\lambda a_{pqs}a_{k-p,l-q,m-s}+b_{pqs}
b_{k-p,l-q,m-s}],
\end{array}
\end{equation}
where we have employed the notation $(0,0,0)\prec(p,q,s)\prec(k,l,m)$
to mean that $0\leq p\leq k$, $0\leq q\leq l$, $0\leq s\leq m$, with
$(p,q,s)\neq (0,0,0)$ and $(p,q,s)\neq(k,l,m)$. Expressed in matrix
form, the left-hand sides become
\begin{equation}\hspace*{-10mm}
\left[\!
\begin{array}{cc}
(k+rl+\bar r m)(k+rl+\bar r m-5) \!\!
& 2\lambda a_{000}\vspace{2mm}\\
2\lambda a_{000}& \!\!\! (k+rl+\bar r m-2)(k+rl+\bar r
m-3)+\frac{6}{\lambda}\end{array}\!\right]\!\!
\left[\begin{array}{c}a_{klm}\\b_{klm}\end{array}\right],
\end{equation}
and the coef\/f\/icient matrix is singular precisely when $k+rl+\bar r m$
has one of the values $-1, 6, r, \bar r$, i.e., when
\begin{equation}
(k,l,m)=(-1,0,0), (6,0,0), (0,1,0), (0,0,1).
\end{equation}
These are the resonances of (B1), and compatibility must be checked
at these values, conf\/irming both the self-consistency of (B1) and the
presence of four arbitrary constants.
The resonance at $(k,l,m)=(-1,0,0)$ is trivially compatible, and
corresponds to the arbitrariness of $t_0$. At $(k,l,m)=(0,1,0)$, we
f\/ind that
\begin{equation}
a_{010}\quad \mbox{is arbitrary,} \qquad
b_{010}=-\frac{r(r-5)}{2\lambda a_{000}}a_{010},
\end{equation}
and at $(k,l,m)=(0,0,1)$, that
\begin{equation}
a_{001}\quad \mbox{is arbitrary,} \qquad
b_{001}=-\frac{\bar r(\bar r-5)}{2\lambda
a_{000}}a_{001}.
\end{equation}
At $(k,l,m)=(6,0,0)$, the compatibility condition is
\begin{equation}
\lambda a_{000}a_{400}A+2\lambda^2a_{000}(a_{200}b_{400}
+a_{400}b_{200})-3b_{400}B -6(\lambda a_{200}a_{400}-b_{200}b_{400})=0.
\end{equation}
The coef\/f\/icients which af\/fect (B8) are
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle a_{000}=\pm\frac{3}{\lambda}\sqrt{2+\frac{1}{\lambda}},\qquad
b_{000}=-\frac{3}{\lambda},
\vspace{3mm}\\
\displaystyle a_{100}=a_{300}=a_{500}=0,\qquad b_{100}=b_{300}=b_{500}=0,
\vspace{3mm}\\
\displaystyle a_{200}=\frac{a_{000}\left(\frac{A}{\lambda}+B\right)}
{12\left(1+\frac{1}{\lambda}\right)},\qquad
b_{200}=\frac{B-A\left(2+\frac{1}{\lambda}\right)}
{4\lambda\left(1+\frac{1}{\lambda}\right)},
\vspace{3mm}\\
\displaystyle a_{400}=\frac{a_{200}\left(1+\frac{3}{\lambda}\right)
(A+2\lambda b_{200})-\lambda a_{000}\left(Bb_{200}+\lambda a^2_{200}-
b^2_{200}\right)} {10\left(4+\frac{3}{\lambda}\right)},
\vspace{3mm}\\
\displaystyle b_{400}=-\frac{\lambda a_{000}a_{200}(A+2\lambda b_{200})
-2\left(Bb_{200}+\lambda a^2_{200}-b^2_{200}\right)}
{10\left(4+\frac{3}{\lambda}\right)},
\end{array}
\end{equation}
from which (B8) is conf\/irmed. The system governing $a_{600}$ and
$b_{600}$ is
\begin{equation}
\left(\begin{array}{cc}6&2\lambda a_{000}
\vspace{2mm}\\
2\lambda a_{000}&
12+\frac{6}{\lambda}\end{array}\right)
\left(\begin{array}{c}a_{600}\\
b_{600}\end{array}\right)= \left(\begin{array}{c}-Aa_{400}-2\lambda(a_{200}b_{400}
+a_{400}b_{200})\vspace{3mm}\\
-Bb_{400}-2(\lambda a_{200}a_{400}-b_{200}b_{400})\end{array}\right),
\end{equation}
and reduces to
\begin{equation}
6a_{600}+2\lambda a_{000}b_{600}=-A
a_{400}-2\lambda(a_{200}b_{400}+a_{400}b_{200}),
\end{equation}
showing that $a_{600}$ is arbitrary, and $b_{600}$ is def\/ined in
terms of $a_{600}$.
Next, we demonstrate that (2.1) is equivalent to (2.2). The series
which def\/ines $x$ in (2.1) may be broken up into three parts: $m=l$,
$m<l$, and $m>l$. Thus,
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}a_{kll}\tau^{k-2+5l}
+\sum_{k=0}^{\infty}\sum_{l=1}^{\infty}
\sum_{m=0}^{l-1}a_{klm}\tau^{k-2+5m+(l-m)r}
\vspace{3mm}\\
\displaystyle \qquad+\sum_{k=0}^{\infty}\sum_{m=1}^{\infty}
\sum_{l=0}^{m-1}a_{klm}\tau^{k-2+5l+(m-l)\bar r},
\end{array}
\end{equation}
where use has been made of the fact that $r+\bar r=5$. The f\/irst sum
in (B12) is expressible as
\begin{equation}
\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}a_{kll}\tau^{k-2+5l}=
\sum_{i=0}^{\infty}\sum_{k+5l=i}a_{kll}\tau^{i-2}
=\sum_{i=0}^{\infty}a_{i0}\tau^{i-2},
\end{equation}
where
\begin{equation}
a_{i0}=\sum_{k+5l=i}a_{kll}.
\end{equation}
The second sum in (B12) may be manipulated as follows:
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle \sum_{k=0}^{\infty}\sum_{l=1}^{\infty}\sum_{m=0}^{l-1}a_{klm}
\tau^{k-2+5m+(l-m)r}\vspace{3mm}\\
\displaystyle \qquad =\sum_{k=0}^{\infty}\sum_{l=1}^{\infty}
\sum_{j=1}^{l}a_{k,l,l-j}\tau^{k-2+5(l-j)+jr}\qquad (\mbox{letting}\quad
j=l-m)\vspace{3mm}\\
\displaystyle \qquad =\sum_{k=0}^{\infty}\sum_{j=1}^{\infty}\sum_{l=j}^{\infty}
a_{k,l,l-j}\tau^{k-2+5(l-j)+jr}\qquad(\mbox{reversing the order of summation})
\vspace{3mm}\\
\displaystyle \qquad =\sum_{k=0}^{\infty}\sum_{j=1}^{\infty}
\sum_{l=0}^{\infty}a_{k,l+j,l}\tau^{k-2+5l+jr}\qquad(\mbox{replacing $l$
by $l+j$})\vspace{3mm}\\
\displaystyle\qquad =\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}
\sum_{k+5l=i}a_{k,l+j,l}\tau^{i-2+jr}\vspace{3mm}\\
\displaystyle \qquad =\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}a_{ij}\tau^{i-2+jr},
\end{array}\hspace{-7.6pt}
\end{equation}
where
\begin{equation}
a_{ij}=\sum_{k+5l=i}a_{k,l+j,l}.
\end{equation}
Similarly, the third term in (B12) becomes
\begin{equation}
\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}\bar a_{ij}\tau^{i-2+j\bar r},
\end{equation}
where
\begin{equation}
\bar a_{ij}=\sum_{k+5m=i}a_{k,m,m+j}.
\end{equation}
Combining (B13), (B15) and (B17), we obtain
\begin{equation}
x=\sum_{i=0}^{\infty}\sum_{j=0}^{\infty}a_{ij}\tau^{i-2+jr}
+\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}\bar
a_{ij}\tau^{i-2+j\bar r}.
\end{equation}
Similarly,
\begin{equation}
y=\sum_{i=0}^{\infty}\sum_{j=0}^{\infty}b_{ij}\tau^{i-2+jr}
+\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}\bar
b_{ij}\tau^{i-2+j\bar r}.
\end{equation}
Def\/ining $\bar a_{i0}=\bar b_{i0}=0$ for all $i\geq 0$, (B19), (B20)
coincide with (2.2).
Finally, the solution given in [3] is
\begin{equation}
x=\sum_{j=0}^{\infty}\sum_{i=0}^{\infty}A_{ji}\tau^{i-2+j(r-2)}+
\sum_{j=1}^{\infty}\sum_{i=0}^{\infty}\bar
A_{ji}\tau^{i-2+j(\bar r-2)}
\end{equation}
(their equation (3.5a), in our notation), with a similar expression
for $y$. To obtain the form (B21) from (B19), write
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle
x=\sum_{i=0}^{\infty}\sum_{j=0}^{\infty}a_{ij}\tau^{i-2+jr}
+\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}\bar
a_{ij}\tau^{i-2+j\bar r}
\vspace{3mm}\\
\displaystyle \qquad =\sum_{i=0}^{\infty}\sum_{j=0}^{\infty}a_{ij}\tau^{i+2j-2+j(r-2)}
+\sum_{i=0}^{\infty}\sum_{j=1}^{\infty}\bar
a_{ij}\tau^{i+2j-2+j(\bar r-2)}
\vspace{3mm}\\
\displaystyle \qquad =\sum_{j=0}^{\infty}\sum_{i=2j}^{\infty}a_{i-2j,j}\tau^{i-2+j(r-2)}
+\sum_{j=1}^{\infty}\sum_{i=2j}^{\infty}\bar
a_{i-2j,j}\tau^{i-2+j(\bar r-2)}
\vspace{3mm}\\
\displaystyle \qquad =\sum_{j=0}^{\infty}\sum_{i=0}^{\infty}A_{ji}\tau^{i-2+j(r-2)}
+\sum_{j=1}^{\infty}\sum_{i=0}^{\infty}\bar
A_{ji}\tau^{i-2+j(\bar r-2)},
\end{array}
\end{equation}
where
\begin{equation}
A_{ji}=\left\{\begin{array}{ll}a_{i-2j,j},&\mbox{if }\ i\geq
2j\\0,&\mbox{otherwise}\end{array}\right\} ,\qquad \bar
A_{ji}=\left\{\begin{array}{ll}\bar a_{i-2j,j},&\mbox{if }\ i\geq
2j\\0,&\mbox{otherwise}\end{array}\right\},
\end{equation}
and similarly for $y$.
\section*{Appendix C}
\setcounter{section}{3}\setcounter{equation}{0}
\noindent
First, we conf\/irm that the solution (4.2), namely,
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle x=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}
\sum_{m=0}^{\infty}a_{klm}\tau^{k+\bar\alpha+(2+\bar\alpha)m+\bar rl},
\vspace{3mm}\\
\displaystyle y=\sum_{k=0}^{\infty}\sum_{l=0}^{\infty}\sum_{m=0}^{\infty}b_{klm}
\tau^{k-2+(2+\bar\alpha)m+\bar rl},
\end{array}
\end{equation}
of the system (1.4), corresponding to the case(ii)-singularities, is
self-consistent and contains four arbitrary constants.
Substitution of (C1) into (1.4) gives
\begin{equation}
a_{000}\quad \mbox{ is arbitrary},\qquad b_{000}=6,
\end{equation}
and for $(k,l,m)\neq(0,0,0)$, the coef\/f\/icient recursion relations
\begin{equation}
\hspace*{-5pt}\begin{array}{l}
\displaystyle [k+(2+\bar\alpha)m+\bar r l][k+(2+\bar\alpha)m+\bar r(l-1)]a_{klm}+
2\lambda a_{000}b_{klm}\vspace{3mm}\\
\displaystyle \qquad =-Aa_{k-2,l,m}-2\lambda\sum_{(0,0,0)\prec(p,q,r)
\prec(k,l,m)}a_{k-p,l-q,m-r}b_{pqr},
\vspace{3mm}\\
\displaystyle [k+(2+\bar\alpha)m+\bar r l+1][k+(2+\bar\alpha)m+\bar rl-6]b_{klm}
\vspace{3mm}\\
\displaystyle \qquad=-Bb_{k-2,l,m}-\lambda\sum_{p=0}^{k}\sum_{q=0}^{l}
\sum_{r=0}^{m}a_{k-p,l-q,m-r-2}a_{pqr}\vspace{3mm}\\
\displaystyle \qquad +\sum_{(0,0,0)\prec(p,q,r)\prec(k,l,m)}b_{k-p,l-q,m-r}b_{pqr},
\end{array}
\end{equation}
where we have def\/ined the ``$\prec$'' symbol as in Appendix~B.
Expressed in matrix form, the left-hand sides become
\begin{equation}
\hspace*{-10mm}
\left[\!\! \begin{array}{cc}
\mbox{\scriptsize $[k+(2+\bar \alpha)m+\bar r l][k+(2+\bar\alpha)m+
\bar r(l-1)]$}
&2\lambda a_{000}
\vspace{2mm}\\
0& \mbox{\scriptsize $[k+(2+\bar \alpha)m+\bar r
l+1][k+(2+\bar\alpha)m+\bar rl-6]$} \end{array}\!\!\right]\!\!
\left[\hspace*{-5pt}\begin{array}{c}a_{klm}\\b_{klm}\end{array}\right],
\end{equation}
and the coef\/f\/icient matrix is singular precisely when
$k+(2+\bar\alpha)m+\bar r l$ has one of the values $-1, 6, \bar r$,
i.e., when
\begin{equation}
(k,l,m)=(-1,0,0), (6,0,0), (0,1,0).
\end{equation}
These, together with $(k,l,m)=(0,0,0)$, are the resonances of (C1).
Compatibility at $(0,0,0)$ and the arbitrariness of $a_{000}$ has
already been conf\/irmed. The compatibility of $(-1,0,0)$, ref\/lecting
the arbitrariness of $t_0$, is trivial, as usual. It remains to check
the remaining two.
At $(k,l,m)=(0,1,0)$, we f\/ind that
\begin{equation}
b_{010}=0,\qquad a_{010}\quad \mbox{is arbitrary}.
\end{equation}
The compatibility condition at $(k,l,m)=(6,0,0)$ is
\begin{equation}
-Bb_{400}+\sum_{p=1}^5 b_{6-p,0,0}b_{p00}=0.
\end{equation}
The coef\/f\/icients which af\/fect (C7) are
\begin{equation}
b_{100}=0,\qquad b_{300}=0,\qquad b_{500}=0,\qquad
b_{200}=\frac{B}{2},\qquad b_{400}=\frac{B^2}{40},
\end{equation}
from which (C7) is conf\/irmed. The second equation in (C3) then shows
that $b_{600}$ is arbitrary, and the f\/irst gives $a_{600}$ in terms
of $b_{600}$.
Second, we note that a solution dif\/ferent from (C1) has been given in
[3], but we f\/ind that the coef\/f\/icient recursion relations (their
equation (3.9)) do not follow from the solution (their equation
(3.6)) without neglecting certain terms, precisely those terms which
lead to anomalies when a resummation is performed in an attempt to
prove convergence.
Finally, we remark upon the ``derivation'' of the form of (C1). The
exponent $\bar rl$ is required in order to produce an arbitrary
coef\/f\/icient corresponding to the resonance at $\bar r$. This is
usual. The exponent $\bar \alpha m$ is required in order that all terms
in the system (1.4) can be balanced. But since $Re(\bar\alpha)>-2$,
$\bar\alpha m$ must be replaced by $(2+\bar\alpha)m$ in order to avoid
exponents of $\tau$ with arbitrarily large negative real parts, which
would render the series invalid.
|
\section{Introduction}
\label{sec_1}
Let $M$ denote a compact oriented three-manifold. A {\em plane field}
$\eta$ on $M$ is a subbundle of the tangent bundle
$TM$ which associates smoothly to each point $p\in M$ a
two-dimensional subspace $\eta(p)\subset T_pM$. Unlike line fields,
a plane field cannot always be integrated to yield a two-dimensional
foliation ${\cal F}$.
A plane field is said to be {\em integrable} if it can be ``patched
together'' to yield a foliation whose leaves are tangent to the plane field at
each point. Certainly, such plane fields have strong topological and
geometric properties. On the other hand, the
case where the plane field $\eta$ is nowhere integrable can be equally
important. A maximally nonintegrable (in the sense of Frobenius ---
see \S\ref{sec_5}) plane field on an odd-dimensional
manifold is a {\em contact structure}. Seen as an ``anti-foliation'',
contact structures are rich in geometric and topological properties
which of late have become quite important in understanding
the topology of three-manifolds and the symplectic
geometry of four-manifolds.
Let $X$ be a vector field on $M$. The dynamics of $X$ are often
related to global properties of $M$. If we further specify that $X$
is tangent to a plane field $\eta$ --- that is,
$X(p)\in\eta(p)$ for all $p\in M$
--- then we might expect stronger relationships.
We will consider the ways in which the topology and geometry of
a plane field $\eta$ are coupled to the dynamics of vector fields
contained in $\eta$. The general principal at work here as elsewhere
is that {\em simple dynamics implicate simple topological objects in
dimension three.} We will reassert this by examining the gradient
flows within plane fields.
The examination and classification of gradient flows
has been ubiquitous in the study of manifolds: {\em e.g.}, the h-cobordism
theorem and the resolution of the high-dimensional
Poincar\'e Conjecture.
This paper will add to the typical scenario the constraint of lying
within a plane field. Atypical restrictions on the
dynamics and on the underlying manifold are born out of this.
We note that the problem of understanding gradient fields
constrained to lie within plane fields is by no means unnatural.
The study of mechanical systems with nonholonomic constraints
is precisely the study of flows constrained to lie within a
nowhere integrable distribution ({\em i.e.}, in odd dimensions, a
contact structure). For example, gradient flows for mechanical systems have
been used successfully in the control of robotic systems (see, {\em e.g.},
\cite{KR90}): to maneuver a robot from points $A$ to $B$ through
a physical space replete with obstacles, one establishes a gradient flow on
a suitable configuration space with $B$ as a sink, with $A$ in
the basin of attraction for $B$, and with infinite
walls along the obstacles. In this paper, we show that the
nonholonomic version of this procedure possesses potentially
difficult topological obstructions.
The paper is organized as follows: the remainder of this section
provides a brief sketch of the requisite theory from the dynamical
systems approach to flows. In
\S\ref{sec_2}, we commence our investigation of plane field
flows by examining local and global properties of fixed points:
fixed points will not be isolated, but must (on an open dense
subset of $C^r$ vector fields tangent to $\eta$, $r\geq 1$)
rather appear in {\em links}, or embedded closed curves.
This culminates in a classification of gradient flows on three-manifolds
which can lie within a plane field in \S\ref{sec_3}.
The existence of such flows is
equivalent to the existence of a certain type of
round handle decomposition for the manifold
(see Definition \ref{def_RH}). Surprisingly, this same
restriction appears when considering energy surfaces for
(Bott-) integrable Hamiltonian flows \cite{CMAN}.
\noindent{\bf Theorem:} {\em
Let $M$ be a compact 3-manifold outfitted with a plane field
$\eta$. If $X$ is a nondegenerate\footnote{See
Definition~\ref{def_Nondegenerate}.} gradient field tangent to $\eta$,
then $X$ lies in the boundary of the space of nonsingular
Morse-Smale flows on $M$. Furthermore, the set of fixed
points for $X$ forms the cores of an essential round handle
decomposition for $M$.
}
This leads to the corollary (a stronger form of which is proved
in \S\ref{sec_3}):
\noindent{\bf Corollary:} {\em
Non-gradient dynamics is a generic (residual) property in the class of
$C^r$ ($r\geq4$) vector fields tangent to a fixed $C^r$
plane field on a closed hyperbolic three-manifold.
}
In \S\ref{sec_4}, we consider the manifestation of these restrictions
on a knot-theoretic level for the particular case of the 3-sphere.
\noindent{\bf Theorem:} {\em
For $X$ a nondegenerate gradient plane field flow on $S^3$,
each connected component of the fixed point set of $X$ is a knot
whose knot type is among the class generated from the unknot
by the operations of iterated cabling and connected sum.
}
We proceed with remarks on two cases in which the plane field carries
additional geometric structure: first, the case of an everywhere integrable
plane field, {\em i.e.}, a foliation; and second, the case of a maximally
nonintegrable plane field, {\em i.e.}, a contact structure.
The property of carrying a gradient flow in a foliation forces the
foliation to be taut; hence, there are no
(nondegenerate) gradient flows within a foliation on $S^3$.
More generally, we have the following restrictions on the underlying
three-manifold:
\noindent{\bf Theorem:} {\em
A closed orientable three-manifold containing a nondegenerate gradient
field within a $C^r$ ($r\geq2$) codimension-one foliation must be a surface
bundle over $S^1$ with periodic (or reducibly periodic) monodromy map.
}
The corresponding restrictions do not hold for the contact case. We
demonstrate that gradient fields can always reside within the analogue
of a non-taut foliation: an {\em overtwisted} contact structure.
We close with two questions on the higher dimensional
versions of the results of this paper.
\subsection{The dynamics of flows}
\label{sec_1.1}
Ostensibly, flows within a plane field would appear to be a relatively
restricted class of objects. However, the dynamics of such flows can
exhibit behaviors which range from strictly two-dimensional
dynamics (as when the plane field yields a foliation by compact
leaves) to fully three-dimensional phenomena ({\em e.g.}, an Anosov
flow, which is tangent to a pair of transverse integrable plane
fields). In \S\ref{sec_2}, we show that near a fixed point of
a plane field flow, the dynamics are locally ``stacked'' planar
dynamics. In contrast, it is a simple exercise in homotopy
theory that every nonsingular flow on $S^3$ (or any integral
homology 3-sphere) lies within a plane field.
A few definitions are important for the dynamical systems
theory used in this paper.
The most important aspect of a flow with respect to its
geometry and dynamics is the notion of hyperbolicity.
Recall that an invariant set $\Lambda\subset M$ of a flow
$\phi^t$ is {\em hyperbolic}
if the tangent bundle $\rest{TM}{\Lambda}$
has a continuous $\phi^t$-invariant splitting
into $E^\phi\oplus E^s\oplus E^u$,
where $E^\phi$ is tangent to the flow direction, and $D\phi^t$
uniformly contracts and expands along $E^s$ and $E^u$
respectively: {\em i.e.},
\begin{equation}
\begin{array}{ll}
\norm{D\phi^t(\mbox{{\bf v}}^s)} \leq Ce^{-\la t}\norm{\mbox{{\bf v}}^s} &
\mbox{for } \mbox{{\bf v}}^s\in E^s \\
\norm{D\phi^{-t}(\mbox{{\bf v}}^u)} \leq Ce^{-\la t}\norm{\mbox{{\bf v}}^u} &
\mbox{for } \mbox{{\bf v}}^u\in E^u \\
\end{array}, t>0,
\end{equation}
for some $C\geq 1$ and $\la > 0$.
A flow $\phi^t$ which is hyperbolic on all of $M$ is
called an {\em Anosov flow.}
The existence of hyperbolic invariant sets greatly simplifies
the analysis of the dynamics. The principal tool available
is the Stable Manifold Theorem \cite{HPS70}, which states
that for a hyperbolic invariant set, the distributions
$E^s$ and $E^u$ are in fact tangent to
global {\em stable} and {\em unstable manifolds}:
manifolds, all of whose points have the
same backwards and forwards (resp.) asymptotic behavior.
See any of the standard texts ({\em e.g.}, \cite{GH83})
for further information and examples.
\section{Fixed points}
\label{sec_2}
In analyzing the dynamics and topology of a flow, one examines
dynamical $n$-skeleta of increasing dimension:
first the fixed points, then periodic and connecting orbits,
lastly higher-dimensional invariant manifolds and attractors.
This section concerns the typical distribution of fixed points
for plane field flows.
\begin{lemma}\label{lem_Coords}
Given $\eta$ a $C^r$ plane field on $M^3$ and $p\in M$ there exists a
neighborhood $U\cong\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^3$ of
$p$ along with local coordinates $(x,y,z)$ on $U$ such that
$\eta=\mbox{ker}(\alpha)$, where $\alpha$ is a one-form given by
\begin{equation} \label{eq_normal}
\alpha = dz + g(x,y,z)dy,
\end{equation}
for some function $g$ which vanishes at the origin.
The space $\Gamma^r(\rest{\eta}{U})$ of $C^r$ sections of $\eta$
on $U$ is isomorphic to $C^r(\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R},C^r(\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2,\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2))$,
the space of $C^r$ arcs of $C^r$ planar vector fields.
\end{lemma}
{\em Proof: }
That $\alpha$ exists is easy to derive (and is stated in
\cite{ET97}): choose coordinates $(x,y,z)$ so that $\partial/\partial z$ is
transverse to $\eta$ on $U$. Then, after rescaling, $\eta$ is the kernel of
$dz+f(x,y,z)dx + g(x,y,z)dy$. By a change of variables, one can
eliminate $f$ and remove constant terms in $g$.
Parameterize $U$ as $\{\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2\times\{z\} : z\in\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}\}$.
Given any 1-parameter family of functions $F_z:\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2\rightarrow\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2$,
there is a well-defined vector field on $U$ given by
\begin{equation}\label{eq_Coords}
\begin{array}{l}
\dot{x} = f_1(x,y,z) \\
\dot{y} = f_2(x,y,z) \\
\dot{z} = -g(x,y,z)f_2(x,y,z)
\end{array}
\; \mbox{ where } F_z(x,y)=\left( f_1(x,y,z) , f_2(x,y,z) \right) ,
\end{equation}
which lies within $\eta$ by Equation~\ref{eq_normal}. Similarly, any
vector field on $U$ contained in $\eta$ induces a 1-parameter family
of planar vector fields $F_z:\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2\rightarrow\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2$ by inverting the above
procedure. Since $\partial/\partial z$ is always transverse to $\eta$,
zeros of $F_z$ correspond precisely with zeros of
the induced vector field in $\eta$. Note finally that the correspondence
is natural with respect to the $C^r$-topology
(nearby families of planar vector fields induce nearby plane field
flows and vice versa).
\hfill$\Box$
\begin{prop}\label{prop_Link}
Let $\eta$ be a $C^r$ ($r\geq 1$) plane distribution on
$M$ a compact 3-manifold,
and let $\Gamma(\eta)$ denote the space of $C^r$ sections
of $\eta$. Then on an open dense subset of $\Gamma(\eta)$,
the fixed point set is a smooth finite link of embedded circles.
\end{prop}
{\em Proof: }
From the Transversality Theorem (see \cite[p. 74]{Hir76})
we know there is an open dense subset of sections of $\eta$
which are transverse to the zero section. The proposition
clearly follows.
\hfill$\Box$
\begin{cor}\label{cor_NoHypFP}
Let $X$ be any vector field on $M^3$ contained in the distribution
$\eta$. Then any fixed point of $X$ is nonhyperbolic.
\end{cor}
{\em Proof: }
Hyperbolic fixed points are isolated and persist in
$C^1$-neighborhoods of vector fields; hence, they cannot be
perturbed to yield circles of fixed points.
\hfill$\Box$
To analyze the dynamics near a curve of singularities, we show that
for all but finitely many points, the dynamics are
{\em transversally hyperbolic}; {\em i.e.}, after ignoring the
nonhyperbolic direction along the curve, the flow
is hyperbolic along the tangent plane transverse to the curve.
We then turn to classify the (codimension-1) bifurcations
in the transverse behavior along a curve of singularities.
\begin{prop}\label{prop_TransHyp}
Let $X\subset\eta$ be a $C^r$ ($r\geq2$) section of a $C^r$ plane
field $\eta$. Then on a residual set of such vector fields,
${\mbox{Fix}}(X)$ is a link $L$ which is transversally hyperbolic with
respect to all but finitely many $p\in L$.
\end{prop}
{\em Proof: }
By a standard argument (see \cite[p. 74]{Hir76}) is suffices to show
that there is an open cover $\{U_i\}$ of $M$ for which there is a
residual set of sections of $\eta\vert_{U_i}$ with the desired property.
Cover each $p\in M$ by a chart as in Lemma~\ref{lem_Coords}. On
each chart, consider the map from $\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^3\rightarrow\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2$ induced by
a section of $\eta$. Extend this to a map into the 1-jet space
$J^1(\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^3,\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2)$ to capture information about the linearization
of the flow. One may easily find a codimension three stratified subset $S$ of
$J^1(\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^3,\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2)$ on which a section will both vanish and be
transversally nonhyperbolic. Thus by the Jet Transversality Theorem for
$C^2$ maps we obtain a residual subset of sections of $\eta$ whose
1-jets transversally intersect $S$ at isolated points (which clearly
must lie on $L$).
\hfill$\Box$
\begin{cor}\label{cor_QuadTan}
Under the hypotheses of Proposition~\ref{prop_TransHyp},
the singular link $L$ is transverse to $\eta$ at all but a finite number
of points.
\end{cor}
{\em Proof: }
If the curve of singularities $S$ is tangent to the plane field
$\eta$ at a point $p$, then $p$ is not transversally hyperbolic since the
eigenvalue whose eigenvector points in the direction transverse
to $\eta$ is zero (the vector field can have
no component in the direction transverse to $\eta$).
\hfill$\Box$
It is now a simple matter to classify the points at which the vector
field is not transversally hyperbolic to the equilibria. Thanks
to Lemma~\ref{lem_Coords}, this analysis reduces simply
to bifurcation theory of fixed points in planar vector fields.
In particular, there are precisely two ways in which a (generic)
$X\subset\eta$ can fail to be transversally hyperbolic at a point.
Given any singular point $p\in S$, the transverse dynamics is
characterized by the pair of transverse eigenvalues for the
linearized flow: $\la^x$ and $\la^y$. Transverse hyperbolicity
fails if and only if one or both of these eigenvalues has zero real part.
Generically, this can occur in two distinct ways. First, $\la^x$ and $\la^y$
may be both real, and one of them goes transversally through zero:
this is a {\em saddle-node bifurcation}. Second,
$\la^x$ and $\la^y$ may be a complex conjugate pair of eigenvalues
which together pass through the imaginary axis transversally: this is
a {\em Hopf bifurcation}. Again, these names correspond
with analogous bifurcations of fixed points in planar vector fields.
\begin{figure}[htb]
\begin{center}
\epsfxsize=3.0in\leavevmode\epsfbox{figs/saddle.eps}
\end{center}
\caption{A saddle-node bifurcation of singularities in a plane field
flow.}
\label{fig_SN}
\end{figure}
\begin{prop}\label{prop_SN}
In the unfolding of a $C^r$-generic ($r\geq2$) saddle node bifurcation
on a curve of fixed points in a plane field flow, there is a
quadratic tangency between the plane field and the fixed
point curve, along with a one-parameter family of
heteroclinic connections between fixed points limiting onto the
bifurcation point, as in Figure~\ref{fig_SN}.
\end{prop}
{\em Proof: }
As per Lemma~\ref{lem_Coords}, choose a coordinate system
$(x,y,z)$ on a neighborhood of the bifurcation point $p$
so that $\partial/\partial z$ is everywhere transverse to the plane field
$\eta$. It is also clearly possible (via the Stable Manifold
Theorem) to choose coordinates so that
the $x$-direction corresponds to the eigenvector for the
transversally hyperbolic eigenvalue $\la^x$.
By Lemma~\ref{lem_Coords}, the unfolding of this
codimension-1 fixed point in a plane field flow corresponds to
the codimension-1 unfolding of a generic fixed point in a planar
vector field having one hyperbolic eigenvalue and one
eigenvalue with zero real part. The unfolding of the planar
saddle-node is conjugate to the system \cite{GH83}
\begin{equation}
F_z : \begin{array}{l}
\dot{x} = \la^x x \\
\dot{y} = z - a y^2
\end{array} ,
\end{equation}
for some $a\neq 0$, which, under Equation~\ref{eq_Coords},
corresponds to the vector field within $\eta$
\begin{equation}
\begin{array}{l}
\dot{x} = \la^x x \\
\dot{y} = z - a y^2 \\
\dot{z} = -g(x,y,z) (z - a y^2)
\end{array} .
\end{equation}
The curve of fixed points is thus a parabola tangent to $\eta$
at the bifurcation point.
To show the existence of a family of heteroclinic curves from
one branch of the parabola to the next, note that the
planar vector fields $F_z$ have precisely this 1-parameter
family of orbits. Upon ``suspending'' to obtain a vector field
within $\eta$, the orbits remain, since the expression for
$\dot{z}=-g(x,y,z)(z-a y^2)$ vanishes at $(0,0,0)$; hence,
$\dot{z}$ is bounded near zero in a neighborhood of the bifurcation
value and the integral curves within the invariant
plane $\dot{x}=0$ must connect.
\hfill$\Box$
\begin{figure}[htb]
\begin{center}
\epsfxsize=3.5in\leavevmode\epsfbox{figs/hopf.eps}
\end{center}
\caption{A Hopf bifurcation of singularities in a plane field
flow.}
\label{fig_Hopf}
\end{figure}
\begin{prop}\label{prop_Hopf}
In the unfolding of a $C^r$-generic ($r\geq4$)
codimension-one Hopf bifurcation on a
curve of fixed points in a plane field flow, there is an
invariant attracting or repelling paraboloid which opens
along the curve of fixed points as in Figure~\ref{fig_Hopf}.
\end{prop}
{\em Proof: }
Since only the real portion of the transverse eigenvalues
vanish, the curve of fixed points is transverse to the plane field
in a neighborhood of the bifurcation point $p$.
Hence, choose coordinates as per Lemma~\ref{lem_Coords}
such that the curve of fixed points is the $z$-axis and the
bifurcation point is at $(0,0,0)$. Again, by Lemma~\ref{lem_Coords},
this bifurcation in a plane field flow
corresponds precisely to the codimension-one Hopf bifurcation
of planar vector fields, conjugate to the truncated normal form
\cite{GH83}
\begin{equation}
F_z : \begin{array}{l}
\dot{r} = z r + a r^3 \\
\dot{\theta} = \omega
\end{array} ,
\end{equation}
where we have transformed $(x,y)$ to polar coordinates
and the constant $a$ is (in the codimension-1 scenario) nonzero.
Solving this equation for $\dot{r}=0$ yields the paraboloid
$r = \sqrt{-{z}/{a}}$,
which is either attracting or repelling, depending on the sign
of the coefficient $a$. By translating Lemma~\ref{lem_Coords}
into polar coordinates, it follows that $\dot{z}$ is of
order $r^2$, which is less than $\dot{r}$; hence,
adding the dynamics in the $z$-component
affects neither the existence nor the attracting/repelling nature
of the invariant paraboloid; however, unlike the planar case,
the paraboloid is not necessarily fibered with closed curves. In
general, orbits will spiral about the paraboloid.
\hfill$\Box$
\begin{rem}{\em \label{rem_Bifn}
We note that saddle-node or Hopf bifurcations must occur in
pairs, since the fixed point curves are circles and the index at
a bifurcation changes. However, in
the case where there are no saddle-node
or Hopf bifurcations along the singular curve, the flow is
everywhere transversally hyperbolic, and the index of
the fixed points (source, saddle, or sink) is constant along
the curve.
}\end{rem}
We conclude with the definition of a nondegenerate vector field
tangent to a plane field, and prove that such vector fields are
generic.
\begin{dfn}\label{def_Nondegenerate}{\em
A {\em nondegenerate} section of a plane field $\eta$
is a vector field $X\subset\eta$ whose fixed point set is a link having
transversally hyperbolic dynamics at all but a finite number of
points, at which the degeneracies are codimension one.
}\end{dfn}
\begin{prop}
Nondegenerate fields are generic (residual in the $C^r$ topology
$r\geq4$) within the space of sections to a $C^r$ plane field $\eta$.
\end{prop}
{\em Proof: }
We simply repeat the argument in the proof of Proposition~\ref{prop_TransHyp} using
Propositions \ref{prop_SN} and \ref{prop_Hopf}.
\hfill$\Box$
\section{Round handles and gradients}
\label{sec_3}
Let $X$ be a nondegenerate vector field contained in the plane field $\eta$.
The goal of the remaining sections is to understand restrictions
on the topology of 3-manifolds supporting plane field flows which
are forced by prescribed dynamics. A well-known example of this
occurs in the case of Anosov flows: certain three-manifolds are prohibited
from carrying Anosov dynamics. In contrast, we examine obstructions
associated to the simplest kinds of dynamics: gradient plane field flows.
We show that only certain topologically ``simple'' manifolds support
such dynamics. This will lead us to further knot-theoretic obstructions
based on the singular links in a plane field flow. An old theme is
played out: when the dynamics of $X$ are simple, the links associated
to it are simple.
\begin{lemma}\label{lem_Nonsense}
Let $M$ denote an oriented Riemannian 3-manifold and $X=-\nabla\Psi$ a
$C^r$ ($r\geq2$) gradient vector field which lies within a $C^r$
plane field $\eta$ on $M$.
Then $\Psi$ is constant on each connected component of $\mbox{Fix}(X)$, the
fixed point set of $X$. Furthermore, if $c$ is a regular
value of $\Psi$, then $\Psi\inv(c)$ is a disjoint union of tori
transverse to both $X$ and $\eta$.
\end{lemma}
{\em Proof: }
Each component of $\mbox{Fix}(X)$ is a compact connected set
of critical points for $\Psi$, whose image under $\Psi$ is a compact
connected subset of $\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}$ having measure zero, by the Morse-Sard
Theorem. For $c$ regular, $\Psi\inv(c)$ is a disjoint union of smooth
surfaces, and $X$ is transverse to each component since $X$ is a
gradient field. Hence, the plane field $\eta$ is everywhere transverse
to $\Psi\inv(c)$ and the resulting line field given by the
intersection of $\eta$ and the tangent planes to $\Psi\inv(c)$ in
$\rest{TM}{\Psi\inv(c)}$ is nonsingular. Thus, the Euler
characteristic of each component of $\Psi\inv(c)$ is zero.
The transverse vector field $X$ gives an orientation to
the surface, which excludes from consideration the Klein bottle.
\hfill$\Box$
Grayson and Pugh \cite{GP93} give examples of $C^\infty$
functions on $\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^3$
whose critical points consist of a smooth link, yet for which the level
sets are usually not tori: see Remark \ref{rem_GP}.
The above mentioned restrictions on gradient plane fields
translate into very precise conditions on the topology of the
underlying three-manifold. The fact that the manifold consists of
a finite number of thick tori $T^2\times[0,1]$ glued together
in ways prescribed by $\Psi$ implies that the manifold
can be decomposed into solid tori in a canonical fashion: this
phenomenon was identified and analyzed by Asimov and Morgan
in the 1970's \cite{Asi75,Mor78} in a completely different context.
\begin{dfn}{\em \label{def_RH}
A {\em round handle} (or RH) in dimension three is a solid torus
$H=D^2\times S^1$ with a specified index and exit set
$E\subset T^2=\partial(D^2\times S^1)$ as follows:
\begin{description}
\item[index 0:] $E = \emptyset$.
\item[index 1:] $E$ is either (1) a pair of disjoint annuli on the boundary
torus, each of which wraps once longitudinally; or
(2) a single annulus which wraps twice longitudinally.
\item[index 2:] $E=T^2$.
\end{description}
}\end{dfn}
\begin{dfn}{\em \label{def_RHD}
A {\em round handle decomposition} (or RHD) for a manifold $M$ is a
finite sequence of submanifolds
\begin{equation} \emptyset = M_0\subset M_1\subset\cdots M_n=M ,\end{equation}
where $M_{i+1}$ is formed by adjoining a round handle
to $\partial M_i$ along the exit set $E_{i+1}$ of the round handle.
The handles are added in order of increasing index.
}\end{dfn}
Asimov and Morgan \cite{Asi75,Mor78}
used round handles to classify nonsingular
{\em Morse-Smale} vector fields: that is, vector fields whose
recurrent sets consist entirely of a finite number of hyperbolic
closed orbits with transversally intersecting invariant
manifolds.
\begin{thm}\label{thm_RH}
Let $\eta$ denote a $C^r$ ($r\geq2$) plane field on $M^3$ (compact) with $X\subset\eta$
a $C^r$ nondegenerate gradient vector field.
Then the set of fixed points for $X$ forms the cores of
a round handle decomposition for $M$. Furthermore, the
indices of the fixed points correspond to the indices of
the round handles, and $X$ is transverse to $\partial M_i$ for all $i$.
\end{thm}
{\em Proof: }
Let $L$ denote the set of fixed points for $X=-\nabla\Psi$:
this is an embedded link.
We first show that every fixed point is transversally hyperbolic.
From Remark~\ref{rem_Bifn}, the only non-transversally
hyperbolic points must occur as Hopf bifurcations or
saddle-node bifurcations. Hopf bifurcations are associated to
complex transverse eigenvalues, which cannot exist in a
gradient flow. Similarly, a
saddle-node bifurcation introduces a one-parameter family
of heteroclinic connections as in Figure~\ref{fig_SN}. This
also cannot occur in a gradient flow, since by
Lemma~\ref{lem_Nonsense} we have the function $\Psi$ constant
on the curve of fixed points. The orbits of the flow
which necessarily connect one side to the other cannot
be obtained by flowing down a gradient. Hence,
each singular curve is transversally hyperbolic with constant index.
Choose $N$ a small tubular neighborhood of $L$ in $M$ and let $f$
denote a bump function in $N$ which evaluates to 1 on $L$
and is zero outside of $N$. Orient the link $L$ and perturb
$X$ to the new vector field $X+\epsilon f\frac{\partial}{\partial z}$, where
$\frac{\partial}{\partial z}$ denotes the unit tangent vector along $L$.
This yields a nonsingular flow which has $L$ as a set of
hyperbolic closed orbits and no other recurrence. After a
slight perturbation to remove any nontransverse intersections
of stable and unstable manifolds to $L$, this vector field
is a nonsingular Morse-Smale field with periodic orbit link $L$.
The work of Morgan \cite{Mor78} then implies that $L$ forms
the cores of a round handle decomposition for $M$, where the
index of each handle corresponds to the transverse index of
the curve of fixed points (source, saddle, or sink).
In \cite{Mor78} it is moreover shown that the nonsingular
Morse-Smale vector field is transverse to each $\partial M_i$; since
the neighborhood $N$ is very small, this transversality
remains in effect for $X$.
\hfill$\Box$
\begin{cor}
Gradient flows on plane fields in three-manifolds
lie on the boundary of the space of nonsingular Morse-Smale fields.
\end{cor}
{\em Proof: }
In the proof of Theorem~\ref{thm_RH}, let $\epsilon\rightarrow 0$. This gives a
one-parameter family of nonsingular Morse-Smale flows which
converges to the gradient plane-field flow.
\hfill$\Box$
\begin{cor}\label{cor_Non-Graph}
Non-gradient dynamics is a generic condition in the
space of plane field flows on an irreducible non-graph three-manifold
({\em e.g.}, a hyperbolic 3-manifold).
\end{cor}
{\em Proof: }
By the work of Morgan \cite{Mor78}, round handle decompositions
of irreducible three-manifolds exist only for the class of graph-manifolds.
\hfill$\Box$
Recall that a graph manifold is a three-manifold given by gluing together
Seifert-fibered spaces along essential torus boundaries.
Examples include $S^3$, lens spaces, and manifolds with many
$S^2\times S^1$ connected summands.
The property of being composed of Seifert-fibered pieces ({\em i.e.},
a graph manifold) is relatively rare among three-manifolds,
the ``typical'' irreducible three-manifold being composed of
hyperbolic pieces.
\begin{rem}{\em
We may push Theorem~\ref{thm_RH} a bit further. Let $\phi^t$
be a plane field flow whose chain-recurrent set consists entirely
of transversally hyperbolic curves of fixed points and a finite
set of hyperbolic periodic orbits (note that hyperbolic periodic
orbits can easily live within plane fields, even within nowhere
integrable plane fields). This situation is,
after the class of gradient flows,
the next simplest scenario dynamically. Then, by the same proof, the
connected components of the entire chain-recurrent set must form the
cores of a round-handle decomposition. Hence, the additional
dynamics forced upon plane field flows in a non-graph
manifold is something other than hyperbolic periodic orbits.
}\end{rem}
\section{The link of singularities}
\label{sec_4}
We have shown that fixed points of plane field flows appear in
links. The natural question is which links can
arise as the singular points, and what dependence is there upon
the dynamics of the plane field flow. For nondegenerate gradient
fields, it is an immediate corollary of Theorem~\ref{thm_RH} that
the singular link is a collection of fibers in the
Seifert-fibered portions of a graph manifold. We can be
more specific, however, in the special case of $S^3$. We recall two
standard operations for transforming simple knots into more
complex knots: see Figure~\ref{fig_Knots} for an illustration.
\begin{dfn}{\em
Let $K$ be a knot in $S^3$. Then the knot
$K'$ is said to be a $(p,q)$-{\em cable} of
$K$ if $K'$ lives on the boundary of a tubular neighborhood of $K$,
wrapping about the longitude (along $K$) $p$-times and about the
meridian (around $K$) $q$-times.
Let $K$ and $J$ be a pair of knots in $S^3$. Then the {\em connected sum},
denoted $K\# J$, is defined to be the knot obtained by removing
from each a small arc and identifying the endpoints along a
band as in Figure~\ref{fig_Knots}.
}\end{dfn}
\begin{figure}[htb]
\begin{center}
\epsfxsize=4.5in\leavevmode\epsfbox{figs/knots.eps}
\end{center}
\caption{Operations to generate zero-entropy knots: (left) cabling;
(right) connected sum.}
\label{fig_Knots}
\end{figure}
\begin{dfn}{\em
The {\em zero-entropy knots} are the collection of knots
generated from the unknot by the operations of cabling and
connected sum; {\em i.e.}, it is the minimal class of knots closed under
these operations and containing the unknot.
}\end{dfn}
Zero-entropy knots are relatively rare among all knots:
{\em e.g.}, none of the {\em hyperbolic knots} (such as the
figure-eight knot, whose complement has a hyperbolic structure)
are zero-entropy. The title stems from the often-discovered
fact (see \cite{G97CSF} for history)
that such knots are associated to three-dimensional flows
with topological entropy zero.
\begin{cor}
Given a nondegenerate gradient plane field flow on $S^3$,
every component of the fixed point link is a zero-entropy knot.
\end{cor}
{\em Proof: }
Wada \cite{Wad89} classifies the knot types for
cores of all round handle decompositions
on $S^3$. Each component is a zero-entropy knot.
\hfill$\Box$
\begin{rem}{\em
Much more can be said: Wada in fact classifies all possible {\em links}
which arise as round handle cores on all graph manifolds. This class
of {\em zero-entropy links} is an extremely restricted class, which lends
credence to the motto that simple dynamics implicate simple links in
dimension three. We note this same class of links appears independently
in the study of nonsingular Morse-Smale flows,
suspensions of zero-entropy disc maps,
and in Bott-integrable Hamiltonian flows with two degrees of freedom.
}\end{rem}
\begin{rem}{\em \label{rem_GP}
It is possible to construct gradient flows on $S^3$ (for example)
in which the fixed point set is an embedded link which is {\em not}
a zero-entropy link. Let $L_1$ and $L_2$ denote any pair of links in $S^3$
which each have at least three components. Grayson and Pugh \cite{GP93}
prove the existence of $C^\infty$
functions $\Psi_1, \Psi_2:\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^3\rightarrow\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}$ which
have $L_1$ and $L_2$ as the (respective) sets of critical points.
Moreover, these functions are proper and, for large enough $c\in\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}$,
the inverse image of $c$ is a smooth 2-sphere near infinity. Hence,
we may consider the balls $B_i$ bounded by $\Psi_i\inv(c)$ and glue
them together along the boundaries, obtaining $S^3$. The
resulting function $\Psi$ given by $\Psi_1$ on $B_1$ and $-\Psi_2+2c$
on $B_2$ has as its gradient flow the split (unlinked) sum of
$L_1$ and $L_2$ as its fixed points. Thus, this flow cannot live
within a plane field.
}\end{rem}
It is not ostensibly clear that every zero-entropy link in $S^3$ is
realized as the zero set of a gradient flow within a plane field.
We close this section with a realization theorem for such flows
which shows that, in fact, a particular subclass of round-handle
decompositions (and, hence, zero-entropy links) is realized.
\begin{lemma} \label{lem_Essential}
If $X$ is a nondegenerate gradient field on $M$ contained in
the plane field $\eta$, then each index-1 round-handle $H$ in
the decomposition must be attached to $\partial M_i$ along annuli which are
essential (homotopically nontrivial) in $\partial M_i$.
\end{lemma}
{\em Proof: }
Assume that $M_{i}$ is the $i$th stage in a round handle
decomposition, and that $H$ is an index-1 round handle
with an exit annulus $E$ which is essential in $\partial H$
by definition. By Theorem~\ref{thm_RH}, the intersection of
$\eta$ with $\partial M_i$ is always transverse. Thus,
if $H$ is attached to $M_i$ along an annulus $A\subset\partial M_i$,
then the foliations given by the intersections of $\eta$
with the tangent planes to $A$ and $E$ respectively
must match under the attachment. We claim this is
impossible when $A$ is homotopically trivial in $\partial M_i$.
Define the {\em index} of a smooth (oriented) curve $\gamma$ in
an orientable surface with a (nonsingular, oriented) foliation ${\cal F}$
to be the degree of the map which associates
to each point $p\in\gamma$ the angle between
the tangent vectors to $\gamma$ and ${\cal F}$ at $p$. This index is
independent of the metric chosen and also
invariant under homotopy of $\gamma$ or of ${\cal F}$;
hence, we can speak of the index of an annulus in a surface
with foliation.
When $A$ is homotopically trivial, the index must be
equal to $\pm 1$, since a foliation is locally a product.
However, the index of the exit annulus $E\subset\partial H$ must be
zero as follows.
Under the gradient field $X$, the core of the 1-handle is a curve
$\kappa$ of fixed points with transverse index 1 whose unstable manifold
$W^u(\kappa)$ intersects $\partial H$ transversally along the core of the
exit set $E$. Deformation retract $E$ to a small neighborhood of
${\cal W}^u(\kappa)\cap\partial H$ --- here, the intersections with $\eta$
are always transverse. Next, homotope the annulus to a neighborhood
of $\kappa$ by integrating the gradient field $X$ backwards in
time. This has the effect of taking the annulus transverse
to $W^u(\kappa)$ and sliding along $W^u(\kappa)$ back to $\kappa$.
Since $X$ points outwards along $W^u(\kappa)$, the image of the
annulus $E$ under the homotopy is always transverse to $X$, and
hence to $\eta$. The fact that $\eta\pitchfork\kappa$ then implies that
the foliation on $E$ induced by $\eta$ must be homotopic through nonsingular
foliations to a product foliation by intervals on the annulus,
which implies that the longitudinal annulus $E$ has index zero. Note that
this works for exit sets $E$ which wrap any number of times about
the longitude of $H$ (to cover both types of index-1 round handles).
\hfill$\Box$
Hence, any round-handle decomposition which is realizable
as a gradient plane field flow must have all 1-handles attached
along essential annuli. We call such a round-handle
decomposition {\em essential}.
\begin{thm}\label{thm_Realized}
Let $M$ be a compact 3-manifold with $L$ an indexed link. Then
$L$ is realized as the indexed set of zeros for some nondegenerate
gradient plane field flow on $M$ if and only if $L$ is the indexed
set of cores for an essential RHD on $M$.
\end{thm}
{\em Proof: }
The necessity is the content of Lemma~\ref{lem_Essential}. Given
any essential RHD, we construct a corresponding plane field gradient
flow. One may begin with the fact proved by Fomenko that
any essential round-handle decomposition can be generated by
a vector field $X$ integrable via a Bott-Morse function
$\Psi:M\rightarrow\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}$ with all critical sets being circles
(see \cite{CMAN} for a
detailed exposition). After choosing a metric on $M$ we claim that
$-\nabla\Psi$ lives within the plane field $\eta$ orthogonal to $X$.
Indeed, away from $L$ the plane field $\eta$ will be spanned by
$\nabla\Psi$ and $\nabla\Psi\times X$ since these are linear
independent vectors orthogonal to $X$ (recall $X$ is tangent
to the level sets of $\Psi$). Thus $-\nabla\Psi$ clearly
lies in $\eta$ on the complement of $L$. Along $L$ the
gradient $-\nabla\Psi$ lies in $\eta$ since it is zero.
\hfill$\Box$
The above construction may be modified so as to force the plane field to
twist monotonically along orbits of the gradient field, by a careful choice of the
vector field $X$. This implies that all of the permissible round handle decompositions
may be realized by a totally nonintegrable plane field (a contact structure). Totally
integrable plane fields are not so flexible, as will be illustrated in the next
section.
\section{Flows on foliations and contact structures}
\label{sec_5}
\subsection{Foliations}
In the case where our given plane field has some geometrical
property, we may further
restrict the types of round-handle decompositions which may
contain a gradient flow. For example, if the plane field $\eta$
is integrable, it determines a foliation on the manifold. In this
subsection, we note that, in this case, $S^3$ cannot support such
a gradient flow. This result, which is an obvious corollary of
Novikov's Theorem on foliations, generalizes to other
three-manifolds.
Recall from the theory of foliations on three-manifolds
(see, {\em e.g.}, \cite{God91})
that a {\em Reeb component} is a foliation of the solid torus
$D^2\times S^1$ that consists of the boundary $T^2$ leaf
along with a one-parameter family of leaves, each homeomorphic
to $\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^2$ and limiting onto the boundary with nontrivial
holonomy, as in Figure~\ref{fig_Reeb}. A codimension-one
foliation of a three-manifold is {\em taut} if there do not
exist Reeb components or ``generalized'' Reeb components (see,
{\em e.g.}, \cite{ET97} for definitions). An equivalent definition
of taut is that given any leaf $\L$ there exists a closed
curve through $\L$ transverse to the foliation.
\begin{figure}[htb]
\begin{center}
\epsfxsize=3.45in\leavevmode\epsfbox{figs/reeb.eps}
\hspace{0.2in}
\epsfxsize=1.15in\leavevmode\epsfbox{figs/otdisc.eps}
\end{center}
\caption{A Reeb component in a foliation on a 3-manifold
(left) can be perturbed into an overtwisted contact
structure (right).}
\label{fig_Reeb}
\end{figure}
It is straightforward to show that gradient fields must lie within
taut foliations:
\begin{thm}\label{thm_Taut}
Let $\cal F$ denote a codimension-1 foliation on a compact
three-manifold $M$ which contains a nondegenerate gradient
vector field $X$. Then ${\cal F}$ is taut.
\end{thm}
{\em Proof: }
Assume that $X=-\nabla\Psi$ is a nondegenerate gradient field on a foliation
$\cal F$. For $\L$ a leaf of ${\cal F}$, the restriction of $X$ to $\L$
must also be a gradient flow.
In the case where $\L$ is compact, there must be a nondegenerate
fixed point of $X$ on $\L$ which lies on a circle of fixed points
transverse to ${\cal F}$ (note that in the case of a boundary torus in
a Reeb component, this is an immediate contradiction).
In the case where $\L$ is not a compact leaf, choose some nontrivial
path $\gamma\subset\L$ whose endpoints are directly above one
another in a local product chart. Then, by perturbing $\gamma$
to be transverse to ${\cal F}$, we may close it up to a transverse loop
through $\L$.
\hfill$\Box$
This result can be greatly improved by considering the holonomy
of the foliation. Recall that the {\em holonomy} of any closed curve
$\gamma:S^1\rightarrow\L$ in a leaf $\L$ of a codimension-one
foliation ${\cal F}$ is the germ of the Poincar\'e
map associated to the characteristic foliation on an annulus
transverse to $\L$ along $\gamma$.
The holonomy of a curve is an invariant of its homotopy class within
the leaf. A foliation has {\em vanishing holonomy} if the holonomy of
every curve $\gamma$ is trivial (the identity).
\begin{thm}\label{thm_Foliation}
Any closed orientable three-manifold $M$ containing a nondegenerate
gradient field within a ($C^r$ for $r>1$) codimension-one foliation
is a surface bundle over $S^1$.
\end{thm}
{\em Proof: }
Suppose that $M$ admits a foliation ${\cal F}$ which supports
a nondegenerate gradient field. Then, by Theorem~\ref{thm_RH},
$M$ has a round handle decomposition where all the regular
tori are transverse to the foliation. The foliation on each
round handle is equivalent to the product foliation by discs on
$D^2\times S^1$, since these solid tori are filled with leaves
transverse to the boundary each having a gradient flow with a single
fixed point. We show in subsequent steps that the foliation ${\cal F}$
may be modified within the round handle structure so that the new
foliation ${\cal F}'$ has no holonomy. Once we show this,
the celebrated theorem of Sacksteder implies that this foliation
must be topologically conjugate to the kernel of a closed
nondegenerate 1-form on $M$ \cite{Sac65}. The existence of this
1-form implies, via the theorem of Tischler
\cite{Tis70}, that $M$ must be a surface bundle over $S^1$.
We illustrate in Figure~\ref{fig_Holonomy} below
that it {\em is} possible to have gradient fields within a
foliation having holonomy, so it is truly necessary to develop
the following modification procedure, which makes use of
``shearing'' the foliation along 1-handles ({\em cf. } \cite{EHN81}).
\begin{figure}[htb]
\begin{center}
\epsfxsize=2.0in\leavevmode\epsfbox{figs/holonomy.eps}
\end{center}
\caption{A foliated RHD on $T^2\times S^1$ with holonomy along
each 1-handle: the
2-handle has been removed and the $S^1$-factor cut open, revealing
a pair of 1-handles attached to a 0-handle with inverse attaching maps.}
\label{fig_Holonomy}
\end{figure}
Denote by $M_0$ the (disjoint) union of all the 0-handles and
by $M_i$ ($1\leq i\leq N$) the subsequent stages in the decomposition:
\[
M_i = \left(\left. M_{i-1}\sqcup H_i\right) \right/ \phi_i,
\]
where $H_i$ is the $i$th 1-handle and $\phi_i:E_i\hookrightarrow\partial M_{i-1}$
is the attaching map on the exit set $E_i\subset H_i$.
Recall that each exit set $E_i$ is either one or two annuli and
that the boundary of each $M_i$, $\partial M_i$, is the disjoint union
of a collection of tori.
For each 1-handle $H_i$, let $W_i^u$ denote the (2-dimensional)
unstable manifold to the core of $H_i$. Modify the
round handle structure so that each $H_i$ is very
``thin'' -- that is, each $H_i$ is restricted to a small neighborhood of
$W^u_i$, appending the ``leftover'' portion to the neighboring 2-handles.
Denote by
\[
{\mbox{Bd}}_i = \partial M_0\bigcup_{j=1}^i W^u_j ,
\]
the 2-complex given by the union of all the 0-handle boundaries and
unstable manifolds of the 1-handles in $M_i$.
{\em Claim 1:}
${\cal F}$ has vanishing holonomy if the restriction of ${\cal F}$ to Bd$_N$
has vanishing holonomy.
{\em Proof 1:} Let $\gamma$ denote a loop within a leaf
$\L$ of ${\cal F}$. Then the restriction of $\L$ to each $k$-handle
is a collection of disjoint discs whose boundaries lie in the union of
0- and 1-handles. Push $\gamma$ to these boundaries and, since
$\partial H_i$ is very close to $W^u_i$, perturb $\gamma$ to
lie within $W^u_i$ for each $H_i$ it intersects.
Since Bd$_N$ is transverse to ${\cal F}$, we may choose the transverse
annulus $A$ containing $\gamma$ to lie within this set.
\hfill$\Box$$_1$
In what follows, we consider holonomy on the 2-complex
Bd$_N$, keeping in mind that the 1-handles are actually thin
neighborhoods of the 2-cells $W^u_i$. The holonomy on each
component of $\partial M_i$ is equivalent to that on the corresponding
piece of Bd$_N$ since each $H_i$ has a product foliation.
{\em Claim 2:} The maps $\{\phi_i\}_1^N$ may be isotoped so that
the induced foliation ${\cal F}'$ on Bd$_N$ is without holonomy.
{\em Proof 2:}
It suffices to show that the foliation restricted to each
$\partial M_i$ is without holonomy (a product foliation): we
proceed by induction on $i$.
On the boundary of $M_0$ the foliation ${\cal F}$ restricts to a product
foliation by circles.
Assume as an induction hypothesis a lack of holonomy on $\partial M_{i-1}$.
There are three cases to consider: (1) $H_i$ is an orientable handle
with attaching circles in the same component of $\partial M_{i-1}$;
(2) $H_i$ is orientable with attaching circles in two
distinct components of $\partial M_{i-1}$; and (3) $H_i$ is
nonorientable.
Case (1): Let $C_\pm$ denote the circles in the selected
component $T$ of $\partial M_{i-1}$ along which $W^u_i$ is attached.
Note $C_\pm$ divides $T$ into two annuli $A_0$ and $A_1.$
After fixing a diffeomorphism from $C_+$ to $C_-$ there is a
``handle holonomy map''
$f_H:C_+\rightarrow C_-$ which is the diffeomorphism given by sliding
along leaves on $\phi_i(H_i)$. There are corresponding
``boundary holonomy maps'' $f_j:C_+\rightarrow C_-$ given by sliding
along leaves on $A_j$. Isotope $\phi_i$ on $C_+$ so that
$f_H$ equals $f_0$ up to a rigid rotation
(which is necessary in order to add subsequent handles along curves
transverse to ${\cal F}$ --- see Claim 3).
The holonomy on the two new components of $\partial M_i$ is determined
by taking the transverse
curve $C_+$ (actually one must take a parallel copy of $C_+$ that
sits in $\partial M_i$) as a section. These holonomy maps factor as
$f_1\inv\circ f_H$ and $f_H\inv\circ f_0$;
however, the holonomy along $C_+$ within $\partial M_{i-1}$ is a map of the
form $f_1\inv\circ f_0$, which, by induction, is a rigid rotation.
Hence, up to rotations, $f_0=f_1$. Since we chose $f_H=f_0$
up to rotations the holonomy on $\partial M_i$ vanishes.
Case (2): If $H_i$ connects two disconnected boundary components
of $M_{i-1}$, then the holonomy along $H_i$ will always cancel
with itself as follows. Denote by $f_H:C_+\rightarrow C_-$ the handle holonomy
maps as before. Then the global holonomy map along $\partial M_i$
is of the form $g_+\circ f_H\circ g_- \circ f_H\inv$,
where $g_+:C_+\rightarrow C_+$ and $g_-:C_-\rightarrow C_-$ are holonomy self-maps
along loops in the two components of $\partial M_{i-1}$, and hence by
induction, identity maps.
Case (3): If $H_i$ has connected exit set, the proof follows as
in Case (1), since the handle must connect a single component
of $\partial M_{i-1}$ to itself: isotope $\phi_i$ so that the handle holonomy
map equals the holonomy map along the boundary up to a rigid
rotation.
\hfill$\Box$$_2$
{\em Claim 3:}
This ``linearization'' of ${\cal F}$ does not affect the topology of $M$.
{\em Proof 3:}
Throughout the addition of the
1-handles, nothing about the topology of $M$ has changed, since
the handle structure is identical --- we modify only the foliation.
However, after attaching the last 1-handle, the characteristic
foliation on the boundary tori must be linear and rational, in order
to glue in the 2-handles respecting the product foliation on their
boundaries. The slopes of ${\cal F}$ restricted to $\partial M_N$ completely
determine the topology of $M$ after adding the 2-handles (these
are Dehn filling coefficients).
Hence, we must be able to linearize all of the attaching maps for the
1-handles without changing the boundary slopes at the end of the
sequence. To do so, we preserve at every stage the {\em rotation
number} of the holonomy maps $h_i$ which slide the attaching
curves of $H_i$ along $\partial M_i$. Recall that to every
diffeomorphism $f:S^1\rightarrow S^1$ is
associated a rotation number $\rho_f\in\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}/\mbox{\bbb Z}} %{\mbox{\bf Z}\hspace{-.06in}{\bf Z}$ which measures
the average displacement of orbits of $f$ (see, {\em e.g.}, \cite{GH83}).
When modifying $\phi_i$ to $\tilde\phi_i$ in the above
procedure, we may compose $\tilde\phi_i$ with a rigid rotation
by the angle necessary to preserve the rotation number of the
holonomy map $h_i$ acting on the attaching curves in $\partial M_{i-1}$
(without adding further Dehn twists).
This shearing maintains the average slope of the boundary foliation
at each stage without adding holonomy. Hence, at the end
of the 1-handle additions, when the original foliation had all
boundary components with linear foliations of a particular fixed
slope, the modified foliation also has linear boundary foliations
with the same slope. Thus, adding the 2-handles is done using the
same surgery coefficients, yielding the original manifold $M$
with a foliation having trivial holonomy.
\hfill$\Box$$_3$
Claims 1-3 complete the proof of Theorem~\ref{thm_Foliation}.
\hfill$\Box$
\begin{rem}{\em
Of course, not every surface bundle over $S^1$ may support a
gradient field within a foliation: there is still the restriction that
$M$ be a graph-manifold. This translates precisely into a condition
on the monodromy map of the fibration --- the monodromy must
be of periodic (or reducibly periodic) type with respect to the
Nielsen-Thurston classification of surface homeomorphisms.
Any pseudo-Anosov piece in the monodromy forces
hyperbolicity, contradicting the graph condition.
It is not hard to see that any such bundle can be
given a gradient field lying within each fiber $F$ of the bundle by
choosing a Morse function $\phi:F\rightarrow\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}$ which is equivariant with
respect to the monodromy map.
}\end{rem}
\begin{rem}{\em
All of the results of this section apply not only to
gradient flows, but also to {\em gradient-like flows}, or
flows for which there exists a function which decreases strictly
along non-constant flowlines. The reason why nondegenerate gradient-like
flows in foliations determine round-handle decompositions whereas
for general plane fields they do not lies in the fact that
the Hopf bifurcation of Proposition~\ref{prop_Hopf} cannot take
place among gradient-like flows in the integrable case, while
it can in the nonintegrable.
}\end{rem}
\subsection{Contact structures}
In contrast to the case of an integrable plane field, one may
consider the class of {\em contact structures}, which has
attracted interest in the fields of symplectic geometry and
topology, knot theory, mechanics, and hydrodynamics.
\begin{dfn}\label{def_Contact}{\em
A {\em contact form} on a three-manifold $M$ is a one-form
$\alpha\in\Omega^1(M)$ such that the Frobenius integrability
condition fails everywhere: that is,
\begin{equation}
\alpha\wedge d\alpha \neq 0.
\end{equation}
A {\em contact structure} on $M$ is a plane field $\xi$ which is
the kernel of a locally defined contact form: that is,
\begin{equation}
\xi_p = \{ \mbox{{\bf v}} \in T_p : \alpha(\mbox{{\bf v}})=0 \},
\end{equation}
for each $p\in M$.
}\end{dfn}
Contact structures are thus maximally nonintegrable: the plane
field is locally twisted everywhere.
One may think of a contact structure as being an {\em anti-foliation},
which leads one to suspect that the topology of the manifold
may be connected to the geometry of the structure, as is often
the case with foliations. Indeed, the contrast between
foliations with Reeb components and those without Reeb components
is reflected in the {\em tight / overtwisted dichotomy} in contact
geometry (due primarily to Eliashberg \cite{Eli92} and Bennequin \cite{Ben82}).
\begin{dfn}\label{def_Tight}{\em
Given a contact structure $\xi$ on $M$ and an embedded
surface $F\subset M$, then the {\em characteristic foliation}
$F_\xi$ is the (singular) foliation induced by the (singular) line field
$\{T_pF\cap\xi_p : p\in F\}$.
A contact structure $\xi$ is {\em overtwisted} if there exists
an embedded disc $D\in M$ such that $D_\xi$ has a limit cycle,
as in Figure~\ref{fig_Reeb}(right).
A contact structure is {\em tight} if it is not overtwisted.
}\end{dfn}
The classification of contact structures follows along lines similar
to that of codimension one foliations with or without Reeb
components. An infinite number of homotopically distinct overtwisted
contact structures exist on every closed orientable
three-manifold \cite{Mar71,Lut77} and are algebraically classified up to
homotopy \cite{Eli89}. Tight structures, on the other hand, are
quite mysterious: {\em e.g.}, it is unknown whether they exist on all
three-manifolds.
Several examples of the similarity between tight contact structures and
Reebless foliations are provided by the recent work of Eliashberg
and Thurston \cite{ET97}. For example, Reebless foliations can be
perturbed into tight contact structures and foliations with Reeb
components can be perturbed into overtwisted structures ({\em cf. }
Figure~\ref{fig_Reeb}). Also, both Reebless foliations
and tight structures satisfy a strong inequality restricting Euler
classes. Tight structures are somewhat more general than
Reebless foliations since the former can exist on $S^3$ \cite{Ben82}
while the latter cannot \cite{Nov67}. Likewise, overtwisted
structures are slightly more general than their foliation
counterparts via the following observation, to be contrasted with
Theorem~\ref{thm_Taut}:
\begin{prop}
Any nondegenerate gradient field $X$ which lies within a tight
contact structure $\xi$ on $M^3$ also lies within an overtwisted
contact structure $\xi'$ on $M^3$.
\end{prop}
{\em Proof: }
The canonical way to turn a tight structure into an
overtwisted structure is by performing a {\em Lutz twist}
\cite{Lut77,Mar71} on a simple closed curve $\gamma$
transverse to $\xi$. We execute a version of this twisting which
respects a gradient field.
Given a gradient field $X\subset\xi$, choose a curve $\gamma$ of fixed
points of index zero (sinks). Translate
the function $\Psi$ whose gradient defines $X$ so that
$\rest{\Psi}{\gamma}\equiv 0$. Since $\gamma$ is an index
zero curve, $\Psi$ increases as one moves radially away from $\gamma$.
Let $N$ denote a tubular neighborhood of $\gamma$ whose boundary
is a connected component of $\Psi\inv(\epsilon)$ for some $\epsilon>0$.
Place upon $N$ the natural cylindrical coordinates $(\Psi,\theta,z)$.
In analogy with Lemma~\ref{lem_Coords}, we may choose $\theta$ and
$z$ so that $\rest{\xi}{N}$ is the kernel of the locally defined 1-form
\begin{equation}
\alpha = g(\Psi,\theta,z)d\theta + dz ,
\end{equation}
for some function $g$ with $g(0,\theta,z)=0$.
The contact condition implies that
\begin{equation}\label{eq_grow}
\frac{\partial g}{\partial\Psi} > 0 .
\end{equation}
Replacing this structure locally with the kernel of the form
\begin{equation}
\alpha' =
\sin\left(\frac{\pi}{4}+\frac{2\pi g}{g(\epsilon,\theta,z)}\right)
g\,d\theta
+ \cos\left(\frac{\pi}{4}+\frac{2\pi g}{g(\epsilon,\theta,z)}\right)dz ,
\end{equation}
yields a contact structure since
\begin{equation}
\alpha'\wedge d\alpha' = \left[
\cos\left(\frac{\pi}{4}+\frac{2\pi g}{g(\epsilon,\theta,z)}\right)
\sin\left(\frac{\pi}{4}+\frac{2\pi g}{g(\epsilon,\theta,z)}\right)
+ \frac{2\pi g}{g(\epsilon,\theta,z)}
\right] \frac{\partial g}{\partial\Psi}\,d\Psi\wedge d\theta\wedge dz ,
\end{equation}
and this coefficient is positive by Equation~\ref{eq_grow}.
This contact structure agrees with that defined by $\alpha$
along the torus $\Psi=\epsilon$ since
\begin{equation}
\rest{\alpha'}{\Psi=\epsilon}
= \sin\left(\frac{9\pi}{4}\right)g(\epsilon,\theta,z)d\theta
+ \cos\left(\frac{9\pi}{4}\right) dz
= \frac{\sqrt{2}}{2}\rest{\alpha}{\Psi=\epsilon} ,
\end{equation}
and these have the same kernel.
Furthermore, this modified structure contains the
vector field $X=\nabla\Psi$, since $X$ points in the $-d/d\Psi$
direction. Finally, one can easily show that a perturbation of a
constant-$z$ disc has a limit cycle in the characteristic foliation
near $c=\epsilon/2$ ({\em cf. } \cite{Ben82}); hence, this
defines an overtwisted structure containing $X$. This construction
can obviously be done in the $C^\infty$ category using bump
functions.
\hfill$\Box$
\begin{ex}{\em
Consider the flow on $S^3$ (considered as the unit sphere in
$\mbox{\bbb R}} % {\mbox{\rm I} \hspace{-.03in} {\bf R}^4$ with the induced metric) given by the gradient of the
function
\begin{equation}
\Psi = \frac{1}{2}(x_1^2+x_2^2)-\frac{1}{2}(x_3^2+x_4^2),
\end{equation}
the gradient being taken in $S^3$.
One can check that the fixed point set consists of a
pair of unknots linked once in a {\em Hopf link}, as in
Figure~\ref{fig_NS}. The standard tight contact form on $S^3$ is
\begin{equation}
\alpha = \frac{1}{2}\left(
x_1 dx_2 - x_2 dx_1 + x_3 dx_4 - x_4 dx_3 \right) .
\end{equation}
A simple calculation shows that $\alpha$ is a contact form on $S^3$
with $\nabla\Psi\subset\ker\alpha$. However, we may Lutz twist this
structure in a neighborhood of the fixed point links: a family of
such overtwisted forms ($n\in\mbox{\bbb Z}} %{\mbox{\bf Z}\hspace{-.06in}{\bf Z}^+$) is given by \cite{GV83}
\begin{equation}
\alpha_n = \cos\left(\frac{\pi}{4}+n\pi(x_3^2+x_4^2)\right)
(x_1 dx_2 - x_2 dx_1) +
\sin\left(\frac{\pi}{4}+n\pi(x_3^2+x_4^2)\right)
(x_3 dx_4 - x_4 dx_3),
\end{equation}
from which it can be shown that $\nabla\Psi\subset\ker\alpha_n$.
Here, the integer $n$ denotes the number of twists that the
plane field undergoes as an orbit travels from source to sink
in Figure~\ref{fig_NS}.
}\end{ex}
\begin{figure}[htb]
\begin{center}
\epsfxsize=3.0in\leavevmode\epsfbox{figs/ns.eps}
\end{center}
\caption{The gradient field on $S^3$ having a Hopf link of fixed points
exists within both tight and overtwisted contact structures.}
\label{fig_NS}
\end{figure}
\section{Two questions}
\label{sec_6}
This work has focused on the case of gradient flows in plane fields
in dimension three, as the round-handle theory is most interesting here.
However, there are natural questions about gradient flows in
arbitrary distributions for manifolds of dimension greater than
three. We do not present any results in this area, but rather note
that many of the tools remain valid: fixed point sets of a vector
field constrained to a codimension-$k$ distribution consists of a
finite collection of embedded $k$-dimensional submanifolds.
Two problems emerge. In the case of a codimension-one distribution,
nondegenerate gradient fields induce round-handle decompositions.
However, every manifold of dimension greater than three whose Euler
characteristic is zero possesses
an RHD. Are there any such manifolds of dimension greater than three
which do not possess a nondegenerate gradient field tangent to a
codimension-one distribution? Secondly, in the case of
higher codimension distributions, what restrictions exist on
the topology of the fixed point sets? The case of a plane field on
a four-manifold is particularly interesting with respect to the
genera of the (two-dimensional) fixed point sets.
\noindent{\sc ACKNOWLEDGMENTS}
\vspace{0.05in}
This work has been supported in part by the National Science
Foundation [JE: grant DMS-9705949; RG: grant DMS-9508846].
The authors wish to thank Mark Brittenham, John Franks, Will Kazez,
Alec Norton, and Todd Young for their input. Special thanks are due
the referee for constructive remarks.
\small
|
\section{Introduction}
A spherical gravitational wave antenna ideally has equal sensitivity to
gravitational waves from all directions and polarizations and is able to
determine the directional information and tensorial character of a
gravitational wave. The solution for the inverse problem for a noiseless
antenna has been known for some time \cite{Wagoner_Pavia_1976}, and an
analytic solution for an noisy antenna was recently found
\cite{Merkowitz_PRD_1998}. These solutions are quite elegant as they only
require linear algebra to estimate the wave direction and polarization from
the detector outputs.
By monitoring the five quadrupole modes of an elastic sphere, one has a
direct measurement of the effective force of a gravitational wave on the
sphere \cite{Lobo_PRD_1995}. The standard technique for doing so on
resonant detectors is to position resonant transducers on the surface of
the sphere that strongly couple to the quadrupole modes. A number of
proposals have been made for the type and positions of the transducers
\cite{Johnson_PRL_1993,Lobo_EPL_1996,Zhou_PRD_1995}. What all of these
proposals have in common is that the outputs of the transducers are
combined into five ``mode channels'' $g_m(t)$ that are constructed to have
a one-to-one correspondence with the quadrupole modes of the sphere and
thus the spherical amplitudes of the gravitational
wave~\cite{Merkowitz_PRD_1995,Lobo_CQG_1998}.
The mode channels $g_m\/$ can be collected to form a ``detector response''
matrix
\begin{equation}
\bi{A} =
\left[\begin{array}{ccc}
g_{1} - \frac{1}{\sqrt{3}}g_{5} & g_{2} & g_{4} \\
g_{2} & -g_{1} - \frac{1}{\sqrt{3}}g_{5} & g_{3} \\
g_{4} & g_{3} & \frac{2}{\sqrt{3}} g_{5}
\end{array} \right],
\label{eqn:cartesian_strain_tensor}
\end{equation}
which, in the absence of noise in the detector, is equal to the GW strain
tensor, expressed in lab frame coordinates. The latter tensor has the
canonical form
\begin{equation}
\bi{H} = \left[
\begin{array}{ccc}
h_+ & h_\times & 0 \\
h_\times & -h_+ & 0 \\
0 & 0 & 0
\end{array} \right],
\label{eqn:wave_strain}
\end{equation}
in the wave frame, and is related to $\bi{A}$ by an orthogonal
transformation ---a rotation. $\bi{H}$ clearly has an eigenvector,
$\bi{v}_3$, say, with zero eigenvalue which corresponds to the wave
propagation direction. The same therefore applies to $\bi{A}$, and this
enables the determination of the wave direction by a straightforward algebraic
procedure directly from detector data: it is the eigenvector of $\bi{A}$
with null eigenvalue.
Things change when the detector is noisy: noise gets added to
the signal in the mode channels, destroying the equivalence between
the data matrix $\bi{A}$ and the signal matrix $\bi{H}$. However, it
has been shown that under ideal conditions of the noise a modified version
of the above procedure can be used \cite{Merkowitz_PRD_1998}: the eigenvector
of the noisy $\bi{A}$ with eigenvalue {\it closest to zero\/} is the
{\it best\/} approximation to the actual incidence direction of the
gravitational wave.
In this paper we shall be taking an {\it analytic\/} approach to the
diagonalization problem, whereby errors in the estimated GW parameters can
be assessed to any desired degree of accuracy. Inherent in this approach is
the unambiguous definition of the incidence direction estimate, as well as
a Cartesian coordinate convention for it, which rids us of the ambiguities
intrinsically associated to the Euler angle characterization for incidence
directions near the Poles. Errors in these quantities will be shown to be
incidence direction independent. In addition, we shall also address the
problem of estimating the GW amplitudes $h_+$ and $h_\times$, and their
errors. Previous authors \cite{Zhou_PRD_1995,Stevenson_PRD_1997} found it
impossible to give isotropic estimates of these quantities, a very strange
result for a spherical detector. We explain why these results come about
and we show how the problem can be solved by properly including all
the necessary information.
The paper is organized as follows. We begin in section~\ref{sec:A_ev} by
deriving analytic expressions for the eigenvalues of $\bi{A}$. We then
use these expressions to find the first order statistical errors in the
eigenvalues in section~\ref{sec:uncertainties_ev}, followed by the direction
estimation error in section~\ref{sec:dir_error}. Higher order corrections to
these errors are presented in section~\ref{sec:corrections}.
In section~\ref{sec:polarization} we discuss the errors on the polarization
amplitude estimates. We explain why past solutions
have direction dependent errors on these quantities and we describe a
maximum likelihood algorithm, based a hypothesis on the physical nature
of the source, that fulfills the natural property of source location
independence.
\section{Detector response eigenvalues}
\label{sec:A_ev}
The detector response matrix $\bi{A}$ is symmetric and traceless and has
the eigenvalue equation
\begin{equation}
\bi{A}\bi{v}_k = \lambda_k\bi{v}_k, \; k=1,2,3,
\label{eqn:noisy_ev}
\end{equation}
where the eigenvectors are normalized in the usual way
\begin{equation}
\bi{v}_k \cdot \bi{v}_l = \delta_{kl}.
\end{equation}
Expanding equation \eref{eqn:noisy_ev} we find
\begin{equation}
\lambda_k^3 - g^2 \lambda_k - D = 0,
\end{equation}
where we have defined
\numparts
\begin{eqnarray}
g^2 \equiv g_1^2 + g_2^2 + g_3^2 + g_4^2 + g_5^2 \\[1 ex]
D \equiv \det(\bi{A}).
\end{eqnarray}
\endnumparts
Solving this cubic equation we find the eigenvalues of $\bi{A}$ to be
\begin{equation}
\lambda_k = -\frac{2}{\sqrt{3}}\;g\,\cos\theta_k, \ k=1,2,3,
\label{eq:roots}
\end{equation}
where
\begin{equation}
\theta_k = \frac{\theta + 2(k-1)\pi}{3},\ \mbox{and}\
\cos\theta = -\frac{3\sqrt{3}}{2}\;\frac{D}{g^3}.
\label{eq:1.9}
\end{equation}
The $k=3$ eigenvalue is identically zero in the absence of noise, so it
will generally be the one closest to zero in the presence of noise. Random
fluctuations may eventually change this (more likely for low SNR), but we
shall always take the corresponding eigenvector $\bi{v}_3$ as the best
approximation to the direction of the source.
The amplitude of the wave $h$ can be calculated in many ways from the mode
channels (for example, $g\/$ is an estimate for the amplitude), but the
semi-difference of the other two eigenvalues will
give the best estimate \cite{Merkowitz_PRD_1998},
\begin{equation}
h = \frac{1}{2}\,\left(\lambda_2 - \lambda_1\right).
\label{eq:estim_h}
\end{equation}
\section{Eigenvalue errors}
\label{sec:uncertainties_ev}
We assume that the mode channels $g_m$ have uncorrelated noise with zero
mean and variance $\sigma_{g_m}^2 \equiv E\left\{(\delta g_m)^2\right\}$.
The lowest order statistical errors in the eigenvalues are easily
calculated by
\begin{equation}
\sigma_{\lambda_k}^2
=
\sum_{m=1}^5
\left(\frac{\partial\lambda_k}{\partial g_m}\right)^2
\sigma_{g_m}^2,
\label{eq:ev_error}
\end{equation}
where the derivatives of the eigenvalues are given by
\begin{equation}
\frac{\partial\lambda_k}{\partial g_m} =
-\frac{2}{\sqrt{3}}\;\frac{g_m}{g}\,\cos\theta_k +
\frac{\partial D}{\partial g_m}\,\sin\theta_k,
\label{eq:derl}
\end{equation}
and the derivatives of the determinant $D\/$ are
\numparts
\begin{eqnarray}
\frac{\partial D}{\partial g_1}
& = &
-\frac{4}{\sqrt{3}}g_1g_5 - g_3^2 + g_4^2,
\\
\frac{\partial D}{\partial g_2}
& = &
-\frac{4}{\sqrt{3}}g_2g_5 + 2g_3g_4,
\\
\frac{\partial D}{\partial g_3}
& = &
-2g_3\left(g_1-\frac{g_5}{\sqrt{3}}\right) + 2g_2g_4,
\\
\frac{\partial D}{\partial g_4}
& = &
2g_4\left(g_1+\frac{g_5}{\sqrt{3}}\right) + 2g_2g_3,
\\
\frac{\partial D}{\partial g_5}
& = &
\frac{1}{\sqrt{3}}\,\left(-2g_1^2-2g_2^2+g_3^2+g_4^2+2g_5^2\right).
\end{eqnarray}
\endnumparts
If the variances on the mode channels are equal then it is easily seen that
equations \eref{eq:ev_error} and \eref{eq:derl} lead to
\begin{equation}
\sigma_{\lambda_k}^2 = \frac{4}{3}\,\sigma_g^2,\ k=1,2,3,
\label{eq:iso}
\end{equation}
where $\sigma_g^2$ is the variance on any one mode channel. Note that from
equation \eref{eq:iso} all three eigenvalues have equal variance to first
order. Cross correlations between these eigenvalues are also easily
calculated, and for equal mode channel variances they are equal for all
pairs $(\lambda_k,\lambda_k')$:
\begin{equation}
\sigma_{\lambda_k\lambda_k'} = -\frac{2}{3}\,\sigma_g^2,\ k,k'=1,2,3.
\label{eq:isocross}
\end{equation}
Shown in Figs.~\ref{fig:ev_error_1}-\ref{fig:ev_error_3} is this variance
as a function of the ${\rm SNR} = g^2/\sigma_g^2$. Also shown is the
results of a Monte Carlo type simulation of the errors which take into
account the higher order perturbations at low values of SNR. As shown, the
analytic expressions match the simulated errors for high SNR. For low SNR
discrepancies arise between the analytic and simulated values. This is
within expectation since equation \eref{eq:iso} is only accurate for large
SNR. Higher order correction will be considered below.
\begin{figure}
\psfig{file=fig1.eps,height=15cm,rheight=7.8cm,bbllx=-3cm,bblly=-6.4cm,bburx=14.2cm,bbury=21cm}
\caption{The results of a numerical simulation describing the variance of
the first eigenvalue for a range of SNR. The dot-dashed line was computed
by a 1000 trial Monte Carlo simulation for the range of SNR.
The dashed line is the error found from the first order analytic
expression and the solid line includes the second order corrections.}
\label{fig:ev_error_1}
\end{figure}
\begin{figure}
\psfig{file=fig2.eps,height=15cm,rheight=7.8cm,bbllx=-3cm,bblly=-6.4cm,bburx=14.2cm,bbury=21cm}
\caption{The results of a numerical simulation describing the variance of
the second eigenvalue for a range of SNR. The dot-dashed line was computed
by a 1000 trial Monte Carlo simulation for the range of SNR.
The dashed line is the error found from the first order analytic
expression and the solid line includes the second order corrections.}
\label{fig:ev_error_2}
\end{figure}
\begin{figure}
\psfig{file=fig3.eps,height=15cm,rheight=7.8cm,bbllx=-3cm,bblly=-6.4cm,bburx=14.2cm,bbury=21cm}
\caption{The results of a numerical simulation describing the variance of
the third eigenvalue for a range of SNR. The dot-dashed line was computed
by a 1000 trial Monte Carlo simulation for the range of SNR.
The dashed line is the error found from the first order analytic
expression and the solid line includes the second order corrections.}
\label{fig:ev_error_3}
\end{figure}
\section{Direction estimation error}
\label{sec:dir_error}
We assume that the eigenvector $\bi{v}_3$ points in the propagation
direction of the gravitational wave. We want to estimate the fluctuations
in the determination of this direction caused by the presence of noisy
fluctuations in the mode channels. We represent with $\delta$ a difference
between a given quantity and its ideal value if there were no noise (i.e.
$\delta\bi{v}_3$ is the difference between the position calculated from
noisy data and its real position in the sky). We now take equation
\eref{eqn:noisy_ev} for $k=3$ and consider fluctuations in it. If
these are not too large (high SNR) we can retain only first order terms
\begin{equation}
\left[\bi{A}-\lambda_3\right]\delta\bi{v}_3
=
-\left[\delta\bi{A}-\delta\lambda_3\right]\bi{v}_3.
\label{pert}
\end{equation}
This is an equation for $\delta\bi{v}_3$, but the matrix
$\left[\bi{A}-\lambda_3\right]$ is not invertible. The only consequence
of this is that we cannot determine the component of $\delta\bi{v}_3$
which is parallel to $\bi{v}_3$ itself. The orthogonal components (those
parallel to $\bi{v}_1$ and $\bi{v}_2$) can easily be found by
multiplying equation \ref{pert} on the left by $\bi{v}_1$ and $\bi{v}_2$
\begin{eqnarray}
\bi{v}_1\cdot\delta\bi{v}_3
& = &
-\frac{1}{\lambda_1-\lambda_3}\;\bi{v}_1\delta\bi{A}\bi{v}_3,
\\
\bi{v}_2\cdot\delta\bi{v}_3
& = &
-\frac{1}{\lambda_2-\lambda_3}\;\bi{v}_2\delta\bi{A}\bi{v}_3.
\end{eqnarray}
An appropriate assessment of the error on a direction measurement is the
solid angle error $\Delta\Omega$. Since $|\bi{v}_3|=1$, this error is
given by
\begin{equation}
\Delta\Omega = \pi\left|\Delta\bi{v}_3\right|^2.
\label{eq:omega}
\end{equation}
where $|\Delta\bi{v}_3|^2$ is the quadratic error in the
determination of $\bi{v}_3$. To find it we need to calculate the
expectation of the squared modulus of the above fluctuations,
\begin{equation}
\fl\left|\Delta\bi{v}_3\right|^2 = E\left\{
(\bi{v}_1\cdot\delta\bi{v}_3)^2
+ (\bi{v}_2\cdot\delta\bi{v}_3)^2\right\} =
E\left\{\left|
\frac{\bi{v}_1\delta\bi{A}\bi{v}_3}{\lambda_1-\lambda_3}
\right|^2 + \left|
\frac{\bi{v}_2\delta\bi{A}\bi{v}_3}{\lambda_2-\lambda_3}
\right|^2\right\}.
\label{spqr}
\end{equation}
First order calculations only require us to take expectations in
$\delta\bi{A}$, while leaving the rest untouched,
\begin{equation}
\delta\bi{A} = \sum_{m=1}^5 \bi{A}_m\delta g_m,
\end{equation}
where we have defined
\begin{equation}
\bi{A}_m\equiv\frac{\partial \bi{A}}{\partial g_m},
\end{equation}
Explicitly,
\numparts
\begin{eqnarray}
\fl\qquad\ \ \ \ \bi{A}_1
=
\left[\begin{array}{ccc}
1 & 0 & 0 \\
0 & -1 & 0 \\
0 & 0 & 0
\end{array}\right],
\ \bi{A}_2
=
\left[\begin{array}{ccc}
0 & 1 & 0 \\
1 & 0 & 0 \\
0 & 0 & 0
\end{array}\right],
\ \bi{A}_3
=
\left[\begin{array}{ccc}
0 & 0 & 0 \\
0 & 0 & 1 \\
0 & 1 & 0
\end{array}\right], \\[1 em]
\bi{A}_4
=
\left[\begin{array}{ccc}
0 & 0 & 1 \\
0 & 0 & 0 \\
1 & 0 & 0
\end{array}\right],
\ \bi{A}_5
=
\left[\begin{array}{ccc}
-\frac{1}{\sqrt{3}} & 0 & 0 \\
0 & -\frac{1}{\sqrt{3}} & 0 \\
0 & 0 & \frac{2}{\sqrt{3}}
\end{array}\right].
\end{eqnarray}
\endnumparts
We thus have
\begin{equation}
|\Delta\bi{v}_3|^2 =
\sum_{m=1}^5\left[\left(
\bi{v}_1\bi{A}_m\bi{v}_3
\right)^2
+ \left(
\bi{v}_2\bi{A}_m\bi{v}_3
\right)^{2}
\right]\,\frac{\sigma_{g_m}^2}{g^2}.
\label{eq:deltav}
\end{equation}
Again, setting the variances on the mode channels equal to $\sigma_g^2$,
the sum in equation \eref{eq:deltav} can easily be done, giving
\begin{equation}
|\Delta\bi{v}_3|^2 = 2\,\frac{\sigma_g^2}{g^2}.
\label{eq:dv3}
\end{equation}
From equation \eref{eq:dv3} we see that the error in the incidence direction
is independent of this direction, as expected of an omnidirectional antenna.
Substituting this into \eref{eq:omega} we find
\begin{equation}
\Delta\Omega = \frac{2\pi}{SNR}.
\label{eq:soe}
\end{equation}
\begin{figure}
\psfig{file=fig4.eps,height=15cm,rheight=7.8cm,bbllx=-3cm,bblly=-6cm,bburx=14.2cm,bbury=21.4cm}
\caption{The results of a numerical simulation describing the solid angle
direction estimation error $\Delta\Omega$ of a source direction measurement
due to a finite signal-to-noise ratio. The dot-dashed line was computed by
a 1000 trial Monte Carlo simulation for the range of SNR. The dashed line
is the error found from the first order analytic expression and the solid
line includes the second order corrections.}
\label{fig:solid_angle_error}
\end{figure}
This expression is in perfect agreement with the solid angle estimation
error found by Zhou and Michelson who used a maximum likelihood technique
to estimate the wave direction \cite{Zhou_PRD_1995}. This is not
surprising as the two methods of estimating the wave direction have been
shown to be equivalent (though the assumptions behind each are quite
different) \cite{Merkowitz_PRD_1998}. The advantage of our approach is
that, by using unit vectors (Cartesian components), we are all the time
free from the anomalously high errors and correlations intrinsically
associated to the Euler angle parametrization.
Shown in Fig.~\ref{fig:solid_angle_error} is the solid angle estimation
error as a function of the SNR. Also shown is the results of a Monte Carlo
type simulation of the errors. As shown, the analytic expressions match
the simulated errors for high SNR. Deviations however appear for lower
values of SNR. This again is due to the insufficiency of the first order
analytical estimates of the errors. In the next section we present
improved theoretical estimates of the errors by going one order beyond the
first in the calculations of variances.
\section{Higher order corrections}
\label{sec:corrections}
In order to improve our theoretical understanding of the error behaviours of
the Monte Carlo simulations displayed in
Figs.~\ref{fig:ev_error_1}-\ref{fig:solid_angle_error} we need to go one
step beyond the linear error terms of equations \eref{eq:ev_error}
and \eref{eq:deltav}. This requires calculations of higher order
derivatives for the added terms, and then the resulting general expressions
become quite complicated. They somewhat simplify for equal mode channel
variances, but are still rather cumbersome. For example, the next order
correction to the eigenvalues, assuming the mode channel noises are
zero-mean independent Gaussian processes, is given by (see appendix)
\begin{eqnarray}
\fl\sigma_{\lambda_k}^2 = \frac{4}{3}\,\sigma_g^2 + \nonumber\\
+\fl\left(\sum_{m,m'=1}^5\,\frac{\partial\lambda_k}{\partial g_m}\;
\frac{\partial^3\lambda_k}{\partial g_m\,\partial g_{m'}\,
\partial g_{m'}} + \frac{1}{2}\,
\frac{\partial^2\lambda_k}{\partial g_m\,\partial g_{m'}}\,
\frac{\partial^2\lambda_k}{\partial g_m\,\partial g_{m'}}\right)\;
\sigma_g^4,\ \ \ k=1,2,3,
\label{eq:ev_error_2nd}
\end{eqnarray}
After rather long algebra it is found that
\begin{equation}
\sigma_{\lambda_k}^2 = \frac{4}{3}\,\sigma_g^2 -
2\,\left(1+\sin^2\theta_k\right)\;\frac{\sigma_g^4}{g^2}.
\label{eq:four}
\end{equation}
It turns out that the series for $\sigma_{\lambda_k}^2$ converges {\it very
slowly\/} for low SNR, so that equation \eref{eq:ev_error_2nd} only
constitutes an improvement on (\ref{eq:ev_error}) for a rather limited range
of SNR ---see Figs.~\ref{fig:ev_error_1}-\ref{fig:ev_error_3}. Equation
\eref{eq:four} shows that the errors in $\lambda_1$ and $\lambda_2$ split
from the error in $\lambda_3$ for low values of SNR, and reproduces the
observed behaviour that $\sigma_{\lambda_3}^2$ falls below
$\sigma_{\lambda_1}^2$ and $\sigma_{\lambda_2}^2$. It is a reasonable
approximation to $\sigma_{\lambda_3}^2$ for SNR between 30 and 6, but it is
not quite as good as regards $\sigma_{\lambda_1}^2$ and $\sigma_{\lambda_2}^2$
for those values of SNR. Higher order terms would be required for an
improvement, but these imply still much longer calculations of derivatives of
the eigenvalues up to the fifth order, as can be seen in equation
(\ref{eq:sigma_f}) of the appendix.
Similar corrections can be applied to the incidence direction error estimate
of equation \eref{eq:dv3}. They appear to be given by
\begin{equation}
|\Delta\bi{v}_3|^2 = 2\,\frac{\sigma_g^2}{g^2}\,
\left(1+\frac{\sigma_g^2}{g^2}\right)^{-1}
\label{eq:dve_2nd}
\end{equation}
for equal mode channel variances, $\sigma_g^2$. Solid angle errors can be
directly inferred from here:
\begin{equation}
\Delta\Omega = \frac{2\pi}{SNR}\;\left(1+SNR^{-1}\right)^{-1}.
\label{eq:soe_2nd}
\end{equation}
As shown in Fig.~\ref{fig:solid_angle_error}, the above theoretical
prediction is a better approach to the behaviour observed in the numerical
simulations than is equation \eref{eq:soe}. For SNR less than 2,
equation \eref{eq:soe_2nd} is also insufficient.
It is important to remind ourselves that for very low SNR the uncertainties
on the direction estimation are so large that the measurement is almost
meaningless. Zhou and Michelson decided that a minimum SNR of 10 in energy
was necessary for a direction measurement \cite{Zhou_PRD_1995}. Looking at
Fig.~\ref{fig:solid_angle_error} this lower limit seems reasonable, thus it
is only necessary to have accurate analytical expressions down to that level.
\section{Polarization amplitudes}
\label{sec:polarization}
We now come to the discussion of errors in the polarization amplitudes $h_+$
and $h_\times$. It is easily shown that the uncertainty on the measurement
of the polarization amplitudes is direction independent if the source
position is known ahead of time \cite{Zhou_PRD_1995}, but it has been
claimed that in the unknown direction case there is a strong direction
dependency \cite{Stevenson_PRD_1997}. This is disturbing given that a
spherical detector is equally sensitive to waves of all polarization and
direction. We argue in this section that the difficulties to find isotropic
estimates of $h_+$ and $h_\times$ are ultimately due to the use of unsuitable
criteria to set up those estimates. We discuss a more natural procedure,
based on data set processing, that leads to a solution of this problem.
To understand why a simple estimate of the errors leads to direction
dependencies, let us look at the basics of the solution to the inverse
problem. We are given 3 eigenvectors and 3 eigenvalues, but these are
not independent. The eigenvectors are orthogonal, so we actually only
get 2 pieces of information from them. We use that information to
determine the direction of the wave. Next, the strain tensor is traceless
so the third eigenvalue can be determined from the other two, therefore,
we only get 2 pieces of information from them. We use one of them
(actually a combination of 2) to get the wave amplitude. Past reasoning
suggested that we can use the last piece of information to determine the
polarization angle. This is wrong. The last eigenvalue does not tell us
the polarization, but rather is related to the scalar component of the GW
(or lack thereof). We assume this to be zero for GR, so this can be
interpreted as a measurement of the non-zeroness of this component. The
fact that we decide that this should be zero {\it a priori} does not give
us additional information about the polarization, only the level of noise in
our system. Without any additional information we have an under-determined
system which will lead to direction dependent errors as seen in
reference~\cite{Stevenson_PRD_1997}.
The additional piece of information needed is the polarization angle
$\alpha$: the angle between the GW's axes and the eigenvectors $\bi{v}_1$
and $\bi{v}_2$ perpendicular to the incidence direction $\bi{v}_3$. To
this end, we submit our diagonal form of the matrix $\bi{A}$ to a rotation
of angle $\alpha$ about $\bi{v}_3$ to obtain best estimates of $h_+$ and
$h_\times$ by the formulas
\numparts
\begin{eqnarray}
h_+ & = & h\cos 2\alpha, \label{eq:+} \\
h_\times & = & h\sin 2\alpha, \label{eq:x}
\end{eqnarray}
\endnumparts
where the pure and uncorrelated noise term in $h_+$ was dropped out. The
GW amplitude $h\/$\,$\equiv$\,$(h_+^2+h_\times^2)^{1/2}$ has been shown
to have a best estimate given by equation \eref{eq:estim_h}. $h\/$ can be
determined with isotropic sensitivity since that is the case with
$\lambda_1$ and $\lambda_2$, as we have just seen. For a
{\it fixed\/}~$\alpha$, equations \eref{eq:+} and \eref{eq:x} give us an
estimate of $h_+$ and $h_\times$ in the presence of noise in the detector.
We can use the results of section \ref{sec:uncertainties_ev} to see that
\numparts
\begin{eqnarray}
\sigma_+^2 & = & \cos^2 2\alpha\,
\sigma_g^2 \label{eq:sigma+} \\
\sigma_\times^2 & = & \sin^2 2\alpha\,
\sigma_g^2 \label{eq:sigmax} \\
\sigma_{+\times} & = & -\frac{1}{2}\,\sin 4\alpha\,
\sigma_g^2 \label{eq:crossx+}
\end{eqnarray}
\endnumparts
and these errors are indeed isotropic, for they only depend on $\sigma_g$.
In the absence of further information on the specific {\it physical\/}
nature of the source, {\it any\/} polarization angle $\alpha\/$ is valid,
for the canonical form of the tensor~(\ref{eqn:wave_strain}) is invariant
to rotations about the third axis. A particular choice of $\alpha\/$ is
thus a matter of taste in this case, and equations
\eref{eq:sigma+}--\eref{eq:crossx+} give the correct error estimates.
A common way \cite{Wagoner_Pavia_1976} to resolve the arbitrariness
in~$\alpha\/$ is to set the first Euler angle in the rotation relating
the lab frame to the wave frame equal to zero.
However, this is a very much observer dependent criterion, for detectors at
different locations would claim different values for $h_+$ and $h_\times$,
even if they agreed to be seeing the {\it same\/} source. Errors
in $h_+$ and $h_\times$ based on such criterion have been shown e.g.\ by
\cite{Stevenson_PRD_1997} to be strongly direction dependent, which is
certainly not surprising. It is however paradoxical that a spherical detector
should prefer certain directions to others to detect a GW signal, therefore
this must be reassessed. We now propose a more consistent solution.
It is clear from the above discussion that any criterion to resolve the
arbitrariness in $\alpha\/$, therefore to estimate $h_+$ and $h_\times$,
should be established {\it relatively to the GW source\/}, be it known
ahead of time or based on a hypothesis to be checked {\it a posteriori\/}.
Let us, for concreteness, consider a {\it coalescing binary system\/} as
the GW source \cite{EugViv_PLA_1996}. The signal generated by such a system
is given by somewhat complicated functions of the space-time variables and
a number of system parameters; it will not be necessary for our purposes to
consider in detail the explicit form of such functions (see for example
\cite{Clemente_PhD_1994}), it will suffice to use formal expressions
indicating the signal dependencies:
\numparts
\begin{eqnarray}
h_+ & = & h_+(\bi{r},t;{\cal K}), \\ \label{eq:cb+}
h_\times & = & h_\times(\bi{r},t;{\cal K}). \label{eq:cbx}
\end{eqnarray}
\endnumparts
Here $\bi{r}\/$ is the source position, and $t\/$ is the time.
${\cal K\/}$ stands for the {\it set of characteristic source parameters\/},
which in this case include the masses of the stars, the inclination of the
orbital plane, the semimajor axis, the eccentricity of the orbit, the
periastron position, etc. Note that these amplitudes are referred to a set
of {\it source\/} axes, so they are independent of the detector's location.
The usual way to estimate the parameters ${\cal K\/}$ is to resort to
classical Statistics \cite{Helstrom_1968}, as has been done for example
in \cite{Krolak_PRD_1993} for interferometric detectors or in
\cite{Stevenson_PRD_1997} for spherical detectors. The fundamental
quantity required by such method is the {\it likelihood function\/},
$\Lambda$, which is a functional of the (unknown) signal parameters
{\it and\/} the detector data.
We then proceed as follows: we construct the likelihood function
associated to the hypothesis that equations \eref{eq:+} and \eref{eq:x}
be a fit to equations \eref{eq:cb+} and \eref{eq:cbx} for suitable values
of the parameters $\cal K\/$. It will thus have the generic form
\begin{equation}
\Lambda = \Lambda(h;\alpha;{\cal K}) \label{eq:lambda}
\end{equation}
Standard manipulations of $\Lambda$ yield both best estimates of the signal
parameters ${\cal K\/}$ {\it and\/} of the polarization angle~$\alpha\/$,
as well as errors and cross correlations between any pair of these ---it is
recalled that such are identified as the coefficients of the {\it covariance
matrix\/}, which is the inverse of the matrix of second derivatives of
$\Lambda$ \cite{Helstrom_1968}.
We shall not attempt to give a detailed discussion of this process here.
The important point to stress is that in the approach
just proposed, we have managed to have $h\/$ as the only combination of
actual data entering the likelihood function $\Lambda$. Errors and cross
correlations between parameter estimates will thus ultimately be functions
only of the errors and cross correlations between the eigenvalue estimates,
$\lambda_1$ and $\lambda_2$, which we have proved in
section~\ref{sec:uncertainties_ev} to be direction independent.
So not only $\alpha\/$ but also the source parameters ${\cal K\/}$ can be
determined with isotropic sensitivity by means of a spherical GW detector.
The same therefore applies to the GW amplitudes $h_+$ and $h_\times$, as
indeed expected.
The {\it quantitative\/} estimation of the errors in $h_+$ and $h_\times$
cannot however be given explicitly until the full parameter estimation
problem has been completely solved, as interactions between all those
estimates will strongly affect one another.
\section{Discussion}
With analytic expressions for the uncertainties on the eigenvectors and the
eigenvalues of the mode channel matrix~(\ref{eqn:cartesian_strain_tensor})
we can turn our attention to the physical interpretation of these values.
An unambiguous selection of $\lambda_3$ can be made on the basis that it
is the {\it third root\/} in equation \eref{eq:roots}. This will usually
be the one closest to zero. Then the other two represent the amplitude
measurement. As proved in \cite{Merkowitz_PRD_1998}, the best estimate
of the GW amplitude is the semi-difference of these two,
$(\lambda_2-\lambda_1)/2$.
The third eigenvalue ideally should be zero if general relativity is
correct. Once noise is introduced this is no longer the case, but the
variance on this eigenvalue gives us a level of the ``non-zeroness''. One
can imagine setting a threshold on this eigenvalue that is a function
of $\sigma_{\lambda_3}$ (a function of the SNR) to veto any candidate
events that have an excessive $\lambda_3$. Many non-GW sources are likely
to produce a non-zero $\lambda_3$, therefore becoming easily identified
and discarded.
The errors in both eigenvalues and eigenvectors are direction independent.
The last step in the analysis is the splitting of the GW amplitude $h\/$
into the usual $h_+$ and $h_\times$ components. We have shown that this
can be accomplished by making suitable reference to the source properties,
whereby isotropic sensitivity to these quantities obtains. This solves the
paradox of the anisotropies in the determination of $h_+$ and $h_\times$,
and stresses the fact that the most fundamental magnitudes to estimate
from the detector data are the eigenvalues and eigenvectors of the mode
channel matrix: these are {\it source independent\/}, and any further
progress in signal deconvolution explicitly requires reference to the
source properties, be them known ahead of time or be them stated in the
form of a hypothesis to test.
\ack{
We are grateful for the hospitality of the ROG group at the Laboratori
Nazionali di Frascati (Rome), where this work began and partly developed.
One of us (SMM) thanks E. Mauceli for helpful discussions. Two of us (JAL
and MAS) acknowledge financial support from the Spanish Ministry of
Education, contract number PB96-0384, and the Institut d'Estudis Catalans.}
|
\section*{1.Introduction}
~~ In the renormalization of quantum field theories, to extract finite
physical results from higher order perturbative calculations, a certain
subtraction scheme is necessary to be used so as to remove the divergences
occurring in the calculations. There are various subtraction schemes in the
literature, such as the minimal subtraction (MS)$^1$, the modified minimal
subtraction ($\overline{MS}$) $^2$, the momentum space subtraction (MOM)$^3$%
, the on-mass-shell subtraction (OS) $^{4,5}$, the off-mass shell
subtraction (OMS)$^6$ and etc.. However, there exists a serious ambiguity
problem$^{3,6.7}$ that different subtraction schemes in general give
different physical predictions, conflicting the fact that the physical
observables are independent of the subtraction schemes. Ordinarily, it is
argued that the ambiguity appears only in finite order perturbative
calculations, while the exact result given by the whole perturbation series
is scheme- independent. Though only a finite order perturbative calculation
is able to be done in practice, one still expects to get unambiguous results
from such a calculation. To solve the ambiguity problem, several
prescriptions, such as the minimal sensitivity principle$^7$, the effective
charge method$^8$ and some others$^9$ , were proposed in the past. By the
principle of minimal sensitivity, an additional condition has to be
introduced and imposed on the result calculated in a finite order
perturbative approximation so as to obtain an optimum approximant which is
least sensitive to variations in the unphysical parameters. In the effective
charge method, the use of a coupling constant is abandoned. Instead, an
effective charge is associated with each physical quantity and used to
determine the Gell-Mann-Low function in the renormalization group equation
(RGE). In this way, a renormalization-scheme-invariant result can be found
from the RGE. In Ref.(9), the authors developed a new perturbation approach
to renormalizable field theories. This approach is based on the observation
that if the perturbation series of a quantity R is not directly computable
due to that the expansion coefficients are infinite, the unambiguous result
of the quantity can be found by solving the differential equation $Q\frac{%
dR(Q)}{dQ}=F(R(Q))$ where F(R) is well-defined and can be expanded as a
series of R. Particularly, the finite coefficients of the expansion of F(R)
are renormalization-scheme-independent and in one-to-one correspondence with
the ones in the ordinary perturbation series of the R. The prescriptions
mentioned above are somehow different from the conventional perturbation
theory in which the coupling constant is commonly chosen to be the expansion
parameter for a perturbation series.
In this paper, we wish to deal with the ambiguity problem from a different
angle. It will be shown that the ambiguity problem can be directly tackled
in the conventional perturbation theory by the renormalization group method$%
^{10-15}$. The advantage of this method is that the anomalous dimension in a
RGE is well-defined although it is computed from the renormalization
constant which is divergent in its original definition. In comparison of the
renormalization group method with the aforementioned approach proposed in
Ref.(9), we see, both of them are much similar to one another in
methodology. This suggests that the renormalization group method is also
possible to yield the theoretical results which are free from the ambiguity.
The possibility relies on how to choose a good subtraction scheme which
gives rise to such renormalization constants that they lead to unique
anomalous dimensions. The so-called good subtraction scheme means that it
must respect necessary physical and mathematical requirements such as the
gauge symmetry, the Lorentz invariance and the mathematical convergence
principle. The necessity of these requirements is clear. Particularly, in
the renormalization group approach to the renormalization problem, the
renormalization constants obtained from a subtraction scheme not only serve
to subtract the divergences, but also are directly used to derive physical
results. Apparently, to guarantee the calculated results to be able to give
faithful theoretical predictions, the subtraction procedure necessarily
complies with the basic principles established well in physics and
mathematics. Otherwise, the subtraction scheme should be discarded and thus
the scheme-ambiguity will be reduced. Let us explain this viewpoint in some
detail from the following aspects: (1) For a gauge field theory, as one
knows, the gauge-invariance is embodied in the Ward identity. This identity
is a fundamental constraint for the theory. Therefore, a subtraction scheme,
if it is applicable, could not defy this identity. As will be demonstrated
in latter sections, the Ward identity not only establishes exact relations
between renormalization constants, but also determines the functional
structure of renormalization constants; (2) The Lorentz-invariance is partly
reflected in the energy-momentum conservation which holds at every vertex.
Since the renormalization point in the renormalization constants will be
eventually transformed to the momentum in the solutions of RGEs, obviously,
in order to get correct functional relations of the solutions with the
momentum, the energy-momentum conservation could not be violated by the
subtraction of vertices; (3) Why the convergence principle is required in
the renormalization calculation? As we know, the renormalization constants
are divergent in their original appearance. Such divergent quantities are
not well-defined mathematically and hence are not directly calculable$^9$%
because we are not allowed to apply any computational rule to do a
meaningful or unambiguous calculation for this kind of quantities. The
meaningful calculation can only be done for the regularized form of
renormalization constants which are derived from corresponding regularized
Feynman integrals in a subtraction scheme. The necessity of introducing the
regularization procedure in the quantum field theory may clearly be seen
from the mathematical viewpoint as illustrated in Appendix A. Therefore, in
renormalization group calculations, the correct procedure of computing
anomalous dimensions is starting from the regularized form of
renormalization constants. The limit operation taken for the regularization
parameter should be performed after completing the differentiation with
respect to the renormalization point. Since the anomalous dimensions are
convergent, the limit is meaningful and would give definite results.
Obviously, the above procedure of computing the anomalous dimensions agrees
with the convergence principle; (4) In comparison with the other subtraction
schemes, the MOM scheme appears to be more suitable for renormalization
group calculations. This is because this scheme naturally provides not only
a renormalization point which is needed for the renormalization group
calculation, but also a renormalization boundary condition for a
renormalized quantity (a wave function, a vertex or a propagator), which
will be used to fix the solution of the RGE for the renormalized quantify.
Based on the essential points of view stated above, it may be found that a
renormalized quantity can be unambiguously determined by its RGE and thus a
renormalized S-matrix element can be given in an unique form without any
ambiguity. To illustrate this point, we limit ourselves in this paper to
take the QED renormalization as an example to show how the ambiguity can be
eliminated. As one knows, the QED renormalization has been extensively
investigated by employing the OS, $\overline{MS}$ and MOM schemes in the
previous works$^{4-6,10-26}$. But, most of these studies are concentrated on
the large momentum (short distance) behaviors of some quantities for the
sake of simplicity of the calculation. In this paper, we restudy the QED
renormalization with the following features: (1) The renormalization is
performed in a mass-dependent scheme other than in the mass-independent
scheme which was adopted in many previous works$^{20-26}$. The
mass-dependent scheme obviously is more suitable for the case that the mass
of a charged fermion can not be set to be zero; (2) The subtraction is
carried out in such a MOM scheme that the renormalization point is mainly
taken to be an arbitrary time-like momentum other than a space-like momentum
as chosen in the conventional MOM scheme. The time-like MOM scheme actually
is a generalized mass-shell scheme (GMS). The prominent advantage of the GMS
scheme is that in this scheme the scale of renormalization point can
naturally be connected with the scale of momenta and the results obtained
can directly be converted to the corresponding ones given in the OS-scheme ;
(3) The subtraction is implemented by fully respecting the necessary
physical and mathematical principles mentioned before . Therefore, the
results obtained are faithful and free of ambiguity; (4) The effective
coupling constant and the effective fermion mass obtained in the one-loop
approximation are given exact and explicit expressions which were never
found in the literature. These expressions exhibit physically reasonable
infrared and ultraviolet behaviors. We will pay main attention to the
infrared (large distance) behavior because this behavior is more sensitive
to identify whether a subtraction is suitable or not for the QED
renormalization.
~~The rest of this paper is arranged as follows. In Sect.2, we sketch the
RGE and its solution and show how a S-matrix element can be free of
ambiguity. In Sect.3, we briefly discuss the Ward identity and give a
derivation of the subtraction version of the fermion self-energy in the GMS
scheme. In Sect.4, we derive an exact expression of the one-loop effective
coupling constant and discuss its asymptotic property. In Sect.5, the same
thing will be done for the effective fermion mass. The last section serves
to make some comments and discussions. In Appendix A, we show a couple of
mathematical examples to help understanding the regularization procedure
used in the renormalization group calculations.
\setcounter{section}{2}
\section*{2.Solution to RGE and S-matrix element}
\setcounter{equation}{0}
~~ Among different formulations of the RGE (see the review given in Ref.(14))%
$^{10-15}$, we like to employ the approach presented in Ref.(15). But, we
work in a mass-dependent renormalization scheme, therefore, the anomalous
dimension in the RGE depends not only on the coupling constant, but also on
the fermion mass. Suppose $F_R$ is a renormalized quantity. In the
multiplicative renormalization, it is related to the unrenormalized one $F$
in such a way
\begin{eqnarray}
F=Z_FF_R
\end{eqnarray}
where $Z_F$ is the renormalization constant of $F$. In GMS scheme, the $Z_F$
and $F_R$ are all functions of the renormalization point $\mu =\mu _0e^t$.
Differentiating Eq.(2.1) with respect to the $\mu $ and noticing that the $F$
is independent of $\mu $, we immediately obtain a RGE satisfied by the
function $F_R^{}$
\begin{eqnarray}
\mu \frac{dF_R}{d\mu }+\gamma _FF_R=0
\end{eqnarray}
where $\gamma _F$ is the anomalous dimension defined by
\begin{eqnarray}
\gamma _F=\mu \frac d{d\mu }\ln Z_F
\end{eqnarray}
We first note here that the anomalous dimension can only depends on the
ratio ${\sigma =\frac{m_R}\mu }$, ${\gamma }_F{=\gamma }_F{(g}_R{,\sigma ),}$
because the renormalization constant is dimensionless. Next, we note,
Eq.(2.2) is suitable for a physical parameter (mass or coupling constant), a
propagator, a vertex, a wave function or some other Green function. If the
function ${F_R}$ stands for a renormalized Green function, vertex or wave
function, in general, it not only depends explicitly on the scale $\mu $,
but also on the renormalized coupling constant $g_R$, mass $m_R$ and gauge
parameter $\xi _R$ which are all functions of $\mu $, $F_R=F_R(p,g_R(\mu
),m_R(\mu ),\xi _R(\mu );\mu )$ where $p$ symbolizes all the momenta.
Considering that the function ${F_R}$ is homogeneous in the momentum and
mass, it may be written, under the scaling transformation of momentum $%
p=\lambda p_0$ , as follows
\begin{eqnarray}
F_R(p;g_R,m_R,\xi _R,\mu )=\lambda ^{D_F}F_R(p_0;g_R,\frac{m_R}\lambda ,\xi
_R;\frac \mu \lambda )
\end{eqnarray}
where $D_F$ is the canonical dimension of $F$. Since the renormalization
point is a momentum taken to subtract the divergence, we may set $\mu =\mu
_0\lambda $ where $\lambda =e^t$ which is taken to be the same as in $%
p=p_0\lambda $. Noticing the above transformation, the solution of the RGE
in Eq.(2.2) can be expressed as$^{15}$
\begin{eqnarray}
F_R(p;g_R,m_R,\xi _R,\mu _0) &=&\lambda ^{D_F}e^{\int_1^\lambda \frac{%
d\lambda }\lambda \gamma _F(\lambda )}F_R(p_0; \nonumber \\
&&g_R(\lambda ),m_R(\lambda )\lambda ^{-1},\xi _R(\lambda );\mu _0)
\end{eqnarray}
where $g_R(\lambda ),m_R(\lambda )$ and $\xi _R(\lambda )$ are the running
coupling constant, the running mass and the running gauge parameter,
respectively. The solution written above shows the behavior of the function $%
F_R$ under the scaling of momenta.
How to determine the function $F_R(p_0;\cdots ,\mu _0)$ on the RHS of
Eq.(2.5) when the $F_R(p_0,...)$ stands for a wave function, a propagator or
a vertex? This question can be unambiguously answered in MOM scheme, but was
not answered clearly in the literature $^{27,28}$. Noticing that the
momentum $p_0$ and the renormalization point $\mu _0$ are fixed, but may be
chosen arbitrarily, we may, certainly, set $p_0^2=\mu _0^2$. With this
choice, by making use of the following boundary condition satisfied by a
propagator, a vertex or a wave function
\begin{equation}
F_R(p_0;g_R,m_R,\xi _R,\mu )\mid _{P_0^2=\mu ^2}=F_R^{(0)}(p_0;g_R,m_R,\xi
_R)
\end{equation}
where the function $F_R^{(0)}(p;g_R,m_R,\alpha _R)$ is of the form of free
propagator, bare vertex or free wave function and independent of the
renormalization point (see the examples given in the next section) and
considering the homogeneity of the function $F_R$ as mentioned in Eq.(2.4),
we may write
\begin{eqnarray}
\lambda ^{D_F}F_R &(p_0;g_R(\lambda ),m_R(\lambda )\lambda ^{-1},\xi
_R(\lambda ),\mu _0)\mid &_{p_0^2=\mu _0^2} \nonumber \\
=F_R^{(0)} &&(p;g_R(\lambda ),m_R(\lambda ),\xi _R(\lambda ))
\end{eqnarray}
where the renormalized coupling constant, mass and vertex in the function $%
F_R^{(0)}(p,...)$ become the running ones. With the expression given in
Eq.(2.7), Eq.(2.5) will finally be written in the form
\begin{eqnarray}
F_R(p;g_R,m_R,\xi _R)=e^{\int_1^\lambda \frac{d\lambda }\lambda \gamma
_F(\lambda )}F_R^{(0)}(p;g_R(\lambda ),m_R(\lambda ),\xi _R(\lambda ))
\end{eqnarray}
For a gauge field theory, it is easy to check that the anomalous dimension
in Eq.( 2.8) will be cancelled out in S-matrix elements. To show this point
more specifically, let us take the two-electron scattering taking place in
t-channel as an example. Considering that a S-matrix element expressed in
terms of unrenormalized quantities is equal to that represented by the
corresponding renormalized quantities, the scattering amplitude may be
written as
\begin{equation}
S_{fi}=\overline{u}_R^{\alpha ^{^{\prime }}}(p_1^{^{\prime }})\Gamma _R^\mu
(p_1^{^{\prime }},p_1)u_R^\alpha (p_1)iD_{R\mu \nu }(k)\overline{u}_R^{\beta
^{^{\prime }}}(p_2^{^{\prime }})\Gamma _R^\nu (p_2^{^{\prime
}},p_2)u_R^\beta (p_2)
\end{equation}
where $k=p_1^{\prime }-p_1=p_2^{}-p_2^{\prime }$; $u_R^\alpha (p)$, $\Gamma
_R^\mu (p^{\prime },p)$ and $iD_{R\mu \nu }(k)$ represent the fermion wave
function, the proper vertex and the photon propagator respectively which are
all renormalized. The renormalization constants of the wave function, the
propagator and the vertex will be designated by $\sqrt{Z_2},Z_3$ and $%
Z_\Gamma $ respectively. The constant $Z_\Gamma $ is defined as
\begin{equation}
Z_\Gamma =Z_2^{-1}Z_3^{-\frac 12}
\end{equation}
because the vertex in Eq.(2.9) contains a coupling constant in it.
According to the formula given in Eq.(2.8), the renormalized fermion wave
function, photon propagator and vertex can be represented in the forms as
shown below. For the fermion wave function, we have
\begin{equation}
u_R^\alpha (p)=e^{\int_1^\lambda \frac{d\lambda }\lambda \gamma _F(\lambda
)}u_{R\alpha }^{(0)}(p,m_R(\lambda ))
\end{equation}
where
\begin{equation}
u_{R\alpha }^{(0)}(p,m_R(\lambda ))=\left( \frac{E+m_R(\lambda )}{%
2m_R(\lambda )}\right) ^{\frac 12}\left( \frac{\overrightarrow{\sigma }.%
\overrightarrow{p}}{E+m_R(\lambda )}\right) \varphi _\alpha (\overrightarrow{%
p})
\end{equation}
is the free wave function in which $m_R(\lambda )$ is the running mass and
\begin{equation}
\gamma _F=\frac 12\mu \frac d{d\mu }\ln Z_2
\end{equation}
is the anomalous dimension of fermion wave function. For the renormalized
photon propagator, we can write
\[
iD_{R\mu \nu }(k)=e^{\int_1^\lambda \frac{d\lambda }\lambda \gamma
_3(\lambda )}iD_{R\mu \nu }^{(0)}(k)
\]
where
\begin{equation}
iD_{R\mu \nu }^{(0)}(k)=-\frac i{k^2+i\varepsilon }[g_{\mu \nu }-(1-\alpha
_R(\lambda ))\frac{k_\mu k_\nu }{k^2}]_{}
\end{equation}
is the free propagator with $\alpha _R(\lambda )$ being the running gauge
parameter in it and
\begin{equation}
\gamma _3(\lambda )=\mu \frac d{d\mu }\ln Z_3
\end{equation}
is the anomalous dimension of the propagator. For the renormalized vertex,
it reads
\begin{equation}
\Gamma _R^\mu (p^{\prime },p)=e^{\int_1^\lambda \frac{d\lambda }\lambda
\gamma _\Gamma (\lambda )}\Gamma _R^{(0)\mu }(p^{\prime },p)
\end{equation}
where
\begin{equation}
\Gamma _R^{(0)\mu }(p^{\prime },p)=ie_R(\lambda )\gamma ^\mu
\end{equation}
is the bare vertex containing the running coupling constant (the electric
charge) $e_R(\lambda )$ in it and
\begin{equation}
\gamma _\Gamma (\lambda )=\mu \frac d{d\mu }\ln Z_\Gamma =-\mu \frac d{d\mu
}\ln Z_2-\frac 12\mu \frac d{d\mu }\ln Z_3
\end{equation}
is the anomalous dimension of the vertex here the relation in Eq.(2.10) has
been used. Upon substituting Eqs.(2.11), (2.14) and (2.17) into Eq.(2.9) and
noticing Eqs.( 2.13), (2.16) and (2.19), we find that the anomalous
dimensions in the S-matrix element are all cancelled out with each other. As
a result, we arrive at
\begin{equation}
S_{fi}=\overline{u}_{R\alpha ^{^{\prime }}}^{(0)}(p_1^{\prime })\Gamma
_R^{(0)\mu }(p_1^{\prime },p_1)u_{R\alpha }^{(0)}(p_1)iD_{R\mu \nu }^{(0)}(k)%
\overline{u}_{R\beta ^{^{\prime }}}^{(0)}(p_2^{\prime })\Gamma _R^{(0)\nu
}(p_2^{\prime },p_2)u_{R\beta }^{(0)}(p_2)
\end{equation}
This expression clearly shows that the exact S-matrix element of the
two-electron scattering can be represented in the form as given in the
lowest order (tree diagram) approximation except that all the physical
parameters in the matrix elements are replaced by their effective (running)
ones. For other S-matrix elements, the conclusion is the same. This is
because any S-matrix element is unexceptionably expressed in terms of a
number of wave functions, propagators and proper vertices each of which can
be represented in the form as shown in Eq.(2.8) and the anomalous dimensions
in the matrix element , as can be easily proved, are all cancelled out
eventually. This result and the fact that any S-matrix element is
independent of the gauge parameter ( This is the so-called gauge-invariance
of S-matrix which is implied by the unitarity of S-matrix elements) indicate
that the task of renormalization for a gauge field theory is reduced to find
the running coupling constant and the running mass by their RGEs. These
running quantities completely describe the effect of higher order
perturbative corrections.
\setcounter{section}{3}
\section*{3. Ward Identity}
\setcounter{equation}{0}
In QED renormalization, the following Ward identity plays crucial role $^6$%
\begin{eqnarray}
\Lambda _\mu (p^{\prime };p)_{\mid p^{\prime }=p}=-\frac{\partial \Sigma (p)%
}{\partial p_\mu }
\end{eqnarray}
where $\Lambda _\mu (p^{\prime },p)$ represents the vertex correction which
is defined by taking out a coupling constant e and $\Sigma (p)$ denotes the
fermion self-energy. Firstly, we show how the above identity determines the
subtraction of the fermion self-energy $\sum (p)$. According to the Ward
identity, the divergence in $\Lambda _\mu (p^{\prime },p)$ should be, in the
GMS scheme, subtracted at a time-like (Minkowski) renormalization point for
the momenta of the external fermion lines, $p^2=p^{\prime 2}=\mu ^2$ which
implies $\not p=\not p^{\prime }=\mu $. When $\mu =m$ (the fermion mass), we
will come to the subtraction in the OS scheme. In the case of $\mu \neq m$,
the subtraction is defined on a generalized mass shell. At the
renormalization point $\mu $, we have
\begin{eqnarray}
\Lambda _\mu (p^{\prime },p)\mid _{{\not p}^{\prime }={\not p}=\mu }=L\gamma
_\mu
\end{eqnarray}
Thus, the vertex correction may be represented as
\begin{eqnarray}
\Lambda _\mu (p^{\prime },p)=L\gamma _\mu +\Lambda _\mu ^c(p^{\prime },p)
\end{eqnarray}
where $L$ is a divergent constant depending on $\mu $ and $\Lambda _\mu
^c(p^{\prime },p)$ is the finite correction satisfying the boundary
condition
\begin{eqnarray}
\Lambda _\mu ^c(p^{\prime },p)\mid _{{\not p}^{\prime }={\not p}=\mu }=0
\end{eqnarray}
On inserting Eq.(3.3) into Eq.(3.1) and integrating the both sides of
Eq.(3.1) over the momentum $p^\mu $, we get
\begin{eqnarray}
\Sigma (p)-\Sigma (\mu )=-({\not p}-\mu )L-\int_{p_0^\mu }^{p^\mu }dp^\mu
\Lambda _\mu ^c(p,p)
\end{eqnarray}
where the momentum $p_0$ is chosen to make ${\not p}_0=\mu $. Since the last
term on the RHS of Eq.(3.5) vanishes when $p^\mu \rightarrow p_0^\mu $, we
may write
\begin{eqnarray}
\int_{p_0^\mu }^{p\mu }dp^\mu \Lambda _\mu ^c(p,p)=({\not p}-\mu )C(p^2)
\end{eqnarray}
where $C(p^2)$ is a convergent function satisfying the following boundary
condition
\begin{eqnarray}
C(p^2)\mid _{p^2=\mu ^2}=0
\end{eqnarray}
which is implied by Eq.(3.4). Substituting Eq.(3.6) into Eq.(3.5) and
setting
\begin{eqnarray}
\Sigma (\mu )=A
\end{eqnarray}
and
\begin{eqnarray}
L=-B
\end{eqnarray}
the self-energy is finally written as
\begin{eqnarray}
\Sigma (p)=A+({\not p}-\mu )[B-C(p^2)]
\end{eqnarray}
where the constants A and B have absorbed all the divergences appearing in
the $\Sigma (p)$. The above derivation shows that the subtraction given in
Eq.(3.10) is uniquely correct in the GMS scheme as it is compatible with the
Ward identity. According to the subtraction in Eq.(3.3), the full vertex can
be written as
\begin{eqnarray}
\Gamma _\mu (p^{\prime },p) &=&\gamma _\mu +\Lambda _\mu (p^{\prime },p)
\nonumber \\
&=&Z_1^{-1}\Gamma _\mu ^R(p^{\prime },p)
\end{eqnarray}
where $Z_1$ is the vertex renormalization constant defined as
\begin{eqnarray}
Z_1^{-1}=1+L
\end{eqnarray}
and $\Gamma _\mu ^R(p^{\prime },p)$ is the renormalized vertex represented
by
\begin{eqnarray}
\Gamma _\mu ^R(p^{\prime },p)=\gamma _\mu +\Lambda _\mu ^R(p^{\prime },p)
\end{eqnarray}
which satisfies the boundary condition
\begin{eqnarray}
\Gamma \mu ^R(p^{\prime },p)\mid _{{\not p}^{\prime }={\not p}=\mu }=\gamma
_\mu
\end{eqnarray}
Based on the subtraction given in Eq.(3.10), the full fermion propagator may
be renormalized in such a way
\begin{eqnarray}
iS_F(p)=\frac i{{\not p}-m-\Sigma (p)+i\varepsilon }=Z_2iS_F^R(p)
\end{eqnarray}
where $Z_2$ is the propagator renormalization constant defined by
\begin{eqnarray}
Z_2^{-1}=1-B
\end{eqnarray}
and $S_F^R(p)$ denotes the renormalized propagator represented as
\begin{eqnarray}
S_F^R(p)=\frac i{{\not p}-m_R-\Sigma _R(p)+i\varepsilon }
\end{eqnarray}
which has a boundary condition as follows
\begin{equation}
S_F^R(p)\mid _{p^2=\mu ^2}=\frac i{{\not p}-m_R}
\end{equation}
In Eq.(3.17), $m_R$ and $\Sigma _R(p)$ designate the renormalized mass and
the finite correction of the self-energy respectively. The renormalized mass
is defined by
\begin{eqnarray}
m_R=Z_m^{-1}m
\end{eqnarray}
where $Z_m$ is the mass renormalization constant expressed by
\begin{eqnarray}
Z_m^{-1}=1+Z_2[Am^{-1}+(1-\mu m^{-1})B]
\end{eqnarray}
Particularly, from Eqs.(3.9), (3.12) and (3.16). it is clear to see
\begin{eqnarray}
Z_1=Z_2
\end{eqnarray}
This just is the Ward identity obeyed by the renormalization constants.
~~Let us verify whether the Ward identity is fulfilled in the one-loop
approximation. The Feynman integrals of one-loop diagrams in QED have been
calculated in the literature by various regularization procedures$%
^{3-6,14,15},$. In the GMS scheme, the fermion self-energy depicted in
Fig.(1a), according to the dimensional regularization procedure, is
regularized in the form
\begin{eqnarray}
\Sigma (p) &=&-\frac{e^2}{(4\pi )^2}(4\pi M^2)^\varepsilon \Gamma
(1+\varepsilon )\int_0^1dx\{\frac 1{\varepsilon \Delta (p)^\varepsilon
}[2(1-\varepsilon ) \nonumber \\
&&\times (1-x){\not p}-(4-2\varepsilon )m+(1-\xi )(m-2x{\not p})]-2(1-\xi )
\nonumber \\
&&\times (1-x)\frac{x^2p^2{\not p}}{\Delta (p)}\}
\end{eqnarray}
where $\varepsilon =2-\frac n2$,
\begin{eqnarray}
\Delta (p)=p^2x(x-1)+m^2x
\end{eqnarray}
and M is an arbitrary mass introduced to make the coupling constant e to be
dimensionless in the space of dimension n. According to the definition shown
in Eq.(3.8) and noticing
\begin{eqnarray}
p^2{\not p}=({\not p}-\mu )[p^2+\mu ({\not p}+\mu )]+\mu ^3
\end{eqnarray}
one can get from Eq.(3.22)
\begin{eqnarray}
A &=&-\frac{e^2}{(4\pi )^2}(4\pi M^2)^\varepsilon \Gamma (1+\varepsilon
)\int_0^1dx\{\frac 1{\varepsilon \Delta (\mu )^\varepsilon } \nonumber \\
&&\times [2\mu [1+(\xi -2)x-\varepsilon (1-x)]-(3+\xi -2\varepsilon )m]
\nonumber \\
&&-2(1-\xi )(1-x)x^2\frac{\mu ^3}{\Delta (\mu )^{1+\varepsilon }}
\end{eqnarray}
where
\begin{eqnarray}
\Delta (\mu )=x[\mu ^2(x-1)+m^2]
\end{eqnarray}
On substituting Eqs.(3.22) and (3.25) in Eq.(3.10), it is found that
\begin{eqnarray}
B &=&[\Sigma (p)-A](p-\mu )^{-1}\mid _{{\not p}=\mu } \nonumber \\
&=&-\frac{e^2}{(4\pi )^2}(4\pi M^2)^\varepsilon \Gamma (1+\varepsilon
)\int_0^1dx\{\frac 1{\varepsilon \Delta (\mu )^\varepsilon }[2(1-\varepsilon
) \nonumber \\
&&\times (1-x)-2(1-\xi )x]+\frac{2\mu ^2}{\Delta (\mu )^{1+\varepsilon }}%
[2(1-\varepsilon )x(x-1)^2 \nonumber \\
&&+5(1-\xi )x^2(x-1)+\frac m\mu (3+\xi -2\varepsilon )x(x-1)]-\frac{4\mu ^4}{%
\Delta (\mu )^{2+\varepsilon }} \nonumber \\
&&\times (1-\xi )(1+\varepsilon )(x-1)^2x^3\}
\end{eqnarray}
~~For the diagram of one-loop vertex correction shown in Fig.(1b), according
to the definition written in Eq.(3.2), it is not difficult to obtain, in the
n-dimensional space, the regularized form of the constant L
\begin{eqnarray}
L &=&\frac{e^2}{(4\pi )^2}(4\pi M^2)^\varepsilon \Gamma (1+\varepsilon
)\int_0^1dx\{\frac{2x}{\varepsilon \Delta (\mu )^\varepsilon }[\varepsilon
(\varepsilon -\frac 32) \nonumber \\
&&+\xi (1-\frac 12\varepsilon )]-\frac x{\Delta (\mu )^{1+\varepsilon
}}[2\mu ^2(\varepsilon -1)(x-1)^2 \nonumber \\
&&-(1-\xi )(x^2-x-1)\mu ^2+\varepsilon (1-\xi )(x-1)\mu ^2-4m\mu
[(2-\varepsilon ) \nonumber \\
&&\times (x-1)+\frac 12(1-\xi )(2-3x)+\frac 12\varepsilon (1-\xi
)(x-1)]+m^2[2(\varepsilon \nonumber \\
&&-1)-(1-\varepsilon )(1-\xi )(x-1)]]+(1-\xi )\frac{(1+\varepsilon )}{\Delta
(\mu )^{2+\varepsilon }}(x-1)x^3\mu ^2 \nonumber \\
&&\times (m+\mu )^2\} \\
&& \nonumber
\end{eqnarray}
With the expressions given in Eqs.(3.27) and (3.28), in the approximation of
order $e^2$, the renormalization constants defined in Eq.(3.12) and (3.16)
will be represented as $Z_1=1-L$ and $Z_2=1+B$. In the limit $\varepsilon
\to 0$, these constants are divergent, being not well-defined. So, to verify
the Ward identity in Eq.(3.21), it is suitable to see whether their
anomalous dimensions satisfy the corresponding identity
\begin{equation}
\gamma _1=\gamma _2
\end{equation}
where ${\gamma _i(\mu )=\lim_{\varepsilon \rightarrow 0}\mu \frac d{d\mu
}\ln Z_i(\mu ,\varepsilon )}$ (i=1,2). For the renormalization group
calculations, in practice, the above identity is only necessary to be
required. Through direct calculation by using the constants in Eqs.(3.27)
and (3.28), it is easy to prove
\begin{eqnarray}
\gamma _1 &=&\gamma _2 \nonumber \\
&=&-\frac{e^2}{(4\pi )^2}\{6\xi -6(3+\xi )\sigma +12\xi \sigma ^2+6(3+\xi
-2\xi \sigma ) \nonumber \\
&&\times \sigma ^3\ln \frac{\sigma ^2}{\sigma ^2-1}+4[2\xi -(3+\xi )\sigma
]\frac 1{\sigma ^2-1}
\end{eqnarray}
where ${\sigma =\frac m\mu }$. This identity guarantees the correctness of
the one-loop renormalizations in the GMS scheme. In the zero-mass limit $%
(\sigma \rightarrow 0)$, the identity in Eq.(3.30) reduces to the result
given in the MS scheme
\begin{eqnarray}
\gamma _1=\gamma _2=\frac{\xi e^2}{8\pi ^2}
\end{eqnarray}
This result can directly be derived from such expressions of the constants B
and L that they are obtained from Eqs.(3.27) and (3.28) by setting m=0. It
is easy to see that in these expressions, only the terms proportional ${%
\varepsilon }^{-1}$ give nonvanishing contributions to the anomalous
dimensions. However, in the case of $m\neq 0$, the terms proportional to ${%
\varepsilon }^{-1}$ in Eqs.(3.27) and (3.28) give different results. In this
case, to ensure the identity in Eq.(3.29) to be satisfied, the other terms
without containing ${\varepsilon }^{-1}$ in Eqs.(3.27) and (3.28) must be
taken into account.
\setcounter{section}{4}
\section*{4.Effective Coupling Constant}
\setcounter{equation}{0}
~~ The RGE for the renormalized coupling constant may be immediately written
out from Eq.(2.2) by setting ${F=e,}$
\begin{eqnarray}
\mu \frac d{d\mu }e_R(\mu )+\gamma _e(\mu )e_R(\mu )=0
\end{eqnarray}
In the above, the anomalous dimension ${\gamma _e(\mu )}$ as defined in
Eq.(2.3) is now determined by the following renormalization constant$^5$
\begin{eqnarray}
Z_e=\frac{Z_1}{Z_2Z_3^{\frac 12}}=Z_3^{-\frac 12}
\end{eqnarray}
where the identity in Eq.(3.21) has been considered. The photon propagator
renormalization constant ${Z_3}$ is, in the GMS scheme, defined by
\begin{eqnarray}
Z_3^{-1}=1+\Pi (\mu ^2)
\end{eqnarray}
where ${\Pi (\mu ^2})$ is the scalar function appearing in the photon
self-energy tensor ${\Pi _{\mu \nu }=(k}_\mu {k}_\nu {-k^2g_{\mu \nu })\Pi
(\mu ^2)}$. In view of Eq.(4.2), we can write
\begin{eqnarray}
\gamma _e=\lim_{\varepsilon \to 0}\mu \frac d{d\mu }\ln Z_e=-\frac
12\lim_{\varepsilon \to 0}\mu \frac d{d\mu }\ln Z_3
\end{eqnarray}
For the one-loop diagram represented in Fig.(1c), according to Eq.(4.3), it
is easy to derive the regularized form of the constant ${Z_3}$ by the
dimensional regularization procedure
\begin{eqnarray}
Z_3=1+\frac{e^2}{4\pi ^2}(4\pi M^2)^\varepsilon (2-\varepsilon )\frac{\Gamma
(1+\varepsilon )}\varepsilon \int_0^1\frac{dxx(x-1)}{[\mu
^2x(x-1)+m^2]^\varepsilon }
\end{eqnarray}
Substituting Eq.(4.5) into Eq.(4.4), it is found that
\begin{eqnarray}
\gamma _e=-\frac{e^2}{12(\pi )^2}\{1+6\sigma ^2+\frac{12\sigma ^4}{\sqrt{%
1-4\sigma ^2}}\ln \frac{1+\sqrt{1-4\sigma ^2}}{1-\sqrt{1-4\sigma ^2}}\}
\end{eqnarray}
where $\sigma =\frac m\mu $. In this expression, the charge e and the mass m
are unrenormalized. In the approximation of order $e^2$, they can be
replaced by the renormalized ones $e_R$ and $m_R$ because in this
approximation, as pointed out in the previous literature$^{15}$, the lowest
order approximation of the relation between the $e(m)$ and the $e_R(m_R)$ is
only necessary to be taken into account. Furthermore, when we introduce the
scaling variable $\lambda $ for the renormalization point and set $\mu
_0=m_R $ (which can always be done since the ${\mu _0}$ is fixed, but may be
chosen at will), we have $\sigma =\frac{m_R}{\mu _0\lambda }=\frac 1\lambda $%
. Thus, with the expressions of Eq.(4.6), Eq.(4.1) may be rewritten in the
form
\begin{eqnarray}
\lambda \frac{de_R(\lambda )}{d\lambda }=\beta (\lambda )
\end{eqnarray}
where
\begin{eqnarray}
\beta (\lambda ) &=&-\gamma _e(\lambda )e_R(\lambda ) \nonumber \\
&=&\frac{e_R^3(\lambda )}{12\pi ^2}F_e(\lambda )
\end{eqnarray}
in which
\begin{eqnarray}
F_e(\lambda )=1+\frac 6{\lambda ^2}+\frac{12}{\lambda ^4}f(\lambda )
\end{eqnarray}
\begin{eqnarray}
f(\lambda ) &=&\frac \lambda {\sqrt{\lambda ^2-4}}\ln \frac{\lambda +\sqrt{%
\lambda ^2-4}}{\lambda -\sqrt{\lambda ^2-4}} \nonumber \\
&=&\cases{ \frac{2\lambda }{\sqrt{4-\lambda ^2}}cot^{-1}\frac \lambda
{\sqrt{4-\lambda ^2}},&if $ \lambda \leq 2$ \cr \frac{2\lambda
}{\sqrt{\lambda ^2-4}}coth^{-1}\frac \lambda {\sqrt{\lambda^2 -4}},&if $
\lambda \geq 2$\cr}
\end{eqnarray}
Upon substituting Eqs.(4.8)-(4.10) into Eq.(4.7) and then integrating
Eq.(4.7) by applying the familiar integration formulas, the effective
(running) coupling constant will be found to be
\begin{eqnarray}
\alpha _R(\lambda )=\frac{\alpha _R}{1-\frac{2\alpha _R}{3\pi }G(\lambda )}
\end{eqnarray}
where ${\alpha _R(\lambda )=\frac{e_R^2(\lambda )}{4\pi }}$, $\alpha
_R=\alpha _R(1)$ and
\begin{eqnarray}
G(\lambda ) &=&\int_1^\lambda \frac{d\lambda }\lambda F_e(\lambda )
\nonumber \\
&=&2+\sqrt{3}\pi -\frac 2{\lambda ^2}+(1+\frac 2{\lambda ^2})\frac 1\lambda
\varphi (\lambda )
\end{eqnarray}
in which
\begin{eqnarray}
\varphi (\lambda ) &=&\sqrt{\lambda ^2-4}\ln \frac 12(\lambda +\sqrt{\lambda
^2-4}) \nonumber \\
&=&\cases{ -\sqrt{4-\lambda^2 }\cos ^{-1}\frac \lambda 2,&if $ \lambda \leq
2$ \cr \sqrt{\lambda ^2-4}\cosh ^{-1}\frac \lambda 2,&if $ \lambda \geq
2$\cr}
\end{eqnarray}
As mentioned in Sect.2, the variable $\lambda $ is also the scaling
parameter of momenta, $p=\lambda p_0$ and it is convenient to put ${p_0}^2={%
\mu _0}^2$ so as to apply the boundary condition. Thus, owing to the choice $%
\mu _0=m_R$, we have ${p_0}^2={m_R}^2$ and $\lambda =(\frac{p^2}{m_R^2}%
)^{\frac 12}$. In this case, it is apparent that when $\lambda =1$,
Eq.(4.11) will be reduced to the result given on the mass shell, $\alpha
_R(1)=\alpha _R=\frac 1{137}$ which is identified with that as measured in
experiment.
~~ The behavior of the $\alpha _R(\lambda )$ are exhibited in Figs.(2) and
(3). For small $\lambda $, Eq.(4.11) may be approximated by
\begin{eqnarray}
\alpha _R(\lambda )\approx \frac 34\lambda ^3
\end{eqnarray}
It is clear that when $\lambda \rightarrow 0$, the $\alpha _R(\lambda )$
tends to zero. This desirable behavior, which indicates that at large
distance (small momentum) the interacting particles decouple, is completely
consistent with our knowledge about the electromagnetic interaction. For
large momentum (small distance), Eq.(4.11) will be approximated by
\begin{eqnarray}
\alpha _R(\lambda )\approx \frac{\alpha _R}{1-\frac{2\alpha _R}{3\pi }\ln
\lambda }
\end{eqnarray}
This result was given previously in the mass-independent MS scheme. In the
latter scheme, the $\beta -$ function in Eq.(4.8) is only a function of the $%
e_R(\lambda )$ since $F_e(\lambda )=1$ due to $m=0$ in this case. But, in
general, the mass of a charged particle is not zero. Therefore, the result
given in the MS scheme can only be viewed as an approximation in the large
momentum limit from the viewpoint of conventional perturbation theory.
Fig.(3) shows that the $\alpha _R(\lambda )$ increases with the growth of $%
\lambda $ and tends to infinity when the $\lambda $ approaches an extremely
large value $\lambda _0\approx e^{\frac{2\pi }{3\alpha _R}}=e^{287}$ (the
Landau pole). If the $\lambda $ goes from $\lambda _0$ to infinity, we find,
the $\alpha _R(\lambda )$ will become negative and tends to zero. This
result is unreasonable, conflicting with the physics. The unreasonableness
indicates that in the region $[\lambda _0,\infty )$, the QED perturbation
theory and even the QED itself is invalid$^5$.
\setcounter{section}{5}
\section*{5.Effective Fermion Mass}
\setcounter{equation}{0}
~~The RGE for a renormalized fermion mass can be directly read from Eq.(2.2)
when we set $F=m_R$. Noticing $\mu \frac d{d\mu }=\lambda \frac d{d\lambda }$%
, this equation may be written in the form
\begin{eqnarray}
\lambda \frac{dm_R(\lambda )}{d\lambda }=-\gamma _m(\lambda )m_R(\lambda )
\end{eqnarray}
where the anomalous dimension $\gamma _m(\lambda )$, according to the
definition in Eq.(2.3), can be derived from the renormalization constant
represented in Eq.(3.20). At one-loop level, by making use of the constants
A, B and $Z_2$ which were written in Eqs.(3.25), (3.27) and (3.16)
respectively, in the approximation of order $e^2$, it is not difficult to
derive
\begin{eqnarray}
\gamma _m(\lambda )=\lim_{\varepsilon \to 0}\mu \frac d{d\mu }\ln Z_m=\frac{%
e_R^2}{(4\pi )^2}F_m(\lambda )
\end{eqnarray}
where
\begin{eqnarray}
F_m(\lambda ) &=&2\xi \lambda +6[3+2\xi -\frac{3(1+\xi )}\lambda +\frac{2\xi
}{\lambda ^2}] \nonumber \\
&&\ -\frac{12(1+\xi )\lambda }{1+\lambda }+6[3+\xi -\frac{3(1+\xi )}\lambda +%
\frac{2\xi }{\lambda ^2}]\frac 1{\lambda ^2}\ln \left| 1-\lambda ^2\right|
\end{eqnarray}
here the relation $\sigma =\frac 1\lambda $ has been used. Inserting
Eq.(5.2) into Eq.(5.1) and integrating the latter equation, one may obtain
\begin{eqnarray}
m_R(\lambda )=m_Re^{-S(\lambda )}
\end{eqnarray}
This just is the effective (running) fermion mass where $m_R=m_R(1)$ which
is given on the mass-shell and
\begin{eqnarray}
S(\lambda )=\frac 1{4\pi }\int_1^\lambda \frac{d\lambda }\lambda \alpha
_R(\lambda )F_m(\lambda )
\end{eqnarray}
In the above, the bare charge appearing in Eq.(5.2) has been replaced by the
renormalized one and further by the running one shown in Eq.(4.11). If the
coupling constant in Eq.(5.5) is taken to be the constant defined on the
mass-shell, the integral over $\lambda $ can be explicitly calculated. The
result is
\begin{eqnarray}
S(\lambda )=\frac{\alpha _R}{4\pi }[\varphi _1(\lambda )+\xi \varphi
_2(\lambda )]
\end{eqnarray}
where
\begin{eqnarray}
\varphi _1(\lambda )=3(1-\lambda )\{\frac 2\lambda +[\frac 2{\lambda
^3}-\frac 1{\lambda ^2}(1+\lambda )]\ln \left| 1-\lambda ^2\right| \}
\end{eqnarray}
and
\begin{eqnarray}
\varphi _2(\lambda ) &=&2\lambda +5\ln \lambda -\frac{20}3\ln \frac
12(1+\lambda )-\frac{38}{3\lambda } \nonumber \\
&&\ -\frac{11}{2\lambda ^2}-\frac{55}6+[\frac 56-\frac{17}{6\lambda ^2}%
+\frac 5{2\lambda ^3}-\frac 1{2\lambda ^4}] \nonumber \\
&&\ \times \ln \left| 1-\lambda ^2\right|
\end{eqnarray}
in which
\begin{eqnarray}
\ln \left| 1-\lambda ^2\right| =\cases{ 2[\ln (1+\lambda )-\tanh
^{-1}\lambda ],&if $\lambda \leq 1 $ \cr 2[\ln (1+\lambda )-coth^{-1}\lambda
],&if $\lambda \geq 1 $\cr}
\end{eqnarray}
As we see, the function $S(\lambda )$ and hence the effective mass $%
m_R(\lambda )$ are gauge-dependent. The gauge-dependence is displayed in
Fig.(4). The figure shows that for $\xi <10,$ the effective masses given in
different gauges are almost the same and behave as a constant in the region
of $\lambda <1$, while, in the region of $\lambda >1,$ they all tend to zero
with the growth of $\lambda $. But, the $m_R(\lambda )$ given in the gauge
of $\xi \neq 0$ goes to zero more rapidly than the one given in the Landau
gauge $(\xi =0)$. To be specific, in the following we show the result given
in the Landau gauge which was regarded as the preferred gauge in the
literature$^{3,29}$,
\begin{eqnarray}
m_R(\lambda )=m_Re^{-\frac{\alpha _R}{4\pi }\varphi _1(\lambda )}
\end{eqnarray}
In the limit $\lambda \rightarrow 0$,
\begin{eqnarray}
m_R(\lambda )\rightarrow m_Re^{-\frac{3\alpha _R}{4\pi }}=1.001744m_R
\end{eqnarray}
This clearly indicates that when $\lambda $ varies from 1 to zero, the $%
m_R(\lambda )$ almost keeps unchanged. This result physically is reasonable.
Whereas, in the region of $\lambda >1$, the $m_R(\lambda )$ decreases with
increase of $\lambda $ and goes to zero near the critical point $\lambda _0$%
. This behavior suggests that at very high energy, the fermion mass may be
ignored in the evaluation of S-matrix elements.
\setcounter{section}{6}
\section*{6.Comments and Discussions}
\setcounter{equation}{0}
~~In this paper, the QED renormalization has been restudied in the GMS
scheme. The exact and explicit expressions of the one-loop effective
coupling constant and fermion mass are obtained in the mass-dependent
renormalization scheme and show reasonable asymptotic behaviors. A key point
to achieve these results is that the subtraction is performed in the way of
respecting the Ward identity, i.e. the gauge symmetry. For comparison, it is
mentioned that in some previous literature$^{15,19}$, the fermion
self-energy is represented in such a form
\begin{eqnarray}
\Sigma (p)=A(p^2)\not p+B(p^2)m
\end{eqnarray}
If we subtract the divergence in the $\Sigma (p)$ at the renormalization
point $p^2=\mu ^2$, the fermion propagator is still expressed in the form as
written in Eqs.(3.15) and (3.17); but, the renormalization constants $Z_2$
and $Z_m$ are now defined by
\begin{eqnarray}
Z_2^{-1}=1-A(\mu ^2),Z_m^{-1}=Z_2[1+B(\mu ^2)]
\end{eqnarray}
The one-loop expressions of the $A(\mu ^2)$ and $B(\mu ^2)$ can directly be
read from Eq.(3.22). Here, we show the one-loop anomalous dimension of
fermion propagator which is given by the above subtraction
\begin{eqnarray}
\gamma _2 &=&\lim_{\varepsilon \to 0}\mu \frac d{d\mu }\ln Z_2 \nonumber \\
&=&\frac{\xi e^2}{4\pi ^2}[\frac 12+\sigma ^2-\sigma ^4\ln \frac{\sigma ^2}{%
\sigma ^2-1}]
\end{eqnarray}
In comparison of the above $\gamma _2$ with the $\gamma _1$ shown in
Eq.(3.30), we see, the Ward identity in Eq.(3.29) can not be fulfilled
unless in the zero-mass limit $(\sigma \to 0)$. Since the Ward identity is
an essential criterion to identify whether a subtraction is correct or not,
the subtraction stated above should be excluded from the mass-dependent
renormalization.
~~Another point we would like to address is that in the Ward identity shown
in Eq.(3.1), the momenta p and $p^{\prime }$ on the fermion lines in the
vertex are set to be equal. According to the energy-momentum conservation,
the momentum k on the photon line should be equal to zero. correspondingly,
the subtraction shown in Eq.(3.2) was carried out at the so-called
asymmetric points, $p^{\prime }{}^2=p^2=\mu ^2$ and $k^2=(p^{\prime }-p)^2=0$%
. These subtraction points coincide with the energy- momentum conservation,
i.e., the Lorentz-invariance. Nevertheless, the subtraction performed at the
symmetric point $p^{\prime 2}=p^2=k^2=-\mu ^2$ was often used in the
previous works $^{3,19,20}$. This subtraction not only makes the calculation
too complicated, but also violates the energy- momentum conservation which
holds in the vertex. That is why the symmetric point subtraction is beyond
our choice.
~~As mentioned in Introduction, the GMS subtraction is a kind of MOM scheme
in which the renormalization points are chosen to be time-like. In contrast,
in the conventional MOM scheme$^{3,6,29}$, the renormalization points were
chosen to be space-like, i.e. $p_i^2=-\mu ^2$ which implies $\not p_i=i\mu $%
. In this scheme, the one-loop result for the anomalous dimension $\gamma _e$
can be written out from Eq.(4.6) by the transformation $\sigma \to -i\sigma $%
. That is
\begin{eqnarray}
\gamma _e=-\frac{4e^2}{3(4\pi )^2}\{1-6\sigma ^2+\frac{12\sigma ^4}{\sqrt{%
1+4\sigma ^2}}\ln \frac{\sqrt{1+4\sigma ^2}+1}{\sqrt{1+4\sigma ^2}-1}\}
\end{eqnarray}
which is identical to that given in Refs.(6) and (29). In comparison of
Eq.(6.4) with Eq.(4.6), we see, the coefficients of $\sigma ^2$ in Eq.(6.4)
changes a minus sign. Substituting Eq.(6.4) into Eq.(4.1) and solving the
latter equation, we obtain the running coupling constant which is still
represented in Eq.(4.11), but the function $G(\lambda )$ in Eq.(4.11) is now
given by
\begin{eqnarray}
G(\lambda ) &=&\frac 2{\lambda ^2}-2-\frac{\sqrt{\lambda ^2+4}}\lambda
(\frac 2{\lambda ^2}-1)\ln \frac 12(\lambda +\sqrt{\lambda ^2+4}) \nonumber
\\
&&+\sqrt{5}\ln \frac 12(1+\sqrt{5})
\end{eqnarray}
This expression is obviously different from the corresponding one written in
Eqs.(4.12) and (4.13). At small distance $\left( \lambda \rightarrow \infty
\right) $, Eq.(6.5) still gives the approximate expression presented in
Eq.(4.15). However, at large distance, as shown in Fig.(2), the $\alpha
_R(\lambda )$ behaves almost a constant. When $\lambda \rightarrow 0$, it
approaches to a value equal to 0.99986$\alpha _R$, unlike the $\alpha
_R(\lambda )$ given in Eqs.(4.11)-(4.13) which tends to zero. Let us examine
the effective mass. As we have seen from Sect.5, the one-loop effective
fermion mass given in the GMS scheme is real. However, in the space-like
momentum subtraction, due to $\not p=i\mu $, the effective mass will contain
an imaginary part. This result can be seen from the function $F_m(\lambda )$
whose one-loop expression given in the usual MOM scheme can be obtained from
Eq.(5.3) by the transformation $\lambda \to i\lambda $ and therefore becomes
complex. The both of subtractions may presumably be suitable for different
processes of different physical natures. But, if the effective mass is
required to be real, the subtraction at space-like renormalization point
should also be ruled out.
~~As pointed out in Sect.4, the effective coupling constant shown in
Eq.(4.15) which was obtained in the MS scheme is only an approximation given
in the large momentum limit from the viewpoint of conventional perturbation
theory. Why say so? As is well-known, the MS scheme is a mass-independent
renormalization scheme in which the fermion mass is set to vanish in the
process of subtraction. The reasonability of this scheme was argued as
follows$^{1,15}$ .The fermion propagator can be expanded as a series
\begin{equation}
\frac 1{\not p-m}=\frac 1{\not p}+\frac 1{\not p}m\frac 1{\not p}+\frac 1{%
\not p}m\frac 1{\not p}m\frac 1{\not p}+\cdot \cdot \cdot
\end{equation}
According to this expansion, the massive propagator $\frac 1{\not p-m}$ may
be replaced by the massless one $\frac 1{\not p}.$ At the same time, the
fermion mass, as the coupling constant, can also be treated as an expansion
parameter for the perturbation series. Nevertheless, in the mass-dependent
renormalization as shown in this paper, the massive fermion propagator is
employed in the calculation and only the coupling constant is taken to be
the expansion parameter of the perturbation series. Thus, in order to get
the perturbative result of a given order of the coupling constant in the
mass-dependent renormalization, according to Eq.(6.6), one has to compute an
infinite number of terms in the MS scheme. If only the first term in
Eq.(6.6) is considered in the MS scheme, the result derived in the this
scheme, comparing to the corresponding one obtained in the mass-dependent
renormalization, can only be viewed as an approximation given in the large
momentum limit . Even if in this limit, a good renormalization scheme should
still be required to eliminate the ambiguity and give an unique result. To
this end, we may ask whether there should exist the difference between the
MS scheme and the $\overline{MS}$ scheme ${^2}$? As one knows, when the
dimensional regularization is employed in the mass-independent
renormalization, the MS scheme only subtracts the divergent term having the $%
\varepsilon $-pole in a Feynman integral which is given in the limit $%
\varepsilon \rightarrow 0$ and uses this term to define the renormalization
constant. While, the $\overline{MS}$ scheme is designed to include the
unphysical terms ${\gamma -\ln 4\pi }$ (here $\gamma $ is the Euler
constant) in the definition of the renormalization constant. The unphysical
terms arise from a special analytical continuation of the space-time
dimension from n to 4. When the two different renormalization constants
mentioned above are inserted into the relation $e=Z_3^{-\frac 12}e_R$, one
would derive a relation between the two different renormalized coupling
constants given in the MS and $\overline{MS}$ schemes if the higher order
terms containing the $\varepsilon $-pole are ignored. It would be pointed
out that the above procedure of leading to the difference between the MS and
$\overline{MS}$ schemes is not appropriate because the procedure is based on
direct usage of the divergent form of the renormalization constants. As
emphasized in the Introduction, according to the convergence principle, it
is permissible to use such renormalization constants to do a meaningful
calculation. The correct procedure of deriving a renormalized quantity is to
solve its RGE whose solution is uniquely determined by the anomalous
dimension (other than the renormalization constant itself) and boundary
condition. In computing the anomalous dimension, the rigorous procedure is
to start from the regularized form of the renormalization constant. In the
regularized form, it is unnecessary and even impossible to divide a
renormalization constant into a divergent part and a convergent part. Since
the anomalous dimension is a convergent function of $\varepsilon $ due to
that the factor $\frac 1\varepsilon $ disappears in it, the limit $%
\varepsilon \rightarrow 0$ taken after the differentiation with respect to
the renormalization point would give an unambiguous result. Especially, the
unphysical factor $\left( 4\pi \right) ^\varepsilon \Gamma (1+\varepsilon )$
appearing in Eqs.(3.25), (3.27) and (4.5) straightforwardly approaches 1 in
the limit. Therefore, the unphysical terms $\gamma -\ln 4\pi $ could not
enter the anomalous dimension and the effective coupling constant. Even if
we work in the zero-mass limit or in the large momentum regime, we have an
only way to obtain the effective coupling constant as shown in Eq.(4.15) in
the one-loop approximation. That is to say, it is impossible, in this case,
to result in the difference between the MS and $\overline{MS}$ schemes and
also the difference between the MOM and the MS schemes.
The above discussions suggest that the ambiguity arising from different
renormalization prescriptions may be eliminated by the necessary physical
and mathematical requirements as well as the boundary conditions. It is
expected that the illustration given in this paper for the QED one-loop
renormalization performed in a mass-dependent scheme would provide a clue on
how to do the QED multi-loop renormalization and how to give an improved
result for the QCD renormalization.
\begin{center}
{\bf Acknowledgment}
\end{center}
The authors wish to thank professor Shi-Shu Wu for useful discussions. This
project was supposed in part by National Natural Science Foundation.
\renewcommand{\thesection}{\Alph{section}}\setcounter{section}{1}%
\setcounter{section}{1}\setcounter{equation}{0}
\begin{center}
{\bf Appendix: Illustration of the regularization procedure by a couple of
mathematical examples}
\end{center}
We believe that if a quantum field theory is built up on the faithful basis
of physical principles and really describes the physics, a S-matrix element
computed from such a theory is definite to be convergent even though there
occur divergences in the perturbation series of the matrix element. The
occurrence of divergences in the perturbation series, in general, is not to
be a serious problem in mathematics. But, to compute such a series, it is
necessary to employ an appropriate regularization procedure. For example,
for the following convergent integral
\begin{equation}
f(a)=\int_0^\infty dxe^{-ax}
\end{equation}
which equals to $\frac 1a$, if we evaluate it by utilizing the series
expansion of the exponential function
\begin{equation}
e^{-ax}=\sum_{n=0}^\infty \frac{(-a)^n}{n!}x^n
\end{equation}
as we see, the integral of every term in the series is divergent. In this
case, interchange of the integration with the summation actually is not
permissible. When every integral in the series is regularized, the
interchange is permitted and all integrals in the series become calculable.
Thus, the correct procedure of evaluating the integral by using the series
expansion is as follows
\[
f(a)=\lim_{\Lambda \rightarrow \infty }\sum_{n=0}^\infty \frac{(-a)^n}{n!}%
\int_0^\Lambda dxx^n
\]
\begin{eqnarray}
\ &=&\lim_{\Lambda \rightarrow \infty }\sum_{n=0}^\infty \frac{(-a)^n}{n!}%
\frac{\Lambda ^{n+1}}{n+1}=\lim_{\Lambda \rightarrow \infty }\frac
1a(1-e^{-a\Lambda }) \nonumber \\
\ &=&\frac 1a
\end{eqnarray}
This example is somewhat analogous to the perturbation series in the quantum
field theory and suggests how to do the calculation of the series with the
help of a regularization procedure. Unfortunately, in practice, we are not
able to compute all the terms in the perturbation series. In this situation,
we can only expect to get desired physical results from a finite order
perturbative calculation. How to do it? To show the procedure of such a
calculation, let us look at another mathematical example. The following
integral
\begin{equation}
F(a)=\int_0^\infty dx\frac{e^{-(x+a)}}{(x+a)^2}
\end{equation}
where $a>0$ is obviously convergent. When the exponential function is
expanded as the Taylor series, the integral will be expressed as
\begin{eqnarray}
F(a) &=&\int_0^\infty dx\sum_{n=0}^\infty \frac{(-1)^n}{n!}%
(x+a)^{n-2}=\int_0^\infty \frac{dx}{(x+a)^2}-\int_0^\infty \frac{dx}{x+a}
\nonumber \\
&&\ \ +\int_0^\infty dx\sum_{n=2}^\infty \frac{(-1)^n}{n!}(x+a)^{n-2}
\end{eqnarray}
Clearly, in the above expansion, the first term is convergent, similar to
the tree-approximate term in the perturbation theory of a quantum field
theory, the second term is logarithmically divergent, analogous to the
one-loop-approximate term and the other terms amount to the higher order
corrections in the perturbation theory which are all divergent. To calculate
the integral in Eq.(A.4), it is convenient at first to evaluate its
derivative,
\begin{equation}
\frac{dF(a)}{da}=-\int_0^\infty dx\frac{(x+a+2)}{(x+a)^3}e^{-(x+a)}
\end{equation}
which is equal to $-e^{-a}/a^2$ as is easily seen from integrating it over x
by part. In order to get this result from the series expansion, we have to
employ a regularization procedure. Let us define
\begin{equation}
F_\Lambda (a)=\int_0^\Lambda \frac{e^{-(x+a)}}{(x+a)^2}
\end{equation}
Then, corresponding to Eq.(A.5), we can write
\begin{eqnarray}
\frac{dF_\Lambda (a)}{da} &=&-2\int_0^\Lambda \frac{dx}{(x+a)^3}%
+\int_0^\Lambda \frac{dx}{(x+a)^2}+\sum_{n=2}^\infty \frac{(-1)^n}{n!}%
(n-2)\int_0^\Lambda dx(x+a)^{n-3} \nonumber \\
\ &=&\sum_{n=0}^\infty \frac{(-1)^n}{n!}[(\Lambda +a)^{n-2}-a^{n-2}]=\frac{%
e^{-(\Lambda +a)}}{(\Lambda +a)^2}-\frac{e^{-a}}{a^2}
\end{eqnarray}
From the above result, it follows that
\begin{equation}
\frac{dF(a)}{da}=\lim_{\Lambda \rightarrow \infty }\frac{dF_\Lambda (a)}{da}%
=-\frac{e^{-a}}{a^2}
\end{equation}
This is the differential equation satisfied by the function F(a). Its
solution can be expressed as
\begin{equation}
F(a)=F(a_0)-\int_{a_0}^ada\frac{e^{-a}}{a^2}
\end{equation}
where $a_0$ $>0$ is a fixed number which should be determined by the
boundary condition of the equation (A.9). Now, let us focus our attention on
the second integral in the first equality of Eq.(A.8). This integral is
convergent in the limit $\Lambda \rightarrow \infty $ and can be written as
\begin{equation}
\frac{dF_1(a)}{da}=\lim_{\Lambda \rightarrow \infty }\int_0^\Lambda dx\frac
1{(x+a)^2}=\int_0^\infty dx\frac 1{(x+a)^2}=\frac 1a
\end{equation}
Integrating the above equation over a, we get
\begin{equation}
F_1(a)=F_1(a_0)+\ln \frac a{a_0}
\end{equation}
If setting
\begin{equation}
F_1(a_0)=\ln \frac{a_0}\mu
\end{equation}
where $\mu $ is a finite number, Eq.(A.12) becomes
\begin{equation}
F_1(a)=\ln \frac a\mu
\end{equation}
This result is finite as long as the parameter $\mu $ is not taken to be
zero and can be regarded as the contribution of the divergent integral $%
F_1(a)$ appearing in the second term of Eq.(A.5) to the convergent integral
F(a). The procedure described above for evaluating the function F(a) much
resembles the renormalization group method and the approach proposed in
Ref.(9). It shows us how to calculate a finite quantity from its series
expansion which contains divergent integrals.
\newpage
|
\section*{Special Schubert conditions}
For background on the Grassmannian, Schubert cycles, and the Schubert
calculus, see any of~\cite{Hodge_Pedoe,Griffiths_Harris,Fulton_tableaux}.
Let $m,p\geq 1$ be integers.
Let $\gamma$ be a rational normal curve in ${\mathbb R}^{m+p}$.
For $k>0$ and $s\in\gamma$, let $K_k(s)$ be the $k$-plane
osculating $\gamma$ at $s$.
For every integer $a>0$, let $\tau_a(s)$ be the special Schubert cycle
consisting of $p$-planes $H$ in ${\mathbb C}^{m+p}$
which meet $K_{m+1-a}(s)$ nontrivially and let
$\tau^a(s)$ be the special Schubert cycle consisting of $p$-planes $H$
in ${\mathcal C}^{m+p}$ meeting $K_{m-1+a}(s)$ improperly:
$\dim H\cap K_{m-1+a}(s)> a-1$.
These cycles $\tau_a(s)$ and $\tau^a(s)$ each have
codimension $a$ and $\tau^1=\tau_1$.
Recall that the Grassmannian of $p$-planes in ${\mathbb C}^{m+p}$ has
dimension $mp$.
For any Schubert condition $w$, let $\sigma_w(s)$ be the Schubert cycle of
type $w$ given by the flag osculating $\gamma$ at $s$
and set $|w|$ to be the codimension of $\sigma_w(s)$.
\begin{thm}
Let $a_1,\ldots,a_n$ be positive integers with $a_1+\cdots+a_n=mp$.
For each $i=1,\ldots,n$ let $\sigma_i(s)$ be either $\tau^{a_i}(s)$ or
$\tau_{a_i}(s)$.
Then there exist real points $0,\infty,s_1,\ldots,s_n\in{\gamma}$ such that
for any Schubert conditions $w,v$, and integer $k$ with
$|w|+|v|+a_k+\cdots+a_n=mp$, the intersection
\begin{equation}\label{eq:special}
\sigma_w(0)\cap \sigma_v(\infty)\cap
\sigma_k(s_k)\cap\cdots\cap\sigma_n(s_n)
\end{equation}
is transverse with all points of intersection real.
\end{thm}
Our proof is inspired by the Pieri homotopy algorithm of~\cite{HSS}.
We prove this in the case that each
$\sigma_i(s)=\tau_{a_i}(s)$.
This is no loss of generality, as the cycles $\tau_a(s)$ and $\tau^a(s)$
share the properties we need.
We use the following two results of Eisenbud and Harris~\cite{EH83},
who studied such intersections in their theory of limit linear systems.
For a Schubert class $w$,
$w*a$ be the index of summation in the Pieri formula in the
cohomology of the Grassmannian~\cite{Fulton_tableaux},
$$
\sigma_w\cdot \tau_a\ =\ \sum_{v\in w*a} \sigma_v.
$$
\begin{prop}\label{prop:only}
\mbox{ }
\begin{enumerate}
\item (Theorem 2.3 of~\cite{EH83}) Let $s_1,\ldots,s_n$ be distinct points on
$\gamma$ and $w_1,\ldots,w_n$ be Schubert
conditions.
Then the intersection of Schubert cycles
$$
\sigma_{w_1}(s_1)\cap\sigma_{w_2}(s_2)\cap\cdots\cap\sigma_{w_n}(s_n)
$$
is proper in that it has dimension $mp-|w_1|-\cdots-|w_n|$.
\item (Theorem 8.1 of~\cite{EH83}) For any Schubert condition $w$,
integer $a>0$, and $0\in\gamma$,
we have
$$
\lim_{t\rightarrow 0}\left( \sigma_w(0)\cap\tau_a(t)\rule{0pt}{11pt}\right)\
=\ \bigcup_{v\in w*a} \sigma_v(0),
$$
the limit taken along the rational normal curve, and as schemes.
\end{enumerate}
\end{prop}
\noindent{\bf Proof of Theorem~1.}
We argue by downward induction on $k$.
The initial case of $k=n$ holds as Pieri's formula implies the intersection
is a single, necessarily real, point.
Suppose it holds for $k$, and let $w,v$ satisfy
$|w|+|v|+a_{k-1}+\cdots+a_n=mp$.
\noindent{\bf Claim}: The cycle
$$
\sum_{u\in w*a_{k-1}}
\sigma_u(0)\cap\sigma_v(\infty)\cap
\tau_{a_k}(s_k)\cap\cdots\cap\tau_{a_n}(s_n)
$$
is free of multiplicities.
If not, then two summands, say $u$ and $u'$, have a point in common
and so
\begin{equation}\label{eq:bad_intersect}
\sigma_u(0)\cap\sigma_{u'}(0)\cap\sigma_v(\infty)\cap
\tau_{a_k}(s_k)\cap\cdots\cap\tau_{a_n}(s_n)
\end{equation}
is nonempty.
However, $\sigma_u(0)\cap\sigma_{u'}(0)$
is a Schubert cycle of smaller dimension.
Thus the intersection~(\ref{eq:bad_intersect})
must be empty, by Proposition~\ref{prop:only}~(1).
\medskip
From the claim and Proposition~\ref{prop:only}~(2), there is an
$\epsilon_{w,v}>0$
such that if $0<t\leq\epsilon_{w,v}$, then
$$
\sigma_w(0)\cap \sigma_v(\infty)\cap
\tau_{a_{k-1}}(t)\cap\tau_{a_k}(s_k)\cap\cdots\cap\tau_{a_n}(s_n)
$$
is transverse with all points of intersection real.
Set
$$
s_{k+1}\ :=\ min\{\epsilon_{w,v}\mid
mp=|w|+|v|+a_{k-1}+\cdots+a_n\}.\qed
$$
\noindent{\bf Remark. }
Eisenbud and Harris~\cite{EH83} prove Proposition~\ref{prop:only}~(2) for
any nondegenerate arc $\gamma$.
Theorem~1 may be similarly strengthened.\medskip
\section*{Consequences}
Since small real perturbations of the points $s_1,\ldots,s_n$ cannot create
or destroy real points in a transverse intersection, Theorem~1 has the
following consequence.
\begin{cor}\label{cor:open}
There is an open subset (in the classical topology on $\gamma^n$)
consisting of $n$-tuples $(s_1,\ldots,s_n)\in\gamma^n$ such
that~(\ref{eq:special}) is transverse with all points of intersection real.
\end{cor}
Theorem 1 proves part of a conjecture of Shapiro and Shapiro.
\begin{conj}[Shapiro and Shapiro]\label{conj:shapiro}
Let $m,p\geq 1$ and $w_1,\ldots,w_n$ be Schubert conditions on $p$-planes in
$(m+p)$-space.
If $s_1,\ldots,s_n\in\gamma$ are real points so that
\begin{equation}\label{eq:intersection}
\sigma_{w_1}(s_1)\cap\sigma_{w_2}(s_2)\cap\cdots\cap\sigma_{w_n}(s_n)
\end{equation}
is zero-dimensional, then all points of intersection are real.
\end{conj}
When the points $s_1,\ldots,s_n$ are are distinct and
$|w_1|+\cdots+|w_n|=mp$,~(\ref{eq:intersection}) is
zero-dimensional~\cite{EH83}.
Besides Theorem~1, there is substantial and compelling evidence for the
validity of
Conjecture~\ref{conj:shapiro}~\cite{FRZ,Verschelde,RS98,Sottile_shapiro}.
In every instance (choice of $w_1,\ldots,w_n$ and distinct real points
$s_1,\ldots,s_n\in\gamma$) checked, all points in~(\ref{eq:intersection})
are real.
This includes some when the $w_i$ are not special Schubert
conditions~\cite{Sottile_shapiro} and some spectacular
computations~\cite{FRZ,Verschelde}.
For some sets of Schubert conditions, we have proven that for every choice of
distinct real points $s_1,\ldots,s_n\in\gamma$, all points
in~(\ref{eq:intersection}) are real.
In every known case, the intersection scheme is reduced when the parameters
$s_1,\ldots,s_n$ are distinct and real.
We conjecture this is always the case.
\begin{conj}\label{conj:reduced}
Let $m,p\geq 1$ and $w_1,\ldots,w_n$ be Schubert conditions with
$mp=|w_1|+\cdots+|w_n|$.
If $s_1,\ldots,s_n\in\gamma$ are distinct and real, then
the intersection scheme~(\ref{eq:intersection}) is reduced.
\end{conj}
The number of real points in the scheme~(\ref{eq:intersection})
is locally constant on the set
of parameters for which it is reduced.
Thus for special Schubert conditions, Conjecture~\ref{conj:reduced}
would imply Conjecture~\ref{conj:shapiro}, by Theorem 1.
In fact, more is true.
\begin{thm}
Conjecture~\ref{conj:reduced} implies Conjecture~\ref{conj:shapiro}.
\end{thm}
\noindent{\bf Proof. }
By Remark~3.4 to Theorem~3.3 of~\cite{Sottile_shapiro}, if
Conjecture~\ref{conj:shapiro} holds when $a_1=\cdots=a_{mp}=1$, then it
holds for any collection of Schubert conditions, if the
cycles~(\ref{eq:intersection}) are reduced for any choice of
$w_1,\ldots,w_n$ with $mp=|w_1|+\cdots+|w_n|$ and general
$s_1,\ldots,s_n\in\gamma$.
Both of these conditions are supplied by Conjecture~\ref{conj:reduced}.
\qed
\medskip
When $n=mp$ so that $a_1=\cdots=a_{mp}=1$,~(\ref{eq:special}) is a special
case of the pole placement problem in systems theory~\cite{Byrnes}.
A physical system (e.g.~a mechanical linkage), called a {\it plant} with
$m$ inputs and $p$ measured outputs
whose evolution is governed by a system of linear differential equations
is modeled by a
$m\times(m+p)$-matrix of univariate polynomials $[N(s): D(s)]$.
The largest degree of a maximal minor of this matrix is the MacMillan
degree, $n$, of the evolution equation.
For $s_1,\ldots,s_n\in{\mathbb C}$, any $p$-plane $H$ satisfying
$$
H\cap \mbox{\rm row span}[N(s_i): D(s_i)]\ \neq\ \{0\}\qquad
i=1,\ldots,n
$$
gives a constant linear feedback law for which the resulting closed system
has natural frequencies (poles) $s_1,\ldots,s_n$.
In this way, the pencil $K(s)$ of $m$-planes osculating $\gamma$
represents a particular $m$-input $p$-output plant of McMillan degree
$mp$, and the $p$-planes in
$$
\tau_1(s_1)\cap\tau_1(s_2)\cap\cdots\cap\tau_1(s_{mp})
$$
represent
linear feedback laws for which the resulting closed system has
natural frequencies $s_1,\ldots,s_{mp}$.
Since translation along $\gamma$ fixes the system and acts on the feedback
laws by a real linear transformation, we may assume that
$s_1,\ldots,s_{mp}<0$ and so we obtain the existence of a stable
system with only real feedback laws.
\begin{cor}
The plant represented by $K(s)$ can be stabilized by real feedback laws in
such a way that all feedback laws are real.
\end{cor}
|
\section{Introduction}
As is well-known the nonlinear interference effects in the resonant
atom-light interaction can lead to the electromagnetically induced
transparency (EIT) of atomic medium \cite{EIT} as well as to other
interesting phenomena \cite{CPT}.
The key point of all these phenomena is the light-induced coherence
between atomic levels, which are not coupled by dipole transitions.
Recently Akulshin and co-workers have observed
subnatural-width resonances in the absorption on the $D_2$ line
of rubidium vapor under excitation by two copropagating optical waves
with variable frequency offset \cite{EIA}. Surprisingly, that apart
from EIT-resonances with negative sign, they have detected positive
resonances termed in ref.\cite{EIA} as electromagnetically induced
absorption (EIA). Basing on the experimental results and numerical
calculations, authors of ref.\cite{EIA} have deduced \cite{EIA2} that
EIA occurs in a degenerate two-level system when the three conditions
are satisfied: i) The excited-state total angular momentum
$F_e=F_g+1$. ii) Transition $F_g \rightarrow F_e$ is closed. iii)
The ground state is degenerate $F_g > 0$.
In the present paper, motivated by the absence of EIA-resonances in
three-state $\Lambda$ and $V$-systems, we propose a simple
theoretical model for EIA -- four-state $N$-atom.
Namely, we consider an atom with four states $| i\rangle,\;\;
i=1\ldots 4$. The odd states $| 1\rangle$ and $| 3\rangle$ are
degenerate and belong to the ground level, while the even states
$| 2\rangle$ and $| 4\rangle$
(also degenerate) form the excited level. All optical transitions
$| odd\rangle \rightarrow | even\rangle$ are permitted except for
$| 1\rangle \rightarrow | 4\rangle$, that is forbidden. The
control field with frequency $\omega_1$ drives the transitions
$| 1\rangle \rightarrow | 2\rangle$ and $| 3\rangle \rightarrow |
4\rangle$. The weak probe at $\omega_2$ is applied to the $| 3\rangle
\rightarrow | 2\rangle$ transition. An analytical expression for
the probe absorption as a function of frequency offset
$\omega_1-\omega_2$ is found. It is shown that EIA appears due to the
spontaneous transfer of the light-induced low-frequency coherence
from the excited level to the ground one. Obviously, both $\Lambda$
and $V$-systems do not describe such a process. The sign of the
subnatural resonance depends on the branching ratio constant $0 \leq
b\leq 1$ and becomes positive for closed transition $b=1$. Velocity
averaging in the case of Doppler broadening is briefly discussed.
It is worth noting, effects of the
spontaneous coherence transfer on nonlinear resonances in the probe
field spectroscopy were first considered from the general point of
view by S. G. Rautian \cite{cohcasc}.
\section{Formulation of the problem}
Let us consider the resonant interaction of a bichromatic light
field:
\begin{equation} \label{field}
{\bf E}({\bf r},t) =
{\bf E}_{1} \exp[-i\omega_1 t+i({\bf k_1}{\bf r})] +
{\bf E}_{2} \exp[-i\omega_2 t+i({\bf k_2}{\bf r})] + c.c.
\end{equation}
with a four-state atom. This atomic system has four states
$| i\rangle,\;\; i=1\ldots 4$ (see in fig.1). The two odd states
$| 1\rangle$ and $| 3\rangle$ are
degenerate and belong to the ground level with zero energy and zero
relaxation rate. The even states are also degenerate and form the
excited level with the energy $\hbar \omega_0$ and relaxation rate
$\Gamma$. We assume that among optical transitions between the ground
and excited levels $| odd\rangle \rightarrow | even\rangle$ the
transition $| 1\rangle \rightarrow | 4\rangle$ is forbidden due to
some selection rule (for instance, with respect to the momentum
projection). Let the first term in eq.(\ref{field})
(will be refered as a control field) is sufficiently larger
than the second one, which is a probe field.
The control field with the vector amplitude ${\bf E}_1$ and
frequency $\omega_1$ drives simultaneously two transitions
$| 1\rangle \rightarrow | 2\rangle$ and $| 3\rangle \rightarrow |
4\rangle$. The weak probe filed ${\bf E}_2$ at the
frequency $\omega_2$ induces the $| 3\rangle\rightarrow | 2\rangle$
transition. In the rotating frame the Hamiltonian for the free
atom reads
\begin{equation} \label{ham0}
\widehat{H}_0=\hbar \delta_1 |1\rangle\langle 1| +
\hbar \delta_2 |3\rangle\langle 3|+
\hbar (\delta_2 -\delta_1)|4\rangle\langle 4| \,,
\end{equation}
where $\delta_q = \omega_q - \omega_0 -({\bf k}_q {\bf v})$
($q=1,2$) are the detunings including the Doppler shifts.
Using the rotating wave approximation, we write the atom-field
interaction Hamiltonian in the form
\begin{equation} \label{hamAF}
\widehat{H}_{AF}=\hbar \Omega_1 \widehat{Q}_1 +
\hbar \Omega_2 \widehat{Q}_2+h.c. \,.
\end{equation}
Here $\Omega_q$ are the corresponding Rabi frequencies and
the operators $\widehat{Q}_q$ are given by
\begin{eqnarray} \label{Qop}
\widehat{Q}_1 &=& A |2\rangle\langle 1| +
|4\rangle\langle 3|\nonumber\\
\widehat{Q}_2 &=& B |2\rangle\langle 3|\,,\;\;\; A^2+B^2=1 \,,
\end{eqnarray}
where the real numbers $A$ and $B$ govern the relative amplitudes of
transitions in the model under consideration.
In the case of the pure radiative relaxation the optical Bloch
equations for the atomic density matrix $\widehat{\rho}$ read
\begin{equation} \label{OBE}
\frac{d}{dt}\widehat{\rho}+
\frac{i}{\hbar}\left[\widehat{H}_{0}+\widehat{H}_{AF},\,\widehat{\rho}\right]
+\frac{1}{2}\Gamma \left\{\sum_{q=1,2}\widehat{Q}_q \widehat{Q}^{\dagger}_q,
\,\widehat{\rho}\right\}
- b \Gamma \sum_{q=1,2}\widehat{Q}^{\dagger}_q \widehat{\rho}\widehat{Q}_q
=\widehat{R} \, ,
\end{equation}
where the third term on the l.h.s. has a structure of
anticommutator and describes the radiative damping of the
excited-level populations and optical coherences. The last term on
the l.h.s. corresponds to the transfer of the populations and
low-frequency coherences from the excited level to the ground one
under the spontaneous emission. The branching ratio constant $b$ can
vary between $0$ and $1$ and governs a degree in which the atomic
system is open. For example, $b=1$ means that the considered
transition is closed. The r.h.s. of eq.(\ref{OBE}) is a source
describing possible external pumping of levels.
\section{Linear response to the probe field}
As is well-known, under the stationary conditions the intensity
${\cal I}_2$ of the probe field propagating through atomic
medium obeys the equation
$$
\frac{d {\cal I}_2}{d z}=-\hbar \omega_2 {\cal N}\,
( iB\Omega^{*}_2 \langle\rho_{23}\rangle_v+c.c ) \; ,
$$
where the $z$-axis is directed along ${\bf k}_2$, ${\cal N}$ is
the atomic density, and $\langle\ldots\rangle_v$ means averaging
over the atomic velocity. Thus, the probe field absorption is
proportional to the real part of the product $i\Omega^{*}_2
\rho_{23}$. The steady-state off-diagonal element $\rho_{23}$ can be
written as
\begin{equation} \label{opcoh}
\rho_{23}=[\Gamma/2-i\delta_2]^{-1}
\{-iB\Omega_2(\rho_{33}-\rho_{22})
-iA\Omega_1\rho_{13}+i\Omega_1\rho_{24}\} \,.
\end{equation}
The last two terms in the curly brackets in eq.(\ref{opcoh})
(proportional to $\Omega_1$)
describe modifications of the absorption due to the light-induced
low-frequency coherences. It is well-known that in both $\Lambda$ and
$V$ systems the coherences $\rho_{13}$ and $\rho_{24}$ have such
phases that the absorption of the probe field is reduced at the
two-photon resonance $\delta_2=\delta_1$. We note that these two
terms in eq.(\ref{opcoh}) have opposite signs. In the case under
consideration an additional term in equation for $\rho_{13}$ arises
from the spontaneous transfer $\rho_{24} \rightarrow \rho_{13}$
($d\rho_{13}/dt =\ldots + b A \Gamma \rho_{24}$). Hence, the phase of
$\rho_{24}$ giving the transparency through the term
$i \Omega_1 \rho_{24}$ can give the icrease of absorption through
the term $-i A \Omega_1 \rho_{13}$.
For the sake of clarity, in the following analyses we use two
approximations: i) The first order in the probe amplitude
$\Omega_2$; and ii) The low-saturation limit for the control field,
i.e. $\Omega_1 < \Gamma$. In this case instead of eq.(\ref{opcoh}) we
write \begin{equation} \label{linresp}
\rho^{(1)}_{23}=[\Gamma/2-i\delta_2]^{-1}
\{-iB\Omega_2 \rho^{(0)}_{33}
-iA\Omega_1\rho^{(1)}_{13}\} \,,
\end{equation}
where the index over $\rho^{(n)}$ means that this element is taken in
the $n$-th order on $\Omega_2$. The equation (\ref{linresp}) should
be completed by the following equations for the first-order
coherences:
\begin{eqnarray} \label{1order}
i(\delta_1-\delta_2)\rho^{(1)}_{13} &=& iB\Omega_2\rho^{(0)}_{12}
-iA\Omega_1^{*} \rho^{(1)}_{23}
+i\Omega_1\rho^{(1)}_{14}
+ bA\Gamma\rho^{(1)}_{24} \nonumber\\
\left[\Gamma+i(\delta_1-\delta_2)\right] \rho^{(1)}_{24} &=&
-iB\Omega_2\rho^{(0)}_{34} - iA\Omega_1 \rho^{(1)}_{14}
+ i\Omega_1^{*}\rho^{(1)}_{23} \nonumber \\
\left[\Gamma/2+i(2\delta_1-\delta_2)\right]\rho^{(1)}_{14} &=&
i \Omega_1^{*} \rho^{(1)}_{13} \\
\rho^{(0)}_{12} =
\frac{iA\Omega_1^{*}\rho^{(0)}_{11}}{\Gamma/2+i\delta_1}\,; &&
\rho^{(0)}_{34} =
\frac{i\Omega_1^{*}\rho^{(0)}_{33}}{\Gamma/2+i\delta_1} \;.\nonumber
\end{eqnarray}
Here we assume that the term $\widehat{R}$ in
eq.(\ref{OBE}) is diagonal and, consequently, gives
contributions into eqs.(\ref{1order}) implicitly through the
zero-order populations $\rho^{(0)}_{ii}$ only.
From eqs.(\ref{1order}) one can get
the coupled equations for the low-frequency coherences:
\begin{eqnarray} \label{lfcoh}
\left[\frac{|A\Omega_1|^2}{\Gamma/2-i\delta_2}+
\frac{|\Omega_1|^2}{\Gamma/2+i(2\delta_1-\delta_2)}+
i (\delta_1-\delta_2)\right] \rho^{(1)}_{13}
- b A \Gamma \rho^{(1)}_{24}& = &
-\frac{AB \Omega_2 \Omega_1^{*}}{\Gamma/2-i\delta_2}\rho^{(0)}_{33}
-\frac{AB \Omega_2 \Omega_1^{*}}{\Gamma/2+i\delta_1}\rho^{(0)}_{11}
\nonumber \\
\left[\Gamma+i(\delta_1-\delta_2)\right] \rho^{(1)}_{24}
-\left\{\frac{A|\Omega_1|^2}{\Gamma/2-i\delta_2} +
\frac{A|\Omega_1|^2}{\Gamma/2+i(2\delta_1-\delta_2)}\right\}
\rho^{(1)}_{13} &=&
\left\{\frac{B \Omega_2
\Omega_1^{*}}{\Gamma/2-i\delta_2} +
\frac{B \Omega_2 \Omega_1^{*}}{\Gamma/2+i\delta_1}\right\}
\rho^{(0)}_{33} \,.
\end{eqnarray}
The right-hand sides of eqs.(\ref{lfcoh}) are the field interference
terms describing the creation of $\rho^{(1)}_{13}$ and
$\rho^{(1)}_{24}$. In the square bracket of the first line the field
broadening and optical shifts of the ground-level states are present.
Equations (\ref{lfcoh}) are not independent due to the second terms
on l.h.s. of both lines, which correspond to the spontaneous and
induced coherence transfer between levels. In the low-saturation
limit the excited-level coherence $\rho^{(1)}_{24}$ enters into the
equation for the ground-level coherence $\rho^{(1)}_{13}$ through the
term describing the spontaneous coherence transfer. As it can be
seen, this process leads to changes in the position, width and
amplitude of nonlinear resonances connected with the low-frequency
coherence. Since in the present paper we are interesting in the
subnatural-width resonance,
the anzatz $|\delta_1-\delta_2| \ll \Gamma$ is relevant.
Using eqs.(\ref{lfcoh}), we can eliminate the low-frequency
coherence in eq.(\ref{linresp}) and arrive at the final result for
the linear response:
\begin{eqnarray} \label{final}
\rho^{(1)}_{23}&=& \frac{-iB\Omega_2}{\Gamma/2-i\delta_2}
\left\{
\rho^{(0)}_{33} +
\frac{(b-1)\rho^{(0)}_{33}|A\Omega_1|^2}
{|A\Omega_1|^2(1-b)+
|\Omega_1|^2(1-bA^2)\frac{\Gamma/2-i\delta_2}{\Gamma/2+i\delta_1}
+ i(\delta_1-\delta_2)(\Gamma/2-i\delta_2)}+ \right.
\nonumber\\
&+& \left.
\frac{(b\rho^{(0)}_{33}-\rho^{(0)}_{11})|A\Omega_1|^2}
{|A\Omega_1|^2(1-b)\frac{\Gamma/2+i\delta_1}{\Gamma/2-i\delta_2}+
|\Omega_1|^2(1-bA^2)
+ i(\delta_1-\delta_2)(\Gamma/2+i\delta_1)}\right\} \, .
\end{eqnarray}
\subsection{Homogeneous broadening}
Consider first the case of ${\bf v}=0$.
The steady-state zero-order populations $\rho^{(0)}_{11}$ and
$\rho^{(0)}_{33}$ are governed by the equilibrium between the
excitation and relaxation processes in the absence of the probe field.
Obviously, these values can not contain structures with the width
less than $\Gamma$. Then, only the last two terms on the r.h.s. of
(\ref{final}) are responsible for the subnatural-width resonance on
the frequency offset $\delta_1-\delta_2$. If $\delta_1=0$, the sign
of the absorption resonance is determined by the sign of the
expression $(2b-1)\rho^{(0)}_{33}-\rho^{(0)}_{11}$ and, consequently,
depends on both the branching ratio constant and zero-order
populations. For example, in the absence of the spontaneous transfer
of the low-frequency coherence ($b=0$) the resonance is always
negative, that corresponds to EIT. In the opposite case of the closed
transition ($b=1$) the resonance is positive if
$\rho^{(0)}_{33} > \rho^{(0)}_{11}$, i.e. we have EIA (see in fig.2).
The position and the width of the EIA-resonance is determined by
the real and imagine parts of
the linear combination of the complex ground-level optical shifts:
$$
(1-b)\Delta\varepsilon_1-(1-bA^2)\Delta\varepsilon_3^{*} =
(1-b)\frac{|A\Omega_1|^2}{\delta_1-i\Gamma/2 }-
(1-bA^2)\frac{|\Omega_1|^2}{\delta_1+i\Gamma/2} \,.
$$
It is remarkable that the
coefficients of this combination depend on the branching ratio $b$.
\subsection{Doppler broadening}
In the case of atomic gas the EIA-resonance is a sum of structures
with different amlitudes, positions, and width. Here we consider the
result of such velocity averaging in one specific case. Let the
copropagating control and probe fields have the approximately equal
Doppler shifts
$({\bf k}_1{\bf v})\approx({\bf k}_2{\bf v})\approx k v_z$.
Besides, we assume that the transition is closed $b=1$, and
the zero-order populations
$\rho^{(0)}_{11}=0$ and
$\rho^{(0)}_{33}=f_M({\bf v})$ is the Maxwell distribution. The
averaged optical coherence is expressed through the error function:
\begin{eqnarray} \label{averaging}
\langle \rho_{23} \rangle_v &=& -iB\Omega_2\left\{
V(\delta_2)+\frac{|A\Omega_1|^2}{|A\Omega_1|^2+i\Gamma
(\delta_1-\delta_2)-(\delta_1-\delta_2)^2}
\left[V(\delta_2)+V(-\delta_1+\frac{|B\Omega_1|^2}{\delta_1-\delta_2})
\right] \right\} \nonumber \\
V(x) &=& \frac{\sqrt{\pi}}{k \overline{v}}
\exp\left[\left(\frac{\Gamma/2-ix}{k \overline{v}}\right)^2\right]
\left(1+\mbox{erf}\left(\frac{ix-\Gamma/2}{k
\overline{v}}\right)\right) \,,
\end{eqnarray}
where $\delta_q =\omega_q-\omega_0$,
$\overline{v} = \sqrt{2 k_B T/M}$, and $V(x)$ is
the well-known Voight contour. In the case of large Doppler broadening
$k\overline{v} \gg \Gamma$ the probe field absorbtion
$\sim \mbox{Re}\{i\Omega^{*}_2\langle \rho_{23} \rangle_v\}$ as a
function of the frequency offset $\delta_1-\delta_2$ contains two
structures situated at $\delta_1-\delta_2=0$ with different width and
signs. One of them described by the Lorentzian
$|A\Omega_1|^2/(|A\Omega_1|^2+i\Gamma (\delta_1-\delta_2))$
has the width $|A\Omega_1|^2/\Gamma$ and gives the rising of
absorption. The last Voight function in
eq.(\ref{averaging}) $V(-\delta_1+|B\Omega_1|^2/(\delta_1-\delta_2))$
describes a very narrow dip in the absorption spectrum (see. in
fig.3.a). This structure with the width $|B\Omega_1|^2/(k
\overline{v})$ is the result of averaging. For an atom with given
velocity ${\bf v}$ the effective detuning is $\delta_1 - k v_z$ due
to the Doppler shift. As is seen from eq.(\ref{final}) and fig.2.b
the subnatural resonance is optically shifted with respect to the
point $\delta_1-\delta_2=0$. The sum of such shifted resonanes with
amplitudes and width depentent of $v_z$ gives a dip. If
$\delta_1 \neq 0$, the absorption as a function of
$\delta_1-\delta_2$ becomes asymmetric (see in fig.3.b).
\section{Conclusion}
To conclude we note the four-state $N$-type interaction scheme can be
easily organized in real atomic systems. For example, let us consider
a closed $F_g=F \rightarrow F_e=F+1$ transition of the $D_2$ line of
an alkali atom interacting with the $\sigma_{+}$ polarized control
field. In the steady state all atoms are completely pumped into the
stretched states $|F_g,\,m_g=F\rangle$ and $|F_e,\,m_e=F+1\rangle$.
If the probe field has $\sigma_{-}$ polarization, then in the first
order on the probe field amplitude $\Omega_2$ we have $N$-atom with
the states $|1\rangle=|F_g,\,m_g=F-2\rangle$,
$|2\rangle=|F_e,\,m_e=F-1\rangle$, $|3\rangle=|F_g,m_g=F\rangle$,
and $|4\rangle=|F_e,\,m_e=F+1\rangle$.
\acknowledgments
We are grateful to authors of refs.\cite{EIA,EIA2} for
stimulating discussions.
|
\section{Introduction and results}
We use standard notation and notions from Banach space theory, as presented
e.\,g. in \cite{djt}, \cite{lt} and \cite{tj}. If $E$ is a Banach space,
then $B_E$ is its (closed) unit ball and $E'$ its dual; we consider
complex Banach spaces only. As usual ${\cal L}(E,F)$ denotes the Banach space
of all (bounded and linear) operators from $E$ into $F$ endowed with the
operator norm.
\par
For an infinite orthonormal system $B \subset L_2(\mu)$ (over some
probability space $(\Omega,\mu)$) an operator $T:E \rightarrow F$ between
Banach spaces $E$ and $F$ is said to be $B$-summing if there exists a
constant $c>0$ such that for all finite sequences
$b_1, \ldots, b_n$ in $B$ and $x_1, \ldots,x_n$ in $E$
\begin{equation}
\label{bpsumming}
\left( \int_\Omega \norm{\sum_{i=1}^n b_i \cdot Tx_i}^2 d\mu \right)^{1/2}
\le c \cdot \sup_{x' \in B_{E'}} \left ( \sum_{i=1}^n |\langle
x',x_i \rangle|^2 \right )^{1/2};
\end{equation}
we write $\pi_B(T)$ for the smallest constant $c$ satisfying
\eqref{bpsumming}. In this way we obtain the injective and maximal Banach operator ideal
$(\Pi_B, \pi_B)$, which became of interest recently in the theses of
Baur~\cite{baurdiss} and Seigner~\cite{seigner}.
For a
sequence of independent standard Gaussian random variables
the associated Banach operator ideal $\Pi_\gamma$ of all Gaussian-summing
operators was introduced by Linde and Pietsch \cite{linde}, and for
operators acting on finite-dimensional Hilbert spaces
$\pi_\gamma$ is also known as the $\ell$-norm, which turned out to be
important for the study of the geometry of Banach spaces (see e.\,g.
\cite{tj}).
\par
For an infinite subset $\Lambda$ of the character group $\Gamma$ of
some compact abelian group $G$ (which can be viewed as an orthonormal
system in $L_2(G,m_G)$, where $m_G$ denotes the normalized Haar measure
on $G$) Baur in \cite[9.5]{baurdiss} (see also \cite[4.2]{baur99}) gave
the following characterization:
\begin{center}
\em
$\Lambda$ is a Sidon set if and only if $\Pi_\Lambda=\Pi_\gamma$.
\end{center}
Recall that a subset $ \Lambda \subset
\Gamma$ is said to be a {\em Sidon set}
if there exists $\theta>0$ such that for all $Q=\sum_{\gamma \in \Lambda}
\alpha_\gamma \cdot \gamma \in \text{{\rm span}}(\Lambda)$ we have
$
\sum_{\gamma \in \Lambda} |\alpha_\gamma| \le \theta \cdot
\norm{Q}_{L_\infty(G,m_G)},
$
and for $2 < p < \infty$ it is called a {\em $\Lambda(p)$-set} if
there exists a constant $c>0$ such that for all
$\lambda \in \text{{\rm span}} (\Lambda)$ we have $\norm{\lambda}_{L_p(G,m_G)} \le
\norm{\lambda}_{L_2(G,m_G)}$; the infimum over all such constants $c$
is denoted by $K_p(\Lambda)$. Pisier in \cite{pisier78} showed that
$\Lambda$ is a Sidon set if and only if
\begin{center}
\em
$\Lambda$ is a $\Lambda(p)$-set
with $K_p(\Lambda) \le \kappa \sqrt{p}$ for all $2<p<\infty$ and some
$\kappa>0$.
\end{center}
As a natural counterpart of Baur's result we prove the
following characterization of sets which are $\Lambda(p)$-sets for all
$2<p<\infty$, with no control of $K_p(\Lambda)$ as in Pisier's
characterization of Sidon sets above:
\begin{theorem}
\label{khinchar}
For every infinite subset $\Lambda \subset \Gamma$ the following are
equivalent:
\begin{enumerate}[(a)]
\item
$\Lambda$ is a $\Lambda(p)$-set for all $2 < p < \infty$.
\item
$ \lambda(\Pi_\Lambda,u,v) = \lambda(\Pi_\gamma,u,v)$ for all $1 \le u,v \le
\infty$.
\end{enumerate}
\end{theorem}
Here, the limit order $\lambda({\mathcal{A}},u,v)$ of a Banach operator ideal
$({\mathcal{A}},A)$ for $1 \le u,v \le \infty$ is defined as usual
(see e.\,g. \cite[14.4]{pietsch}):
$$
\lambda({\mathcal{A}},u,v):=
\inf \{ \lambda>0 \,| \, \exists \, \rho >0 \, \forall \, n \in \Bbb{N}:
A(\text{{\rm id}}: \ell_u^n \hookrightarrow \ell_v^n) \le \rho \cdot n^\lambda \}.
$$
Note that there exist sets which are $\Lambda(p)$-sets for all $2<p<\infty$
but fail to be Sidon sets (see e.\,g. \cite[5.14]{lopez}). Our proof is
mainly based on complex interpolation techniques, in particular on
formulas for the complex interpolation of spaces of operators due to
Pisier~\cite{pisier90} and Kouba~\cite{kouba} (see also \cite{dm99}).
These techniques also yield asymptotic estimates of the Gaussian-summing
norm of identities between finite-dimensional Schatten classes:
\begin{theorem}
\label{gammaschatten}
For $1 \le u,v \le \infty$
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \asymp
\begin{cases}
n^{1/2+1/v} &\text{ if } 2 \le u \le \infty, \\
n^{1/2+ \max(0,1/2+1/v-1/u)} &\text{ if } 1 \le u \le 2.
\end{cases}
$$
In particular,
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \asymp n^{1/2 + \lambda(\Pi_\gamma,u,v)}.
$$
\end{theorem}
Here, $\mathcal{S}_u^n$ denotes the space of all linear operators $T:\ell_2^n
\rightarrow \ell_2^n$ endowed with the norm $\norm{T}_{\mathcal{S}_u^n}:=
\norm{(s_k(T))_{k=1}^n}_{\ell_u^n}$, where $(s_k(T))_{k=1}^n$ is the sequence
of singular numbers of $T$. Besides the interpolation argument,
our proof uses the close relationship between the
Gaussian-summing norm of the identity operator $\text{{\rm id}}_E$
and the Dvoretzky dimension of a finite-dimensional Banach space $E$
due to Pisier.
\section{Complex interpolation of $\boldsymbol{B}$-summing operators}
Our main tool will be an ``interpolation theorem'' for the
$B$-summing norm of a fixed operator acting between
finite-dimensional complex interpolation
spaces; a similar approach for the $(s,2)$-summing norm was used in
\cite{dm98} to study the well-known ``Bennett--Carl inequalities''
within the context of interpolation theory.
\par
For all information on complex interpolation we refer to \cite{BL}.
Given an interpolation couple $[E_0,E_1]$ of complex Banach spaces and
$0<\theta<1$, the associated complex interpolation space is denoted by
$[E_0,E_1]_\theta$. If $E_0$ and $E_1$ are finite-dimensional Banach spaces
with the same dimensions, we speak of a finite-dimensional interpolation
couple, and in this case we define
$$
d_\theta[E_0,E_1]:=\sup_m \norm{{\cal L}(\ell_2^m,[E_0,E_1]_\theta) \hookrightarrow
[{\cal L}(\ell_2^m,E_0),{\cal L}(\ell_2^m,E_1)]_\theta}.
$$
Pisier~\cite{pisier90} and Kouba~\cite{kouba} derived upper estimates
for $d_\theta[E_0,E_1]$ for particular situations (see also \cite{dm99});
we will use the fact that for $1 \le p_0,p_1 \le 2$
\begin{equation}
\label{dtheta1}
d_\theta[\ell_{p_0}^n,\ell_{p_1}^n] \le \sqrt{2};
\end{equation}
in particular, $\sup_n d_\theta[\ell_1^n,\ell_2^n] < \infty$.
Junge \cite[4.2.6]{junge} gave an analogue for Schatten classes:
\begin{equation}
\label{dtheta2}
\sup_n d_\theta[\mathcal{S}_1^n,\mathcal{S}_2^n] <\infty.
\end{equation}
These ``uniform'' estimates will be crucial for the applications of the
following result:
\begin{proposition}
\label{interpolB}
Let $0 < \theta <1$. Then for two
finite-dimensional interpolation couples $[E_0,E_1]$ and $[F_0,F_1]$, each
$T \in {\cal L}([E_0,E_1]_\theta,[F_0,F_1]_\theta)$ and each orthonormal system
$B \subset L_2(\mu)$
$$
\pi_B(T:[E_0,E_1]_\theta \rightarrow [F_0,F_1]_\theta)
\le d_\theta[E_0,E_1] \cdot \pi_B(T:E_0 \rightarrow F_0)^{1-\theta}
\cdot \pi_B(T:E_1 \rightarrow F_1)^\theta.
$$
\end{proposition}
\proof For the moment set $E_\theta:=[E_0,E_1]_\theta$, $F_\theta:=
[F_0,F_1]_\theta$, and
consider for $\eta=0,\theta,1$ and ${\mathcal{F}}= \{ b_1, \ldots,
b_m \} \subset B$ the mapping
$$
\begin{array}{lccc}
\Phi_\eta^{m,{\mathcal{F}}} :
& {\cal L}(\ell_2^m,E_\eta) & \rightarrow & L_2(\mu,F_\eta)
\\ & S & \mapsto & \sum_{i =1}^m b_i \cdot TSe_i.
\end{array}
$$
Since for each $S= \sum_{j=1}^m e_j \otimes x_j \in {\cal L}(\ell_2^m, E_\eta)$
$$
\norm{S}= \sup_{x' \in B_{E_\eta'}} \left ( \sum_{j=1}^m |\langle x',x_j
\rangle|^2 \right )^{1/2},
$$
we obviously get that
$$
\pi_B(T: E_\eta \rightarrow F_\eta) = \sup \{\norm{\Phi_\eta^{m,
{\mathcal{F}}}} \, | \, m \in \Bbb{N}, {\mathcal{F}}
\subset B \text{ with } |{\mathcal{F}}| = m \}.
$$
For the interpolated mapping
$$
[\Phi_0^{m,{\mathcal{F}}}, \Phi_1^{m,{\mathcal{F}}}]_\theta :
[{\cal L}(\ell_2^m,E_0), {\cal L}(\ell_2^m,E_1)]_\theta \rightarrow
[L_2(\mu,F_0), L_2(\mu,F_1)]_\theta
$$
by the usual interpolation theorem
$$
\norm{[\Phi_0^{m,{\mathcal{F}}}, \Phi_1^{m,{\mathcal{F}}}]_\theta} \le
\norm{\Phi_0^{m,{\mathcal{F}}}}^{1-\theta} \cdot
\norm{\Phi_0^{m,{\mathcal{F}}}}^\theta.
$$
Since $[L_2(\mu,F_0),L_2(\mu,F_1)]_\theta = L_2(\mu,[F_0,F_1]_\theta)$
(isometrically, see \cite[5.1.2]{BL}) we obtain
$$
\norm{\Phi_\theta^{m,{\mathcal{F}}}} \le \norm{{\cal L}(\ell_2^m,
[E_0,E_1]_\theta)
\hookrightarrow [{\cal L}(\ell_2^m,E_0),{\cal L}(\ell_2^m,E_1)]_\theta } \cdot
\norm{[\Phi_0^{m,{\mathcal{F}}}, \Phi_1^{m,{\mathcal{F}}}]_\theta} .
$$
Consequently
\begin{align*}
\pi_B(T & : [E_0,E_1]_\theta \rightarrow [F_0,F_1]_\theta) \\
&= \sup \{\norm{\Phi_\theta^{m, {\mathcal{F}}}} \, | \, m \in \Bbb{N}, {\mathcal{F}}
\subset B \text{ with } |{\mathcal{F}}| = m \} \\
& \le \sup
\{d_\theta[E_0,E_1] \cdot \norm{\Phi_0^{m,
{\mathcal{F}}}}^{1-\theta} \cdot \norm{\Phi_0^{m, {\mathcal{F}}}}^\theta
\, | \, m \in \Bbb{N}, {\mathcal{F}}
\subset B \text{ with } |{\mathcal{F}}| = m \} \\
& \le d_\theta[E_0,E_1] \cdot
\pi_B(T:E_0 \rightarrow F_0)^{1-\theta} \cdot
\pi_B(T: E_1 \rightarrow F_1)^\theta,
\end{align*}
the desired result.
\qed
\par
Since for $0<\theta<1$ and $1 \le p_0,p_1,p_\theta \le \infty$ such that
$1/{p_\theta} = (1-\theta)/{p_0} + \theta/{p_1}$ it holds
$[\ell_{p_0}^n,\ell_{p_1}^n]_\theta= \ell_{p_\theta}^n$ isometrically
(see \cite[5.1.1]{BL}), we obtain together with \eqref{dtheta1} the
following corollary:
\begin{corollary}
\label{interpollimit}
For $0<\theta<1$ let $1 \le u_0,u_1,u_\theta \le 2 $ and $1 \le v_0, v_1,
v_\theta \le \infty$
such that $1/{u_\theta} = (1-\theta)/{u_0} + \theta/{u_1}$ and
$1/{v_\theta} = (1-\theta)/{v_0} + \theta/{v_1}$. Then
$$
\lambda(\Pi_B,u_\theta,v_\theta) \le (1-\theta) \cdot \lambda(\Pi_B,
u_0,v_0) + \theta \cdot \lambda(\Pi_B,u_1,v_1).
$$
\end{corollary}
\section{The proof of Theorem~\ref{khinchar}}
As a generalization of the notion of $\Lambda(p)$-sets, an orthonormal
system $B \subset L_2(\mu)$ is said to be a {\em $\Lambda(p)$-system}
if $B \subset L_p(\mu)$ and there exists a constant $c>0$ such that for
all $f \in \text{{\rm span}} B$ we have $\norm{f}_{L_p(\mu)} \le c \cdot
\norm{f}_{L_2(\mu)}$; the infimum over all such constants $c$ is denoted
by $K_p(B)$. Now one direction of the equivalence in Theorem~\ref{khinchar}
can be formulated for general orthonormal systems:
\begin{proposition}
\label{khintheo}
Let $B \subset L_2(\mu)$ be a
$\Lambda(p)$-system for all $2 < p < \infty$. Then for all
\mbox{$1 \le u,v \le \infty$}
$$
\lambda(\Pi_B,u,v) = \lambda(\Pi_\gamma,u,v) =
\begin{cases}
1/v & \text{ if } 2 \le u \le \infty, \\
\max(0,1/2 +1/v-1/u) & \text{ if } 1 \le u \le 2.
\end{cases}
$$
\end{proposition}
\proof
Although the limit order of $\Pi_\gamma$ is already known by the
results of \cite{linde}, the following proof may also be used to compute
it independently (at least the upper estimates; the lower ones are
somehow simple), but for simplicity we fall back upon this knowledge.
\par
Since $\Pi_2 \subset \Pi_B \subset \Pi_\gamma$ (see \cite[4.15]{pw};
$\Pi_2$ denotes the Banach operator ideal of all $2$-summing operators),
we only have to show that $\lambda(\Pi_B,u,v) \le \lambda(\Pi_\gamma,u,v)$,
and moreover, we conclude that
$\lambda(\Pi_B,u,v)\le \lambda(\Pi_2,u,v) =\lambda(\Pi_\gamma,u,v)$ for all
\mbox{$1 \le u \le \infty$} and $1 \le v \le 2$.
For $2 < v < \infty$ and $2 \le u \le \infty$ it can be easily seen
that
$$
\Pi_B(\ell_u^m \hookrightarrow \ell_v^m) \le K_v(B) \cdot m^{1/v}
$$
(just copy the proof of \cite[3.11.11]{pw}), hence---together with the continuity
of the limit order, see \cite[14.4.8]{pietsch}---we obtain
$$
\lambda(\Pi_B,u,v) \le
1/v = \lambda(\Pi_\gamma,u,v)
$$
for all $2\le u,v\le \infty$. Now the case
$1 \le u \le 2 \le v \le \infty$ follows from Corollary~\ref{interpollimit}:
For $1 \le u \le 2$ choose $u_0:=1$, $u_1:=2$, $v_0:=2$, $v_1:=\infty$,
$\theta:=2/{u'}$ and
$v_u$ such that $1/{v_u} =1/u-1/2$. Then
$$
\lambda(\Pi_B,u,v_u) \le (1-\theta) \cdot \lambda(\Pi_B,1,2)
+ \theta \cdot \lambda(\Pi_B,2,\infty) = 0.
$$
For arbitrary $1 \le u \le 2 \le v \le \infty$
factorize through $\ell_{v_u}^m$:
$$
\pi_B(\ell_u^m \hookrightarrow \ell_v^m) \le m^{\max(0,1/v +1/2 -1/u)} \cdot
\pi_B(\ell_u^m \hookrightarrow \ell_{v_u}^m),
$$
hence $\lambda(\Pi_B,u,v) \le \max(0,1/v +1/2-1/u)= \lambda(\Pi_\gamma,
u,v)$.
\qed
\\[10pt]
The reverse implication in Theorem~\ref{khinchar}
follows from \cite{baurdiss}: (b) trivially implies $\lambda(\Pi_B, \infty,
\infty)=0$, and the comments after \cite[7.12]{baurdiss} then tell us that
$\Pi_p \subset \Pi_\Lambda$ for all $2<p<\infty$. This in turn gives by
\cite[9.6]{baurdiss} (see also \cite[5.1]{baur99})
that $\Lambda$ is a $\Lambda(p)$-set for all $2<p<\infty$.
\par
Note that the last argument requires the setting of characters on a
compact abelian group; Baur has recently informed us that her results are
also valid for the non-abelian case, and therefore our
Theorem~\ref{khinchar} as well.
\section{The proof of Theorem~\ref{gammaschatten}}
\proof
For $1 \le v \le 2$ by \cite{tom} $\mathcal{S}_v$ is of cotype~$2$, hence
$$
n^{1/v + \min(1/2,1-1/u)} = \pi_2(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n)
\ge \pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n)
\ge \cotype{2}(\mathcal{S}_v)^{-1} \cdot \pi_2(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n)
$$
(see e.\,g. \cite[Corollary 3]{dm98}) for $1 \le u \le \infty$ and
$1 \le v \le 2$; here $\cotype{2}(\mathcal{S}_v)$ denotes the Gaussian
cotype 2 constant of $\mathcal{S}_v$. We are left with
the case $2 \le v \le \infty$; first let $u=v=\infty$.
Then by \cite[4.15.18]{pw}
(a result of Pisier, see also \cite[4.4]{pisier89} and \cite{pisierLN})
and \cite[3.3]{figiel} for each $\varepsilon>0$
$$
\pi_\gamma(\mathcal{S}_\infty^n \hookrightarrow \mathcal{S}_\infty^n) \asymp \sqrt{D(\mathcal{S}_\infty^n,
\varepsilon)}
\asymp n^{1/2},
$$
where $D(X,\varepsilon)$ denotes the Dvoretzky
dimension of a Banach space $X$, i.\,e. the largest $m$ such that there
exists an $m$-dimensional subspace $X_m$ of $X$ with Banach--Mazur
distance $d(X_m,\ell_2^m) \le
1+\varepsilon$ (see \cite[4.15.15]{pw}). Now the general case
$2 \le u,v \le \infty$ follows by factorization:
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \le
n^{1/v} \cdot \pi_\gamma(\mathcal{S}_\infty^n \hookrightarrow \mathcal{S}_\infty^n) \asymp n^{1/v+1/2},
$$
and conversely
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \ge n^{1/v-1/2} \cdot
\pi_\gamma(\mathcal{S}_2^n \hookrightarrow \mathcal{S}_2^n) = n^{1/v+1/2}.
$$
The case $1 \le u \le 2 \le v \le \infty$ is done by interpolation:
We have (recall that \mbox{$\pi_2(\mathcal{S}_1^n \hookrightarrow \mathcal{S}_2^n)=n^{1/2}$})
$$
\pi_\gamma(\mathcal{S}_1^n \hookrightarrow \mathcal{S}_2^n) \asymp \pi_\gamma(\mathcal{S}_2^n \hookrightarrow \mathcal{S}_\infty^n)
\asymp n^{1/2},
$$
hence for $1 < u <2 < v_u < \infty$ and $0 < \theta <1$ such that
$1/{v_u} =1/u -1/2$ and $\theta=2/{u'}$
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_{v_u}^n) \le d_\theta[\mathcal{S}_1^n,\mathcal{S}_2^n] \cdot
\pi_\gamma(\mathcal{S}_1^n \hookrightarrow \mathcal{S}_2^n)^{1-\theta} \cdot \pi_\gamma(\mathcal{S}_2^n \hookrightarrow
\mathcal{S}_\infty^n)^\theta \asymp n^{1/2};
$$
recall that $\sup_n d_\theta[\mathcal{S}_1^n,\mathcal{S}_2^n] < \infty$ by \eqref{dtheta2},
and that $[\mathcal{S}_1^n,\mathcal{S}_2^n]_\theta = \mathcal{S}_u^n$ and
$[\mathcal{S}_2^n, \mathcal{S}_\infty^n]_\theta = \mathcal{S}_{v_u}^n$ hold isometrically (this
can be deduced from e.\,g. \cite[Satz 8]{pt} and the complex reiteration
theorem \cite[4.6.1]{BL}).
The remaining estimates now follow easily from
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \ge \pi_\gamma(\ell_2^n \hookrightarrow \ell_2^n)=
n^{1/2},
$$
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \ge \pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_2^n)
\cdot \norm{\mathcal{S}_v^n \hookrightarrow \mathcal{S}_2^n}
$$
and
$$
\pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_v^n) \le \pi_\gamma(\mathcal{S}_u^n \hookrightarrow \mathcal{S}_{v_u}^n)
\cdot \norm{\mathcal{S}_{v_u}^n \hookrightarrow \mathcal{S}_v^n}.
$$
\qed
\par
The contents of this article are part of the author's thesis written
at the Carl von Ossietzky University of Oldenburg under the supervision
of Prof. Dr. Andreas Defant.
\providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}
|
\section{Introduction}
Big Bang Nucleosynthesis (BBN) remains one of the most successful
probes of early times in the hot big bang cosmology. Standard BBN
(SBBN) assumes the Standard Model proposition of only three massless
light neutrinos, leaving the primordial element abundances
characterized by only one parameter, the baryon-to-photon ratio
$\eta$. Now, the evidence for neutrino mass from the Super-Kamiokande
atmospheric neutrino observations is overwhelming \cite{superk}. In
addition, the solar neutrino deficit and the LSND experiment
\cite{solar} are hinting at the presence of a fourth light neutrino
that would necessarily be ``sterile'' due to the $Z$ decay width
\cite{zdecay}. The role of neutrino masses and their mixing bring
another aspect to the physical evolution of the Early Universe, when
neutrinos played a large role. The effects of neutrinos on big bang
nucleosynthesis can be in part parameterized by the effective number
of neutrinos $N_{\nu}$ ({\it i.e.} the expansion rate), and in the weak
physics which determine the neutron-to-proton ratio. In many works
$N_{\nu}$ is regarded as the sole determinant of neutrino effects in
BBN. However, we find that the effect of neutrino mixing on BBN is
not well described by this one parameter. Characterizing the effects
of neutrino mixing on BBN by $N_{\nu}$ can become overly complicated,
since the parameter has been used to describe several distinct
physical effects, including lepton number asymmetry and change in
energy density. Another feature we find is that matter-enhanced
mixing of neutrinos in the Early Universe can alter $\nu_e$ spectra to
be non-thermal, which uniquely affects the nucleon weak rates by
lifting Fermi blocking of neutron decay by neutrinos and suppression
of neutrino capture on neutrons.
The observation of the abundances of primordial elements has greatly
increased in precision within the past few years
\cite{schburl,burtyt98a,burtyt98b,oss,bonmol}. In SBBN, each observed
primordial abundance corresponds to a value of the baryon-to-photon
ratio $\eta \approx 2.79\times 10^8\,\Omega_b^{-1}h^{-2}$, where
$\Omega_b$ is the fractional contribution of baryon rest mass to the
closure density and $h$ is the Hubble parameter in units of $100\rm
km\,s^{-1}\,Mpc^{-1}$. The theory is apparently successful since the
inferred primordial abundances correspond within errors to a single
value of $\eta$. SBBN took a new turn in 1996 when it was found that
the deuterium abundance at high redshifts may be significantly lower
than that inferred through the cosmic helium abundance \cite{dtxfsb}.
The advent of high-resolution spectroscopy of quasars allowed the
measurement of deuterium abundance in high-redshift clouds between us
and distant quasars. The ``early days'' of these high-redshift
spectral deuterium measurements were filled with controversy due to a
disparity in the inferred deuterium to hydrogen (D/H) ratio between
two observational groups working with different quasars. More
recently, the number of quasars with low D/H has grown. The deuterium
abundance has converged to $\sim$10\% precision: $({\rm D/H})_p \simeq (3.4
\pm 0.25) \times 10^{-5}$. The $\eta$ value inferred from this
observation is $\eta_D \simeq (5.1 \pm 0.5) \times 10^{-10}$. The
prediction for the corresponding primordial $^4$He mass fraction is
$Y_p(D) \simeq 0.246 \pm 0.0014$.
Based on stellar absorbtion features in metal-poor HII regions, the
observed helium abundance has been argued by Olive, Steigman, and
Skillman (OSS) \cite{oss} to be $Y_p(OSS) \simeq 0.234 \pm 0.002 ({\rm
stat}) \pm 0.005 ({\rm sys})$. The corresponding baryon-to-photon
ratio is $\eta_{OSS} \simeq (2.1 \pm 0.6) \times 10^{-10}$. The
discrepancy between the central values of $Y_p(D)$ predicted by the
Burles and Tytler deuterium observations and $Y_p(OSS)$ has led to
much speculation on non-standard BBN scenarios which could reconcile
the two measurements by decreasing the $Y_p(D)$ predicted by BBN.
However, the systematic and statistical uncertainty in the $Y_p(D)$
and $Y_p(OSS)$ puts them well within their $2\sigma$ errors.
Furthermore, Izotov and Thuan \cite{izotov} exclude a few potentially
tainted HII regions used in OSS from their analysis and arrive at a
significantly higher $Y_p \simeq 0.244 \pm 0.002 ({\rm stat})$. The
systematic uncertainty claimed in observed $Y_p$ may also be
underestimated \cite{matfulboyd}.
It has been suggested that this dubious $Y_p$ and D/H discrepancy with
standard BBN could be hinting at the presence of sterile neutrino
mixing in the Early Universe \cite{footvolk}. However, we find that
the overall effect of active-sterile mixing in the Early Universe
would only increase $Y_p$ and thus increase the possible disparity.
Active-sterile mixing can produce an electron lepton number asymmetry
$L(\nu_e)$ before and during BBN \cite{footvolk,xdchaos}. If the sign
of $L(\nu_e)$ is the same throughout the Universe, the helium
abundance would change to be more favorable ($\delta Y_p < 0$) or less
favorable ($\delta Y_p > 0$) with observation. Taking into account
the effects of energy density, lepton number generation and alteration
of neutrino spectra, we find that $\delta Y_p$ can be large in the
positive direction, but limited in the negative direction \cite{lept}.
In particular, we find that matter-enhanced transformations can leave
non-thermal $\nu_e$ spectra which significantly affect the weak rates
through the alleviation of Fermi-blocking and suppression of neutrino
capture on neutrons \cite{asf}. Furthermore, causality limits the
size of the region with a certain sign of $L(\nu_e)$ to be within the
size of the horizon during these early times ($\sim 10^{10}\;\rm cm$
at weak freeze-out) \cite{xdcaus}. The sign of the asymmetry will
vary between regions, and the net effect of transformation on $Y_p$ will
then be the average of both the $L(\nu_e) > 0$ and $L(\nu_e) < 0$ BBN
yields, which leaves $\delta Y_p > 0$.
\section{Matter-Enhanced Mixing of Neutrinos and Neutrino Spectra\\
in the Early Universe}
Two possible schemes exist for producing a nonzero $L(\nu_e)$
involving matter-enhanced (MSW) transformation of neutrinos with
sterile neutrinos. These two schemes differ greatly in the energy
distribution of the asymmetry in the spectra of the
$\nu_e/\bar{\nu}_e$ the asymmetry. The resonant transformation of one
neutrino flavor to another depends on the energy of the neutrino. In
the case of BBN, the active neutrinos all start off as thermal, and
sterile neutrinos are not present. As the Universe expands, decreases
in density and cools, the position of the MSW resonance moves in
energy space from lower to higher neutrino energies. The energy of
the resonance is \cite{lept}
\begin{equation}
\left({E\over T}\right)_{\rm res} \approx {| \delta m^2|/{\rm eV^2}
\over 16 (T/{\rm MeV})^4 L(\nu_\alpha)}
\end{equation}
where $(E/T)_{\rm res}$ is the neutrino energy normalized by the
ambient temperature $T$. The asymmetry $L(\nu_\alpha)$ is generated as
the resonance energy moves up the neutrino spectrum. As the Universe
cools, the resonance energy increases. This produces a distortion of
the neutrino or anti-neutrino spectrum. There are two plausible cases
for producing an asymmetry $L(\nu_e)$ that would affect the production
of primordial Helium through the weak rates:
\begin{itemize}
\item {\bf Direct Two Neutrino Mixing:} An asymmetry in $L(\nu_e)$ is
created directly through a $\nu_e \rightarrow \nu_s$ or $\bar{\nu}_e
\rightarrow \bar{\nu}_s$ resonance. In this case, the resonance starts at
low temperatures for low neutrino energies, when neutrino-scattering
processes are very slow at re-thermalization. This case leaves the
$\nu_e$ or $\bar{\nu}_e$ spectrum distorted, with the spectral
distortion from a thermal Fermi-Dirac form residing in the low energy
portion of the spectrum. The position of the distortion cut-off moves
through the spectrum before and during BBN. In calculating the
effects of this scenario on BBN, we included the evolving non-thermal
nature of the neutrino spectrum.
\item {\bf Indirect Three Neutrino Mixing:} An asymmetry in
$L(\nu_\tau)$ or $L(\nu_\mu)$ created by a $\nu_\tau \rightarrow \nu_s
(\nu_\mu \rightarrow \nu_s)$ resonance is later transferred into the
$L(\nu_e)$ through a $\nu_\tau \rightarrow \nu_e (\nu_\mu \rightarrow
\nu_e)$ resonance ($L(\nu_e) > 0$). This will happen for the
anti-neutrino flavors for the $L(\nu_e) < 0$ case. In this scenario,
the resonant transitions occur at higher temperatures when
neutrino-scattering is more efficient. Both the $\nu_\tau (\nu_\mu)$
and $\nu_e$ spectra re-thermalize, but since the temperature is below
weak decoupling, the asymmetry in neutrino lepton numbers remain. The
effect of this scenario on BBN can be described by suppression or
enhancement over the entire $\nu_e/\bar{\nu}_e$ spectra.
\end{itemize}
\section{Neutron-Proton Interconversion Rates and Neutrino Spectra}
BBN can be approximated as the freeze-out of nuclides from nuclear
statistical equilibrium in an ``expanding box.'' Nuclear reactions
freeze-out, or fail to convert one nuclide to another when their rate
falls below the expansion rate of the box (the Hubble expansion). As
different reactions rates freeze-out at different temperatures, the
abundances of various nuclides get altered and then become fixed. The
nuclide produced in greatest abundance (other than hydrogen) is
$^4$He. The reactions that affect the $^4$He abundance ($Y_p$) the
most are the weak nucleon interconversion reactions
\begin{equation}
\label{reactions}
n + \nu_e \leftrightarrow p + e^- \;\;\;\;\;
n + e^+ \leftrightarrow p + \bar{\nu}_e \;\;\;\;\;
n \leftrightarrow p + e^- + \bar{\nu}_e .
\end{equation}
These reactions all depend heavily on the density of the electron-type
neutrinos and their energy distributions \cite{alpher}. The two
neutrino transformation scenarios for altering the $\nu_e$ spectrum
described above affect $Y_p$ through these rates and through the
effect of increased energy density on the expansion rate.
\subsection{Direct Two-Neutrino Mixing: $\nu_s \leftrightarrow \nu_e$}
In this situation, either the populations of the $\nu_e$ or
$\bar{\nu}_e$ energy spectra are cutoff at lower energies. The
reaction rates that alter the overall $n\leftrightarrow p$ rates the
most because of the spectral cutoff are
\begin{equation}
\label{rate1}
\lambda_{n\rightarrow p e \nu} = A \int{v_e E_\nu^2 E_e^2 dp_\nu
[1+e^{-E_\nu/k T_\nu}]^{-1} [1+e^{-E_e/k T_\nu}]^{-1}}
\end {equation}
\begin{equation}
\label{rate2}
\lambda_{n \nu \rightarrow p e} = A \int{v_e E_e^2 p_\nu^2 dp_\nu
[e^{E_\nu/k T_\nu} + 1]^{-1} [1+e^{-E_e/k T_\nu}]^{-1}}
\end {equation}
(the notation in these expressions follows that in
Ref. \cite{weinberg}). The rate in Eq. \ref{rate1} is modified with
the $\bar{\nu}_e$ spectra, and the rate in Eq. \ref{rate2} is modified
with the $\nu_e$ spectra.
For a spectral cutoff for $\bar{\nu}_e$, corresponding to
$\bar{\nu}_{s}\leftrightarrow \bar{\nu}_e$ transformations, the
Fermi-blocking term $[1+e^{-E_\nu/k T_\nu}]^{-1}$ in neutron decay
(Eq. \ref{rate1}) becomes unity at low $\bar{\nu}_e$ energies. The
rate integrand is then greatly enhanced when low energy neutrinos
would otherwise block the process (see Figure \ref{intnueb}). In the
case of a spectral cutoff for $\nu_e$, corresponding to
$\nu_{s}\leftrightarrow \nu_e$ transformations, the spectral term
$p^2_\nu[e^{E_\nu/k T_\nu} + 1]^{-1}$ and thus the rate integrand go
to zero for low energy neutrinos (see Fig \ref{intnue}).
\begin{figure}[ht]
\centerline{\epsfxsize 5 truein \epsfbox{abazajian0214fig1.eps}}
\vskip 0 cm
\caption[]{
\label{intnueb}
\small
Plotted are the contours of the amplitude of the rate
integrand for $n \rightarrow p + e^- + \bar{\nu}_e$, with increasing
amplitude towards the center. (a) shows the lack of fermi blocking by
low energy electron anti-neutrinos at higher temperatures. (b) shows
the standard BBN integrand.}
\end{figure}
\begin{figure}[ht]
\centerline{\epsfxsize 5 truein \epsfbox{abazajian0214fig2.eps}}
\vskip 0 cm
\caption[]{
\label{intnue}
\small Plotted are the contours of the amplitude of the rate
integrand for $n + \nu_e \rightarrow p + e^-$, with increasing amplitude
towards the upper right: (a) shows the blocking of the integrand for lower
energies due to the suppression of low energy electron neutrinos; (b) shows
the standard BBN integrand.}
\end{figure}
The resultant change in $Y_p$ for this ``two neutrino'' case is shown
in Figure \ref{dycut}. For $\bar{\nu}_e \rightarrow \bar{\nu}_s$
transformations, $L(\nu_e) > 0$, the neutron decay rate in
Eq. \ref{rate1} is enhanced, fewer neutrons are present, and $Y_p$ is
decreased. For $\nu_e \rightarrow \nu_s$ transformations, $L(\nu_e) <
0$, the $n\,\nu_e\rightarrow p\, e$ rate in Eq. \ref{rate2} is
suppressed, more neutrons are present, and $Y_p$ is increased. The
effect of the rate in Eq. \ref{rate2} is much greater than that of the
rate in Eq. \ref{rate1} on the neutron-to-proton ratio and $Y_p$. This
causes a larger magnitude change in the neutron-to-proton ratio for the
$L(\nu_e) < 0$ case, and thus a larger magnitude change in $Y_p$ in
the positive direction than in the negative.
The sign of the asymmetry $L(\nu_e)$ is randomly determined
\cite{xdchaos}. Because of causality, different horizons will have an
asymmetry of random sign \cite{xdcaus}, fifty-percent positive and
fifty-percent negative. The effect of the asymmetry on $Y_p$ will
then be the average of these two cases, which is also plotted in
Figure \ref{dycut}.
\begin{figure}[ht]
\centerline{\epsfxsize 3.5 truein \epsfbox{abazajian0214fig3.eps}}
\vskip 0 cm
\caption[]{
\label{dycut}
\small
Plotted are $\delta Y_p$ vs. $\delta m^2$ for the direct two-neutrino
transformation scenario. The top line is for $L(\nu_e) < 0$; the bottom
is for $L(\nu_e) > 0$; the middle line is the average of both cases
enforced by causality.
}
\end{figure}
\subsection{Indirect Three-Neutrino Mixing}
In the case that an asymmetry is produced in $\nu_e/\bar{\nu}_e$
indirectly through an initial transformation between $\nu_\tau
\rightarrow \nu_s$ or $\nu_\mu \rightarrow \nu_s$, the neutrinos can
re-thermalize through self-scattering at higher temperatures. The
number densities remain asymmetric, but the asymmetry is spread
uniformly over the entire energy spectra. The effect on the weak rates
is not as drastic as the above case, but is still present. In this
case, the increase in energy density due to the thermalization of
sterile neutrinos also affects $Y_p$ through the increase in the
expansion rate. Thus, for $L(\nu_e) > 0$, the suppression of $Y_p$ is
counteracted by the increased expansion rate. The increase in energy
density always causes a positive change in $Y_p$, while the change
caused by the weak nucleon rates depends on the sign of $L(\nu_e)$,
thus the overall resulting change in $Y_p$ is greater in the positive
direction than in the negative (see Figure \ref{dyoverall}). For the
$L(\nu_e) > 0$ case, the negative change $\delta Y_p$ has a minimum at
$\delta m^2 \approx 100\,\rm eV^2$ since at this point the positive
effect of energy density begins to dominate the negative effect of the
modification of the weak rates.
\begin{figure}[ht]
\centerline{\epsfxsize 3.5 truein \epsfbox{abazajian0214fig4.eps}}
\vskip 0 cm
\caption[]{
\label{dyoverall}
\small
Plotted are $\delta Y_p$ vs. $\delta m^2$ for the indirect three-neutrino
transformation scenario. The top line is for $L(\nu_e) < 0$; the bottom
is for $L(\nu_e) > 0$; the middle line is the average of both cases
enforced by causality.
}
\end{figure}
\section{Conclusions}
The characterization of the effects of neutrino transformations on BBN
through a single parameter $N_\nu$ hides varied physical effects of
neutrino mass and mixing that are non-trivially related. These
effects include an increase in energy density and modification in the
nucleon weak rates. The modification of these rates is through either
an asymmetry that is distributed over the entire spectrum, or an
asymmetry confined to low energy $\nu_e/\bar{\nu}_e$. We find that
the distribution of the $L(\nu_e)$ in the spectra is important to the
weak rates, the neutron-to-proton ratio, and the production of $^4$He.
In addition, the change in the predicted $Y_p$ is always greater in
the positive direction than in the negative, thus when averaged over
causal horizons, $\delta Y_p$ is always positive. The changes $\delta
Y_p$ are comparable to the uncertainty in current $Y_p$ measurements,
thus resonant neutrino mixing can measurably affect the BBN
predictions. It is important to note that there may remain
non-trivial portions of parameter space in non-standard
BBN---including, but not limited to, neutrino mixing, spatial
variations in $\eta$, and massive decaying particles---that fit within
the observed uncertainties of the primordial abundances.
K.~A., X.~S. and G.~M.~F. are partially supported by NSF grant
PHY98-00980 at UCSD.
|
\section{Introduction}
The present series of papers (Paper I \cite{GE}) establishes the 1+3
covariant kinetic theory formalism of Ellis, Treciokas and Matravers
\cite{ETMa,ETMb} in a form that makes possible an investigation
of the Cosmic Microwave Background anisotropies in the non-local context
of emission of radiation at the surface of last scattering in the early
universe, and its reception here and now (the `Sachs-Wolfe effect'
and its later extensions). These papers aims to provide the link
between the exact (non-linear) theory \cite{MGE} and the linearised
threading formalism (this paper and to a lesser degree \cite{cl}),
to the linearised foliation formalism, based on Bardeens' approach
\cite{Bardeen} to cosmological perturbations \cite{HS95a}.
The Ellis, Treciokas and Matravers papers introduced
a covariant approach to kinetic theory based on a 1+3 covariant
representation of Cosmic Microwave Background anisotropies in terms of Projected Symmetric
Trace-Free (PSTF) tensors orthogonal to a preferred time-like vector
field $u^a$ \cite{T1,P}. The benefits of the approach have been
briefly summarized in Paper I \cite{GE} (see also Challinor \& Lasenby
\cite{cl,cl2}) -- we seek clarity by providing a direct formal
derivation of the standard results as well as providing the
background necessary to include the non-perturbative
corrections discussed by Maartens, Gebbie and Ellis \cite{MGE}.
In essence, we provide (a) clear definitions of the variables used,
(b) 1+3 Covariant and Gauge-Invariant (CGI) variables and equations,
(c) a sound basis from which to proceed to non-linear calculations
(as introduced in \cite{MGE}), and (d) the possibility of using any
desired coordinate and tetrad system for evaluating the variables and
solving the equations in specific cases (because the general equations
and variables used are covariant).
This approach has been used in a previous series of papers
\cite{MES,SME,MES1} to look at the local generation of anisotropies
in freely-propagating radiation caused by anisotropies and
inhomogeneities in any universe model, providing a proof that
near-spatial homogeneity in a region $U$ follows from radiation
near-isotropy in that region (an `Almost-Ehlers, Geren and Sachs'
theorem, generalizing the important paper by Ehlers, Geren,
and Sachs \cite{EGS}). By contrast, the present series looks at
non-local (integrated) anisotropy effects in the context of the
standard model of cosmology -- given the observational justification
provided by the almost-Ehlers, Geren and Sachs theorem \cite{SAG}.
Paper I \cite{GE} of the present series of papers \cite{GE}) dealt with the
CGI irreducible representation of cosmic background radiation
anisotropies by PSTF tensors, and their relation to observable
quantities (specifically, the angular correlation functions), first
within a general framework and then and specialized to almost-Friedmann-Lema\^{\i}tre
universe models \footnote{In order to be clear on the use of these
names, Robertson-Walker refers to the Robertson-Walker {\it geometry} whatever
the dynamics, while Friedmann-Lema\^{\i}tre refers to such a geometry which additionally
obeys the Friedmann-Lema\^{\i}tre {\it dynamics} implied by imposing
the Einstein Field Equations. An {\it almost-Friedmann-Lema\^{\i}tre universe} is a
universe model governed by the Einstein field equations, whose
difference from a Friedmann-Lema\^{\i}tre\ universe is at most $O(\epsilon )\,$ in
terms of a small parameter $\epsilon $\thinspace \cite {BE}.}
\cite{MES}, as well as dealing with multipole and mode expansions and
the relation to the usual formalisms in the literature
\cite{WS,GSS,HS95a,HS95b,EBW}.
In this paper (Paper II), we use the CGI approach to study cosmic
background radiation anisotropies in almost-Friedmann-Lema\^{\i}tre models in an analytic
manner, by time-like integration of the almost-Friedmann-Lema\^{\i}tre differential
relations. Our emphasis is on the canonical linearised model for
cosmic background radiation anisotropies
\cite{SW,W83,AS,S89,GSS,HS95a,HS95b,WHb}, systematically developing the CGI
approach and providing a comprehensive and transparent
analytic link to the alternative analytic gauge-invariant treatments based on
the Bardeen gauge-invariant variables. We develop these results both
in mode and multipole form, emphasizing the different
physical processes and assumptions and demonstrating how these are
dealt with in the CGI context. This requires the covariant mode form of
the integrated Boltzmann equations, based on the recursion relations
for almost-Friedmann-Lema\^{\i}tre mode functions \cite{GE}, enabling a direct mirroring
of standard treatments based on Wilson's seminal paper
\cite{W83,GSS,HS95a,HS95b,WHb}, except carried out in a CGI fashion,
thus forming a sound basis for extension to non-linear effects.
Indeed one of the advantages of the CGI approach in the context of the
generic multipole divergence equations is the ability to include
non-linear corrections to the almost-Friedmann-Lema\^{\i}tre treatment. Towards this end,
the relationship between the covariant mode formulations and the
almost-Friedmann-Lema\^{\i}tre covariant multipole treatment are given with this in mind,
based on the relations in Paper I \cite{GE}. A key point here is that
there is nothing new about the linear formulation itself, however
recovering the standard analytic linear results from the
the exact theory \cite{MGE} in a straightforward way, is new.
Moreover, the results presented here provide the foundation for
a non-linear extension of this approach which is outlined in a
paper by Maartens, Gebbie and Ellis \cite{MGE}.
We emphasize that in our treatment, $\langle \tau_{A_\ell}\tau^{A^\ell} \rangle$ (the
multipole form of the angular correlation function) is given for small
temperature anisotropies irrespective of the form of the
geometry \cite{MGE}, making it the natural representation for the
inclusion of non-linear dynamic effects, while the analysis for
$|\tau_{\ell}|^2$ (the mode coefficient mean square) is specifically
for almost-Friedmann-Lema\^{\i}tre models \cite{GE}. The non-linear extension of the
almost-Friedmann-Lema\^{\i}tre treatment given here will be based on the multipole-to-mode
relations, leaving the use of mode expansion to the latest possible stage.
Our focus is on the era following spectral decoupling (near $500$ eV).
A complex series of interactions take place at the various epochs of
the expansion of the universe. The kinetic equations developed in Paper
I and \cite{MGE} can represent almost any such interactions provided we use
appropriate interaction (`collision') terms; the issue is how to obtain
simplified models that are reasonably accurate in the various epochs. We
will consider only two kinds of representation here: namely
\begin{enumerate}
\item Thomson scattering, valid at late times when particle and photon
numbers are conserved and the matter is non-relativistic (during
decoupling, an alternative approach is to use a visibility function).
\item An effective two-fluid approach, obtained by truncation of
the hierarchy and valid at earlier times when strong interactions
take place establishing equilibrium or close to equilibrium conditions
between the components, i.e. when the interaction time-scale is much
less than the expansion time-scale; an alternative description is to use
a single imperfect fluid \cite{RT}.
Both descriptions are valid when the matter is relativistic.
The detailed form of interactions does not need to be represented
in this case, because the state of the matter depends only on
its equilibrium condition, characterized by its equation of state.
\end{enumerate}
At some times either form is valid and they can then be
related to each other. We do not attempt here to give a description
of earlier non-equilibrium eras when processes such as pair
production, nucleosynthesis, etc, occur, nor do we consider issues
such as inflation and reheating after inflation, and the differences
between the inflation sourced perturbations as opposed to
those based on other phase transitions.
Thus our models will be valid after the end of any period of inflation
that may have occurred and after strong non-equilibrium processes have
ceased. During this era the processes which result from initial
fluctuations left over from earlier non-equilibrium epochs, determine the
final cosmic background radiation anisotropy.
Specifically, we deal with four eras of interest. Going backwards in time
from the present, they are, firstly, free streaming from last scattering to
here and now, in the matter dominated almost-Friedmann-Lema\^{\i}tre context; secondly, slow
decoupling during a matter dominated era, which is when the
cosmic background radiation spectrum freezes out; thirdly, the late
tight-coupling era after matter-radiation equality, during which
structure formation begins; and fourthly, the
main tight coupling era after any inflationary epoch and before
matter-radiation equality, during which acoustic modes occur
in the tightly-coupled fluid, the initial matter perturbations having
been seeded by earlier conditions (for example, inflation). We then
show how to put these CGI results together to determine the major
features of the expected anisotropy spectrum. We develop sound models
of the dominant effects in each of the eras we consider, but there
will always be a need for refinement of these models by taking into
account further effects (in particular polarisation and the effect of
neutrinos).
Although the effect of the neutrinos is crucial, and can been subsumed into
the gravitational variables. We do not provide nor discuss further the
additional hierarchy of neutrino moment equations that would then need to be
included. The exact massless neutrino evolution equations are given
in a previous paper \cite{MGE} and it is easy to show that the resulting
linearised equations are essentially the same as those for massless radiation
without collisionals, hence in this paper we focus on the photon equations.
In more detail:
\begin{itemize}
\item {\it Free-Streaming} : We find the CGI integral solutions
to the almost-Friedmann-Lema\^{\i}tre multipole divergence equations with no collision
term, and use them to project the initial data from decoupling to
the current sky. We explicitly do this for instant decoupling.
Neither the Vishniac, Rees-Sciama, thermal Sunyaev-Zel'dovich nor
lensing effects are considered here -- we will focus
on the CGI model of the dominant processes, and these
further effects will introduce detailed modifications. However, a
comprehensive understanding of such secondary higher order effects
relies on a derivation of the anisotropy effects given here.
\item {\it Slow-Decoupling} : Here we consider modification of the
previous results when slow decoupling of the interactions due to
Thomson scattering is taken into account. We consider the damped integral
solution for slow recombination, and as an alternative description
modify the integral solutions appropriately with a visibility function
carrying the functional dependence of the varying electron fraction, in
a matter dominated context. Effectively, recombination is complete before
the radiation decouples. This means that the surface of last scattering
is found a little after the end of recombination \footnote{A note on
nomenclature: By {\it decoupling} we have in mind the situation when the
interaction rate per particle becomes less that the expansion rate.
By {\it last scattering surface} we mean the surface upon which the diffusion
scale is equal to the horizon scale, after which it is larger than that scale
and the free-streaming approximation is sufficient. The photons will decouple
from the thermal plasma near $0.2$ eV, and from the matter after
recombination has effectively ended, near $0.3$ eV. Free-streaming is
considered to be a good approximation from about $0.26$ eV.}.
It is during this era that photon diffusion damping scale effects
become important -- the damping scale is affected by the duration of
this era. This will be investigated in the context of
almost-Friedmann-Lema\^{\i}tre universes after matter radiation equality.
\item {\it Tight-coupling}: This is the key to the entire
treatment. We give the CGI version of the tight-coupling
approximation of Hu {\it et al} \cite{HS95a,HS95b,HWa2}. In the
almost-Friedmann-Lema\^{\i}tre treatment, the slow decoupling and free-streaming era's will
only `damp' and `project' the spectrum formed at the end of
tight-coupling onto the current sky \footnote{This useful feature of the
almost-Friedmann-Lema\^{\i}tre models is due to the homogeneity and isotropy conditions in
the background Friedmann-Lema\^{\i}tre universe and is not generic \cite{MGE}.}.
We consider two different tight-coupling regimes.
\begin{itemize}
\item The {\it late tight coupling era} is separated,
conceptually, from the early tight coupling era by matter-radiation
equality, after which time the matter perturbations have
effectively decoupled and (CDM-based) structure formation begins.
In this era, strong interactions have ceased and a Thomson scattering
description can be used. We first carry out the near tight-coupling
treatment of this era based on Peebles \& Yu \cite{PY,Kaiser},
and then reduce these equations to the covariant tight-coupling
equations equivalent to those of Hu {\it et al}
\cite{HS95a,HS95b,HWa2}. This provides the basis for understanding the
acoustic signatures in the temperature anisotropies within the CGI approach.
\item The {\it early tight coupling era} occurs between the matter
and radiation eras (after strong interactions have ceased), when
a Thomson scattering description will also be sufficient. This
era is characterized by acoustic oscillations in a tightly coupled
fluid; for calculation convenience this can be represented as a single
dissipative fluid \cite{RT}, or for a slightly more sophisticated
treatment by tightly coupled two-fluid models \cite{DBE92,dunsby}.
We give a CGI derivation of the harmonic oscillator equation providing
the primary source terms in the standard model of Doppler peak
formation by acoustic oscillations.
\end{itemize}
\end{itemize}
We put these results together in sections 7 and 8, where the equations are
integrated to give the CGI treatment in the $K=0$ (flat background)
case in terms of an integral solution.
The primary sources of the temperature anisotropies (the acoustic and Doppler
contributions near last scattering resulting in `Doppler Peaks' today) are
demonstrated \footnote{following the analytic treatment of Hu \& Sugiyama
\cite{HS95a,HS95b}}. This recovers the Sachs-Wolfe family of effects for
flat background Robertson-Walker geometries, but derived from a CGI kinetic theory viewpoint
as opposed to the photon-propagation description used in the original
Sachs-Wolfe paper.
The form of the angular correlation function is determined for the
primary effects (although not given explicitly in terms of the
matter power spectrum). The normalization to standard CDM (Adiabatic
CDM) is presented in terms of CGI variables. This demonstrates
the basic effects in the CGI formalism, and links our approach to
the standard literature, see for example \cite{WHb} and references
therein, where further details of this era are given.
It is important to note that the integrations considered here
are carried out along time-like curves, even though the cosmic
background radiation reaches us along null curves. These are
alternative approaches that are equivalent in the context of
linearisations about Friedmann-Lema\^{\i}tre models; differences will however occur
if we include non-linear corrections. Briefly, the key point
about cosmic background radiation integrations is that there are
two ways in which to proceed:
Firstly a {\it null-cone integration}, following the radiation from last
scattering to the present day\footnote{which can be parametrized either
by a null cone parameter, a projected spatial coordinate, or a projected time
coordinate}, and secondly a {\it time-like integration}
along the matter flow lines (as here). In the latter case one is
(at least implicitly) thinking of a small comoving box containing matter and
radiation \cite{ERH} which is similar to all other small boxes at the
same time, and where one has assumed that the radiation leaving is
exactly balanced by the radiation entering (from neighbouring boxes),
whether in tight-coupling (when it is a local assumption) or in the
free-streaming era (when it is a non-local assumption). In effect one
integrates behaviour in such a box in a small domain about our own
world-line from tight-coupling through decoupling to the present day;
to do this, one does not need to know about the behaviour of
null-geodesics (the integration is along time-like geodesics). Before
decoupling the matter and radiation evolve as a unit while after they
need to be integrated separately (giving the corresponding transfer
functions in each case). Then this is related to observations by,
first, conceptually shifting copies of the small box at the time of
last scattering from our world line to all points where the past
null-cone intersects the surface of last scattering at that time; this
can be done because spatial homogeneity says that these boxes are
essentially the same (a Copernican assumption is used here, justified
by the almost-Ehlers-Geren-Sachs theorem \cite{MES,MES1}); second, by
then relating distances on the last scattering surface to observed
angles by using the area distance relation, relating physical
distances at last scattering to angular size in the sky.
The next paper in the series, Paper III \cite{DGE}, deals with the
explicit relationship between null-cone and time-like integrations.
Further papers will look at non-linear extensions of the results given
here.
\section{Linearised Covariant Mode Equations}
To study details of cosmic background radiation anisotropy generation
we need both a spatial Fourier decomposition, defining wavelengths
of perturbations, together with the angular harmonic decomposition
relating anisotropies to observed angles in the sky. The CGI versions
of both decompositions were given in Paper I \cite{GE}, giving the
relationship between the mode and multipole variables.
The dynamic relations obeyed by these quantities, determining
the cosmic background radiation spatial and angular structure,
can be obtained from the Boltzmann equation in two ways: via
multipole divergence equations or via the integrated Boltzmann
equations. In each case the general equation
needs to be mode-analyzed to obtain the spatial structure.
In the first case, the almost-Friedmann-Lema\^{\i}tre multipole divergence equations are
obtained by systematically linearising the non-linear multipole
divergence equations for small temperature anisotropies given in
Maartens, Gebbie and Ellis \cite{MGE}. The mode
form of these equations (\ref{li_ibe_1}-\ref{li_ibe_l}) \cite{MGE}
can then be obtained by mode analysis, see
(\ref{PSTF-tau1}-\ref{PSTF-tau2}) below.
By contrast, the more common procedure is to directly construct the mode
form of the integrated Boltzmann equations from the linearised
integrated Boltzmann equations by substituting (\ref{dEdv})
into (\ref{ibe}) and integrating over the energy shell with
respect to $E^2 dE$ (For a more detailed relativistic kinetic
theory description of these equations see \cite{MGE} and \cite{cl2}):
\begin{eqnarray}
\int_0^{\infty} E^2 d E \Big[ E (u^a+e^a)
\nabla_a f - (\frac{1}{3}\Theta + A_a e^a + \sigma_{ab} e^a e^b )
E^2 \frac{\partial f}{\partial E} \Big]
\approx \int_0^{\infty} E^2 dE C[f]\;.
\label{ibe-aflrw}
\end{eqnarray}
Upon using the CGI definition of directional
bolometric brightness \cite{MGE}:
\begin{equation}
T(x)\left[1+\tau(x,e)\right]=\left[{4\pi\over r}\int E^3
f(x^i,E,e^a)\mbox{d} E \right]^{1/4}\,, \label{r28}
\end{equation}
the covariant equivalent of the standard formulae given in the
Wilson-Silk approach \cite{W83} can be found, where as usual
the partial energy derivative is removed by a integration by parts
and application of the regularity conditions. This is the approach
we develop now, showing later the relation to the approach based
on the multipole divergence equations.
\subsection{The Mode Equations}
The optical depth $\kappa$ is given in terms
of the Thompson scattering cross-section $\sigma_T$, the
number density of electrons $n_e$, and the fraction of electrons
ionized $x_e$:
\begin{eqnarray}
\kappa(t_0,t) = \int_{t}^{t_0} \sigma_T n_e(t')
x_e(t') d t' = \int \dot{\kappa} d t'\;,
\end{eqnarray}
where $\eta$ is the conformal time ($dt = a d\eta$)
in the Friedmann-Lema\^{\i}tre background.
Starting with the almost-Friedmann-Lema\^{\i}tre integrated Boltzmann equations
(\ref{ibe-aflrw}) for the temperature anisotropies (\ref{r28}) with
the collisional term for isotropic scattering in terms of the optical
depth, and the expansion replaced by substituting from
the radiation energy conservation equation (\ref{EFE-mc},\ref{mono_temp}),
we obtain the linearised integrated Boltzmann equations:
\begin{eqnarray}
- {\dot \tau} \approx e^a \mbox{D}_a \tau - \ts{1 \over 3} \mbox{D}_a \tau^a
+ ({\mbox{D}_a \ln T} + A_a) e^a + \sigma_{ab} e^a e^b
- {\dot \kappa (e^a v_a^B - \tau)}\,. \label{li_ibe}
\end{eqnarray}
We take a mode expansion (see Paper I \cite{GE}) for $\tau$ (the temperature
anisotropy), $A_a$ (the acceleration),
$\mbox{D}_a (\ln T)$ (the spatial-temperature perturbation),
$\sigma_{ab}$ (the shear), and the gradient of the radiation
dipole, $\mbox{D}^a \tau_a$, based on solutions $Q(x)$ of the scalar
Helmholtz equation:
\begin{equation}
\mbox{D}^a \mbox{D}_a Q=- {k^2 \over a^2} Q
\end{equation}
in the (background) space sections, where the $Q$'s are
covariantly constant scalar functions (i.e. to linear order
$\dot{Q} \approx 0$) corresponding to a wavenumber $k$. The
physical wavenumber is defined by $k_{phys}(t)=k/a(t)$.
These functions define tensors $Q_{A_{\ell}}(k^\nu ,x^i)$ that are
PSTF (in the case of scalar perturbations they are given by
the PSTF covariant derivatives of the eigenfunctions
$Q$) and so allow us to define the {\it mode functions} \cite{GE}:
\begin{eqnarray}
Q_{A_{\ell}}=\left({- {k \over a}}\right)^{-{\ell}}\mbox{D}_{\langle A_{\ell}
\rangle}Q\,~~~\mbox{and}~~~
G_{\ell}[Q]= O^{A_{\ell}} Q_{A_{\ell}}\;,
\label{SQdef-modef}
\end{eqnarray}
where $O^{A_{\ell}}=e^{\langle A_{\ell} \rangle}$ is the trace-free part
of $e^{A_{\ell}}$. We can expand any given function $f(x,e)$ in
terms of these functions, thus for scalar perturbations (see
(\ref{scalar-potential1} -\ref{scalar-potential2}),
the mode coefficients of the temperature anisotropy are
constructed as follows:
\begin{eqnarray}
\tau (x,e)=\sum_{\ell=1}^\infty \,\sum_k\,\tau _{\ell}(t,k)\,G_{\ell}[Q]\;.
\label{expn}
\end{eqnarray}
Note that $\tau_0$ is identically zero, because (\ref{r28})
defines the temperature $T$ gauge-invariantly as the all-sky average in the
real (lumpy) universe (it is not defined in terms of a background model).
We can then write:
\begin{eqnarray}
\mbox{D}^a \tau_a = \sum_k {k \over a} \tau_1(k,t) Q\,,~~~~
\mbox{D}_a (\ln T) + A_a= \sum_k [\ts{k \over a} \delta T(k,t) + A(k,t) ] Q_a\,,
\end{eqnarray}
and
\begin{eqnarray}
\sigma_{ab} = \sum_k \sigma(k,t) Q_{ab}\,.
\end{eqnarray}
The mode coefficient of the velocity of the baryons $v_B$ relative
to the reference frame $u^a$ is given by
$$v_a^B = \sum_k v_B(k,t) Q_a\,.
$$
The equations are linearised at $O(v^2)$,
$O(\epsilon v)$ and $O(\epsilon^2)$ \cite{MGE}, so we can
use, for example, ${u}_B^a \approx u^a + v_B^a$ to give the baryon
relative velocity.
The equations here are gauge-invariant (given relative to a
unique physically-based choice of the 4-velocity vector $u^a$) and valid
for any choice of mode functions, but the detailed result of their
translation back into the spacetime representation
$\tau (x,e)$ via (\ref{expn}) will depend on the harmonic functions
$Q(x^i)$ chosen\footnote{
In effect there are two major choices, namely plane wave solutions and
spherical solutions; the former occur naturally in galaxy formation studies
and the latter in observational analysis, so the relation between the two
(see Paper I \cite{GE}) is a central feature of analyzing
null-cone observations.}.
We substitute these expansions into the linearised integrated
Boltzmann equations and then use the recursion relation
\cite{W83,GSS,HS95a,GE} for mode functions $G_{\ell}[Q](e^a,x^i)$
with wave-number $k$ :
\begin{eqnarray}
e^a \mbox{D}_a[G_{\ell}[Q]]=\frac{k}{a} \Big[ {{\frac{{\ell}^2}
{(2{\ell}+1)(2{\ell}-1)}}\left( {1-{\frac{K}{k^2}}(\ell^2-1)}\right)
G_{\ell-1}[Q] - G_{\ell+1}[Q]}\Big]\;,
\label{mode_recursion}
\end{eqnarray}
where $K$ is the curvature constant of the background space sections.
With the mode decompositions of each term in (\ref{li_ibe})
for each wave number\footnote{
There should be a summation over wave numbers in the following equations.
However we follow the established custom in omitting this summation and any
explicit reference to the assumed wave number $k$.}, on using the recursion
relation (\ref{mode_recursion}) and separating out the different harmonic
components, the almost-Friedmann-Lema\^{\i}tre mode equations are found from (3)\footnote{
Note that these equations are valid for any choice of $Q$, including
both spherical and plane wave harmonic functions.}:
\begin{eqnarray}
-\dot{\tau}_{\ell} &\approx & \frac{k}{a} \left[ {{\frac{(\ell+1)^2}{(2
\ell+3)(2\ell+1)}}\left(
{1-{\frac K{k^2}}((\ell+1)^2-1)}\right) \tau _{\ell+1}-\tau _{\ell-1}}\right]
+\dot{\kappa}\tau _{\ell}\;, \ell \geq 3, \label{mibec-1} \\
-\dot{\tau}_2 &\approx& \frac{k}{a} \left[ {\frac 9{35}\left( {1-{\frac{8K}{k^2}}}
\right) \tau _3-\tau _1}\right]+ [\sigma] +\dot{\kappa}\tau _2,
\label{mibec-2} \\
-\dot{\tau}_1 &\approx & \frac{k}{a} \left[ {\frac 4{15}\left(
{1-{\frac{3K}{k^2}}}\right) \tau _2}\right]
+ [\ts{k \over a} \delta T + A] - \dot{\kappa}(v_B-\tau _1)\;.
\label{mibec-3}
\end{eqnarray}
The above equations demonstrate the up and down cascading effect
whereby lower order terms generate anisotropies in the higher order
terms, and vice versa, in a wavelength-dependent way; curvature
affects the down-cascade but not the up one. These equations can be
compared to the equations of Hu \& Sugiyama, in particular
(eqn. 6, p. 2601) \cite{HS95b}. They are identical if we use the
Newtonian frame (discussed in the following sections), and so have
the same physical content, but are more general since they are
valid with respect to a general frame $u^a$.
\subsection{From Multipole Equations to Mode Equations}
The relationship between the angular harmonic and mode expansions
are given in Paper I \cite{GE}. We start by writing the CGI harmonic
coefficients $\tau_{A_\ell}$ in terms of the mode functions
(\ref{SQdef-modef}):
\begin{equation}
\tau_{A_\ell} = \sum_k \tau_{\ell}(k,t) Q_{A_{\ell}} \approx \sum_k
\tau_{\ell}(t,k) (-{k \over a})^{-\ell} \mbox{D}_{\langle A_{\ell} \rangle} Q\;.
\label{multi-mode}
\end{equation}
Then the angular harmonic expansion for $f$ becomes the mode expansion
(\ref{expn}). On taking the multipole integrals of $f$ as in
Paper I \cite{GE}, they too are then mode-expanded by
(\ref{multi-mode}); so the linearised divergence relations
for these multipoles given in
\cite{MGE} become mode equations, equivalent to the almost-Friedmann-Lema\^{\i}tre mode
equations (\ref{mibec-1}-\ref{mibec-3}) derived above.
In detail: The almost-Friedmann-Lema\^{\i}tre multipole divergence equations are \cite{MGE}:
\begin{eqnarray}
&\;& -\left( { {\dot{T} \over T}+ \case1/3 \Theta }\right) \simeq +\case1/3
\mbox{D}^c\tau _c, \label{mono_temp} \\
(-\dot{\tau}_a) &\simeq& \mbox{D}_a \ln T +A_a+\case2/5\mbox{D}^c\tau_{ac}
-\sigma _T n_e (v^B_a-\tau _a), \label{li_ibe_1} \\
(-\dot{\tau}_{ab}) &\simeq& \sigma _{ab}+ \mbox{D}_{\langle a} \tau _{b \rangle}
+\case3/7 \mbox{D}^c\tau_{abc}+\sigma _T n_e \tau _{ab}, \label{li_ibe_2} \\
(-\dot{\tau}_{A_{\ell}}) &\simeq& \mbox{D}_{\langle a_{\ell}} \tau _{A_{\ell-1} \rangle}
+{\frac{(\ell+1)}{(2\ell+3)}} \mbox{D}^c\tau _{A_{\ell} c}
+\sigma _Tn_e\tau _{A_{\ell}}\;. \label{li_ibe_l}
\end{eqnarray}
Now the following identities are used (dropping the k-summation):
\begin{eqnarray}
O^{A_{\ell}} \mbox{D}^c \tau_{A_{\ell} c} &\approx& \tau_{\ell+1} {(\ell+1)
\over (2 \ell+1)}
\left({ + {k \over a} }\right) \left[{ 1 - {K \over k^2} \ell(\ell+2)} \right]
O^{A_{\ell}} Q_{A_{\ell}}\,, \label{PSTF-tau1}\\
O^{A_{\ell}} \mbox{D}_{\langle a_{\ell}} \tau_{ A_{\ell-1} \rangle} &\approx& \tau_{\ell-1}
\left({ - {k \over a}} \right) O^{A_{\ell}} Q_{A_{\ell}}\;,
\label{PSTF-tau2}
\end{eqnarray}
where the first relation arises from the use of the identity
\footnote{This has also been derived by Challinor and Lasenby \cite{cl2}
and is found from the PSTF recursion relations and the generalized
Helmholtz equation (which is in turn found from the constant
curvature Ricci identity) \cite{GE}.}:
\[
\mbox{D}^c \mbox{D}_{\langle c A_{\ell} \rangle} Q = {(\ell+1) \over (2 \ell +1)} \left( {
- {k^2 \over a^2}} \right) \left[ {1 - {K \over k^2} \ell(\ell+2)}
\right] \mbox{D}_{\langle A_{\ell} \rangle} Q\;.
\]
These are substituted directly into the multipole equations after taking a
mode expansion of those equations and then dropping the k-space
summation, leading again to the equations (\ref{mibec-3}). The point to
note is that while one does not explicitly need the multipole equations
in order to find the almost-Friedmann-Lema\^{\i}tre mode equations (which can be derived
from the linearised integrated Boltzmann equations as shown above),
in order to examine non-linear effects one can obtain the necessary
equations by proceeding as here from the non-linear multipole
divergence equations, to obtain higher approximations of the mode
equations and the mode-mode couplings.
\subsection{The Einstein Equations}
The key quantities which link the radiation evolution through to the
matter in the spacetime geometry are the shear $\sigma_{ab} =
u_{\langle a;b \rangle}$, the acceleration $A_a=u_{a;b} u^b$ and the expansion
$\Theta$. These couple the multipole divergence equations to the
Einstein field equations (which are given in Appendix G,
see (\ref{h-constraint})-(\ref{Friedmann-Lemaitre_R})).
The shear and acceleration, are related to the electric part of
the Weyl tensor $E_{\langle ab\rangle}$, the anisotropic pressure
$\pi_{\langle ab \rangle}$ and matter spatial gradients (see
(\ref{h-constraint}) -(\ref{Friedmann-Lemaitre_R})) while the CGI spatial
gradient of the expansion is linked to the divergence of the shear,
heat flux vector $q_a$ and the vorticity vector $\omega_a$:
\begin{eqnarray}
- \case1/2 (\rho+p) \sigma_{ab} &\approx& (\dot{E}_{ab} + \case1/2
\dot{\pi}_{ab} ) + 3 H ( E_{ab} + \case1/2 \pi_{ab} ) - H \pi_{ab}
-\left\{ {\case1/2 \mbox{D}_{\langle a} q_{b \rangle} } \right\}, ~~\label{s-1}\\
(\rho + p) A_a &\approx& - \mbox{D}_a p - \mbox{D}^b \pi_{ab} - \left\{ {
\dot{q}_a + 4 H q_a} \right\}, \label{a-1}\\
\case1/3 \mbox{D}_a \Theta &\approx& \case1/3 (\mbox{D}^b \sigma_{ab})
- \left\{ {\case1/3 q_a + \mbox{curl}\, \omega_a} \right\} \label{exp-1}\;.
\end{eqnarray}
These equations are valid for general almost-Friedmann-Lema\^{\i}tre perturbations.
In the restricted case of scalar perturbations, we set the vorticity to
zero\footnote{We can obtain scalar equations even when the vorticity is
not zero, by taking the total divergence of these equations; we will
not pursue that case here.} and non-zero quantities can be written in
terms of potentials \cite{stewart}. In particular
\begin{eqnarray}
E_{ab} &\approx& \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_E
= \case1/2 \mbox{D}_{\langle a} \mbox{D}_{b \rangle}( \Phi_A - \Phi_H), \\
\pi_{ab} &\approx& \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_{\pi}
= - \mbox{D}_{\langle a} \mbox{D}_{b \rangle} (\Phi_H + \Phi_A)\;,
\end{eqnarray}
where the potentials $\Phi_A$ and $\Phi_H$ are analogous to
the GI potentials used by Bardeen \cite{BDE92}. The following useful
combinations can be found :
\begin{eqnarray}
E_{ab} - \case1/2 \pi_{ab} \approx \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_A,~~\mbox{and}~~
E_{ab} + \case1/2 \pi_{ab} \approx - \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_H.
\label{B_potentials}
\end{eqnarray}
Using the Einstein field equations, the total flux, $q_a$, can also
be expressed covariantly in terms of these potentials:
\begin{eqnarray}
H q_a &\approx& \mbox{D}^b \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_H
- \case1/3 \mbox{D}_a \rho, \label{flux_phi} \\
H \mbox{D}_{\langle a} q_{b \rangle} &\approx& \mbox{D}_{\langle a} \mbox{D}_{b \rangle}
\left[ {\case2/3 ( \mbox{D}^2 \Phi_H + (\rho - 3 H^2) \Phi_H)
- \case1/3 \rho} \right]\;. \label{PSTF_flux_phi}
\end{eqnarray}
This then allows us to write the shear and acceleration in terms of the
scalar potentials and perturbation variables:
\begin{eqnarray}
\case1/2 (\rho + p) \sigma_{ab} &\approx& (\mbox{D}_{\langle a} \mbox{D}_{b \rangle}
\Phi_H)^{\dot{}} + 3 H \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_H - H \mbox{D}_{\langle a}
\mbox{D}_{b \rangle}(\Phi_H + \Phi_A)
+ \left\{ {\case1/2 \mbox{D}_{\langle a} q_{b \rangle}} \right\}\;,
\label{shear} \\
(\rho + p) A_a &\approx& - \mbox{D}_a p
- \mbox{D}^b \mbox{D}_{\langle a} \mbox{D}_{b \rangle} (\Phi_H + \Phi_A)
-\left\{ {\dot{q}_a + 4 H q_a} \right\}\;. \label{acceleration}
\end{eqnarray}
\subsection{Frame Transformations and Gauge Fixing} \label{frame-gauge}
There is freedom associated with the choice of reference velocity
$u^a$, which we call a {\it frame choice}. This is to be distinguished
from the choice of coordinates in the realistic universe model, which
can be done independently of the choice of $u^a$. It is equivalent to
the choice of time-like world-lines mapped into each other by the
perturbation gauge chosen (i.e. the mapping between the background
model and the realistic lumpy universe model, see for example Bruni
and Ellis \cite{BE}), but is independent of the choice of time surfaces in
that mapping. Given a particular covariantly defined choice for
this velocity, the frame choice is physically specified and the
equations are covariantly determined and gauge invariant under the
remaining gauge freedom \footnote{Gauge fixing requires in addition a
specification of the correspondence of time surfaces in the realistic
and background models (effectively specified by determining the choice
of surfaces of constant time in the realistic universe model) and of
points in initial space-like surfaces.}. In simple situations this choice
will be unique, however in more complex situations several choices of
this velocity are possible, each leading to a somewhat different CGI
description.
When we restrict ourselves to a particular frame in order to
simplify calculations, we can straightforwardly make the appropriate
simplifications in the general equations to see what the implications are
(for example setting $q^a = 0$ for the {\it energy frame}, the quantities
in the braces in equations (\ref{shear},\ref{acceleration}) above
vanish). However it is also useful to explicitly transform from one frame to
another and examine the resulting effect on dynamic and kinematic quantities.
Under a frame transformation $\tilde u^a \approx u^a + v^a$, $|v^a| \ll 1$
(restricting our analysis to non-relativistic relative velocities
\footnote{It is important to recall that gauge invariance is only guaranteed
if the choice of velocity $u_a$ coincides {\it exactly} with its value
in the background spacetime. This is not difficult
in practice, as appropriate physically defined frames ${\tilde u}_a$ will
necessarily obey this condition because of the Robertson-Walker symmetries.}),
the following relations \cite{MGE} hold:
\begin{eqnarray}
{\tilde\sigma}_{ab} &\approx& \sigma_{ab} + \mbox{D}_{\langle a} v_{b \rangle},
\label{s-trans}\\
{\tilde{A}}_a &\approx& A_a + \dot{v}_a + H v_a, \label{a-trans} \\
{\tilde{\Theta}} &\approx& \Theta + \mbox{div}\, v, \label{exp-trans} \\
{\tilde{q}}_a &\approx& q_a - (\rho + p) v_a, \label{q-trans} \\
{\tilde{\omega}}_a &\approx& \omega_a - \case1/2 \mbox{curl}\, v_a\,. \label{vort-trans}
\end{eqnarray}
The quantities $\rho$, $p$, $\pi_{ab}$, $E_{ab}$ and $H_{ab}$,
remain unchanged to linear order in almost Friedmann-Lema\^{\i}tre universes
(e.g. ${\tilde \pi}_{ab} \approx \pi_{ab}$ and
${\tilde E}_{ab} \approx E_{ab}$), and the
temperature anisotropies ($\tau_{A_\ell}$) for
$\ell > 1$ are similarly invariant for the small
velocity transformations.
The baryon and radiation (dipole) relative velocities change
according to:
\begin{eqnarray}
\tilde v_a^{B} \approx v_a^B - v_a\;, \\
\tilde \tau_a \approx \tau_a - v_a\;. \label{tau-trans}
\end{eqnarray}
Also, the projection tensor $h_{ab}$ changes if we boost to
the frame $\tilde u_a$ giving ${\tilde h}_{ab}$, hence any spatial
gradients need to be modified and ${\tilde{\mbox{D}}}_a$ will be the
totally projected derivative in that frame. The consequence is that
the perturbation variables change accordingly: Thus for any species
I, \cite{MGE}
\begin{eqnarray}
{\tilde{\mbox{D}}}_a \ln \rho_I \approx \mbox{D}_a \ln \rho_I
- 3 H v_a (\rho^I + p^I)/ \rho^I\;.
\end{eqnarray}
For example $I=R$ and $I=B$ give the equations for radiation and
baryons respectively, implying:
\begin{eqnarray}
{\tilde {\mbox{D}}}_a \ln T \approx\mbox{D}_a \ln T - v_a H\,, ~~~
{\tilde {\mbox{D}}}_a \ln \rho_M \approx \mbox{D}_a \ln \rho_M - 3 H v_a\,.
\label{pert-trans}
\end{eqnarray}
These equations allow us to determine the required transformation to obtain
desired properties of a particular choice $\tilde u^a$.
The almost-Friedmann-Lema\^{\i}tre multipole divergence equations
(\ref{li_ibe_1}-\ref{mono_temp})
are valid in any frame; in particular, if a frame transformation
is performed as above, they can be given in terms of the resulting
variables in the new frame, ${\tilde u}_a$, with whatever restrictions result.
While various choices of $\tilde u^a$ are available in a
multi-fluid description of the early universe \cite{MGE},
there are three particularly useful choices.
\begin{enumerate}
\item The {\it energy frame}: ${\tilde q}_a =0$ is preferred
when dealing with two coupled particle species, as in the two
fluid scenario \cite{dunsby}. This is useful as the Einstein field
equations are simplified to a form which takes on a similar structure
to the matter dominated equations, and even in the strong interaction
case may be expected to be unaffected by collisions because of
energy-momentum conservation (this choice is dealt with in more detail
below in the context of scalar perturbations).
\item The {\it zero acceleration frame} (or CDM frame): $\tilde u_a
= u^{C}_a\approx u_a + v_a^{C}$.
{}From the CDM velocity equation \cite{cl2,MGE} and (\ref{a-trans}) we then
find: $\dot{v}_a^{C} + H v_a^{C} + A_a \approx 0$~~$\Rightarrow$~~
${\tilde A}_a \approx 0$ \cite{MB,cl2,MGE}. This choice is
particularly useful in multi-species situations, as this frame
will be geodesic right through tight-coupling, slow decoupling
and into the free-streaming era.
\item The {\it Newtonian frame}: ${\tilde u}_a = n_a$ in which
the vorticity and shear of the reference frame vanishes:
${\tilde \sigma}_{ab} \approx\mbox{D}_{\langle a} n_{b \rangle} =0$, when
such a frame can be found. This frame is only consistent
in restricted cases \cite{HvEE}, but is particularly useful in
making comparisons with much of the analytic literature
\cite{HS95a,HS95b,GSS} and in making connections with the local
physics in terms of Newtonian analogues in Eulerian coordinates.
For example, the matter shear can be then thought of in terms of distortion
due to the relative velocities (\ref{s-trans}): $\sigma_{ab} \approx
-\mbox{D}_{\langle a} v^N_{b \rangle}$.
\item The {\it constant expansion frame}: $\tilde{\mbox{D}}_a \tilde \Theta =0$.
This choice is sometimes useful when discussing perturbations on small
scales.
\end{enumerate}
These various choices will simplify the equations in significant ways,
and enable us to recover many of the standard results. It should be
noted however, that the covariant equations we have given above are
general and do not require either gauge or coordinate restrictions to
be meaningful, and the covariant quantities have in them a
natural invariant geometric meaning. We will therefore retain
the covariant form and do not restrict ourselves to any particular
frame nor gauge choice for most of this paper, however we retain the
freedom to make such a choice when useful. If and when we do pick
a particular frame, this will be explicitly stated along with
the reason for doing so.
\subsubsection{The Newtonian Frame Link
to the Bardeen Variables} \label{newtg}
Here we demonstrate the direct link between our variables in the scalar case,
and those used in the Newtonian gauge, in terms of the Bardeen variables.
{}From (\ref{s-trans}) and (\ref{a-trans}) we find easily that for
$\tilde{u}^a = n^a$ where $\mbox{D}_{\langle a} n_{b \rangle}=0$, $n_a = u_a + v_a^N$
(the consistency of this choice is discussed in \cite{HvEE}):
\begin{eqnarray}
0 &\approx& \sigma_{ab} - \mbox{D}_{\langle a} v_{b \rangle}^N, \label{n-shear} \label{n-t1}\\
{\tilde A}_a &\approx& {A}_a + \dot{v}_a^N + H v_a^N, \label{n-t2}\\
{\tilde {\mbox{D}}}_a \ln T &\approx& \mbox{D}_a \ln T - H v_a^N, \label{n-t3}\\
{\tilde \tau}_a &\approx& \tau_a - v_a^N, \label{n-t4}\\
{\tilde \Theta} &\approx& \Theta + \mbox{div}\, v^N, \label{n-t5}\\
{\tilde q}_a &\approx& q_a + (\rho + p) v_a^N. \label{n-t6}
\end{eqnarray}
The effect of this frame transformation is to modify the $\ell=1$ and $2$
multipole divergence equations (\ref{mono_temp}-\ref{li_ibe_2}):
\begin{eqnarray}
- \dot{\tilde{\tau}}_a &\approx& {\tilde {\mbox{D}}}_a \ln T + {\tilde A}_a
+ \case2/5 \mbox{D}^c \tau_{ab} - \sigma_T n_e(v_a^B -\tau_a), \\
- \dot{\tau}_{ab} &\approx& \mbox{D}_{\langle a} {\tilde \tau}_{b \rangle}
+ \case3/7 \mbox{D}^c \tau_{abc} + \sigma_T n_e \tau_{ab}\;.
\end{eqnarray}
The $\ell >2$ equations (\ref{li_ibe_l}) remain unchanged, however the
field equations as well the perturbation equations need to modified, if
necessary using the transformation relations (\ref{a-trans} -\ref{tau-trans}).
For example, (\ref{mono_temp}) becomes
\begin{eqnarray}
({\tilde {\mbox{D}}}_a \ln T)^{\dot{}} + H ({\tilde {\mbox{D}}}_a \ln T +
{\tilde {A}}_a) \approx - \case1/3 {\tilde {\mbox{D}}}_a {\tilde \Theta} -
\case1/3 \mbox{D}_a (\mbox{D}^c \tau_c), \label{tempc}
\end{eqnarray}
which can easily be checked to be invariant under the frame transformations
${\tilde u}_a \approx u_a + v_a$.
{}From the shear evolution equation (\ref{EFE-dots}) :
\begin{eqnarray}
\mbox{D}_{\langle a} {\tilde A}_{b \rangle} \approx E_{ab} - \case1/2 \pi_{ab},
~~\Rightarrow~~ {\tilde A}_a \approx \mbox{D}_a \Phi_A\;,
\end{eqnarray}
the momentum constraint (\ref{mom-flux1}) and the propagation
equation for the electric part of the Weyl tensor (\ref{E-dot}) one
finds respectively that:
\begin{eqnarray}
\case1/3 \mbox{D}_{\langle a} {\tilde {\mbox{D}}}_{b \rangle} {\tilde \Theta}
\approx - \case1/2 \mbox{D}_{\langle a} {\tilde q}_{b \rangle}, ~~~~
\case1/3 \mbox{D}_{\langle a} {\tilde {\mbox{D}}}_{b \rangle} {\tilde \Theta}
\approx \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \dot{\Phi}_H
- H \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_A\;. \label{ng-expansion}
\end{eqnarray}
This gives the scalar monopole equation for the
temperature perturbation:
\begin{eqnarray}
\mbox{D}_{\langle a} \mbox{D}_{b \rangle} (\ln T)^{\dot{}} \approx - \mbox{D}_{\langle a} \mbox{D}_{b \rangle}
\dot{\Phi}_H - \case1/3 \mbox{D}_{\langle a} \mbox{D}_{b \rangle} (\mbox{D}^c \tau_c)\;.
\end{eqnarray}
It is important to note here is that in the shear-free frame we
can interpret the acceleration directly in terms of the
$\Phi_A$ potential, in other words, in terms of its Newtonian analog,
while $\Phi_H$ can be interpreted as a curvature perturbation.
In terms of the potentials used by \cite{HS95a,D96} one can
identify $\Psi = \Phi_H$ and $\Phi=\Phi_A$.
The above formulation is useful in linking the covariant work to the usual
GI treatments. So for example we take the mode expansion of
the potentials, one finds on dropping the k-index on the right,
\begin{eqnarray}
A_{\alpha} &\approx& ( \Phi_{A | \alpha} + V_{S|\alpha}' + H V_{S|\alpha} ) =
({V'_S }^{(0)} + {a' \over a} V_S^{(0)} - k \Phi_A) Y_{\alpha}^{(0)}, \\
\sigma_{\alpha \beta} &\approx& a (\mbox{D}_{\langle \alpha} \mbox{D}_{\beta \rangle} V_S)
= - a k V_S^{(0)} Y_{\alpha \beta}^{(0)},
\end{eqnarray}
where the prime ($'$) denotes the time-derivative with respect
to the conformal time, $Y$ are the eigenfunctions
of ${Y^{|\alpha}}_{|\alpha} = - k^2 Y$ and following Kodama and Sasaki
\cite{KS,BDE92}, the bar ($_{|\alpha}$) denotes spatial derivatives
with respect to surfaces of constant curvature in the background.
Furthermore, one can identify $V_S$ as a relative velocity.
\subsubsection{The Energy Frame}
In order to be clear on the consequences and subtleties involved in
fixing the frame, here we give the source terms
in the {\it energy frame} ($\tilde{u}^a = u_a^E \Rightarrow
\tilde{q}_a = 0$) for scalar perturbations.
The important point is that this is a physical
frame, uniquely defined by the local physics.
The equations (\ref{shear}) and (\ref{acceleration}) then take on the form :
\begin{eqnarray}
(\rho + p) {\tilde \sigma}_{ab} &\approx&
2 (\mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_H)^{\dot{}} + 4 H \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_H
- 2 H \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi_A + \mbox{D}_{\langle a} v_{b \rangle}^E, \\
(\rho+p) {\tilde A_a} &\approx&
- \mbox{D}_a p - \mbox{D}^b \mbox{D}_{\langle a} \mbox{D}_{b \rangle} (\Phi_H+ \Phi_A) + \dot{v}^E_a + H v^E_a.
\end{eqnarray}
Many CGI treatments use this frame \cite{dunsby}, and have the
advantage that the equations take on a form which is similar
to those for the matter dominated case, but can still be used
near to matter-radiation equality.
\subsubsection{Matter Domination}
During matter domination (we have in mind the CDM dominated case) there is
a unique physically relevant frame defined by the matter 4-velocity, hence
without restricting the almost-Friedmann-Lema\^{\i}tre universe further there is natural
frame, $u^a$, in which the variables will be gauge invariant and the above
relations hold.
In this frame, equations (\ref{shear}) and (\ref{acceleration}) become
\begin{eqnarray}
a^3 \rho_{\!_M} \sigma_{ab} \approx - 2 (a^3 \mbox{D}_{\langle a } \mbox{D}_{b \rangle}
\Phi_H)^{\dot{}},~~
\mbox{and}~~ A_a &\approx& 0\;.
\end{eqnarray}
Here $\rho \approx \rho_{\!_M}$ is now the density of the matter content
only. The key point is that to retain a consistent linearisation scheme
as well as retaining gauge invariance, we now have three smallness
parameters:
$v$ (non-relativistic relative velocities), $\eta$ (radiation-baryon
ratio is at least $10^{-2}$), $\epsilon$ (the universe is almost Friedmann-Lema\^{\i}tre
when $\epsilon$ is at least $10^{-5}$) \cite{MES,MGE,SAG}. It
follows that $\rho_{\!_R}$ (the radiation energy density) is now $O(\eta)$
and we can drop all terms at
least $O(\eta \epsilon)$, $O(\epsilon^2)$ and $O(\eta^2)$ such as,
for example, $p \sigma_{ab} \approx 0$ or $\case4/3 \rho_{\!_R} \tau_{ab} \simeq
\pi_{ab} \approx 0$. This is how the anisotropic pressure is
eliminated to the order of the calculation in this scheme.
The link to the matter distribution in the spacetime comes through
the mode coefficients $\sigma(k,t)$ (of the shear), $A(k,t)$ (of the
acceleration), and $\delta T(k,t)$ (of the temperature perturbation).
A mode analysis leads to a particular solution of the linearised
Einstein field equations
due to scalar modes as in (\ref{scalar-potential1} - \ref{scalar-potential2}):
\begin{eqnarray}
&& E_{ab} \approx \bar \Phi Q_{ab} = {k^2 \over a^2} \Phi_H(t,k) Q_{ab},\\
&& \sigma_{ab} \approx - \ts\frac23 (H_0^2 \Omega_0)^{-1} k^2
(a \Phi_H(k,t))^{\cdot} Q_{ab}\,, \\
&&\mbox{D}_a \ln \rho_{\!_M} \approx \ts\frac23 k
(H_0^2 \Omega_0)^{-1} \Phi_H(k,t) \left[{k^2-3K}\right] Q_a\,.
\label{tf-Newtonian}
\end{eqnarray}
For a matter dominated open model ($K \neq 0$) where $a_0=+1$ we have
$H_0^2 \approx K / (\Omega_0 -1)$.
If we add the {\it adiabatic assumption} (see Appendix \ref{sec-adb}) we find
\begin{eqnarray}
\mbox{D}_a \ln T \approx + \case1/3 \mbox{D}_a \ln \rho_M\,,
\end{eqnarray}
where we have used that $\rho_{\!_B} \approx 3 H_0^2 \Omega_0 a^{-3}$ in the
background. One can then put the mode coefficients, in the matter
dominated scalar adiabatic almost-Friedmann-Lema\^{\i}tre models, into the form :
\begin{eqnarray}
&&\delta T(k,t) \approx \frac13 (H^2_0 \Omega_0)^{-1} (a \Phi_H)
\left[ {\ts\frac23(k^2 - 3 K)}\right]\,,\\
&& A(k,t) \approx 0 \approx v_{\!_B}(k,t),\\
&&\sigma(k,t) \approx - \frac23 (H_0^2 \Omega_0)^{-1} (a \Phi_H)^{\dot{}}
\left[ {k^2} \right].
\label{adb_sclr_Friedmann-Lemaitre}
\end{eqnarray}
The first expression gives the direct effect of the gravitational potential
on the cosmic background radiation anisotropies
(Sachs-Wolfe effect), and the third the effect of
the time variation of the potential on these anisotropies.
These are investigated in detail in later sections.
The matter dominated Einstein field equations are at
least $O(\epsilon \eta)$ and fix the form
of the shear, the acceleration and the temperature
perturbations $D_a \ln T$ as they enter the integrated
Boltzmann equations (which is how the geometry enters into these
equations). The hierarchy itself is $O(\epsilon)$ and although the
radiation variables do not enter the almost-Friedmann-Lema\^{\i}tre (matter)
Einstein field equations, they
remain non-zero, and therefore there are still temperature anisotropies,
$\tau_{A_\ell}$. This is an important but subtle point --
matter domination implies the radiation moves as a test field
over the geometry.
\subsection{Linearisations, Approximations and Scales} \label{sec-linear}
In this section we discuss the various linearisation
and approximations (already mentioned in the last
section), that will be used in this paper.
\subsubsection{Almost-Friedmann-Lema\^{\i}tre Linearisation}
Here we drop all terms that are at least ${\cal O}(\epsilon^2)$.
The implication of this is that one can only consider small velocity boosts,
large ones would break the linearity about the Friedmann-Lema\^{\i}tre background -- hence
we include $v^2 = |v^a v_a| \ll 1$ as an almost-Friedmann-Lema\^{\i}tre limit dropping the
additional terms that are at least ${\cal O} (v \epsilon,v^2)$.
The important subtlety here is that $\epsilon_\ell$ is in fact the
temperature anisotropy smallness parameter as related to the temperature
moment mean squares $|\tau_{A_\ell}| \propto \epsilon_{\ell}$. The almost-Friedmann-Lema\^{\i}tre
limits on the geometry, $\epsilon$, (which define $\sigma_{ab}$, $A_a$ and
$D_a \Theta$ (for example) as ${\cal O}(\epsilon)$ in appropriate dimensionless
units \cite{SME,SAG} are related to $\epsilon_{\ell}$ via the
almost-Ehlers-Geren-Sachs theorem. In other words, limits on the
temperature anisotropies, $\epsilon_\ell$, put bounds on the size of
the smallness parameter $\epsilon$, given that a weak
Copernican principle holds. Furthermore, the limits on $\epsilon$
in turn place consistency limits on the size of the $v/c$ boosts
that are applicable (here in units of $c=1$). Thus at least
almost-Friedmann-Lema\^{\i}tre means keeping terms that are at most :
\begin{equation}
\mbox{Almost-Friedmann-Lemaitre}~~ \approx {\cal O}(\epsilon,v).
\label{almost-Friedmann-LemaitreRW}
\end{equation}
\subsubsection{Matter Dominated Linearisation}
This is based on the radiation-baryon ratio,
$\eta \propto \ts{\rho_R \over \rho_M}$.
We keep every ${\cal O}(\eta)$ but in the almost-Friedmann-Lema\^{\i}tre case of matter
domination we then drop everything that is at least
${\cal O}(\eta v, \eta \epsilon, \epsilon v,\eta^2, \epsilon^2, v^2)$.
We then have that matter dominated almost-Friedmann-Lema\^{\i}tre means keeping
terms that are at most :
\begin{equation}
\mbox{Matter dominated almost-Friedmann-Lemaitre} ~~\approx {\cal
O}(\epsilon,v,\eta). \label{almost-Friedmann-Lemaitre}
\end{equation}
\subsubsection{Expansion in Thompson Scattering Time}
We will introduce a perturbative scheme in the Thompson scattering
time, $t_c = (\sigma_T n_e)^{-1}$, and will consider terms up to
${\cal O}(t_c^3)$ during the tight-coupling calculation -- such an
expansion will be used to generate equations near to tight-coupling,
the limiting case being when $t_c =0$. Additionally an equivalent
scheme can be constructed in terms of the differential optical depth
$\kappa'$. This scheme is useful in the slow-decoupling era,
{\it { i.e. }} in expansions where $\kappa'$ and $(\kappa')^2$ are
sufficiently small to be ignored when compared to terms of order
$\kappa$. This approximation allows one to additionally
consider the case when $\kappa' e^{-\kappa} \ll e^{-\kappa}$.
\subsubsection{Small and Large Scales}
We will find it convenient to introduce the notion of small and
large scales. We will do this in two heuristically equivalent ways. The
first scheme is based on the parameter $\epsilon_H$, where the Hubble
expansion is of order $\epsilon_H$, and is used when considering
situations outside the Hubble flow; thus in the almost-Friedmann-Lema\^{\i}tre
small-scale case one would ignore all terms at least ${\cal O}
(\epsilon_H^2,\epsilon^2,v^2,v \epsilon_H, v \epsilon,\epsilon
\epsilon_H)$. This scheme is useful since it can be used
without a mode expansion. It is ideal for making qualitative statements
without the details which arise when introducing mode functions; specifically
avoiding the complexity of mode-mode coupling in the small scale non-linear
situation. By small-scale almost-Friedmann-Lema\^{\i}tre we have in mind keeping only terms
that are at most:
\begin{equation}
\mbox{Small-scale almost-Friedmann-Lema\^{\i}tre}~~\approx {\cal O}(\epsilon_H,\epsilon,v)\;.
\label{small_scale}
\end{equation}
A second and more precise scheme is based on the Hubble scale
$\lambda_H=a_E/k_H$ defined near the time of emission (E), allowing one
to use $k/k_H > 1$ and $k/k_H <1$ as characterizing large and small
scales respectively.
\section{Covariant Integral Solutions}
This section has three aims: (i) Reproducing the integral solution of the
free-streaming mode equations and modifying them in order to take into
account Thomson scattering, using the CGI variables \cite{HS95b,SeZalb}.
We carry out a time-like integration, instead of a null-cone
integration corresponding to the original Sachs-Wolfe paper
\cite{SW}, restricting ourselves to scalar perturbations
with adiabatic modes only and assuming for the most part a $K=0$
almost-Friedmann-Lema\^{\i}tre background universe. (ii) Showing in the CGI
formulation how fluctuations at last scattering time result
in measurable cosmic background radiation anisotropies. (iii) Demonstrating
how the solution can be related to standard formalisms by choosing
specific frames; in particular we consider the Newtonian frame
based on a shear free congruence.
The basic equation we are concerned with in this
section is the integrated Boltzmann equation (\ref{li_ibe}).
In covariant form it is given by
\begin{eqnarray}
\dot{\tau}(x,e) + e^a \mbox{D}_a \tau(x,e) + {\cal B}(x,e) \simeq
C[x,e]\;, \label{IBE-aflrw}
\end{eqnarray}
where the {\it gravitational source term}, ${\cal B}$, and {\it Thompson
source term}, $C$, for damping by Thompson scattering are respectively
given by:
\begin{eqnarray}
{\cal B} = -\ts{1 \over 3} \mbox{D}^a \tau_a + (\mbox{D}_a \ln T
+ A_a) e^a + \sigma_{ab} e^a e^b,~~~ C[x,e] \approx \dot{\kappa}
(e^a v_a^B - \tau)\,,
\label{IBE-source}
\end{eqnarray}
and in the almost-Friedmann-Lema\^{\i}tre situation in mind\footnote{One can compare
this to the formulation of Durrer \cite{D96,DZ96} (eqns 3.5 and 18).
To see how this is linked to the
Bardeen potentials $\Psi$ and $\Phi$, we can use
$E_{ab} = \case1/2 \mbox{D}_{\langle a} \mbox{D}_{b \rangle} (\Psi - \Phi)$ \cite{DZ96}. Notice
that Durrer's integral solutions take on the Sachs-Wolfe form and can be
compared with the treatments in \cite{dunsby,Mag}, while ours follow the
form of \cite{W83,SeZalb}.}, the Einstein field equations give:
\begin{eqnarray}
3 \rho_{\!_M}^{-1} (\mbox{div}\, E)_a \approx\mbox{D}_a \ln \rho_{\!_M}, ~~
\sigma_{ab} \approx -2 [ (\rho_{\!_M}^{-1} E_{ab})^{{\dot{}}}
- \mbox{curl}\, (\rho_{\!_M}^{-1} H_{ab})]\,.
\end{eqnarray}
The mode expanded form of this equation for the flat case
($K=0$) can be written in the compact form as follows:
\begin{eqnarray}
\tau _\ell^{\prime }+k\left[ {{\frac{(\ell+1)^2}{(2\ell+1)(2\ell+3)}}
\tau _{\ell+1}-\tau_{\ell-1}}\right] +\kappa ^{\prime }\tau
_\ell\simeq {\cal S}_B\;,
\label{modeibe}
\end{eqnarray}
where
\begin{eqnarray}
{\cal S}_B= - \left[ {a{\cal B}_0\delta _{\ell0}+(a{\cal B}_1
+\kappa^{\prime }v_B)\delta _{\ell1}+a{\cal B}_2\delta _{\ell2}}\right]\;.
\label{source-calB}
\end{eqnarray}
This combines (\ref{mibec-1}-\ref{mibec-3}) in a single
equation for $\tau _\ell(\eta ,k)$ (see \cite{GE,MGE}), where
$\tau = \sum_\ell \tau_\ell G_\ell[Q]$ is
written in terms of conformal time $\eta $ ($dt = a d \eta$),
rather than proper time $t$. It is valid for all $\ell \ge 0$, with
$\tau_0=0$ a solution as required, consistent with the definitions we
introduced above (${\cal B}_0$ cancels the dipole term on the right in
this case).
In what follows, we will deal with the integral solutions to (\ref{modeibe})
given the source terms (\ref{source-calB}).
The paradigm is to match an (almost-Friedmann-Lema\^{\i}tre) era of free-streaming
to one of tight-coupling. We will construct the homogeneous solution
(without gravity or scattering) first, then include the
gravitational effects to construct the free-streaming solution
({\it { i.e. }} after decoupling to the present day) and finally include
Thompson scattering to find the integral solution including
scattering (which can be used during slow decoupling or to
include effects of reionization).
Diffusion damping is included in this full solution with Thompson
scattering, which in general has to be solved numerically, however
it is helpful to introduce various analytical approximations
for the different stages described by the solution;
this will be done later, where the visibility function
approximation is used and the damping scale derived. Additional effects,
such as the anisotropic correction and polarization correction, have to
be dealt with separately.
\subsection{Integral Solutions (Flat Almost-Friedmann-Lema\^{\i}tre Case)}
Here we wish to find the general solution to (\ref{modeibe}) without
collision terms, i.e. with $\kappa^{\prime} = 0$, integrating along time-like
curves using conformal variables. In order to do this we first find the
solution to the homogeneous version of the above equation
(i.e. for no gravity and no scattering), second, an integral solution
of the inhomogeneous equation with gravity taken into account from
the homogeneous solution, and third, the general solution for
free-streaming. In the next sections we consider the effect
of Thompson scattering ($\kappa \neq 0$), and the transition from
tight coupling to free-streaming.
The approach here is similar to the Seljak-Zaldariagga treatment
\cite{SeZalb,SZ97a}, however, they have taken the Sachs-Wolfe like
formulation of the integrated Boltzmann equations, which is an
integration down the null cone, and integrated out the angular
dependence over Legendre polynomials (angular averaging) in order
to construct the mode coefficients; then the conformal radial
distance, $\chi $, is written in terms of the conformal time $\eta$,
leading to an integral solution dependent only on the conformal time.
Thus, formally they have carried out a null-integration. By contrast,
what is carried out here is in effect an integration of the
integrated Boltzmann equations down the matter world-lines, thus this
is a time-like integration, onto an initial surface (`last scattering').
The corresponding initial data near our past world line on that
surface can then be related to data on the intersection of the past
null-cone with the surface by means of a suitable homogeneity assumption.
The time-like nature of the integration is often not made particularly
clear in the literature, but solutions of the generic multipole divergence
equations (from which the almost-Friedmann-Lema\^{\i}tre mode hierarchy of temperature
anisotropies are derived) are usually based on time-like integrations
in the relativistic kinetic theory \cite{MGE}.
There is no effective difference between the time-like and null-integrations
at linear order. This is because, at linear order, one needs to only integrate
up one null geodesic, and then decompose the resulting temperature into
its various moments. This is equivalent to making the multipole decomposition
first and integrating these up one time-like world-line.
This equivalence does not hold in the exact case, so when trying to
include the effects of non-linearity, assuming such an equivalence
could lead to misleading results.
\subsubsection{Finding the Homogeneous Solution}
The $\ell =0,1$ and $2$ multipole divergence equations and hence
mode form of the integrated Boltzmann equations are exceptional,
given that $\tau_0=0$. The point of the integral solutions is to cast
the exceptional equations $\ell=0,1,2$ into a form that allows
analytic investigation. We now find the covariant homogeneous solutions.
Consider the homogeneous equation (valid for $\ell \ge 1$) :
\begin{equation}
{\tau _\ell^{(0)}}^{\prime }+k\left[ {{\frac{(\ell+1)^2}{(2\ell+1)
(2\ell+3)}}\tau_{\ell+1}^{(0)}-\tau _{\ell-1}^{(0)}}\right] =0\;,
\label{homog1}
\end{equation}
for the background $K=0$ case\footnote{Or for the open or
closed cases by following \cite{HS95a} \cite{GSS}.}
without damping. The functions
\begin{equation}
\tau _\ell^{(0)}(k,\eta )=(2\ell+1)\beta _\ell^{-1}j_\ell(k\eta )
\label{solution}
\end{equation}
are solutions of (\ref{homog}), if the coefficients $\beta _\ell$ obey the
recursion relations \cite{GE}:
\begin{equation}
(\ell+1)\beta _\ell=(2\ell+1)\beta _{\ell+1},\;\beta _\ell(2\ell-1)
=\ell\beta _{\ell-1}\;.
\label{recur}
\end{equation}
This can be shown by multiplying (\ref{homog1}) through by
$\alpha_\ell = \beta_\ell(2\ell+1)^{-1}$ and comparing the result with the
recursion relation for spherical Bessel functions.
If the function $j_0(k\eta )$ satisfies the equation (\ref{homog1})
for $\ell=0$ (there are of course no terms with $\ell<1$) then the
rest of the equations ($\ell\geq 1)$ will be satisfied because of the
recursion relations :
\begin{eqnarray}
-(2\ell+1) (\alpha_\ell \tau_\ell)' \simeq k [ (\ell+1) (\alpha_{\ell+1}
\tau_{\ell+1}) - \ell (\alpha_{\ell-1} \tau_{\ell-1})]\;.
\end{eqnarray}
The freedom in $\beta_\ell(k)$ occurs in $\beta_0(k)$
and $\beta_1(k)$. Given that the Bessel function is finite at the
origin: $j_\ell(0) = \delta_{\ell0}$,
these can be chosen to satisfy $\beta_0 = \beta_1 = +1$, the rest are
generated through the recursion relations on $\beta_\ell$
and then determine the solution $\tau_\ell^{(0)}(k,\eta )$.
The arbitrarily specifiable initial data is later fixed by introducing
an integral solution ((\ref{no2.1}) below) containing arbitrary functions
$C_A(\eta)$ (see (\ref{no2.2})) which are determined by the Einstein
field equations through
${\cal B}_I(\eta)$.
The corresponding mode functions are
\begin{eqnarray}
\tau ^{(0)}(x,e)=\sum_{\ell=1}^\infty \beta _\ell^{-1}(2\ell+1)
\,j_\ell(k\eta)O^{A_\ell}Q_{A_\ell}|_{\!_{\sf FLAT}}\;,
\label{modeff}
\end{eqnarray}
(cf. (\ref{expn}))
and the corresponding multipole coefficients can then be found:
\begin{eqnarray}
\tau _{A_\ell}|_{\!_{\sf FLAT}}\simeq \beta _\ell^{-1}(2\ell+1)\,j_\ell(k\eta )
Q_{A_\ell}|_{\!_{\sf FLAT}}\;.
\end{eqnarray}
Notice that this differs by a factor of $i^{-\ell}$ from Wilson
\cite{W83} since we are using plain mode functions instead of
plane waves, although these can be easily related. Note there is no
explicit mode mixing in this approximation, but such mixing is implicitly
determined by the recursion relations (\ref
{recur}). This shows that we should be careful with any truncation
procedure we propose (see Appendix E). This procedure can be
easily extended to the open case using the recursion relations for the open
mode functions (see Appendix K).
\subsubsection{Construction of the Integral Solution}
Given that we have the solution $\tau _\ell^{(0)}(k,\eta )$ to the homogeneous
equation of the form (\ref{homog}), now consider the equation with given
gravitational source terms, but still without damping:
\begin{eqnarray}
~\tau _\ell^{\prime }+k\left[ {{\frac{(\ell+1)^2}{(2\ell+1)(2\ell+3)}}
\tau _{\ell+1}-\tau_{\ell-1}}\right] = - \left[ {
a{\cal B}_1\delta _{\ell1}+a{\cal B}_2{\ \delta _{\ell2}}} \right].~~~
\label{inhomog}
\end{eqnarray}
What is important to notice here, is that this equation is valid
for $\ell \geq 1$, not $\ell \geq 0$ as in (\ref{modeibe});
indeed $\tau_0 = 0$.
We need to find a particular solution to this equation.
We proceed as follows. Consider the ansatz in terms of $A_{\ell}$ using
$\delta \eta= \eta- \eta'$, along with the Liebnitz rule for
differentiation of integrals:
\begin{eqnarray}
\tau_\ell^P(\eta) = \int_0^{\eta} d \eta' A_\ell(\eta,\eta')~~\Rightarrow
~~ {\partial\tau_l^P \over \partial\eta}(\eta) = \int_{0}^{\eta} d \eta' {\partial \over \partial
\eta} A_\ell(\eta,\eta') + A_\ell(\eta,\eta)\;.
\label{no2.1}
\end{eqnarray}
Now we define the kernel, $A_{\ell}$, as in \cite{W83,SZ97a}:
\begin{equation}
{A_\ell(\eta ,\eta ^{\prime })
=C_0(\eta)\tau _\ell^{(0)}(\delta \eta)+C_1(\eta){\frac \partial
{\partial \eta ^{\prime }}}\tau _\ell^{(0)}(\delta \eta)
+C_2(\eta){\frac{\partial ^2}{%
\partial {\eta ^{\prime }}^2}}\tau _\ell^{(0)}(\delta \eta)}\;.
\label{no2.2}
\end{equation}
It can then be shown from (\ref{no2.1}) and (\ref{no2.2}) that :
\begin{eqnarray}
{\tau _\ell^{\prime }}^P+k\left[
{{\frac{(\ell+1)^2}{(2\ell+1)(2\ell+3)}}
\tau_{\ell+1}^P-\tau _{\ell-1}^P}\right]={\cal S}_C\;,
\end{eqnarray}
where
\begin{eqnarray}
{\cal S}_C=C_0(\eta )\tau _\ell^{(0)}(0)+C_1(\eta
){\tau_\ell^{(0)}}^{\prime }
(0)+C_2(\eta ){\tau _\ell^{(0)}}^{\prime \prime }(0)\;,
\label{int_soln_undamped}
\end{eqnarray}
given $\tau _\ell^P(\eta )$ as in (\ref{no2.1}-\ref{no2.2}). Thus
(\ref{inhomog}) is satisfied by our ansatz provided the coefficients
$C_0(\eta ),C_1(\eta ),$ and $C_2(\eta )$ in the integral solution are
found in terms of the CGI variables
${\cal B}_0(\eta)$, ${{\cal B}_1(\eta),}$ and ${\cal B}_2(\eta)$
determined by the Einstein field equations.
This will be considered next, when we put the parts of the
solution together to obtain (\ref{C-B-coef}).
\subsubsection{Inclusion of Damping}
Here we extend the previous solution (\ref{inhomog}),
where the relationship between the coefficients in the integrated
Boltzmann equations (\ref{LEIB001})
can be read off from (\ref{int_soln_undamped}),
including damping through $\kappa ^{\prime }$.
We notice that if $\tau _\ell^{(0)}$ is a solution to (\ref{homog}), then
\begin{equation}
\tau _\ell^{*}(\eta )=e^{-\kappa (\eta )}\tau _\ell^{(0)}(\eta )
\end{equation}
will be a solution to
\begin{equation}
{\tau _\ell^{*}}^{\prime }+k\left[ {{\frac{(\ell+1)^2}{(2\ell+1)(2\ell+3)}}
\tau_{\ell+1}^{*}-\tau _{\ell-1}^{*}}\right] +\kappa ^{\prime }
\tau _\ell^{*}\simeq 0\;.
\end{equation}
Similarly we find that for the integral solution (\ref{no2.1}),
of (\ref{inhomog}), the expressions
\begin{equation}
\tau_\ell^{*P}(\eta )=\int_0^\eta d\eta ^{\prime }e^{-\kappa (\eta ^{\prime
})}A_\ell(\eta ,\eta ^{\prime })~~\mbox{or}~~\;\tau _\ell^{*P}(\eta )
=e^{-\kappa (\eta)}\int_0^\eta d\eta ^{\prime }
A_\ell(\eta ,\eta ^{\prime })\;,
\label{no3}
\end{equation}
will be particular solutions to (\ref{modeibe}), given the correct
choice of $C_0$, $C_1$ and $C_2$. Hence we can modify the solutions
of the previous section to include Thompson scattering by simply including
the damping terms as in these equations. Hence the extended equations
include the special case of free streaming, when for some
interval of time $\kappa = 0$; thus they can extend all the way
from late tight coupling to the present day, if we include a
suitably time-dependent scattering coefficient $\kappa$.
In more detail: we have that $\tau_\ell^{(0)}(\eta_*) =0$. This is simply
due to the fact that during tight-coupling there are no higher
moments, just the monopole.
Here we assume that free-streaming begins after some
$\eta_*$. The slow decoupling solution will modify this assumption. As
before we have (now using $\delta \eta^* = \eta- \eta_*$)
\begin{eqnarray}
\tau_\ell^{(0)}(\eta) = (2\ell+1) \beta_\ell^{-1}
j_\ell(k \delta \eta^*)
\Rightarrow
\tau_\ell^{*P}(\eta) \simeq \int_0^{\delta \eta^*}
d \eta' A_\ell(\eta, \eta')\;,
\label{homog}
\end{eqnarray}
where the initial data is now given by $C_I(\eta'- \eta_*)$; notice
that we do not introduce an additional $\eta_*$ as we will be using the
solution of $\tau^{(0)}$ already including the initial conditions
\footnote{We could have used $\tau_\ell^{(0)}(\delta \eta) \rightarrow
\tau_\ell^{(0)}(\delta \eta+ \eta_*)$ along with the original
homogeneous solution unchanged.}. Once again we have the integral
solution $\tau_\ell^P(\eta)$ integrated from $0$ to $\delta \eta^*$
such that the anisotropies are now determined by $\tau_\ell^P(\eta_0)$.
Evaluating the integral from $0$ to $\Delta \eta_*$
(for $\Delta \eta = \eta_0 -\eta'$), we find
\begin{eqnarray}
\tau_l(\eta_0) &=& \int_{\eta_*}^{\eta_0} d \eta' e^{-\kappa(\eta'-\eta_*)}
\left[ C_0(\eta'- \eta_*) \tau^{(0)}_\ell(\Delta \eta)\right.\nonumber\\
&+&\left. C_1(\eta'-\eta_*){\tau_\ell^{(0)}}'(\Delta \eta) +
C_2(\eta'-\eta_*) {\tau_\ell^{(0)}}''(\Delta \eta)
\right]\;,
\label{damp-soln-1}
\end{eqnarray}
where the damping term now enters explicitly.
\subsubsection{The Complete Solution}
We can now construct the general solution to (\ref{modeibe})
with $\kappa^{\prime} \neq 0$ by putting
the previous results together. The homogeneous seed solution $\tau
_\ell^{(0)}(\eta )$ is given by (\ref{solution}). The particular integral
solution $\tau _\ell^{P}(\eta )$ is given, in terms of $\delta \eta =
\eta -\eta^{\prime }$ by (\ref{no3}). The general solution is then given by
\begin{equation}
\tau _\ell(\eta )=e^{-\kappa (\eta )}\tau _\ell^{(0)}(\eta )
+\tau _\ell^{*P}(\eta )\;.
\label{no1}
\end{equation}
Substituting this into the general equation (\ref{modeibe}) and using
the radial eigenfunctions evaluated at zero (in particular $j_\ell(0)=\delta
_{\ell0}$ along with the recursion relation) gives
\begin{eqnarray}
C_0 \delta _{\ell0}+C_1 {\ \beta _1^{-1}k\delta _{\ell1}}
+C_2k^2 {5\beta _2^{-1}\left( {\frac 2{15}\delta _{\ell2}+\frac 13\delta
_{\ell0}}\right) } = - \left[ {
(a{\cal B}_1+\kappa ^{\prime}v_B)\delta _{\ell1}
+a{\cal B}_2\delta _{\ell2}} \right],
\label{find_coef}
\end{eqnarray}
relating the functions determining the solution to the time-dependent
coefficients in the equation.
{}From (\ref{find_coef}) the functions in the integral
solution are found in terms of the dynamical CGI variables:
\begin{eqnarray}
-C_0(\eta )\simeq
+\frac 52a{\cal B}_2,~~~-C_1(\eta )\simeq +\frac
1k\left( {a{\cal B}_1+\kappa ^{\prime } v_B}\right)\;,~~-C_2(\eta )\simeq
+\frac 1{k^2}{a{\cal B}}_2\;.
\label{C-B-coef}
\end{eqnarray}
For scalar perturbations, the term $C_2$ is effectively the coefficient of
the shear; the term $C_1$ is the coefficient of the gradient of the
temperature and the acceleration and the coefficient $C_0$ also
contributes to the shear (see (\ref{efe2.1b})). Note that these
are CGI with respect to $u^a$.
These functions are all evaluated at a time $\eta$,
which takes all values from $\eta_d$, the time of decoupling, to $\eta_0$,
the time of observation. In the null-cone formulation of the
problem this input of new information corresponds to the way the null
geodesics from the point of emission to the observer keep crossing new
matter and hence encounter new information. Because we are
integrating on a time-like curve, this information is represented here as
varying with time along that curve; and in some simple circumstances, the
values at later times are determined fully by the values at earlier times
(as happens, for example, in the original Sachs-Wolfe case: $K = 0$,
$p = 0$, and only growing scalar modes are considered).
\subsection{Integration by Parts}
In order to deal easily with the initial data it is now useful to
write the general solution for $K=0$ in terms of the present time,
$\eta _0$, and the initial time, $\eta _*$, by integrating with
respect to conformal time and defining
$\Delta \eta_* =\eta _0-\eta _*$\footnote{The relationship between the
conformal time $\eta $ and the radial distance $\chi $
is $d\chi =-d\eta$ so $\chi = \eta _0 -\eta $ which follows for the
homogeneity and isotropy in the background.}. We could fix the conformal
time by setting $\eta_0=0$ here. In order to recover the results
of \cite{cl,SZ97a} one would take $\eta_* \rightarrow 0$, however, we
would like to recover the results as close to \cite{HS95a} and so retain
their conventions where possible.
Notice that from $d\tau^{(0)} /dt=0$, we have $\tau ^{(0)}(x(\eta ),e(\eta ))
=\tau ^{(0)}(x(\eta_*),e(\eta _*))$,
and $j_{\ell}(0)= \delta_{\ell0}$ (we have chosen the solution
to be finite at origin). We choose the initial conditions
$\tau_{\ell}(\eta_*)=0$ \cite{W83} and $\tau_0(\eta)=0$ and using the
parameter freedom in the homogeneous solution, set $\beta_0(k)=+1$
and $\beta_1(k)=+1$.
The homogeneous solution is now fixed as in (\ref{homog}),
for $\delta \eta^* = \eta -\eta_*$ and this in turn sets the integral
solution to (\ref{no3}) \cite{W83}:
\begin{eqnarray}
\tau_\ell^{*P}(\eta) = \int_0^{\delta \eta^*} e^{- \kappa(\eta')}
A_\ell(\eta, \eta') d \eta'\;,
\end{eqnarray}
where we still have the freedom of setting the initial data for the integral
solution from the $C_I(\eta)$'s which are fixed by the Einstein
field equations. Putting this all together we find
\begin{eqnarray}
\tau_\ell(\eta_0) = \tau^{(0)}_\ell(\eta_*) + \tau_\ell^{*P}(\eta_0)
= \tau_\ell^{*P}(\eta_0) = \int_0^{\Delta \eta_*} e^{-\kappa}
A_l (\eta_0, \eta') d \eta'\;.
\end{eqnarray}
On changing the integration to from $\eta_*$ to $\eta_0$ in
(\ref{damp-soln-1}), integrating by parts, and using the initial conditions
(once again $\tau^{(0)}_{\ell}(\eta_*) = 0$) we find :
\begin{eqnarray}
\tau_\ell(\eta_0) &\simeq& \left[ {C_2'(\eta_*) - C_1(\eta_*) } \right]
e^{-\kappa} \tau^{(0)}_\ell(\eta_0) - C_2(\eta_*) e^{-\kappa}
{\tau_\ell^{(0)}}'(\eta_0), \nonumber \\
&+& \int_{\eta_*}^{\eta_0} d \eta' e^{-\kappa} \left[ {
C_0(\eta') - C_1'(\eta') + C_2''(\eta')} \right] \tau^{(0)}_\ell(\eta_0
+ \eta_* - \eta'), \nonumber \\
&+& \int_{\eta_*}^{\eta_0} d \eta' (\kappa' e^{-\kappa}) \left[ {C_1(\eta')
-2 C_2(\eta')} \right]
\tau^{(0)}_\ell(\eta_0 + \eta_* - \eta'), \ldots \nonumber \\
&+& \int_{\eta_*}^{\eta_0} d \eta' \left( {(\kappa')^2 - \kappa''} \right)
e^{-\kappa} C_2(\eta') \tau^{(0)}_\ell(\eta_0 + \eta_* - \eta')\;.
\label{int_soln_damping}
\end{eqnarray}
The initial data for the solution $\tau _\ell^{(0)}(k,\eta )$ is the set of
constants $C_I(\eta_*)$ which are determined by ${\cal B}_A(k)
=\left\{ {\cal B}_0(k),{\cal B}_1(k),{\cal B}_2(k)\right\} $; these
must be matched to the initial distribution function on an appropriate
initial surface $\Sigma$ (for example, the `surface of last scattering'
which can be covariantly and gauge invariantly defined). This then determines
the solution up to the present day (and after). We are free to chose
any $Q$'s as long as they solve the Helmholtz equation in the
background. The choice of Q then explicitly determines $G_\ell[Q]$,
for example we are free to choose Q to be the spherical or plane-wave
basis. In practice we naturally use two sets of mode functions $G_\ell[Q]$,
matching those for the null-cone (given in a spherical basis) to those in
some initial surface (given in terms of a plane-wave basis). The matching
of these two sets of harmonics is then given by the relations usually
written into the construction of the mode coefficients (see (\ref{modeff})).
This matching is based on mode functions $G_l[Q]$ in the Robertson-Walker background,
which is acceptable because of the homogeneity assumption. By using
$G_\ell[Q]$ we do not actually need the explicit form of the $Q$'s.
Equation (\ref{int_soln_damping}) shows (r.h.s. of the first line)
how major parts of the cosmic background radiation anisotropy are
determined directly from the set of initial conditions (at last
scattering, for the freely propagating radiation). The integrated
effect arises through the coefficients $C_I(\eta)$ as integrated down
time-like geodesics in the remaining terms on the right hand side. In general
there is a non-linear coupling through the field equations between
the matter, the radiation and the acceleration and shear terms that
arise in the integrated part. The situation is much simpler when this
back-reaction can be neglected; for this reason it is convenient to
consider the case of matter domination, during which the radiation
can be considered as a test-field propagating on the background
determined by the matter content.
However we can also consider the general set of linearised field
equations (see Appendix \ref{almost-Friedmann-Lemaitre-efe}) and the coupling to the
radiation via the source terms, first the gravitational source,
${\cal B}$, and second, the scattering source, $C[\tau]$,
(\ref{IBE-source}), in (\ref{IBE-aflrw}). In the following sections
we look at the various approximations that can be applied at different
epochs.
\section{Free-streaming}
Using the integral solution (\ref{int_soln_damping}) we construct
the almost-Friedmann-Lema\^{\i}tre free-streaming projection of the initial conditions
near last scattering to here and now (the determination of these initial
conditions is demonstrated in latter sections) and the integrated
secondary contributions arising during the period after last-scattering
until now (we have dropped the baryon relative velocity effect using the
instantaneous decoupling assumption):
\begin{eqnarray}
&~&
{\frac{\tau _\ell(\eta _0)\beta _\ell}{(2\ell+1)}}\simeq \left[ {\frac
1k[a{\cal B}_1](\eta_*)
- \frac 53\frac 1{k^2}[a{\cal B}_2]^{\prime}(\eta_*)
-\frac 1k[a{\cal B}_2](\eta_*){\frac \partial {\partial \eta _0}}}\right]
j_\ell(k\Delta \eta_*)
\nonumber \\
&-&
\int_{\eta _*}^{\eta _0}d \eta \left\{ {\frac 56a{\cal B}_2
-\frac 1k(a{\cal B}_1)^{\prime }+\frac 53\frac 1{k^2}(a{\cal B}_2)^{\prime
\prime }}\right\} j_\ell(k\Delta \eta )\;,
\label{freestreaming}
\end{eqnarray}
Where we used as final conditions :
\begin{equation}
[a{\cal B}_1]^{\prime}(\eta _0)=[a{\cal B}_2](\eta _0)=0\;.
\label{init1}
\end{equation}
The first term on the right, the ${\cal B}_1$ term, will generate the
acoustic primary effect on the anisotropies, the second term is the Doppler
contribution due to the radiation dipole (the baryon velocity
contribution which would arise through $C_1$ (\ref{C-B-coef})), the third and
fourth terms give the effect of any shear, near last scattering
(through the initial conditions of ${\cal B}_2$).
The remaining terms represent the integrated Sachs-Wolfe effect.
The above equation will be modified in the following section
to include slow decoupling, but first we demonstrate how to recover
the basic Sachs-Wolfe effect.
\subsection{The Almost-Friedmann-Lema\^{\i}tre Sachs-Wolfe Effect}
We now find the solutions corresponding to the matter dominated,
free-streaming era, with adiabatic modes only, using the Newtonian
frame treatment (see section \ref{newt-sachs-wolfe}). Using the field
equations from \cite{MGE},
or from the Appendix \ref{almost-Friedmann-Lemaitre-efe} and \ref{tf-Newtonian},
we can find the source terms $a {\cal B}_I(k,t)$ for the free-streaming
projection : eqns. \ref{efe2.1a} through \ref{efe2.3a} in Appendix
\ref{app-source-terms}. This applies to the case of instant decoupling.
During matter domination the dipole is negligible, so we ignore it.
The shear contribution is small on large scales, hence we can also
ignore it.
On substituting these equations into the flat almost-Friedmann-Lema\^{\i}tre integral solution
(\ref{int_soln_damping}) with $K=0$ in the source terms
(eqns. \ref{efe2.1a} through \ref{efe2.3a}) we can find the
free-streaming almost-Friedmann-Lema\^{\i}tre solutions for the temperature anisotropies
(\ref{freestreaming}).
Using $(\bar \Phi \rho_{\!_M}^{-1})^{\prime} \sim 0$ \footnote{
In fact one has that for an Eistein-de Sitter background :
$[\bar \Phi \rho_{\!_M}^{-1}](\eta_*) = - (3 H_0^2 \Omega_0)^{-1}
D_* [k^2 \Phi_A(k,0)]$ and $
(\Omega_0 D_*/a_*)^2 \approx \Omega_0^{1.54}$}
we find the CGI kinetic theory equivalent of the Sachs-Wolfe formula
for cosmic background radiation anisotropies in terms of matter
inhomogeneities at last scattering at various wavelengths, together
with an integral term. In various cases (see \cite{HS-astro}
for references to more general treatments), in particular the
matter-dominated spatially flat solutions
with only growing scalar modes, the integral terms vanish and we
obtain:
\begin{eqnarray}
{\frac{\tau _\ell^{SW}(\eta _0,k)
\beta _\ell}{(2\ell+1)}}\approx
\frac23 [\bar\Phi \rho_{\!_M}^{-1}](\eta_*) j_\ell(k\Delta \eta_* ).
\label{tau_l_SW}
\end{eqnarray}
This gives the (approximate)
projection of the large scale potential
inhomogeneities at last scattering onto the sky today;
the mean-squares $|\tau _\ell|^2$ can then constructed, using
the results from Paper I \cite{GE}.
The effect arise from the terms $\mbox{D}_a \ln T \approx \frac13 \mbox{D}_a \ln \rho
\approx \rho_{\!_M}^{-1} \mbox{D}^b E_{ab}$,
having used the adiabatic assumption.
This recovers the standard Sachs-Wolfe result \cite{SW,dunsby} on
large scales, where the potential fluctuations are just due to
primordial initial inhomogeneities that are unchanged by intervening
physics.
We discuss the origin of these fluctuations in later sections --
they are given by solving these equations before decoupling, which for
example implies the existence of acoustic oscillations. These potential
fluctuations are what seed structure formation through the production of
matter perturbations undergoing gravitational collapse beneath the Jeans
scale. The matter perturbations effectively decouple near
matter-radiation equality, making the large scale temperature
anisotropies the key link between the radiation anisotropies
now to the potential fluctuations then (near radiation decoupling), and so to
the matter power spectrum both on large and small scales today.
Notice that these equations will hold for any choice of 4-velocity that is
close to the matter 4-velocity, i.e. there is still frame-freedom
associated with this freedom of choice. As has been remarked various times,
there are several possible physical choices for this 4-velocity (which will
all agree at late times); the interpretation of the physical meaning of the
cosmic background radiation anisotropy sources will change depending
on this choice. The important difference about the derivation
of the Sachs-Wolfe effect here as opposed to other treatments is that
{\bf (i)} this result {\it is found in the total matter frame},
{\bf (ii)} the integration {\it is explicitly time-like},
so it is not treated mathematically as a projection along null rays
but rather as the evolution of the anisotropies of radiation
in a small comoving box, as explained in the introduction. Thus
the initial data here is not at the intersection of the past light
cone and the last scattering surface, but rather at the intersection
of the world line of the observer and the last scattering surface.
This analysis can be compared to primary anisotropy source term of the
gauge-invariant treatment used in \cite{HS95b}. The subtle difference between
the Bardeen variable gauge-invariant approach and the CGI approach used
here is that the Doppler source, which in their case arises through
${\cal B}_0$, now enters through the integral terms
only; there is no direct Doppler contribution at last scattering from
the first term in the integrated solution.
\section{Slow Decoupling}
To deal with slow-decoupling, we return to the general damped solution,
(\ref{int_soln_damping}), and introduce the slow decoupling approximation
\begin{equation}
\kappa^{\prime \prime} \ll1, ~~(\kappa^{\prime})^2 \ll 1\,,
\end{equation}
to find, on substituting in from the coefficient relations (\ref{C-B-coef}):
\begin{eqnarray}
{\tau_{\ell}(\eta_0) \beta_{\ell} \over (2 \ell +2)} &\simeq&
e^{-\kappa} \left[ {+ \frac{1}{k} [a {\cal B}_1](\eta_*)
+ \frac{1}{k^2} \frac53 [a{\cal B}_2]^{\prime}(\eta_*)
- \frac{1}{k} [a {\cal B}_2](\eta_*) {\partial \over \partial \eta_0} }
\right] j_{\ell}(k \Delta \eta_*) \nonumber \\
&\;& - \int_{\eta_*}^{\eta_0} d \eta' e^{-\kappa} \left\{
{ \frac56 a {\cal B}_2- \frac{1}{k}
(a {\cal B}_1)^{\prime} + \frac53 \frac{1}{k^2} (a {\cal
B}_2)^{\prime \prime} }\right\} j_{\ell}(k \Delta \eta) \nonumber \\
&\;& + \frac{1}{k}(\kappa^{\prime} e^{-\kappa}) {v_B}(\eta_*)
j_{\ell}(k \Delta \eta_*) + \int_{\eta_*}^{\eta_0}
d \eta' ( v_B^{\prime} \kappa^{\prime} + \kappa^{\prime \prime} v_B)
e^{-\kappa}) \frac{1}{k} j_{\ell}(k \Delta \eta) \nonumber \\
&\;& + \int_{\eta_*}^{\eta_0} d \eta' (\kappa' e^{-\kappa}) \left[ {
- \frac{1}{k} (a {\cal B}_1) + 2 \frac{1}{k^2} (a {\cal B}_2)
} \right] j_{\ell}(k \Delta \eta)\;.
\label{slowdec_1}
\end{eqnarray}
We see that damping effects are controlled by $e^{-\kappa}$
and $\kappa^{\prime} e^{-\kappa}$.
A further approximation would be to take
$e^{-\kappa} \gg \kappa^{\prime} e^{-\kappa}$,
so that we need only consider the free-streaming like
solutions, which we then convolve with the damping factor,
defined as a combination
of the visibility function and the diffusion damping envelope -- this
is done later using the damping envelope as derived from the dispersion
relations in section (\ref{visible}).
Later we will explicitly recover these equations in the Newtonian
frame of section (\ref{newtg}).
\subsection{Silk Damping}
Diffusion damping will occur and introduce a damping scale, the Silk
scale, giving a cut-off in the matter perturbations, and there will be
a corresponding diffusion damping effect in the photons. This is naturally
included in our general damped solutions in terms of the exponential envelope
implied by the equations, which can be demonstrated heuristically.
The cut-off arising through photon diffusion occurs when the
term involving $\kappa'$ in (\ref{modeibe}) dominates the other terms;
that is, when for any $\ell$, $k$ is large enough that
\begin{eqnarray}
k \left[ { {\frac{1}{4}\tau _{\ell+1}-\tau_{\ell-1}}} \right] \approx
\frac34 k \tau_{\ell} \ll \kappa^{\prime}\tau _\ell\;,
\label{modeibe2}
\end{eqnarray}
the approximation assuming that the damped modes are roughly of the same
magnitude (independent of $\ell$ when this condition is satisfied).
This then implies an exponential decay in the relevant modes:
\begin{equation}
k \ll \frac43 \kappa^{\prime} ~~\Rightarrow ~~
\tau _\ell^{\prime } \approx - \kappa ^{\prime }\tau _\ell~~
\Rightarrow ~~ \label{modeibe3}
\tau _\ell(\eta) \approx \exp(- \kappa ^{\prime}\eta)\tau _\ell(0)\;.
\end{equation}
Thus small scales will be heavily damped by this process and long wavelengths
unaffected, leading to a wavelength-dependent damping envelope.
The resulting cut-off in perturbation amplitude at a critical wavelength
at last scattering will result in a corresponding cut-off in cosmic
background radiation anisotropy amplitudes observed at a critical
angular scale.
A more detailed examination undertaken later will show the explicit
wavelength dependence of this cut-off effect.
\subsection{The Visibility Function} \label{visible}
An alternative approach to the slow decoupling solution
(\ref{slowdec_1}), is to argue that the dominant contribution during slow-decoupling arises from the visibility function defined by
${\cal V}(k,\eta) \approx \kappa' e^{-\kappa}$ as convolved with the
free-streaming integral solution. The visibility function gives the
probability of a photon last scattering during a small
time interval $d \eta$. From Hu \& Sugiyama \cite{HS95a} it is useful to
define the damping factor (now including diffusion damping, which will be
derived from the coupled baryon-photon equations
in (\ref{dif-damp-correct}) ):
\begin{eqnarray}
{\cal D}(\eta_0,k) = \int_{\eta_*}^{\eta_0} d \eta {\cal V}(\eta,k)
e^{-\left({k/k_D} \right)^2} \approx e^{-(k/k_D)^2}\;. \label{visibe}
\end{eqnarray}
The visibility function will model the changing ionization fraction,
this does not include the diffusion damping, which is added
in by hand above, through the damping scale. It should be realized here
that the Gaussian diffusion damping is {\it naturally included} in
the original almost-Friedmann-Lema\^{\i}tre integral solution (\ref{int_soln_damping}). However,
given that we will use solutions that are first-order
(in the scattering time) and then recover an explicit dispersion relation
for the damping scale, at second-order in the scattering time, it is
convenient to modify the damping factors such that they are re-written
in terms of the visibility function \ref{visibe}. The second order damping
scale of the form used here is explicitly derived in section
\ref{sssec-dispersion}.
Now, we can modify the free-streaming projection by including the damping
factor ${\cal D}$ and the baryon velocity effect (which must be
put in from (\ref{int_soln_damping})). In this approximation, we
can effectively drop the last two lines of (\ref{int_soln_damping})
except for the initial baryon velocity contribution, to obtain:
\begin{eqnarray}
{\frac{\tau _\ell(\eta _0)\beta _\ell}{(2\ell+1)}} &\simeq&
\left[ {C_2^{\prime}(\eta _*)-C_1(\eta _*)}\right] {\cal D}(\eta_0,k)
j_\ell(k\Delta \eta_* )
\nonumber \\ &+& kC_2(\eta _*) {\cal D}(\eta_0,k)
\left[ { {\ \frac \ell{(2\ell+1)}}j_{\ell-1}(k\Delta \eta_*)
- {\frac{(\ell+1)}{(2\ell+1)}}j_{\ell+1}(k\Delta \eta_*)}\right]
\nonumber \\
&+& \int_{\eta _*}^{\eta _0}d\eta {\cal V} e^{-(k/k_D)^2} \left\{
{C_0(\eta)-C_1^{\prime}(\eta)+
C_2^{\prime \prime }(\eta)}\right\} j_\ell(k \Delta \eta)\;,
\label{slowdecoupling}
\end{eqnarray}
where the coefficients $C_1$, $C_2$ and $C_3$ are given
by (\ref{C-B-coef}). This has the effect of taking the previous
more general solution (\ref{int_soln_damping}) and specializing it
to the most important regime as far as decoupling is concerned,
thus giving a major improvement on the sharp decoupling approximation,
while avoiding the complications of the complete integral solution
given above. A similar correction is made using the visibility
function in the integrated part of the solution, in order to best
deal with a changing ionization fraction, given that we will
once again only be using almost-Friedmann-Lema\^{\i}tre solutions, that are either
first order or zero-th order in the scattering times (discussed below).
\subsection{Slow Decoupling in the Conformal Newtonian Frame}
Here we cast the above derived solutions (\ref{freestreaming},\ref{slowdec_1},
\ref{slowdecoupling}) based on the integral solution (\ref{int_soln_damping})
into the CGI Newtonian frame (based on the shear-free frame described
in section \ref{newtg}) for the case of scalar perturbations, in terms
of the Bardeen like scalar potentials $\Phi_H$ and $\Phi_A$.
The vanishing shear condition ${\tilde \sigma}_{ab} \approx 0$ implies
that $\tilde C_0 (\eta)\approx 0$ and $ {\tilde C}_2(\eta)\approx 0$,
hence we can use these conditions directly to find the slow-decoupling
anisotropy solution for an almost-Friedmann-Lema\^{\i}tre model in the Newtonian frame:
\begin{eqnarray}
{\frac{\tau _\ell(\eta _0)\beta _\ell}{(2\ell+1)}} &\simeq&
-{{\tilde C}_1(\eta _*)} {\cal D}(\eta_0,k)
j_\ell(k\Delta \eta_* )
- \int_{\eta _*}^{\eta _0}d\eta {\cal V} e^{-(k/k_D)^2}
{{\tilde C}_1^{\prime}(\eta)} j_\ell(k \Delta \eta)\;.
\label{slowdecoupling-ng-0}
\end{eqnarray}
Using the results from section (\ref{newtg}) and the almost-Friedmann-Lema\^{\i}tre relations
(given in Appendix G), it can be shown that the key quantity of interest
${\tilde B}_a \approx {\tilde {\mbox{D}}}_a \ln T +\mbox{D}_a \Phi_A$, in the case of
scalar perturbations, obeys the following relation in the Newtonian frame.
\begin{eqnarray}
D_{\langle a} {\dot{\tilde {\cal B}}}_{b \rangle} \approx\mbox{D}_{\langle a}\mbox{D}_{b \rangle}
(\dot{\Phi}_A - \dot{\Phi}_H) - 2 H\mbox{D}_{\langle a}\mbox{D}_{b \rangle} \Phi_A
- \frac13\mbox{D}_{\langle a}\mbox{D}_{b \rangle} (\mbox{D}^c {\tilde \tau}_c)
\label{ng-dot-B}
\end{eqnarray}
Now we need to find the mode coefficients for ${\cal B}$ in terms
of the quantities defined in the Newtonian frame. Writing
\begin{equation}
{\tilde {\mbox{D}}}_a \ln T \approx \frac{k}{a} \delta \tilde T Q_a,
~~\mbox{D}_a \Phi_A \approx \ts{k \over a} \Phi_A Q_a,
~~ \mbox{and}~~ D_a \Phi_H \approx \ts{k \over a} \Phi_H Q_a\;,
\end{equation}
we find that
\begin{eqnarray}
{\tilde {\cal B}}_a \approx \ts{k \over a}(\delta \tilde T + \Phi_A ) Q_a,
~~\Rightarrow~~
{\tilde{\cal B}}_1 \approx \ts{k \over a}(\delta \tilde T + \Phi_A)\;.
\label{calb-ng}
\end{eqnarray}
On mode expanding (\ref{ng-dot-B})
{\footnote{We will use $\mbox{D}_{\langle a} \mbox{D}_{b \rangle} \dot{\Phi}_A \approx
- \dot{\Phi}_A \mbox{D}_{\langle a} \mbox{D}_{b \rangle} Q$, $\mbox{D}^a \tau_c \approx
+ \ts{k \over a} \tau_1 Q$ and $\mbox{D}_{\langle a} {\dot{\cal B}}_{b \rangle} \approx
- \ts{a \over k}(\dot{\cal B}_1 + 2 H {\cal B}_1) \mbox{D}_{\langle a} \mbox{D}_{b \rangle} Q$.}}
and transforming to the conformal time derivative we obtain:
\begin{eqnarray}
(a {\tilde {\cal B}}_1)^{\prime} \approx - a^2 H {\tilde {\cal B}}_1
+ k (\Phi_A^{\prime} - \Phi_H^{\prime}) - 2 H a k \Phi_A + \ts\frac13 k^2
{\tilde \tau}_1.
\label{calb-dot-ng}
\end{eqnarray}
Now using (\ref{C-B-coef}) we find that:
\begin{eqnarray}
- {\tilde C}_1^{\prime}(\eta) - (\kappa' {\tilde v}_B)^{\prime} \approx +
\frac{1}{k} (a {\tilde{\cal B}}_1)^{\prime}. \label{calc-dot-ng-01}
\end{eqnarray}
We can now put this all together, first, from (\ref{C-B-coef})
and (\ref{calb-ng}) to find:
\begin{eqnarray}
- {\tilde C}_1(\eta) \approx \kappa' {\tilde v}_B + (\delta \tilde T + \Phi_A).
\label{calc-ng}
\end{eqnarray}
and second from (\ref{calc-dot-ng-01}), (\ref{calc-ng}) and
(\ref{calb-dot-ng}) to find:
\begin{eqnarray}
- {\tilde C}_1^{\prime}(\eta) \approx (\kappa^{\prime}
{\tilde v}_B)^{\prime} + \left( {(\Phi_A^{\prime} - \Phi_H^{\prime})
- a H (\delta \tilde T + \Phi_A) - 2 a H \Phi_A
+ \ts\frac13 k {\tilde \tau}_1} \right)\;.
\end{eqnarray}
Substituting these results into the integral solution
(\ref{slowdecoupling-ng-0}) we obtain:
\begin{eqnarray}
{\frac{ \tau _\ell(\eta _0) \beta _\ell}{(2\ell+1)}} &\approx&
\left[ {(\delta \tilde T + \Phi_A) + \kappa^{\prime} {\tilde v}_B} \right](\eta_*,k)
{\cal D}(\eta_0,k) j_\ell(k\Delta \eta_*) \nonumber \\
&+& \int_{\eta_*}^{\eta_0} d \eta {\cal V} e^{- (k/k_D)^2}
\left\{ {(\kappa^{\prime} \tilde v_B^{\prime} + \kappa^{\prime \prime} \tilde
v_B) + \frac13 k {\tilde \tau}_1 } \right\} j_{\ell}(k \Delta \eta) \nonumber \\
&+& \int_{\eta _*}^{\eta _0} d \eta {\cal V} e^{-(k/k_D)^2}
\left\{ {(\Phi_A^{\prime} - \Phi_H^{\prime}) - a H ( \delta \tilde T + 3\Phi_A )
}\right\} j_\ell(k \Delta \eta)\;.
\label{slowdecoupling-ng}
\end{eqnarray}
Here the second order terms (both in terms of the scattering time
and in the almost-Friedmann-Lema\^{\i}tre sense) have been dropped. We can then pull
out the canonical solution when we ignore the Doppler contribution,
the initial baryon relative velocity at last scattering
(it is tightly coupled to the radiation velocity and is
thus small already). We also ignore the intermediate scale
integrated effect which contributes to the early-integrated Sachs
Wolfe effect. The result is:
\begin{eqnarray}
{\frac{\tau _\ell(\eta_0)\beta _\ell}{(2\ell+1)}}\approx
\left[\delta \tilde T + \Phi_A\right] {\cal D}(\eta_0,k)
j_\ell(k\Delta \eta_*) + \int_{\eta _*}^{\eta _0} d \eta {\cal V}
e^{-(k/k_D)^2} \left[{\Phi_A^{\prime} - \Phi_H^{\prime}}\right]
j_\ell(k \Delta \eta).~~~
\label{canon-temp}
\end{eqnarray}
This completes the recovery of the standard integral solution
results using the 1+3 CGI approach at linear order -- we have
notationally suppressed the $k$-dependence of the temperature
anisotropy ($\tau_{\ell}(\eta_0) \equiv \tau_{\ell}(\eta_0,k)$).
It corroborates the standard anisotropy derivations based on a
3+1 hypersurface foliation, which uses the Bardeen formalism
in the conformal Newtonian gauge.
\section{Late Tight-coupling}
Here we extend the Thomson scattering analysis of the previous
sections to include a simple model of late-tight coupling and hence
of fast decoupling.
We aim to reproduce in covariant form the Peebles and Yu
{\it near-tight} coupling \cite{PY} and Hu and Sugiyama
{\it tight-coupling approximation} \cite{HS95b} treatments, valid for
the period of late tight-coupling, up to and including decoupling
\footnote{Here we are explicitly making a distinction between the
treatment \cite{HS95b} (what we call {\it tight-coupling approximation})
and that in \cite{PY} (what we call {\it near-tight} coupling.
By {tight-coupling approximation} we mean that
$\dot \kappa^{-1}$ is sufficiently small that it can be ignored
(inducing a contribution of the order of magnitude (say) of at
least $10^{-6}$) when multiplying quantities of linear
order such as the shear) The near-tight coupling includes the
radiation quadrupole in the case of isotropic Thompson scattering
(in the matter frame).}. Remember that we are ignoring the anisotropic
and polarization effects as these can be corrected at the level of the
damping scale.
\subsection{Integrated Boltzmann Equation: Near-tight Coupling}
Here the almost-Friedmann-Lema\^{\i}tre integrated Boltzmann equations is used to construct a set
of multipole divergence equations that describe the radiation near
tight-coupling. These are the intermediate scale equations, valid in the
tight-coupling era. This is done by carrying out a CGI version of
perturbation theory in terms of the scattering time.
\subsubsection{The Scattering-strength Expansion}
The solutions we have considered so far are linearised through a
small-parameter expansion in terms of the anisotropy parameter $\tau$. The
basic idea now, following the method of Peebles and Yu \cite{PY}, is that
additionally a second expansion is constructed in terms of the collision
parameter $t_c=({\sigma _Tn_e})^{-1}$, without truncating the Boltzmann
hierarchy at the order of the calculation, and thus avoiding the problems
inherent in exact truncation \cite{ETMa} (see Appendix E). We thus
find the evolution equations for the energy density, momentum flux, and the
anisotropic flux of the radiation close to tight coupling.
Consider the almost-Friedmann-Lema\^{\i}tre integrated Boltzmann equations (\ref{li_ibe}) and (\ref{IBE-aflrw}) for
isotropic (in the baryon frame) Thompson scattering (\ref{IBE-source})
\cite{MGE}; this is inverted to find :
\begin{eqnarray}
\tau (x^i,e^a)=v^a_B e_a- t_c \left[ {{\cal B}+\dot{\tau}
+e^a\mbox{D}_a\tau }\right]\;.
\label{pert_01}
\end{eqnarray}
We now systematically approximate (\ref{pert_01}) in terms of the
smallness parameter $t_c$. The right-hand side (the scattering term) is
used to find the zero-th order collision-time correction to the total
bolometric temperature, with corresponding temperature anisotropy
given by
\begin{eqnarray}
\tau _{(0)}(x^i,e^a)\approx v^\alpha_B e_\alpha\;.
\end{eqnarray}
The equation is now perturbed about the zero order velocity
perturbation and one can then recover the first and second order
corrections in $t_c$ to the zero-th order temperature anisotropies,
{\it {\ }}to find, $\tau _{(1)}$ and $\tau _{(2)}$ respectively,
where the n-th. order correction is denoted by $\tau _{(n)}.$
We obtain an almost-Friedmann-Lema\^{\i}tre perturbative expansion in $t_c$ :
\begin{eqnarray}
\tau _{(n)}\approx v^a _B e_a-t_c\left[ {{\cal B}+\dot{\tau}_
{(n-1)}+e^a\mbox{D}_a\tau _{(n-1)}}\right]\;.
\label{pert_02}
\end{eqnarray}
The tight-coupling limit is recovered when $t _c=0$. This
treatment is then a consistent (in the sense of the truncation conditions
described in Appendix E) near-to-tight-coupling treatment
in almost-Friedmann-Lema\^{\i}tre universes. (The first, in $n$, three temperature
anisotropies, are given in Appendix C).
\subsubsection{Solid Angle Integration}
Now the temperature anisotropy is integrated over the solid sphere to ensure
the condition that there is no contribution to the bolometric average
$T_{(b)}(x^i)$,
\begin{eqnarray}
\int_{4\pi }\tau (x^i,e^a)d\Omega =0\;.
\label{int_cond_I}
\end{eqnarray}
It should be clear why the second order correction to the temperature
anisotropy is needed even though we intend to keep the expansion only to
first order in $t_c$; the integrations over term the $e^av_a$ will vanish.
Now by integrating $\tau _{(2)}$ (\ref{temp_2nd}) over the solid angle
and using (\ref{int_cond_I}) and orthogonality of $O^{A_\ell}$ the gradient
of the radiation flux is found:
\begin{eqnarray}
\mbox{D}_a\tau ^a\simeq \mbox{D}_av^a_B-\alpha _c\left[ {(\mbox{D}_av^a_B)^{\dot{}}
-(\mbox{D}_a\tau ^a)^{\dot{}}+ \ts{1 \over 4}{ \mbox{D}^2 (\ln \rho_R)}
+(\mbox{D}_a\dot{v}^a_B)+(\mbox{D}_aA^a)}\right]\;.
\label{monopole-gradient}
\end{eqnarray}
By taking spatial gradients of the radiation flux (\ref{radf_0013}) we
find on comparing with (\ref{monopole-gradient}), that in order for
there to be no contributions to scalars,
\begin{eqnarray}
(\mbox{div}\, v)_B^{\dot{}}\approx (\mbox{div}\, \tau)^{\dot{}}\;.
\end{eqnarray}
The above equation (\ref{monopole-gradient}) then becomes
\begin{eqnarray}
(\mbox{div}\, \tau) \simeq (\mbox{div}\, v)_B -t _c\left[ {\ts{1 \over 4}} {{\mbox{D}^2 \ln \rho_R} +
(\mbox{D}_a \dot{v}^a_B)+(\mbox{div}\, A) }\right]\;.
\label{mon-gradient}
\end{eqnarray}
\subsubsection{The Transport Equations}
Finally the individual PSTF multipoles are recovered at a given order
\begin{eqnarray}
\tau _{A_\ell}=\Delta _\ell^{-1}\int_{4\pi }O_{A_\ell}\tau _{(n)}
(x^i,e^a)d\Omega. \label{int_cond_II}
\end{eqnarray}
Here $\Delta _\ell$ is defined as before in Paper I \cite{GE}.
The second order temperature multipoles are now found from (\ref
{int_cond_II}) and integrating $\tau_{(2)}(x,e)$ (\ref{temp_1st}) after
the combination of direction vectors has been replaced by
PSTF tensors (one can use \ref{pert_001}-\ref{pert_003}) :
\begin{eqnarray}
\tau ^b &\approx &v^b_B-t _c\left[ { {\mbox{D}^b {\ln T}}
+A^b+\dot{v}^b_B }\right] + t_c^2 \left[{ ( \mbox{D}^b \ln T + A^b){}^{\dot{}}
+\ddot{v}^b_B - \ts{1 \over 3} \mbox{D}^b \mbox{D}_c \tau^c} \right]\;,
\label{radf_0013} \\
\tau ^{ab} &\approx &-t _c\left[ {\sigma ^{ab}+\mbox{D}^{\langle a}v^{b \rangle}_B}\right]
+ t_c^2 \left[{(\mbox{D}^{\langle a} v^{b \rangle}_B)^{\dot{}} + \mbox{D}^{\langle a}
( \mbox{D}^{b \rangle} \ln T + A^{b \rangle}) + \mbox{D}^{\langle a} \dot{v}^{b \rangle}_B
+ \dot{\sigma}^{ab}}\right]\;,
\label{radf_0014} \\
\tau ^{abc} &\approx& +t_c^2 \left[ { \mbox{D}^{\langle ab} v^{c \rangle}_B } \right]\;, \label{radf_0014b}\\
\tau ^{A_\ell} &\approx &0~~~\forall ~\ell>3\;,
\label{radf_0015}
\end{eqnarray}
where we have dropped terms of ${\cal O}(t_c^3)$.
These are the key results of this section. They are the appropriate
transport equations for the late-tight coupling era, i.e. up to last
scattering, and are essentially equivalent to the
{\it causal transport equations} given by causal relativistic
thermodynamics \cite{RT}.
What we have shown here is that if we are interested in the behaviour
of the photon-baryon systems to first order in the scattering time, a
dissipative fluid approximation is sufficient to describe
the radiation (cf. the papers by Israel and Stewart \cite{Is}), and will
not result in an explicit truncation of the Boltzmann multipole hierarchy,
rather it gives a systematic approximation scheme where we can, to the
appropriate accuracy, ignore the third order and higher terms. This is
significant; one cannot merely drop the higher order moments and
truncate to a fluid description, as the kinetic theory treatment fixes the
transport equations. Here we have consistently decoupled the $l<3$ multipole
equations from the rest of the hierarchy and the consistency of this
decoupling is maintained through (\ref{monopole-gradient})
and (\ref{radf_0013} - \ref{radf_0015}).
The solutions to these equations, which lead to acoustic oscillations
during this period, will then affect the cosmic background radiation
anisotropies by setting initial conditions for the free-streaming solution discussed in the previous section. We give a derivation of these results in the
following section.
\subsection{Late Tight-coupling and the Oscillator Equation}
Here we derive the CGI equivalent of the analytic tight-coupling
approximation used by Hu \& Sugiyama \cite{HS95a,HS95b}. This approach
uses the tight-coupling limit
in order to get rid of the radiation quadrupole during late tight-coupling
\footnote{If there are any anisotropic contributions such as anisotropic
scattering (in the matter frame) or large shear (from gravitational waves)
at that time this sort of approximation should be considered with care -
such phenomena would break tight-coupling.} and covariantly
reproduces Hu and Sugiyama's conclusions about the cosmic background radiation anisotropy due to inhomogeneities, acoustic and Doppler sources (what
they call `primary sources'). This gives the `Sachs-Wolfe effect' due to
the Newtonian potential near last scattering, but not the
`integrated Sachs-Wolfe effect' due to changing potentials after
tight coupling (resulting from more complex matter models and/or spatial
curvature).
\subsubsection{Near Tight-coupling Equations}
We start with the near-tight coupling equations (\ref{radf_0013},
\ref{radf_0014}) and (\ref{radf_0015}) except rewritten to first order and
in terms of the optical depth so as to be closer to the notation of the
better known treatments \cite{HS95b}:
\begin{eqnarray}
\tau _a &\simeq &v_a^B-\dot{\kappa}^{-1}\left[ {\mbox{D}_a(\ln T)+ A_a +\dot{v}_a^B}
\right]\;, \label{L008} \\
\tau _{ab} &\simeq &-\dot{\kappa}^{-1}\left[ {\sigma _{ab}+\mbox{D}_{\langle a}
v_{b \rangle}^B}\right]\;, \label{L0081} \\
\tau ^{A_\ell} &\simeq &0~~\forall \ell > 2\;. \label{L0082}
\end{eqnarray}
Here we have assumed that the collisions are dominated by
Thompson scattering and is therefore isotropic in the matter frame.
The relative velocity of the matter with respect to the
preferred reference frame
is $v_B^a \simeq u^a_B - u^a$.
Rewriting (\ref{L0081}) in terms of the shear of the baryon frame,
we have $\tau_{ab} \simeq -\dot \kappa^{-1} \mbox{D}_{\langle a} u^B_{b\rangle }$,
so the quadrupole is given entirely by the shear of the matter.
Notice that $\pi_{ab} = \rho_R \tau_{ab} \approx 0$, as the case of
matter domination. This condition is not sufficient to ensure
that $\tau_{ab}$ can be ignored in equations when it appears
on its own, even though the quadrupole is small. The key point, which
was discussed in section 2.5, is that there are four principal
linearisations: the almost-Friedmann-Lema\^{\i}tre one at least $O(\epsilon^2)$,
the almost-Friedmann-Lema\^{\i}tre radiation isotropy one $O(\epsilon \eta)$, $O(\eta^2)$,
(implying the previous by the almost-Ehlers-Geren-Sachs theorem),
the non-relativistic assumption $O(v \eta)$, $O(v^2)$,
$O(v \epsilon)$, and the linearisation scheme based on the differential
optical depth. Hence care must be taken when approximations are made
to the equations.
\subsubsection{Tight coupling: Momentum Equations}
The tight-coupling calculation is now carried out, assuming
(\ref{L008}-\ref{L0082}) hold. Consider once again the
radiation energy and momentum conservation equations
($\ell=0$ and $\ell=1$ multipole divergence equations):
\begin{eqnarray}
(\ln T)^{\dot{}}+\frac 13\Theta &\simeq & -\frac13 \mbox{div}\, \tau\;,
\label{L001} \\
-\dot{\tau}_a &\simeq &A_a+\mbox{D}_a(\ln T)-\dot{\kappa}(v_a^B -\tau _a) +
\frac25 \mbox{D}^c \tau_{ac}\;,
\label{L002}
\end{eqnarray}
and the baryon energy and momentum conservation equations
\footnote{This is found to O[1] by substituting the matter energy
conservation equation (dust part of \ref{EFE-ec}) into
matter momentum equation (dust part of \ref{EFE-mc}) all to O[1].}
\begin{eqnarray}
(\ln \rho )^{\dot{}}+\Theta &\simeq & - \mbox{D}^a v_a^B, \label{L003} \\
-{\dot{v}_a^B} &\simeq & + H v_a^B +A_a+R^{-1}\dot{\kappa}
(v_a^B-\tau _a)\;.
\label{L004}
\end{eqnarray}
Here $\dot{\kappa}$ is the optical depth, and the radiation-baryon
ratio in the real universe is given by (using the enthalpy
$h= \rho+p$) by
\begin{eqnarray}
R(x^i)= {h_B(x) \over h_R(x)} = {\rho_M + p_B \over \rho_R + p_R}
\approx \frac 34{\frac{\rho (x^i)}{\rho_R (x^i)}}\Rightarrow
\dot{R}\simeq H R\;.
\end{eqnarray}
This is related to the speed of sound
\footnote{By speed of sound we mean adiabatic sound speed :
$c_s^2 = {\dot{p}_0 \over \dot{\rho}_M + \dot{\rho}_R}$
$= \frac13 {\dot{\mu}_0 \over (\dot{\rho}_M + \dot{\rho}_R)}$
$= \frac13 {1 \over (\dot{\rho}_M/\dot{\rho}_R) +1}$
$= \frac13 {1 \over {(3 \rho_M / 4 \rho_R)} +1}$, for matter
domination $c_s \approx 0$. } in the background via $c_s^2 = (1 /
3(R_0+1))$. The matter momentum equations, (\ref{L004}), give
\begin{equation}
v_a^B\simeq \tau _a-\frac R{\dot{\kappa}}\left[ {\dot{v}_a^B + H v_a^B
+A_a}\right]\;. \label{L005}
\end{equation}
Substituting (\ref{L008}) into (\ref{L005}) and retaining all
terms up to linear order (in the relaxation time) we obtain
\begin{equation}
v_a^B\simeq \tau_a - R \dot{\kappa}^{-1} [\dot{\tau_a} + A_a + H
\tau^a] + {\cal O}( \dot \kappa^{-2})\;. \label{L005a}
\end{equation}
This is then substituted into (\ref{L002}) in order
to remove the velocity terms, and with a little algebra, we find
\begin{eqnarray}
- \dot \tau_a \simeq \dot u_a + {1 \over 1+R} \mbox{D}_a \ln T + {\dot{R} \over
(1+R)} \tau_a
- \frac25 {\dot \kappa^{-1} \over (1+R)} \mbox{D}^b \mbox{D}_{\langle a} u^m_{b\rangle }\,,
\end{eqnarray}
where the last term has been written in terms of the matter shear.
We can now consider the situation where $\dot \kappa^{-1}$ becomes
sufficiently small (but non-zero) so that the last term can be ignored.
This is possible as the matter shear is already at least first order.
We then find ({\it cf} \cite{HS95b} eqn. (B2 b)):
\begin{eqnarray}
\dot{\tau}_a + {\frac{\dot{R}}{(1+R)}}\tau _a + {\frac 1{(1+R)}}\mbox{D}_a(\ln
T) \simeq -A_a\;. \label{mom-flux}
\end{eqnarray}
This is the momentum flux equation for the radiation and is a key result.
It can be rewritten as
\begin{equation}
\lbrack (1+R)\tau _a]^{\dot{}}+\mbox{D}_a(\ln T)\simeq -(1+R)A_a\;,
\label{L009a}
\end{equation}
or on taking its divergence as
\begin{eqnarray}
[ a (1+R) (\mbox{D}^a \tau_a) ]^{\cdot} + a (\mbox{D}^2 \ln T) \simeq - (1+R) a
(\mbox{D}^a A_a)\;. \label{L009ab}
\end{eqnarray}
\subsubsection{Spatial Gradients and the Oscillator
Equation on Small Scales}
The basis of this derivation is the `small-scale' assumption which
effectively means that on small enough scales we can ignore
the expansion (see section 2.5). This is just the statement
that the scale of inhomogeneity is less than the Hubble scale
(\ref{small_scale}), so we drop all terms of
$O(\epsilon \epsilon_H)$ (see section 2.5.4).
Our aim is to recover the standard oscillator equation
(the source equation for the acoustic oscillations)
using the 1+3 CGI formalism. Note however that we still have
the freedom to set the relative velocity in the small boost
equations (which we will do in the next section).
Taking the spatial gradient\footnote{Using the identity
$(\mbox{D}_a f)^{\cdot} \simeq \mbox{D}_a \dot{f} - H \mbox{D}_a f + A_a \dot{f}$
we find that $(\mbox{D}_a \ln T)^{\cdot}
\simeq \mbox{D}_a \dot{\ln T} - H (\mbox{D}_a \ln T + A_a)$ from the
almost-Friedmann-Lema\^{\i}tre $\ell=0$ multipole divergence equations and $H = {\dot{a}
\over a}$ \cite{MGE}.} of the radiation energy
conservation equation, (\ref{L001}), we find
\begin{equation}
- \frac 13\mbox{D}_a(\mbox{D}_c\tau ^c)\simeq (\mbox{D}_a\ln T)^{\dot{}}+\frac 13\mbox{D}_a
\Theta + H (\mbox{D}_a \ln T + A_a)\;, \label{L009}
\end{equation}
and taking the divergence of the resulting equation above gives
\begin{eqnarray}
- \frac13 (\mbox{D}^2 (\mbox{D}_c \tau^c) \simeq (\mbox{D}^2 \ln T)^{\cdot} + 2 H (\mbox{D}^2 \ln T)
+\frac13 \mbox{D}^2 \Theta + H (\mbox{div}\, A)\;,
\end{eqnarray}
and this can be written as
\begin{equation}
- \frac 13 ( a^2 \mbox{D}^2 (\mbox{D}_c \tau^c)) \simeq (a^2 \mbox{D}^2 \ln T)^{\cdot}
+\frac 13(a^2 \mbox{D}^2 \Theta ) + H (a^2 \mbox{div}\, A)\;.
\label{L010}
\end{equation}
This is analogous to equation (B3) in \cite{HS95b}. Then using
(\ref{L009ab}) we find
\begin{equation}
\lbrack (1+R) a \mbox{D}_c(\mbox{D}_a\tau ^a)]^{\dot{}}+ H (1+R) a \mbox{D}_c(\mbox{D}^a\tau
_a)+a\mbox{D}_c(\mbox{D}^2 \ln T)\simeq - (1+R)a \mbox{D}_c (\mbox{div}\, A). \label{L011}
\end{equation}
Substituting (\ref{L009}) into (\ref{L011}) and using the fact that
$H \mbox{D}^a A_a \approx {\cal O}(\epsilon_H \epsilon)$, we obtain
\begin{eqnarray}
&- 3[a(1+R)(a\mbox{D}_c\ln T)^{\dot{}}~]^{\dot{}}+a^2\mbox{D}_c(\mbox{D}^2 \ln T) \nonumber
\\ &~~~~~~~~~~~~~~~~\simeq
[a(1+R)(a\mbox{D}_c\Theta )]^{\dot{}} - a(1+R)\mbox{D}_c(\mbox{div}\, A)\;.
\end{eqnarray}
Finally, transforming to conformal time, $dt=ad\eta$ gives
\begin{eqnarray}
&3[(1+R)(a\mbox{D}_c\ln T)^{\prime }]^{\prime} - a^2\mbox{D}_c(\mbox{D}^a(a\mbox{D}_a\ln T)) \nonumber
\\ &~~~~~~~~~~~~~~~~\simeq - [a(1+R)(a\mbox{D}_c\Theta )]^{\prime}
+ a^2(1+R)\mbox{D}_c(\mbox{div}\, A)\;.
\end{eqnarray}
On using the small-scale linearisation scheme described in section
(\ref{small_scale}) we find, on dividing through by $3(1+R)$, the
oscillator equation:
\begin{equation}
(a\mbox{D}_c\ln T)^{\prime \prime } \approx
+ {\frac{a^3}{3(1+R)}}\mbox{D}_c(\mbox{D}^2 \ln T)
- a^2 (\mbox{D}_c \Theta)^{\prime}
+ {\frac{a^2}3}\mbox{D}_c(\mbox{D}^a A_a)\;.
\label{acoustic-source}
\end{equation}
This is the covariant harmonic-oscillator equation which
describes the acoustic modes \footnote{This can also be
obtained from a two-fluid CGI description \cite{dunsby},
as well as from the imperfect-fluid description \cite{RT}
-- the point here is that we have derived it from a self
consistent kinetic theory approach, listing along the way the
necessary physical approximation required to reduce it to the
standard acoustic oscillator form.}. We can compare it to the usual
gauge invariant result found in the Newtonian gauge by transforming
to the shear-free frame $\tilde u_a=n^a$ where $\mbox{D}_{\langle a} n_{b \rangle} =0$.
We will investigate this equation in more detail
in the next section and relate our results to those in the
standard literature (which are expressed in the Newtonian gauge).
\subsubsection{The Newtonian Frame Oscillator Equation}
In this section we recover the harmonic oscillator
equation of Hu and Sugiyama in the Newtonian gauge \cite{HS95a}
from the CGI formalism.
The difference between this and the previous section is that
here we apply the small scale approximation at the very end of
the calculation.
Using the Newtonian frame choice $n^a \approx u^a + v^a_N$,
$\mbox{D}_{\langle a} n_{b \rangle} =0$, equations (\ref{mom-flux}) and
(\ref{mono_temp}) become
\begin{eqnarray}
\dot{\tilde \tau}_a + {\dot{R} \over 1+R} {\tilde \tau}_a + {1 \over 1+R}
{\tilde D}_a (\ln T) \approx - {\tilde A}_a \approx -\mbox{D}_a \Phi_A\;,
\label{ng-dot-tau}\\
( {\tilde D}_a \ln T )^{\dot{}} + H ({\tilde D}_a \ln T + {\tilde A}_a)
+ \frac13 {\tilde D}_a {\tilde \Theta} \approx - \frac13\mbox{D}_a (D_c {\tilde
\tau}^c)\;. \label{ng-pertT}
\end{eqnarray}
These can be re-written and put into the following form by using the
almost-Friedmann-Lema\^{\i}tre Einstein field equations, together with the transformation
relations given in Appendix G:
\begin{eqnarray}
(D_a {\tilde \tau}^a)^{\dot{}} + H (\mbox{D}_a {\tilde \tau}^a)
&\approx& - {\dot{R} \over 1+R} (\mbox{D}_a {\tilde \tau}^a) - {1 \over 1+R}
({\tilde D}^2 \ln T) - (D^2 \Phi_A)\;, \label{D-ng-dot-tau}\\
D_{\langle a} {\tilde D}_{ b \rangle} (\ln T)^{\dot{}} &\approx&
-\mbox{D}_{\langle a}\mbox{D}_{b \rangle} \dot{\Phi}_H
- \frac13\mbox{D}_{\langle a}\mbox{D}_{b \rangle} (D_c {\tilde \tau}^c)\;.
\label{pstf-ng-pertT}
\end{eqnarray}
Taking the time derivative of (\ref{pstf-ng-pertT}) and substituting
into equation (\ref{D-ng-dot-tau}) after first taking PSTF derivatives
we obtain the full equation for $\mbox{D}_{\langle a} \mbox{D}_{b \rangle} \ln T$:
\begin{eqnarray}
( \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \ln T)^{\ddot{}} &+& \dot{H} ( \mbox{D}_{\langle a} \mbox{D}_{b \rangle}
\Phi_A + 2 \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \ln T) + H \left[{ ( \mbox{D}_{\langle a} \mbox{D}_{b \rangle}
\Phi_A)^{\dot{}} + 2 (\mbox{D}_{\langle a} \mbox{D}_{b \rangle} \ln T)^{\dot{}} } \right]
\nonumber \\
&\approx& - (D_{\langle a}\mbox{D}_{b \rangle} \Phi_H)^{\ddot{}} - 2 \dot{H} \mbox{D}_{\langle a}
\mbox{D}_{b \rangle} \Phi_H - 2 H ( \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \Phi)^{\dot{}}
+ {1 \over 1+R} \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \mbox{D}^2 \ln T\nonumber\\
&-& \mbox{D}_{\langle a} \mbox{D}_{b \rangle} D^2 \Phi_A
+ \left[ {{\dot{R} \over (1+R)} - 3 H} \right]
\left( {\mbox{D}_{\langle a}\mbox{D}_{b \rangle} (\ln T)^{\dot{}}
+ \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \dot{\Phi}_H } \right)\;.
\end{eqnarray}
On dropping all terms $O(\epsilon \epsilon_H)$, we once again obtain the 1+3
covariant form of the small scale Newtonian frame oscillator equation
(without using a mode expansion):
\begin{eqnarray}
( \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \ln T)^{\ddot{}}&+&{\dot{R} \over (1+R)}
\mbox{D}_{\langle a}\mbox{D}_{b \rangle} (\ln T)^{\dot{}}
+ {1 \over 1+R} \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \mbox{D}^2 \ln T \nonumber \\
&\approx& - (D_{\langle a}\mbox{D}_{b \rangle} \Phi_H)^{\ddot{}} +
{{\dot{R} \over (1+R)}}{\mbox{D}_{\langle a} \mbox{D}_{b \rangle} \dot{\Phi}_H }
- \mbox{D}_{\langle a} \mbox{D}_{b \rangle} \mbox{D}^2 \Phi_A\;.
\end{eqnarray}
The techniques used to derive the above equation become useful latter
when dealing with the non-linear terms as they avoid the complication
of mode-mode couplings when understanding the qualitative features
of various effects \cite{MGE}.
Upon using the mode decomposition definitions for the temperature
perturbations, radiation dipole and
the scalar Newtonian and curvature perturbations respectively
{\footnote{We use, as before, $
{\tilde D}_a \ln T \approx \ts\frac{k}{a} \delta {\tilde T} Q_a$ ,
${\tilde \tau}_a \approx {\tilde \tau}_1 Q_a$ ,
$D_a \Phi_A =\ts\frac{k}{a} \Phi_A Q_a$ and
$\mbox{D}_a\Phi_H \approx \ts{k \over a} \Phi_H Q_a$. }},
we can write the mode decomposition of (\ref{D-ng-dot-tau}) and
(\ref{pstf-ng-pertT}) as:
\begin{eqnarray}
\dot{\tilde \tau}_1 &\approx& - {\dot{R} \over 1+R} {\tilde \tau}_1
- {1 \over 1+R} \frac{k}{a} \delta {\tilde T} - {\frac{k}{a}} \Phi_A,
\label{mode-dipole} \\
\delta \dot{\tilde T} &\approx& - H \Phi_A - \dot{\Phi}_H
+ \frac{k}{a} \frac13 {\tilde \tau}_1\;.
\label{mode-mono}
\end{eqnarray}
Upon ignoring the expansion coupled term (i.e the small scale
approximation) and including the curvature fluctuations, since this
makes the resulting equations applicable up to the
Jeans length (above which the matter would not be gravitationally
bound), and substituting the
second equation (\ref{mode-mono}) into (\ref{mode-dipole}) we
obtain the well know equation describing the acoustic oscillations in
the radiation \cite{HS95a}:
\begin{eqnarray}
\delta {\tilde T}'' + {R' \over 1+R} \delta {\tilde T}' + k^2 c_S^2 \delta {\tilde T}
\approx - \Phi_H'' -
{R' \over 1+R} \Phi'_H - {k^2 \over 3} \Phi_A\;. \label{temp_osc_ng}
\end{eqnarray}
Here we have used conformal time (since we are now working in the
conformal Newtonian frame).
\subsubsection{The Dispersion Relations and Photon Damping
Scale}\label{sssec-dispersion}
In this section we derive the dispersion relations for small scale
anisotropies, where the focus is once again on developing generic
covariant equations in parallel to the usual gauge invariant
treatments \cite{Kaiser,kosowsky,HS-astro}.
To this end, we begin by iterating the baryon velocity
equation in much the same manner as we iterated the integrated
Boltzmann equations for the radiation.
We begin with the the baryon relative velocity equation (\ref{L004})
which is once again inverted in order for it to take the form
in (\ref{L005}). This is then turned into the basis of an
iteration scheme in terms of the scattering time:
\begin{equation}
{v_{(n)}}^a_B \approx \tau_a - {R \over \dot\kappa} \left[
{ \dot{v}^a_{(n-1)B} + H {v^a_{(n-1)}}_B + A^a } \right]\;. \label{v-iter}
\end{equation}
Using this equation and the zero-th order tight-coupling approximation,
$v^a_{(0)B} \approx \tau_a$, the covariant second order baryon
velocity equation is found:
\begin{eqnarray}
v_a^B \approx \tau_a - { R \over \dot\kappa} \left[ {\dot{v}_a^B + H v_a^B
+ A_a} \right] + {R^2 \over \dot{\kappa}^{2}} \left[{\ddot{v}_a^B + (\dot{H}
+ H^2) v_a^B + \dot{A}_a + H A_a + 2 H \dot{v}_a^B}\right]\;,
\label{v-2nd}
\end{eqnarray}
where we have dropped terms of ${\cal O}(\dot\kappa^{-3})$.
We now consider the small scale version of this equation, by ignoring
terms scaled by the Hubble parameter $H$ and
effects due to the gravitational potentials (see (\ref{small_scale})):
\begin{eqnarray}
v_a^B \approx \tau_a - {R \over \dot\kappa} \left[ {\dot{\tau}_a + A_a}
\right] + {R^2 \over \dot\kappa^2} \ddot{\tau}_a\;.
\label{s_b_vel}
\end{eqnarray}
Expanding the transport equation for the second order radiation quadrupole
(\ref{radf_0014}) to first order and ignoring the shear contribution
which is negligible in the almost-Friedmann-Lema\^{\i}tre small scale limit, we obtain
\begin{equation}
\tau^{ab} \approx - t_c \mbox{D}^{\langle a} v^{b \rangle}_B \approx
- \dot\kappa^{-1} \mbox{D}^{\langle a} \tau^{b \rangle}\;,
\label{tab-s}
\end{equation}
where again we retain only first order terms. We now
substitute (\ref{s_b_vel}) and (\ref{tab-s}) into the radiation dipole
evolution equation (\ref{L002}) to find:
\begin{eqnarray}
- \dot{\tau}_a = ( A_a + \mbox{D}_a \ln T) - R (\dot{\tau}_a + A_a) +
R^2 \dot{\kappa}^{-1} \ddot{\tau}_a - \ts{2 \over 5} \dot{\kappa}^{-1} \mbox{D}^c
{\mbox{D}_{\langle a} \tau_{c \rangle}}\;, \label{rad-dipole-2nd}
\end{eqnarray}
which can be compared to the mode
equation A-11 in Hu and Sugiyama \cite{HS-astro}). This is the
key equation from which we will now proceed to recover the dispersion
relations and hence standard damping scale results. The key-point here
is that the diffusion damping is second order in the scattering time,
while the acoustic oscillations are first order.
We now take a covariant mode expansion of the necessary quantities:
$\tau_a = \tau_1 Q_a$, $A_a = A_k Q_a$ and
$\mbox{D}_a \ln T = (\ts\frac{k}{a}\delta T) Q_a$, together with the
well known result (see Paper I \cite{GE} for the general
case): $\mbox{D}^b Q_{ab} = - \frac23 (a k)^{-1} (-k^2 + 3K) Q_a$.
We consider only the flat case here, so we set $K=0$. Putting these
all into the second order radiation dipole equation (\ref{rad-dipole-2nd})
we obtain:
\begin{eqnarray}
- \dot{\tau}_1 \approx (A + (\ts\frac{k}{a}\delta T)) + R ( \dot{\tau}_1 + A) + R^2
\dot{\kappa}^{-1} \ddot{\tau}_1 + \ts{4 \over 15} \dot{\kappa}^{-1}
{k^2 \over a^2} \tau_1\;. \label{rad_dip_mode_2nd}
\end{eqnarray}
Now we use the WKB approximation:
\begin{equation}
\tau_1 \propto \exp i \int (\omega / a) dt\;,
\end{equation}
and drop terms scaled by $H$ and $a \approx R$. This gives
\begin{eqnarray}
- i (1+R)\ts{\omega \over a} \tau_1 \approx [(\ts\frac{k}{a}\delta T) + (1+R) A]
- R^2 \dot\kappa^{-1} \ts{\omega^2 \over a^2} \tau_1 + \ts{4 \over 15}
\dot{\kappa} ^{-1} \ts{k^2 \over a^2} \tau_1\;. \label{rad-WKB}
\end{eqnarray}
Now, in order to deal with the terms arising from $A_a$ and $\mbox{D}_a \ln T$ we
consider the covariant radiation monopole perturbation equation
(\ref{mono_temp}):
\begin{equation}
(\mbox{D}_a \ln T)^{\dot{}} + \frac13 \mbox{D}_a \Theta + H ( \mbox{D}_a \ln T + A_a)
\simeq - \frac13 \mbox{D}_a (\mbox{D}_c \tau^c)\;. \label{monopole}
\end{equation}
We find, on taking a mode expansion and applying the WKB approximation
again, dropping terms scaled by $H$ and the expansion gradient (the
small scale approximation described in section 2.5), that
\begin{equation}
(\ts\frac{k}{a}\delta T) \approx {\ts \frac13 {k^2 \over a}}
(-i \omega) \tau_1\;.
\label{lnT-WKB}
\end{equation}
Upon substituting (\ref{lnT-WKB}) into (\ref{rad-WKB}), factoring out
the dipole coefficient $\tau_1$, writing the differential optical in
terms of conformal time:
$\dot{\kappa}^{-1} = (a \kappa^{\prime})^{-1}$ and multiplying through by
$i \omega a$ we get:
\begin{equation}
\omega^2 (1+R) + R^2 {\kappa^{\prime}}^{-1} \omega^2 (i \omega)
- {\ts{4 \over 15}} {\kappa^{\prime}}^{-1} k^2 (i \omega) \approx
{\ts{1 \over 3}} k^2\;. \nonumber
\end{equation}
On re-arranging terms we finally obtain
\begin{equation}
\omega^2 \approx \ts{k^2 \over 3(1+R)} + k^2 ({i \omega
{\kappa^{\prime}}^{-1}}) \left( {\ts{R^2 \over 3(1+R)}
+ \ts{4 \over 15}} \right)\;.
\end{equation}
Splitting $\omega$ up into its natural frequency $\omega_0$ and
the diffusion damping term $\gamma$ and then solving the
quadratic for the frequency $w$ we obtain
\begin{equation}
\omega \approx \pm \omega_0 + i \gamma\;,
\end{equation}
where
\begin{eqnarray}
\omega_0 \approx \ts{k \over \sqrt{3 (1+R)}} \approx c_s k,
~~\mbox{and}~~
\gamma \approx k^2 \left( \ts{{\kappa^{-1}}^{\prime} \over 6} \right)
\left[ \ts{R^2 + \frac45 (1+R) \over (1+R)^2} \right] \approx k^2/k_D^2\;.
\label{dif-damp}
\end{eqnarray}
Here $k_D$ is the diffusion damping scale and $c_s$ the baryotropic speed
of sound in the matter. We get oscillations when $k<k_D$ (see the next
section), and diffusion damping when $k> k_D$,
leading to a damping envelope.
The above covariant result is equivalent to the
analytic small scale approximation scheme
of \cite{HS95a,HS95b,HS-astro}, who have
demonstrated its robustness and closeness to more precise numerical
studies\footnote{For numerical integrations of the covariant scalar
temperature anisotropy equations we would refer the reader to \cite{cl2}.}.
This derivation can be easily corrected to include the effect
of anisotropic scattering (which breaks the tight-coupling approximation)
and polarization. This is done
by correcting the scattering terms in the calculation of the damping
scale. We follow the approach of Kaiser
\cite{Kaiser,HS95a,HS95b,HS-astro} and include the
correction $f_2$ to find the modified damping factor:
\begin{equation}
\gamma^* \approx k^2 \left( \ts{{\kappa^{-1}}^{\prime} \over 6} \right)
\left[ \ts{R^2 + \frac45 f^{-1}_2 (1+R) \over (1+R)^2} \right]
\approx k^2/{k^*}_D^2\;,
\label{dif-damp-correct}
\end{equation}
where, first, for the anisotropic effect, $f_2 = \frac{9}{10}$, and
second, to compensate for the polarization, $f_2 =
\frac{3}{4}$. Hence we get the diffusion damping envelope
for $\tau_1$ when $k > k_D$.
\subsubsection{The Temperature Oscillations}
To examine the solution when $k < k_D$, we take a mode expansion,
using the $Q_{A_\ell}$'s defined in Paper I \cite{GE}:
\begin{eqnarray}
a(\mbox{D}_c \ln T)^k = k (\delta T) Q_c \Rightarrow ~a\mbox{D}_c(\mbox{D}^a\mbox{D}_a\ln T)
=(-{k^3 \over a})(\delta T) Q_c\;,
\end{eqnarray}
where the driving term is written in terms of a generic
potential $\Phi_F$ defined by
\begin{eqnarray}
- \frac{a^3}{3} \mbox{D}_c (\mbox{D}^2 \Phi_F) \approx - a^2 (\mbox{D}_c \Theta)^{\prime}
+ {\frac{a^2}3}\mbox{D}_c(\mbox{D}^a A_a) ~~\mbox{and}~~
a^3 \mbox{D}_c (\mbox{D}^2 \Phi_F) \approx + \frac{k^3}{3} \Phi_F Q_c\;.
\end{eqnarray}
{}From (\ref{acoustic-source}) we find the oscillator equation
(once again working in the Newtonian frame {\it cf.} \cite{HS95b}
eqn. (B3) and equation (\ref{temp_osc_ng})).
The solution at first order must be convolved with the damping
envelope, found from the dispersion relations, in order to include
the damping cut-off. We have waited until the form of this damping
is known, so that it can be easily included by simply replacing the
natural oscillator frequency, $\omega_0$, with $\omega=\omega_0
+ i \gamma$, which now includes the damping. We use equation
(\ref{acoustic-source}) in a manifestly gauge invariant form,
to find upon mode expanding
\begin{equation}
(\delta T)^{\prime \prime }+k^2c_s^2(\delta T) \approx
- {\frac{k^2}3}\Phi_F\;,
\end{equation}
where the {\it sound speed} $c_s$ is given by $[3(1+R)]^{-\frac 12}$.
This gives the well known solution (see for example \cite{HS95a,HWa2}
\begin{equation}
(\delta T)(\eta )\approx [(\delta T)(0)+(1+R)\Phi _F]\cos (kr_s)-(1+R)\Phi _F,
\label{acoustic}
\end{equation}
where the {\it sound horizon scale} is given by
$r_s=\int c_sd\eta \approx c_s\eta $.
This describes the source term for the acoustic oscillations with the
isocurvature term dropped
{\footnote{
The other part of this solution comes from $\sin (kr_s)$; this is
the isocurvature part ( $\delta T^{\prime }(0)\neq 0$), giving
the full solution:
$$
(\delta T)(\eta )=\left[ {(kc_s)^{-1}(\delta T)^{\prime}(0)} \right]
\sin (kr_s)+\left[ { (\delta T)(0)+(1+R)\Phi_F} \right]\cos (kr_s)
-(1+R)\Phi_F.
$$
In this paper we consider adiabatic perturbations only so
$(\delta T)^{\prime}(0)=0$.}}.
The last term on the right gives the
Sachs-Wolfe effect, due to the potential; the other terms give the adiabatic
acoustic oscillations.
By putting this back into the momentum flux equation (\ref{mom-flux})
in mode form we obtain
\begin{equation}
\tau_1'(\eta,k) \simeq - {1 \over 1+R} (k \delta T) - a k \Phi_F\;,
\end{equation}
and using the temperature anisotropy mode expansion that
results when (\ref{L0081}) holds, the Doppler contribution
to the temperature anisotropies at decoupling can be found.
These are given by
\begin{equation}
\tau _1(\eta,k)\approx (- k^{-1}) 3[(a\ln T)(0)+(1+R)\Phi _u]c_s\sin (kr_s).
\label{Doppler}
\end{equation}
\section{Temperature Anisotropies from Integral Solutions}
In this section we derive the explicit form of the temperature
anisotropies, using the general $u^a$-frame integral solutions
and the tight-coupling approximation solutions.
Before we do this, there are two important points
that need to be discussed.
Firstly, where does the high-$\ell$ cut-off come from?
The natural frequency of the system is $\omega_0$ and is set by
the sound speed in the tight-coupled system to first order in
Thompson scattering time -- the oscillator equation. The diffusion
damping is a second order effect whose correction is found by
deriving the dispersion relations at second order in the Thompson
expansion for the baryon and photon momentum equations and using
the WKB approximation to find the new oscillator frequency to
be : $\omega \approx \omega_0 + i \gamma$.
The assumption that the anisotropies are sourced by these oscillations
in the radiation is the key to the physics; the initial conditions
before free-streaming are $\tau_{A_\ell} \approx 0$ for $\ell>2$
(where the anisotropic correction at $\ell=2$ is included as a
correction to the damping scale). The free-streaming radiation
transfer function (the spherical Bessel function in k-space) is then
convolved with the initial power sourced by the oscillations in the
average temperature and the dipole (essentially a cosine and sine
function in k-space respectively). As free-streaming continues
power is shifted up into the higher $\ell$'s, as the radiation
transfer functions' maximum is near $\ell \propto k \Delta \eta$
(where $\Delta \eta$ is the elapsed time since last scattering).
This maximum moves into higher
$\ell$ for longer times to give one a sense for how the power is
distributed through the $\ell \pm 1$ moment couplings in the
almost-Friedmann-Lema\^{\i}tre integrated Boltzmann equations. Obviously at high-$\ell$ (which corresponds to
high-$k$) the amount of power surviving the $(k/k_D)^2$ damping will
be very small, and hence as time progresses the peak in the transfer
functions drops off for higher-k, thus giving the high-$\ell$ cut-off.
Given that in free-streaming there is no diffusion damping
to cut-off the high-$k$ power, the truncation of the
hierarchy is only consistent and meaningful if the
significant anisotropy signal is sourced from the initial conditions
at low-$\ell$ near to tight-coupling.
Secondly, why do we find angular variations in the temperature
anisotropies that are entirely due to oscillations in the dipole
and monopole of the radiation -- particularly given that
in relativistic kinetic theory one has time-like integrations
\footnote{Only in the almost-Friedmann-Lema\^{\i}tre sub-case
can one think of the averaged relationship between the
null-projection}?
This is important for non-linear extensions of the almost-Friedmann-Lema\^{\i}tre
model \cite{MGE}, since care should
be taken when treating ensemble averaged quantities integrated down
the null-cone as being generally equivalent to those integrated
down a single time-like world-line.
The issue will become even more complex once the Gaussian
averaging necessary for the construction of the angular correlation
functions is relaxed. It is this, along with the weak Copernican
principle that makes possible the extension of the data here and now,
as integrated in a little box down a time-like world-line to
last scattering, to global statements.
Now we return to the issue of sourcing the temperature anisotropies
from oscillations in the dipole and monopole.
The integral solutions give the full almost-Friedmann-Lema\^{\i}tre solution to the radiation
anisotropies given the appropriate initial data.
Generically it is the transfer of power from low-$\ell$ initial data up
into high-$\ell$, with the accompanying reverse transfer of
power; the $\ell \pm 2$, $\ell \pm 1$ transfer of power where there
is a wall at $\ell=0$ but none at high-$\ell$.
There is strictly no $\ell_{max}$ unless the geometry is exactly
Robertson-Walker \cite{ETMb}. The tight-coupling approximation gives
the monopole and dipole at the $\ell=0$ initial wall.
Diffusion damping in the initial power near last scattering as well as
additional damping through slow decoupling cuts off the transfer of
power. This power is sourced by those initial conditions
and therefore give the temperature anisotropies in terms of the dipole
and monopole alone. Additional integrated effects will
only modify the primary projection, effectively leaving
the cut-off unchanged.
\subsection{Temperature Anisotropies}
The integral solution gives the projection of these conditions at
last scattering onto the current sky. These results are covariant.
The diffusion damping can be found from the dispersion relations arising
from the coupled baryon-photon equations using the WKB method in
tight-coupling which was described in section 6.2. We can now
construct the primary source temperature anisotropies,
by first considering the acoustic and Doppler
contributions in the free-streaming projection (\ref{freestreaming}).
Our solutions for ${\cal B}_0$ and ${\cal B}_1$ are first order in
scattering time expansion hence to include the second order diffusion
cut off, we multiply the solution through by $\exp{\gamma}$
(or $\exp{\gamma^*}$ if we want to include the corrections for the
anisotropic and polarization effects).
We can re-write $\gamma$ or $\gamma^*$ in terms of the damping scale,
$k _D$ ( {\it cf} \cite{HS95b} eqn (A7)-(A8)).
\subsubsection{Sources of Temperature Anisotropies}
In the manifestly gauge invariant integral solution of the temperature
anisotropies in almost-Friedmann-Lema\^{\i}tre spacetime, using the slow decoupling
approximation with scalar perturbations (\ref{slowdec_1},\ref{slowdecoupling}),
there is additional complexity of having to deal with shear terms.
We therefore choose the Newtonian frame in which
the shear contribution does not appear\footnote{Incidently this would also
remove any problems we may have with the introduction of high-$\ell$
truncation as discussed in Appendix E}.
We are then able to recover the canonical scalar treatment
\cite{W83,S89,GSS,HS95a,HS95b}. This can be written out in terms
of the scalar perturbation potentials $\Phi_A$ and $\Phi_H$, the temperature
perturbation, $\delta T$, the radiation dipole $\tau_1$ (through slow
decoupling) and the baryon relative velocity, $v_B$, all within the
covariant approach (including Doppler contributions
(\ref{slowdecoupling-ng})):
\begin{eqnarray}
{\beta_{\ell} \tau_{\ell}(\eta_0) \over (2 \ell+1)}
\approx \underbrace{{\bf S}_{P}(k,\eta_*) j_{\ell} (k \Delta \eta_*)}_{\sf
primary~sources} + \underbrace{\int_{\eta_*}^{\eta_0} d \eta
\left( {{\bf S}_{DISW} + {\bf S}_{ISW}} \right) j_{\ell}(k\Delta \eta)}_{\sf
secondary~sources}\;.
\label{temp-anisotropy}
\end{eqnarray}
Here the source terms are given by:
\begin{eqnarray}
{\bf S}_{P}(\eta,k) &=& {\cal D}(\eta_0,k) \left[{ {(\tilde \delta T} + \Phi_A)
+ \kappa^{\prime} v_B } \right], \label{primary}\\
{\bf S}_{DISW}(\eta,k) &=& {\cal V} e^{-(k/k_D)^2} \left[ {\frac13 k
{\tilde \tau}_1 + (\kappa^{\prime} \tilde v_B^{\prime} +
\kappa^{\prime \prime} \tilde v_B)} \right] \label{doppler} \\
{\bf S}_{ISW}(\eta,k) &=& {\cal V}e^{-(k/k_D)^2}
\left[{ [\Phi_A^{\prime} - \Phi_H^{\prime}](k,\eta) - a H
[ {\tilde \delta T} + 3 \Phi_A](k,\eta)} \right]\;.
\label{secondary}
\end{eqnarray}
In the first line, on the RHS, we have the Sachs-Wolfe and acoustic sourced
projection effects which are the primary sources. The next term describe
secondary Doppler effects during slow decoupling. The last term models
the integrated Sachs-Wolfe (ISW), the late-ISW and early-ISW effects
respectively. We have used $\Delta \eta_* = \eta_0 - \eta_*$
(the conformal time difference between
the surface of last scattering and the time of reception, here and
now) and $\Delta \eta = \eta_0 - \eta$.
In this way we have recovered the solution from the
exact equations in a systematic manner using the 1+3 CGI formalism
rather than recovering the solution from a perturbation
theory about a foliation of Robertson-Walker surfaces of homogeneity. Note that
there is no monopole temperature anisotropy in the CGI approach while
there is one in the canonical treatment.
The angular correlation functions, $C_{\ell}$, for the general small
temperature anisotropy case have been derived in terms of a superposition of
homogeneous and isotropic Gaussian random fields with respect to the
temperature anisotropy multipoles to find the multipole mean-squares,
${\langle \tau_{A_{\ell}}\tau^{A_{\ell}} \rangle}$, in terms of the an
ensemble average (see Paper I \cite{GE}). The relationship
between the multipole mean-squares (in the
almost-Friedmann-Lema\^{\i}tre case) and the mode coefficient
mean-squares $|\tau_{\ell}|^2$ in terms of the Robertson-Walker mode
functions $G_{\ell}[Q]$ are also given in Paper I \cite{GE}. We do
not discuss these further here. It must be emphasized that
the multipole mean-squares are given for general geometries,
while the mode means squares are only for almost-Robertson-Walker geometries. This
is why the application of the non-linear extension is possible;
where the corrections to the almost-Friedmann-Lema\^{\i}tre standard model are calculated
using the multipole formalism of \cite{MGE}.
\subsubsection{Sachs-Wolfe Effect and the Acoustic Source}
\label{newt-sachs-wolfe}
In the Newtonian frame (\ref{primary}) we find that
the Sachs-Wolfe effect arises from a combination of the
$\tilde {\mbox{D}}_a \ln T + \tilde A_a \approx \ts \frac{k}{a} (\delta \tilde T
+ \Phi_A) Q_a$. From the solution to the oscillator equation in the
Newtonian frame (eqn. (\ref{temp_osc_ng})) along with
$r_s \sim c_s \eta$ and ignoring the time evolution of the
potential (for the equivalent canonical version see \cite{HS95a}) we find
\begin{equation}
\delta \tilde T(\eta,k) + \Phi_A(\eta_*,k) \approx [\delta \tilde T(0,k) + (1+R)
\Phi_A(0,k)] \cos (k r_s) - R \Phi_A(\eta_*,k).
\end{equation}
Upon taking the matter dominated limit ($R \approx {\cal O}(\epsilon^2)$) and
then using the adiabatic assumption in tight-coupling $\delta T(0,k)
\sim \frac13 \Delta(0,k) \sim - \frac23 \Phi_{A}(0,k)$
(adiabatic flat CDM model) and finally taking $r^*_s \sim 0$
(by using that $k \eta_* \ll 0$) we recover the usual results as in
\cite{HS95a,HWa}
\begin{equation}
[\delta \tilde T + \Phi_A](\eta_*,k) \approx + \ts\frac13 \Phi_A(0,k)
(1+ 3 R_*) \cos(k r^*_s) + R_* \Phi_A(0,k) \sim \ts\frac13 \Phi_A(0,k).
\label{sachs-wolfe-newtonian}
\end{equation}
Here we have used that $\Phi_A(\eta_*,k) \approx \Phi_A(0,k)$ from
the Einstein de-Sitter result that the potential is constant if we drop the
decaying mode.
The physics of the Sachs-Wolfe effect as opposed to the acoustic
oscillations is then quite clear. There are no oscillations just
an imprint due to an acceleration potential. Although the adiabatic
assumption is invariant to order ${\cal O}(\epsilon)$, the
relationship between the Electric part of the Weyl tensor
and the perturbations are not -- so we will always use the matter
perturbation in the total frame where it can be easily related to
the Newtonian potential and the temperature perturbation in the
Newtonian frame where the oscillator equation takes on a useful
form. In a similar manner we find the explicit form of the
radiation dipole and acoustic oscillations in the slow decoupling
era:
\begin{eqnarray}
&&\delta \tilde T(k,\eta) \approx \ts \frac13
\Phi_A(0,k) (1 +3 R) \cos (k c_s \eta) + (1+R) \Phi_A(0,k), ~\nonumber \\
&&\tilde \tau_1 (k,\eta) \approx \Phi_A(0,k)
(1 + 3 R) c_s \sin (k c_s \eta).
\end{eqnarray}
\subsubsection{Weak-coupling for the Small Scale Solution}
Here we briefly consider the integrated Sachs-Wolfe effect in connection with the
weak-coupling approximation. The idea is that the anisotropies
fall with $\ell$ more rapidly than a simple projection would imply.
What one has in mind is the situation where the anisotropy
contributes across many wavelengths of the fluctuation allowing
cancellations on small scales; the secondary sources, in particular
the term $({{\Phi}^{\prime}}_A - {{\Phi}^{\prime}}_H)$,
varies slowly on small scales \cite{HWa}.
Specifically we consider the situation where
\begin{eqnarray}
\int_{\eta _*}^{\eta _0}d\eta {\cal V} e^{-(k/k_D)^2}
\left\{{(\Phi_A^{\prime} - \Phi_H^{\prime})
}\right\} j_\ell(k \Delta \eta) \approx \sqrt{{\pi \over 2 \ell}}
{1 \over k} (\Phi^{\prime}_A - \Phi^{\prime}_H)(\Delta \eta_*)
{\cal D}(k,\eta_0)\;,
\end{eqnarray}
where we used $\int_0^{\infty} j_{\ell} (k \Delta \eta) d \eta =
[\sqrt{\pi}/2 k] [\Gamma((\ell+1)/2)/\Gamma((\ell+2)/2)]$. The
weak-coupling solution is implied by the assumption that:
\begin{equation}
(\ddot{\Phi}_H - \ddot{\Phi}_A ) \ll k (\dot{\Phi}_H - \dot{\Phi}_A)\;.
\end{equation}
This is nothing more than a useful approximation allowing the direct
construction of analytic solutions. Some care should be taken when
using the weak-coupling approximation when trying to estimate the
early-ISW (the free-streaming analogue of the acoustic driving effect)
and late-ISW (due to non-vanishing curvature or cosmological constant,
which will dominate the expansion rate at latter time) effects. Ideally
one should evaluate the slowly-varying function, which has been taken out
of the integral, at the $\ell$-th peak; $\eta_{\ell}
= \eta_0 - ({\ell + \frac12}) / k$ rather than at
$\Delta \eta_* = \eta_0 - \eta_*$.
In the case of the early-ISW effect, since it satisfies neither the
tight-coupling nor weak-coupling criteria, our approximations schemes
here are not entirely appropriate; its decay time and wavelength
are comparable \cite{HS95a,HS95b,HWa}. However we can use the
weak-coupling in the case of the late-ISW effect because cancellation effects
lead to damping on small scales.
The temperature perturbation decays, and hence
the potential decays on the order of the expansion time near the end of
the matter dominated era. The photons will free-stream across many
wavelengths of the perturbations below the Hubble scale, leading to
cancellation and damping effects.
The important point about the late-ISW effect is that there will
be an imprint due to the exit from the matter dominated era into
a $\Lambda$ or curvature dominated one.
In the context of the equations
here, we can consider $\Lambda$-dominated effects by investigating
the evolution equations for the potentials in a $\Lambda$-dominated case.
In this case the effective expansion changes, given here for late-times
(well after matter-radiation equality):
\begin{equation}
H^2 = a^{-3} \Omega_0 H_0^2 + \frac13 \Lambda
\end{equation}
for the $\Omega_0 + \Omega _\Lambda = +1$ model.
This has however been dealt with in depth in the literature
(see for example \cite{HWa,HS95b}).
\subsubsection{The Mode Coefficients}
The angular correlation functions, measured here and now ($x_0^i$) can then
be computed using the the mode coefficients from Paper I \cite{GE}:
\begin{eqnarray}
C_{\ell} = \frac{2}{\pi} {\beta_{\ell}^2 \over (2 \ell+1)^2} \int_0^{\infty}
{dk \over k} k^3 | \tau_{\ell}(k,\eta_0)|^2\;.
\label{angular-corr-fn}
\end{eqnarray}
The final step is to construct the temperature
anisotropy solutions in terms of the matter power spectra. We discuss
this in the next section.
At linear order, there is no important difference
in our solutions from those found in the canonical
treatment \cite{HS95a,HS95b}, however the advantage is that
we can easily relate our formulation of the temperature anisotropies
to the mean-squares of the multipoles ($\tau_{A_\ell}$) and
hence to the mean-square of generalized
temperature anisotropies, ($\Pi_{A_\ell}$) \cite{MGE}. This was not
attainable in the canonical treatment.
\subsubsection{The Multipole Coefficients}
The mean-square of the multipole moments are related to the almost-Friedmann-Lema\^{\i}tre
mode coefficients by (see Paper I \cite{GE})
\begin{equation}
\left\langle {\tau _{A_\ell}\tau ^{A_\ell}}\right\rangle \approx
{1 \over 2 \pi^2 } \beta _\ell\int k^2 dk|\tau _\ell(k,\eta_0 )|^2\;,
\label{t0_ang_powers}
\end{equation}
giving the {\it angular correlation function} (see Paper I \cite{GE}):
\begin{equation}
C_\ell= \Delta_\ell (2\ell+1)^{-1} \left\langle {\tau _{A_\ell}\tau
^{A_\ell}}\right\rangle\;. \label{t1_ang_powers}
\end{equation}
This relates the matter fluctuation amplitude at last scattering to the
present day via the Newtonian like potential, which in turn may
be related to the matter power spectrum directly.
A plethora of numerical studies of Doppler peak features exist in the
gauge-invariant literature (see for example \cite{hswz} and
\cite{SEf}).
We will now summarize the standard picture of acoustic peak
characteristics before discussing the matter power spectrum.
\subsection{Some `Acoustic peak' Characteristics}
The key features of the standard model of cosmic background
radiation primary sourced Doppler peaks are listed below
\cite{HS95a,HS95b,HWa,HWa2,WHb,Jungmann96}.
The standard model of acoustic peak formation has been given
in its analytic form above.
\begin{enumerate}
\item The $j-th$ {\it peak positions} (as given in the flat
adiabatic case) is given by
\begin{equation}
\ell_j \approx k_j \left| {r_{\theta}(\eta) } \right|
= \left| {r_{\theta} \over r_s}\right| j \pi\;,
\end{equation}
where $r_{\theta}$ is the comoving angular diameter and $r_{s}$
is the sound-horizon near decoupling.
Notice that $j_\ell(k \Delta \eta)$ peaks near $\ell \sim k \Delta \eta$,
where free-streaming projects this physical scale onto the angular scale
$\theta \Delta \eta$ on the current sky. The first peak is dependent on
the driving force which is model independent. It provides a way of
fixing the angular diameter distance -- when using the almost-Friedmann-Lema\^{\i}tre
assumptions.
\item The acoustic {\it relative peak spacings} are given by
\begin{equation}
k_{j+1} - k_j = k_A \approx {\pi \over r_s} \iff \Delta \ell \sim
\ell_A = k_A r_{\theta}\;.
\end{equation}
The peak spacing is fixed by the natural frequency of the
oscillator: $\omega = k c_s$, which is independent of the driving force.
The factor $c_s$ is the photon-baryon sound speed.
\item The peak ratios arise from the angular power spectrum
ratio $C_{jk} = C_j /C_k$ found from $C_\ell$ in terms of $|\tau_\ell|^2$ or
$\langle {\tau_{A_\ell} \tau^{A_\ell}} \rangle$.
\item The {\it damping tail} provides yet another angular diameter
distance test of curvature \cite{HWa} via the damping
scale $k_D$. The peak spacings
$\Delta \ell$ and the damping tail location $\ell_D$ depend only on the
background quantities, they are robust to model changes, assuming that
secondary effects do not overwhelm the signal. The diffusion scale near
decoupling is the angular scale of the wavenumber
$k_D \sim \sqrt{\dot{\kappa} / \Delta \eta}
\sim \sqrt{a / \Delta \eta \dot{a}} \sim \Delta \eta ^{-1}$.
Given that the damping tail of the acoustic oscillations takes on the
form $e^{-(k /k_D)^2}$, $k_D$ can be
found from (as derived previously):
\begin{equation}
k_D^{-2} \approx \frac16 \int_{\eta_*}^{\eta_d}
d \eta {1 \over {\kappa^{\prime}}}
{R^2 + \ts{4 \over 5} f^{-1}_2 (1+R) \over (1+R)^2}\;,
\end{equation}
for no anisotropic stress \cite{HWa,Kaiser}. This can be used to find
that the damping tail location is $\ell_D= k_D r_{\theta}$.
\item The damping tail position to peak scale, $\ell_D /
\ell_A$, is a good measure of the number of peaks and gives an
indication of the delay in recombination independently
of the area distance.
\end{enumerate}
\section{The Matter Power Spectrum}
In this section we related the temperature anisotropies in the
case of the Sachs-Wolfe effect to the matter
power spectrum using the 1+3 CGI formalism. We also deal with basics
of the normalization of the cosmic background radiation angular power spectrum
to the matter power spectrum.
Up to this point everything we have derived has been
found covariantly from general relativity and
from relativistic kinetic theory, however on small scales
we do not have a covariant analytic derivation of the
transfer function, although it is well known for large scales, so
this is the only gap in our treatment \footnote{There is
still some hope that the almost Friedmann-Lema\^{\i}tre Quasi-Newtonian models,
will be useful in this regard \cite{HvEE} providing us
a covariant derivation of the Harrison-Zel'dovich
spectrum.}.
The aim here is to try to predict the matter distribution
today from the cosmic background radiation spectrum. From
the cosmic background radiation spectrum one finds the matter
power spectrum (as fixed by the cosmic background radiation
data today) from which the current
distribution of matter is found (in terms of $\sigma_8$).
Briefly, in the literature there have been two methods
for normalizing the cosmic background radiation data in relation to COBE:
{\it firstly} the $\sigma(10^0)$ normalization where the
r.m.s temperature fluctuation was averaged over a $10^0$
FWHM beam, and {\it secondly} the $Q_{rms-PS}$\footnote{$Q_{rms-PS} \approx
\ts{5 \over 4 \pi}C_2$ and $ C_{\ell}^{\!_{SW}} \sim \ts{6 \over \ell(\ell+1)}
C_2$.} normalization which
uses the best fitting amplitude for a $n=1$ Harrison-Zel'dovich (HZ)
spectrum quoted for the quadrupole, $\langle{Q}\rangle$. The first method
had the advantage of being a model independent way of fitting the data
-- it is observationally determined, however, because of this it
does not provide the most accurate normalization for a specific model
and care must be taken in order to properly deal with
cosmic variance. The second approach is model dependent and
works well only for the HZ spectrum \cite{EBW,BSW}.
Improved, more general normalization schemes have now been
developed and there currently seem to be two favoured ones, both
having variations of the HZ spectrum in mind. It is these two
schemes we now briefly consider here. We emphasis the
so called {\it CDM-models} with $\Omega_0=1$. These models
have a baryon fraction $\Omega_B$ with the rest of the matter
being made up of massive Cold Dark Matter (CDM). Of these the
most easily dealt with is the so called {\it standard-CDM}
model, where initial fluctuations are assumed to have a Gaussian
distribution and have adiabatic, scalar density fluctuations with a HZ
spectrum on large scales:
\begin{eqnarray}
{P}(k) \propto k^{n-1}\;,~~~ n=1\;,~~~ H_0=50\mbox{km}^{-1}\mbox{Mpc}^{-1}\;,
~~~ \Omega_B=0.05\;.
\end{eqnarray}
There are three other popular models: the {\it Adiabatic CDM model} of Peebles,
which is a version of {\it standard-CDM}; the {\it ICDM} model (which
is an isocurvature version of the {\it Adiabatic CDM model};
and lastly the $\Lambda${\it CMD model} which is the
large cosmological constant version of {\it standard-CDM}. More
recently Hu has introduce a Generalized Dark Matter model (GDM)
\cite{HuGDM}.
\subsection{The Power Spectrum}
The Power Spectrum ${\cal P}(k)$ is fully specified by a {\it shape}
and {\it normalization} (see Appendix J for additional details
and references).
The normalization is fixed by the amplitude of the temperature fluctuations
in the cosmic background radiation\footnote{It seems that models normalized to COBE which are
fixed in both the amplitude at small scales and by the shape predicted by
{\it standard-CMD}, are inconsistent with observations \cite{KSper,BSW}.}.
There are two normalization scales, large and
small (see Appendix J).
On small scales the normalization is expressed in terms of $\sigma_8$, the
ratio of the r.m.s mass fluctuations to the galaxy number fluctuations
- both averaged over randomly located spheres of radius $8 \mbox{h}^{-1}
\mbox{Mpc}$; this is the variance of the density field in these spheres.
On large scales the power spectrum is
found by fixing the shape function \cite{E-scot36} which is a
measure of the horizon scale at matter radiation equality. We look
at these now for the case of a flat matter dominated almost-Friedmann-Lema\^{\i}tre model
(the open case is considered in Appendix K).
In the 1+3 CGI approach the emphasis is on the
quantity $a \mbox{D}_a \ln \rho_{\!_M} \approx a \delta_{\!_M} Q_a$. Here
$\delta_{\!_M}= a \delta_{\!_M}(0,k)$ and $\langle \delta_{\!_M}(k,0)
\delta_{\!_M}(k',0) \rangle = (2 \pi)^3 {\cal P}(k) \delta(k-k') /k^2$
for $|\delta_{\!_M}(k,0)|^2 = {\cal P}(k)$.
It becomes preferable in the context here to use the variable
$\Delta$: $\ts\frac{k}{a}\Delta(k,t) = \delta_{\!_M}(k,t)$\footnote{
$\Delta(k)^2 \approx \ts\frac{a^2}{k^2} | \delta_{\!_M}(k)|^2
= \ts\frac{a^2}{k^2} A k^{n+1}$} . We will,
however, be using the power spectrum for the gauge invariant
acceleration potential because we consider the situation of
matter dominated adiabatic perturbations. In the general situation it
is best to use $\bar \Phi$. Note, in the matter dominated situation
we can use that $\Phi_A = - \Phi_H$ and write everything in terms
of either $\Phi_H$ or $\Phi_A$.
The relationship between the power spectrum of the acceleration
potential and the power spectrum of the matter perturbation is
\begin{eqnarray}
| \Phi_A(k,0) |^2 = P_{\Phi_A}(k),~~\mbox{and}~~
| \Delta(k,0) |^2 = P(k), \label{flatpower}
\end{eqnarray}
where these are then related to each other in the total frame via
the constraints for the electric part of the Weyl tensor for the
flat case:
\begin{eqnarray}
P_{\Phi_A}(k,0) = \left( {\ts\frac{3}{2} H_0^2 \Omega_0 \frac{D}{a}}
\right)^2 k^{-4} P(k).
\label{phi-power-D}
\end{eqnarray}
Here $\bar \Phi = - \ts{k^2 \over a^2} \Phi_A$ and as noted before:
\begin{eqnarray}
( \bar \Phi \rho_{\!_M}^{-1})(k,t) \approx -(3 H_0^2 \Omega_0)^{-1} D(t)
[k^2 \Phi_A(k,0)].
\label{phi-power-rel}
\end{eqnarray}
and $D \approx a \approx \eta^2$ for a flat almost-Friedmann-Lema\^{\i}tre dust model. We
will also use $a_0 =+1$.
\subsection{Relating the Power Spectrum to Temperature Anisotropies}
We now combine the CGI
Sachs-Wolfe effect (due to potential fluctuations) and the matter
power spectrum to express the anisotropies in terms of the
matter power spectrum near decoupling.
\subsubsection{Flat Matter Dominated Almost-Friedmann-Lema\^{\i}tre}
We first consider the flat almost-Friedmann-Lema\^{\i}tre model. Using (\ref{flatpower})
and (\ref{sachs-wolfe-newtonian}, \ref{primary} and \ref{temp-anisotropy})
(that is with $K=0$) we obtain
\begin{eqnarray}
{\tau^{\!_{SW}}_\ell(\eta_0) \beta_\ell \over (2\ell+1)} \sim + \ts\frac13
\Phi_A(k,\eta_*) j_\ell(k \Delta \eta)\;, \label{sw-again}
\end{eqnarray}
Here we find the angular power spectrum
from (\ref{angular-corr-fn}), (\ref{sw-again}) and (\ref{phi-power-D}):
\begin{eqnarray}
C^{\!_{SW}}_{\ell} = \frac{2}{\pi} \left({ H_0^2 \Omega_0 \ts\frac{D_*}{a_*}
}\right)^2 \frac{1}{4} \int \frac{dk}{k^2} P(k) j_{\ell}^2( k \Delta \eta_*)\;.
\end{eqnarray}
where $P(k)$ is given by either (\ref{power_sA}) or
(\ref{power_sB}). Now if use that $(\Omega_0 D_*/a_*)^2 \approx \Omega_0^{1.54}$
we find the standard result \cite{EBW}:
\begin{eqnarray}
C^{\!_{SW}}_\ell = \frac{1}{2 \pi} H_0^4 \Omega_0^{1.54}
\int \frac{dk}{k^2} P(k) j_\ell^2( k \Delta \eta_*)\;.
\label{flat-power}
\end{eqnarray}
The next step is to normalize the matter power spectrum to any given
structure formation theory on both small and large scales, this is done
in the next section in the case of the {\it standard-CDM model}.
\subsection{Setting the CDM Normalization}
{}From the previous section we can now relate the angular
power spectrum for the matter dominated flat ($K=0$) almost-Friedmann-Lema\^{\i}tre model to
the current matter power spectrum on small and large scales
using the two different normalizations to the {standard-CDM model}.
In what follows we will use the useful result:
\begin{eqnarray}
\int_0^{\infty} {dz \over z^m} j_{\ell}^2 (z) = {\pi \over 2^{m+2}}
{m! \over (m/2)!}{ (\ell - \frac{m}{2} - \frac{1}{2} )! \over
(\ell + \frac{m}{2} + \frac{1}{2})!}\;.
\label{soln-kj}
\end{eqnarray}
\subsubsection{Large Scales}
{}From (\ref{flat-power}) and $P(k) = A k^{n-1}$ (\ref{power_sA}) we obtain
\begin{eqnarray}
C_{\ell} = {\frac{1}{2 \pi}} A H_0^4 \Omega_0^{1.54}
\int \frac{dk}{k^2} k^{n-1} j_l^2( k \chi)\;,
\end{eqnarray}
and on using (\ref{soln-kj}) for $m=2$ ($n=1$) we find that
\begin{eqnarray}
C_{\ell} = {\frac{A}{2}} H_0^4 \Omega_0^{1.54}
{1 \over (2\ell + 3)(2 \ell +1) (2 \ell -1)}\;.
\label{Cl-CDM-lrgs}
\end{eqnarray}
Of more immediate interest is the angular correlation function
for matter below the horizon scale near decoupling, as this
is what is used to normalize the angular correlation function.
\subsubsection{Small Scales}
For small scales one can use the $\sigma_8$ normalization
via (\ref{flat-power}) and $P(k)=B T^2(k) k^n$ (\ref{power_sB}) and the
parametrized transfer function (\ref{transfer-fn-efs}). This gives
\begin{eqnarray}
C_{\ell} \simeq {\frac{1}{2 \pi}} H_0^4 \Omega_0^{1.54}
\int \frac{dk}{k^2} B k^n T^2(k) j_{\ell}^2( k \Delta \eta_*)\;.
\end{eqnarray}
For example using the {\it standard-CDM model} with $n=1$, $h=0.5$,
$\Omega_B=0.05$ and $\Gamma=0.48$ we can relate $B$ and
$A$ to the scale of fluctuations near horizon crossing and the
quadrupole:
\begin{eqnarray}
B = 2 \pi^2 A = (6 \pi^{2/5})
\langle {Q} \rangle / T_0^2\;.
\end{eqnarray}
To normalize to $\sigma_8$ we can choose $\sigma_8 \simeq 1.3$.
This then allows us, in principle, to invert
the angular correlation function to find the matter
power spectrum once the initial spectrum is known.
We are however more concerned with normalizing the radiation angular
correlation function to CDM on horizon scales near decoupling. In
this regard we once again begin with the matter power
spectrum (we set $T(k)=1$) and on using (\ref{soln-kj})
for $m=1$ we find that:
\begin{eqnarray}
C_{\ell}^{(CDM)} = \left[{ {\frac{1}{2 \pi}} H_0^4 \Omega_0^{1.54}
B {{\pi^{\frac12}} \over 8} }\right] {1 \over{\ell(\ell+1)}}\;.
\label{Cl-CDM-smls}
\end{eqnarray}
\subsubsection{$C_\ell$ Normalized to Adiabatic CDM}
Finally we normalize the angular correlation function
$C_{\ell}$ (found from \ref{slowdecoupling}) to the
potential fluctuations normalized to Adiabatic CDM (\ref{Cl-CDM-smls}) above:
\begin{equation}
D_{\ell} =[{C_{\ell} / C_{\ell}^{(CDM)}} ]
= \left[{ {\frac{1}{2 \pi}} H_0^4 \Omega_0^{1.54}
B {{\pi^{\frac12}} \over 8}}\right]^{-1} \ell (\ell+1)C_{\ell}\;.
\end{equation}
The angular correlation function $C_{\ell}$ can be found
from the primary (\ref{primary}) and secondary (the integrated part of
\ref{slowdecoupling}) source that make up the total:
$C_{\ell}=C_{\ell}^{(P)}+C_{\ell}^{(S)}$. This then allows
one to remove that part of the angular correlation
function arising from the standard Sachs-Wolfe potential fluctuations
leaving a signal which is dominated by the photon primary and
secondary sourced physics. The convention is to
use $D_{\ell}$ rather than $C_{\ell}$ \cite{Jungmann96}, so we
have from (\ref{flat-power} and \ref{Cl-CDM-smls}) for the flat case:
\begin{eqnarray}
D_{\ell} = \ell(\ell+1) C_{\ell} = \int {dk \over k^2} {\cal P}^*(k)
| {j_{\ell}(k \chi)} |^2 ~~~\mbox{using}~~{\cal P}^*(k)= \left( {k \over k_s}
\right)^{n_s + \alpha \ln (k/k_s)}\;,
\label{pstar}
\end{eqnarray}
where $k_s$ is the normalization scale, $\alpha$ gives the deviation
from the power law, and $n_s$ give the scalar power law index. The
angular power per $\ln \ell$ is $(\ell(\ell+1)/4 \pi) C_{\ell}$.
\section{Conclusions}
In this paper we have carried out a covariant analytic
time-like integration reproducing the well known primary
effects generating the `acoustic' peaks measured here and now
for an almost-Friedmann-Lema\^{\i}tre universe with adiabatic scalar perturbations.
We have also demonstrated how, in the CGI formalism,
the angular correlation functions are constructed in terms of
the matter power spectrum and normalized on large
and small scales for {\it standard-CDM}, given appropriate
approximations for the transfer functions.
As pointed out initially, the aim of this paper was to
clarify the link between the standard gauge-invariant
and CGI treatments of cosmic background radiation anisotropies
and provide a strong
basis from which to tackle non-linear and gravitational
wave effects using CGI methods.
Some of the key outstanding issues are:
\begin{enumerate}
\item How does one deal with anisotropic scattering and anisotropic
stresses before and during decoupling within the covariant approach,
specifically in such a manner so that consistency is maintained
when using general relativity, its covariant linearisations
and relativistic kinetic theory.
\item The small anisotropy equations developed in
\cite{MGE} with the application to spacetimes with
arbitrary anisotropy and inhomogeneity have yet to be properly
investigated; these become applicable when one ignores the Copernican
principle that underlies the almost-Ehlers-Geren-Sachs theorem,
which in turn provides
the theoretical basis for using almost-Friedmann-Lema\^{\i}tre spacetime dynamics. An
investigation of their consequences on the cosmic background
radiation may provide an
alternative method of testing the Copernican principle other than
the Sunyeav-Zel'dovich effect or via the use of polarization maps.
\item There is a need to find a working non-Gaussian treatment
from which one can construct a generic characterization
of observables here and now (other than the angular power spectrum
alone (see Paper I \cite{GE}) and, second, finding an
{\it ab initio} covariant analysis of transfer functions extending
the post-Newtonian treatments which use periodic boundary conditions.
\end{enumerate}
In the next paper in the series, Paper III \cite{DGE}, we
hope to establish, in a complete fashion, the relationship
between the null-cone integrations
(favoured in the literature) and the time-like integrations
(found in the exact relativistic kinetic theory).
\section{Acknowledgements}
We acknowledges the partial support of this work
by the NRF and the University of Cape Town. We
would also like to thank the referee for taking the
time to provide helpful and interesting comments. TG
would also like to thank the Dept of Mathematics at
the University of the Witwatersrand for assistance.
|
\section{Introduction. }
Maldacena's $AdS/CFT$ correspondence
\cite{MAL} has spurred interest in Anti-de
Sitter supergravity theories. A particularly interesting case is the
$AdS_3/CFT_2$ correspondence \cite{MALSTR,MAR,BOER,SEIKUTGIV,DMW}.
In this case string theory on $AdS_3\times S^3\times M_4$
with $RR$ flux is
conjectured to be dual to the conformal theory of the Higgs branch of
the D1-D5-brane system, where $M_4$ can be $T^4$ or $K_3$.
It is relevant to understand the
corresponding supergravity on $AdS_3$ for the investigation of this
conjecture. On examining the symmetries of the near horizon geometry
of the D1-D5 system we find that the
specific $AdS_3$ supergravity is the one
associated with the supergroup ${\cal G}= SU(1,1|2)\times SU(1,1|2)$.
In this letter we will construct the pure
( i.e. no coupling to matter )
anti-de Sitter supergravity associated
with the above supergroup.
Construction of this supergravity would be useful in furthering our
understanding of the boundary conformal field theory. It is well known
that gravity in $AdS_3$ is a Chern-Simons theory with the gauge group
$SL(2,R)\times SL(2,R)$ and it can be reformulated as a Liouville
theory at the boundary \cite{CHD}.
A Liouville theory corresponding to
pure $AdS$ supergravity associated with the supergroup ${\cal G}$
is likely to help us
understand the space time conformal field theory constructed in
\cite{SEIKUTGIV} which is also a Liouville theory. Secondly,
construction of this supergravity can help us study backgrounds
in $AdS_3$ other than the pure $AdS_3$ and the $BTZ$ black hole,
which are the
backgrounds being studied presently. These other backgrounds can help us
understand certain properties of the boundary CFT.
In fact the motivation for construction of this $AdS$ supergravity
arose from the problem of supersymmetrically
embedding the conical spacetimes of \cite{DES} in this supergravity.
These spaces provide a one parameter family of metrics
which interpolate between pure $AdS_3$ and the zero mass $BTZ$ black
hole mimicking the spectral flow of the CFT \cite{DMVW}.
To construct the $AdS$ supergravity on ${\cal G}$ we
use the fact that this theory is a Chern-Simons theory associated with
the supergroup ${\cal G}$. That is,
the action can be written formally
as
\begin{equation}
\int_{M_3} tr (\Gamma d\Gamma + \frac{2}{3} \Gamma^3)
\end{equation}
where $\Gamma$ is the connection one-form for the supergroup and $M_3$
is the three-dimensional bosonic submanifold of the
supergroup ${\cal G}$ with local coordinates $x^\mu$.
$\Gamma$ does not depend on other coordinates of the group manifold
\cite{ACHTOW}.
Using this one can fix the field content using the
generators of the supergroup. The equations of motion of the
$AdS$ supergravity on ${\cal G}$ will be the
Maurer-Cartan equations of the supergroup
which can be written down from the corresponding superalgebra.
This letter is organized as follows. In Sec. 2 we write down the
superalgebra on ${\cal G}$ in a form suitable to extract
the field content of the $AdS$ supergravity and list the Maurer-Cartan
equations for the supergroup. In Sec. 3 we present the action for the
$AdS$ supergravity on ${\cal G}$. Then we list down the
supersymmetry transformation laws and show that the action is invariant
under them. In Sec. 4 we state our conclusions.
\section{The $SU(1,1|2) \times SU(1,1|2)$ superalgebra. }
The $SU(1,1|2) \times SU(1,1|2)$ super algebra is the global part of
the small ${\cal N} =(4,4)$ super conformal theory (see for e.g.
\cite{BOER}). The bosonic
generators consist of $L_{-}, L_{+}, L_0$ which form the $SL(2,R)$
part of the algebra and $T^i , T'^i $ which are
the global $SU(2)$ currents.
The supercharges consist of $G_{-1/2}^{\alpha}, G_{1/2}^{\alpha},
G'^{\alpha}_{-1/2}, G'^{\alpha}_{1/2} $ which
transform as fundamentals under the global $SU(2)$ currents. We can
organize the $SL(2,R)\times SL(2,R)$ generators into the generators of
Lorentz transformations $M_{ab}$ and translations $P_a$ of $SO(2,2)$,
the isometry group of $AdS_3$.
The supercharges of the each of the $SU(1,1|2)$ transform as a
Dirac fermion in $AdS_3$ with an internal $SU(2)$ index. Then the
$SU(1,1|2)\times SU(1,1|2)$ super algebra is given by the following
(anti) commutation relations.
\begin{eqnarray}
[P_a , P_b ] = -4 m^2 M_{a b} \; &\;& \;
[P_a , M_{bc}] = - \eta_{ab} P_{c} + \eta_{ac} P_{b} \\ \nonumber
[M_{ab}, M_{cd}] &=& - \eta_{ad} M_{bc} - \eta_{bc} M_{ad}
+ \eta_{ac} M_{bd} + \eta_{bd} M_{ac} \\ \nonumber
[T^{i}, T^{j} ] = 4 i \epsilon_{ijk} T^{k} \; &\;& \;
[T'^{i}, T'^ {j} ] = 4 i \epsilon_{ijk} T'^{k}
\\ \nonumber
[P_a, G] = m \gamma_a G \; &\;& \;
[M_{ab}, G] = \frac{\gamma_{ab}}{2} G \\ \nonumber
[P_{a}, G'] = -m \gamma_{a} G' \; &\;& \;
[M_{ab}, G'] = \frac{\gamma_{ab}}{2} G' \\ \nonumber
[T^{i}, G] = -2 \sigma^{i} G \; &\;& \;
[T^{i}, G'] = -2 \sigma^{i} G' \\ \nonumber
\{ G^{\alpha}, G^{\dagger \beta} \}
&=& \left[\delta^{\alpha \beta}
(-\frac{P_a \gamma^a}{4} +m \frac{M_{cd}\gamma^{cd}}{4} )
-m \frac{ \sigma^i_{\alpha\beta}
T^{i}}{4} \right]\gamma_{0} \\ \nonumber
\{ G'^{\alpha}, G'^{\dagger \beta} \}
&=& \left[ \delta^{\alpha \beta}
(- \frac{P_a \gamma^a}{4} - m \frac{ M_{cd}\gamma^{cd} }{4} )
+m \frac{ \sigma^i_{\alpha\beta}
T'^{i} } {4} \right] \gamma_{0} \\ \nonumber
\end{eqnarray}
Our conventions are as follows. $a,b,c \ldots$ denote the tangent
space indices of the $AdS$ space. The tangent space metric has
signature $(-1,1,1)$. $i,j,k$ take values from $1$ to $3$
corresponding to the generators of $SU(2)$.
The supercharges $G$ and $G'$ are Dirac fermions in $AdS$ space
which transform in the fundamental of the two $SU(2)$'s respectively.
$1/m$ refers to the radius of $AdS_3$ which is related to
the cosmological constant.
The $\gamma$ matrices are given by
\begin{equation}
\gamma_0 = \left( \begin{array} {cc}
0 &1 \\ -1 & 0 \end{array} \right), \;\;
\gamma_1 = \left( \begin{array}{cc}
0 & 1 \\ 1 & 0 \end{array} \right), \;\;
\gamma_2 = \left( \begin{array}{cc}
1 & 0 \\ 0 & -1 \end{array} \right)
\end{equation}
$\sigma^i$ denotes the Pauli matrices and $\alpha , \beta $
denotes
the $SU(2)$ indices. We define
$ \gamma_{ab}=(1/2)[\gamma_a, \gamma_b] $. The following
formulae are useful
\begin{equation}
\gamma_{ab}= \epsilon_{abc}\gamma^c , \;\;\; \epsilon_{123}=1,
\;\;\; \epsilon_{abc}\epsilon^{abc}= -3!
\end{equation}
To write the Maurer-Cartan equations we introduce the gauge field
one-forms
\begin{eqnarray}
\omega_{ab}= dx^\mu \omega_{ab\mu}, \; &\;& \; e_a= dx^\mu e_{a\mu} \\
\nonumber
\psi^\alpha = dx^\mu \psi_\mu^\alpha , \; & \;& \;
\psi^{'\alpha} = dx^\mu \psi_\mu^{'\alpha} \\ \nonumber
A^i=dx^\mu A_\mu^i , \; &\;& \;
A'^i=dx^\mu A'^i_\mu
\end{eqnarray}
associated with the generators $M_{ab}, P_{a}, G^\alpha , G'^\alpha ,
T^i , T'^i $ respectively. In other words
\begin{equation}
\Gamma = \omega_{ab}M^{ab} + e_a P^a + \bar{\psi}G + \bar{G}\psi +
\bar{\psi'}G' + \bar{G}'\psi' + A^i T^i + A'^i T'^i
\end{equation}
Thus the field content of the
$SU(1,1|2)\times SU(1,1|2)$ supergravity consists of a vielbein
$e_{a\mu},$ two $SU(2)$ gauge fields $A_\mu^i ,\, A'^i_\mu $ and
two gravitini $\psi_\mu^{\alpha},\, \psi'^{\alpha}_\mu$ transforming
as the fundamental of the two $SU(2)$'s. The gravitini are Dirac
fermions. $\omega_{ab\mu}$ is the spin connection.
The Maurer-Cartan equations for the one-forms turn out to be
\begin{eqnarray}
\label{Maucar}
de_a &=& -\eta^{bc} e_b \omega_{ca} + \frac{ \bar{\psi} \gamma_a
\psi }{4} + \frac{ \bar{\psi' } \gamma_a \psi'}{4} \\ \nonumber
d\omega_{ab} &=& -4 m^2 e_a e_b - \eta^{cd}\omega_{ac}
\omega_{db} +m \frac{ \bar{\psi} \gamma_{ab} \psi } {2}
-m \frac{ \bar{\psi' \gamma_{ab}} \psi' }{2} \\ \nonumber
d\psi &=& m e_a\gamma^a \psi -
\frac{\omega_{ab}\gamma^{ab} \psi }{4} + 2A^i\sigma^i \psi \\
\nonumber
d\psi' &=& - m e_a\gamma^a \psi'
- \frac{\omega_{ab}\gamma^{ab} \psi' }{4} +2A'^i\sigma^i \psi' \\
\nonumber
dA^i &=& 2i \epsilon_{ijk} A^j A^k -
m \frac{ \bar{\psi}\sigma^i \psi } {4} \\ \nonumber
dA'^i &=& 2i \epsilon_{ijk} A'^j A'^k +
m \frac{ \bar{\psi'}\sigma^i \psi' } {4} \\ \nonumber
\end{eqnarray}
Here $\bar{\psi}$ is defined as $\psi^{\dagger}\gamma_0$.
\section{ The Action. }
The Maurer-Cartan equations of the supergroup $SU(1,1|2)\times
SU(1,1|2)$ written above give the equations of motion of the
supegravity we require. These equations can be obtained as the
Euler-Lagrange equations of the following action.
\begin{eqnarray}
\label{action}
S &=&\!
\int d^3 x \left[ \right. \frac{eR}{2} + 4m^2e \\ \nonumber
-\frac{\epsilon^{\mu\nu\rho} }{2} \bar{\psi}_{\mu} {\cal D}_{ \nu}
\psi_{\rho} \!\!&-&\!\! \frac{2\epsilon^{\mu\nu\rho} } { m } ( A_{\mu}^i
\partial_{\nu} A_{\rho}^i - \frac{ 4i\epsilon_{ijk} }{3} A_{\mu}^i
A_{\nu}^j A_{\rho}^k ) \\ \nonumber
-\frac{\epsilon^{\mu\nu\rho} }{2} \bar{\psi'}_{\mu} {\cal D'}_{ \nu}
\psi'_{\rho} \!\!&+&\!\! \frac{2\epsilon^{\mu\nu\rho}} { m }( A'^i_{\mu}
\partial_{\nu} A'^i_{\rho} - \frac{ 4i\epsilon_{ijk} }{3} A'^i_{\mu}
A'^j_{\nu} A'^k_{\rho}) \left. \right] \\ \nonumber
\end{eqnarray}
Where ${\cal D}_{\nu } = \partial_{\nu} + \frac{\omega_{ab
\nu }\gamma^{ab} }{4} - m e_{a \nu}\gamma^{a} -2 A_{ \nu}^i\sigma^i $
and
${\cal D'}_{\nu } = \partial_{\nu} + \frac{\omega_{ab
\nu }\gamma^{ab} }{4} + m e_{a \nu}\gamma^{a} -2 A'^i_{ \nu}\sigma^i
$.The above Lagrangian reduces to
that of \cite{ISQTOW} if the gauge group is $U(1)$. This makes it easy
to generalize the supersymmetry transformations for the non-abelian
case. The first equation in (\ref{Maucar}) is the following
constraint which defines $\omega_{ab\mu}$
\begin{equation}
\partial_{[\mu} e_{a \nu]} + \omega_{ab[\mu}e^{b}_{\nu]}=
\frac{1}{4}(\bar{\psi}_{[\mu}\gamma_a \psi_{\nu]}+
\bar{\psi'}_{[\mu}\gamma_a \psi'_{\nu]})
\end{equation}
This is also the equation of motion obtained by variation of the
action in (\ref{action}) by
$\omega_{ab\mu}$ treating the spin connection and the
vielbein as independent fields. Thus we use the ``1.5 order formalism
'' in verifying the invariance of the action under supersymmetry
transformations.
The supersymmetry transformations under which the action is invariant
are
\begin{eqnarray}
\delta e^a_{\mu} &=& \frac{1}{4}\bar{\epsilon}\gamma^a \psi_{\mu}
+\frac{1}{4}\bar{\epsilon'}\gamma^a \psi'_{\mu} + cc \\ \nonumber
\delta \psi_{\mu} &=& {\cal D}_{\mu}\epsilon \\ \nonumber
\delta \psi'_{\mu} &=& {\cal D'}_{\mu}\epsilon' \\ \nonumber
\delta A^i_{\mu} &=& -\frac{m}{4} \bar{\epsilon}\sigma^i \psi + cc \\
\nonumber
\delta A'^i_{\mu} &=& \frac{m}{4} \bar{\epsilon'} \sigma^i \psi' + cc
\end{eqnarray}
here $cc$ denotes complex conjugate.
To verify these supersymmetry transformations it is helpful to write
the curvature and the cosmological constant term in the
action (\ref{action}) as
\begin{equation}
\int d^3x \left(
-\frac{1}{2} e_{a\rho}\epsilon^{\rho\mu\nu}\epsilon^{abc}
(\, \partial_{[\mu}\omega_{bc\nu]} + \omega_{bd[\mu}\omega^d_{c\nu]}\, )
-\frac{2m^2}{3}
\epsilon^{\mu\nu\rho}\epsilon^{abc}e_{a\mu}e_{b\nu}e_{c\rho} \right)
\end{equation}
In course of the verification one arrives at the following
typical four-fermion term
\begin{equation}
\frac{m}{4} \left[
-\epsilon^{\mu\nu\rho}(\, \bar{\psi}_{\mu}\gamma_a \psi_\nu
\, ) (\, \bar{\epsilon}\gamma^a \psi_\rho \, )
+ \epsilon^{\mu\nu\rho} (\, \bar{\psi}_\mu \sigma^i \psi_\nu \, )
(\, \bar{\epsilon}\sigma^i \psi_\rho \, ) \right]
\end{equation}
It can be shown that such terms cancel by means of a Fierz indentity
for non-commuting fermions which is
\begin{equation}
(\,\bar{\lambda} \psi \, )(\, \bar{\xi} \chi\, ) = -\frac{1}{2}
\left[\, (\,\bar{\lambda}\chi\, )(\, \bar{\xi}\psi\, ) +
(\, \bar{\lambda}\gamma^a\chi \, )
(\, \bar{\xi}\gamma_a \psi\, )\,\right]
\end{equation}
The additional four-fermion cross-terms simply cancel.
\section{Conclusions.}
In this letter we have constructed the Anti-de Sitter supergravity
associated with the supergroup $SU(1,1|2)\times SU(1,1|2)$. To
construct it we used the fact that the action can be written as the
integral of the Chern-Simons three-form associated with the
supergroup. It is also important to note that
this procedure of constructing $AdS$ supergravity can be generalized
to other three-dimensional $AdS$ supergroups for which the
corresponding supergravity has not yet been constructed.
It would be interesting to couple this pure supergravity
to matter. That such a coupling should
exist is clear from the fact that
the supergravity describing the near horizon geometry of the D1-D5
system contains matter as well. This would give
further understanding of the Maldacena correspondence for this case.
\section{Acknowledgments.}
The author wishes to acknowledge fruitful discussions with
Gautam Mandal and Spenta Wadia
which provided the motivation for construction of the action.
He also gratefully acknowledges discussions with S. Vaidya who
participated in the early phase of this project.
\noindent
{\bf Note added:} After this work was completed we received \cite{TAN}
which overlaps with some portion of this letter.
|
\section{Introduction}
{\bf CPT}~symmetry has an impressive {\em theoretical} pedigree
as an almost inescapable consequence of Lorentz invariant
quantum field theories. Observation of a {\bf CP}~asymmetry
is therefore usually seen as tantamount to the
discovery of a violation of time reversal invariance
{\bf T} . However the {\em experimental} verification
of {\bf CPT}~invariance is much less impressive.
Furthermore the
emergence of superstring theories has opened -- by their
fundamentally non-local structure -- a
{\em theoretical} backdoor through which
{\bf CPT}~breaking might slip in.
This asks for carefully analysing the
empirical basis of {\bf CPT}~invariance and the
degree to which an observable can establish
{\bf T}~violation {\em directly}, i.e. without invoking
the {\bf CPT}~theorem\cite{ellis}.
In addressing this issue, we will rely on as few
other theoretical principles as possible:
since we view the observation of {\bf CPT}~violation
as a rather exotic possibility, we
believe we should accept other theoretical restrictions very
reluctantly only.
Data from the CPLEAR collaboration have provided
{\em direct} evidence for {\bf T}~violation\cite{CPL}. In this note we want
to address the following questions:
\begin{itemize}
\item
To which degree and in which sectors of $\Delta S \neq 0$
dynamics
is {\bf T}~violated?
\item
How accurately is the validity of {\bf CPT}~invariance established
{\em experimentally}?
\item
Which conclusions can be drawn {\em without} invoking the
Bell-Steinberger relation.
\item
Which is the most promising -- or the least hopeless --
observable for finding {\bf CPT}~violations in kaon decays?
\end{itemize}
The reader might wonder why we are insisting on analyzing {\bf T}~
symmetry without assuming the Bell-Steinberger relation. After all,
it is viewed as just a consequence of unitarity.
Yet the following has to be kept in mind: when
contemplating the possibility of {\bf CPT}~violation
-- a quite remote and exotic scenario -- we should
not consider the Bell-Steinberger relation sacrosanct.
The latter is based on the assumption that all relevant
decay channels are known.
Since the major branching fractions have been
measured with at best an error of 1\%, some yet
undetermined decay mode
with a branching fraction of $10^{-3}$ can easily
be hidden \cite{kabir}.
We are
{\em not} arguing that this is a likely scenario -- it is
certainly not! However we do not view it to be more exotic than
{\bf CPT}~violation. Then it does not make a lot of sense to us to
allow for the latter while forbidding the former.
The paper will be organized as follows:
after briefly reviewing the formalism relevant
for $K^0 - \bar K^0$ oscillations in
Sect. \ref{FORMALISM} we list
the direct evidence for {\bf T}~being violated in Sect. \ref{T};
in Sect. \ref{CPT} we analyse the phases of
$\eta _{+-}$ and $\eta _{00}$;
after evaluating what can be learnt from
$K_L \rightarrow \pi ^+\pi ^- e^+e^-$ in Sect. \ref{TODD}, we
give our conclusions in Sect. \ref{SUMMARY}.
\section{Formalism
\label{FORMALISM}}
To introduce our notation and make the paper self-contained
we shall
record here the standard formalism for the neutral
$K$ meson system.
\subsection{$\Delta S=2$ Transitions}
The time dependence of the state $\Psi$, which is a
linear combination
of $K^0$ and $\overline K^0$, is given by
\begin{equation}
i \hbar \frac{\partial}{\partial t} \Psi(t) = {\cal H}\Psi(t)
\; \; \; , \; \; \;
\Psi (t) = \left( \matrix {K^0(t) \cr \overline K^0(t) \cr } \right).
\mlab{Schroed2}
\end{equation}
The $2 \times 2$ matrix ${\cal H}$ can be expressed
through the
identity and the Pauli matrices \cite{lee}
\begin{equation}
{\cal H} \equiv {\bf M} - \frac{i}{2} {\bf \Gamma} =
E_1 {\bf \sigma}_1 + E_2 {\bf \sigma}_2 +
E_3 {\bf \sigma}_3 - i D {\bf 1} .
\mlab{CPTMass}
\end{equation}
with
\begin{eqnarray}
E_1 &=& \Re M_{12}-{i\over 2} \Re \Gamma_{12}\; , \; \;
E_2 = -\Im M_{12}+{i\over 2} \Im \Gamma_{12}
\nonumber\\
E_3 &=& {1\over 2} (M_{11}-M_{22})-{i\over 4}
(\Gamma_{11}-
\Gamma_{22}) , \;
D = {i\over 2} (M_{11}+M_{22})+{1\over 4}
(\Gamma_{11}+\Gamma_{22}).
\mlab{def2}
\end{eqnarray}
It is often convenient to use instead {\em complex} numbers
$ E, \theta$, and $\phi$ defined by
\begin{eqnarray}
E_1=E\, {\rm sin}\theta \, {\rm cos}\phi ,~~
E_2&=&E\, {\rm sin}\theta \, {\rm sin}\phi,~~
E_3=E\, {\rm cos}\theta \nonumber\\
E&=&\sqrt{E_1^2+E_2^2+E_3^2}
\; .
\mlab{6.27}
\end{eqnarray}
The mass eigenstates are given by
\begin{eqnarray}
|K_S \rangle &=& p_1 |K^0\rangle + q_1 |\overline K^0\rangle \nonumber\\
|K_L \rangle &=& p_2 |K^0\rangle - q_2 |\overline K^0\rangle
\mlab{ES}
\end{eqnarray}
with the {\em convention}
${\bf CP} |K^0 \rangle = |\overline K^0 \rangle$ and
\begin{eqnarray}
p_1 &=& N_1 {\rm cos}\frac{\theta}{2} , \;
q_1 = N_1 e^{i \phi}{\rm sin}\frac{\theta}{2} \nonumber\\
p_2 &=& N_2 {\rm sin}\frac{\theta}{2} , \;
q_2 = N_2 e^{i \phi}{\rm cos}\frac{\theta}{2} \nonumber\\
N_1 &=& \frac{1}{\sqrt{|{\rm cos}\frac{\theta}{2}|^2
+ |e^{i \phi}{\rm sin}\frac{\theta}{2}|^2}} \nonumber\\
N_2 &=& \frac{1}{\sqrt{|{\rm sin}\frac{\theta}{2}|^2
+ |e^{i \phi}{\rm cos}\frac{\theta}{2}|^2}}\; .
\end{eqnarray}
The discrete symmetries impose the following
constraints:
\begin{eqnarray}
{\bf CPT}~{\rm or}~ {\bf CP}~{\rm invariance}~~
\Longrightarrow &&\cos\theta =0,
~~M_{11}=M_{22},~~\Gamma _{11}=\Gamma_{22}\nonumber\\
{\bf CP}~{\rm or}~{\bf T}~{\rm invariance}~~\Longrightarrow &&
\phi =0,~~ \Im M_{12}=0={\rm Im}\Gamma_{12}
\mlab{DS2CON}
\end{eqnarray}
\subsection{Nonleptonic Amplitudes}
We write for the amplitudes describing decays into
final states with isospin $I$:
\begin{eqnarray}
T(K^0 \rightarrow [\pi \pi ]_I) &=& A_I e^{i\delta _I},\nonumber\\
T(\overline K^0 \rightarrow [\pi \pi ]_I) &=& \overline A_I e^{i\delta _I}
\end{eqnarray}
where the strong phases $\delta _I$
have been factored out and
find:
\begin{eqnarray}
{\bf CPT} \; \; {\rm invariance} \; \;
\Longrightarrow && A_I = \overline A^*_I \nonumber\\
{\bf CP} \; \; {\rm invariance} \; \;
\Longrightarrow && A_I = \overline A_I \nonumber\\
{\bf T} \; \; {\rm invariance} \; \;
\Longrightarrow && A_I = A^*_I
\end{eqnarray}
The expressions for $\eta_{+-}$ and $\eta_{00}$
\begin{eqnarray}
\eta _{+-} &=& \frac{1}{2} \left( \Delta _0 -
\frac{1}{\sqrt{2}} \omega e^{i(\delta _2 - \delta _0)}
(\Delta _0 - \Delta _2) \right), \nonumber\\
\eta _{00} &=& \frac{1}{2} \left( \Delta _0 +
\sqrt{2} \omega e^{i(\delta _2 - \delta _0)}
(\Delta _0 - \Delta _2) \right), \nonumber\\
\Delta _I &=& \frac{1}{2} \left( 1 - \frac{q_2}{p_2}
\frac{\overline A_I}{A_I} \right) \; , \;
|\omega | \equiv \left|\frac{A_2}{A_0}\right| \simeq \frac{1}{20},
\mlab{ETA}
\end{eqnarray}
are valid
{\em irrespective} of {\bf CPT}~symmetry.
\subsection{Semileptonic Amplitudes}
The general amplitudes for semileptonic $K$ decays
can be expressed as follows:
\begin{eqnarray}
{\langle} l^+\nu \pi^-|{\cal H}_W|K^0{\rangle} &=&F_l(1-y_l)\nonumber\\
{\langle} l^+\nu \pi^-|{\cal H}_W|\overline K^0{\rangle}
&=&x_l F_l(1-y_l)\nonumber\\
{\langle} l^-\overline\nu \pi^+|{\cal H}_W|K^0{\rangle} &=&
\overline x_l^*F_l^*(1+y_l^*)\nonumber\\
{\langle} l^-\overline\nu \pi^+|{\cal H}_W|\overline K^0{\rangle}
&=&F_l^*(1+y_l^*).
\mlab{9.1000}
\end{eqnarray}
with the selection rules
\begin{tabbing}
{\hskip 1cm}\=$\Delta S=\Delta Q$ rule:\hskip 2cm\=
$x_l=\overline x_l=0$\\
\>{\bf CP}~invariance: \>$x_l=\overline x_l^*;~~~F_l=F_l^*;
~~~y_l=- y_l^*$\\
~\> {\bf T}~invariance:\> ${\rm Im}~F={\rm Im}~y_l=
{\rm Im}~x_l={\rm Im}~\overline x_l=0$\\
~\> {\bf CPT}~invariance:\> $y_l=0,~x_l=\overline x_l$.
\end{tabbing}
\section{Direct Evidence for {\bf T}~Violation
\label{T}}
The so-called Kabir test\cite{kabir2} represents a quantity that
probes {\bf T}~violation without reference to
{\bf CPT}~symmetry:
\begin{equation}
A_{{\bf T}} \equiv \frac{\Gamma (K^0 \rightarrow \overline K^0) -
\Gamma (\overline K^0 \rightarrow K^0)}
{\Gamma (K^0 \rightarrow \overline K^0) +
\Gamma (\overline K^0 \rightarrow K^0)}
\mlab{KABIR}
\end{equation}
A nonvanishing $A_{\bf T}$ requires
\begin{equation}
M_{12}-\frac{i}{2}\Gamma_{12}\ne
M_{21}-\frac{i}{2}\Gamma_{21}.
\end{equation}
which constitutes {\bf CP}~as well as {\bf T}~ violation.
Associated production flavor-tags the
{\em initial} kaon. The flavor of the
{\em final} kaon is inferred from semileptonic
decays; i.e., we measure the
{\bf CP}~asymmetry
\begin{equation}
A_{{\bf CP}} \equiv \frac{\Gamma(K\rightarrow l^-\nu\pi^+)-
\Gamma(\overline K\rightarrow l^+\nu\pi^-)}
{\Gamma(K\rightarrow l^-\nu\pi^+)+
\Gamma(\overline K\rightarrow l^+\nu\pi^-)}
\mlab{1}
\end{equation}
Yet a violation of
{\bf CPT}~invariance and/or of the $\Delta S = \Delta Q$ rule
can produce an asymmetry in the latter
-- $A_{{\bf CP}} \neq 0$ -- without one being
present in the former -- $A_{{\bf T}} =0$.
These issues have to be tackled first.
There is nothing new in our remarks on this subject;
we add them for clarity and completeness.
Analysing the asymmetries in
$\Gamma(\overline K^0(t)\rightarrow l^+\nu K^-)$ vs.
$\Gamma( K^0(t)\rightarrow l^- \bar \nu K^+)$ and
$\Gamma(\overline K^0(t)\rightarrow l^-\bar \nu K^+)$ vs.
$\Gamma( K^0(t)\rightarrow l^+ \nu K^-)$
for large times $t$ CPLEAR has found \cite{CPLEARCPT}
\begin{equation}
\Re\cos\theta= (6.0\pm 6.6\pm 1.2)\times 10^{-4}.
\mlab{cp1}
\end{equation}
>From the decay rate evolution they have inferred
\begin{eqnarray}
\Im\cos\theta= (-3.0\pm 4.6\pm 0.6)\times 10^{-2},\nonumber\\
\frac{1}{2}\Re(x_l-\bar x_l)= (0.2\pm 1.3\pm 0.3)\times 10^{-2},\nonumber\\
\frac{1}{2}\Im(x_l+\bar x_l)= (1.2\pm 2.2\pm 0.3)\times 10^{-2} \; .
\mlab{CPTSLDATA}
\end{eqnarray}
While there is no sign of {\bf CPT}~violation in any of these
observables, the bounds of \mref{CPTSLDATA} are not overly
restrictive.
Another input is provided by the charge asymmetry
in semileptonic $K_L$ decays for which the general
expression reads as follows:
\begin{eqnarray}
\delta _{\rm Lept} &=&
\frac{\Gamma (K_L \rightarrow l^+ \nu \pi ^-) -
\Gamma (K_L \rightarrow l^- \nu \pi ^+)}
{\Gamma (K_L \rightarrow l^+ \nu \pi ^-) +
\Gamma (K_L \rightarrow l^- \nu \pi ^+)}\nonumber\\
&= &{\rm Im}\, \phi - {\rm Re \, cos}\, \theta -
{\rm Re}\; (x_l - \overline x_l) - 2 {\rm Re}\, y_l.
\end{eqnarray}
{\bf CPT}~violation, if it exists, is most likely to surface in $M_{12}$,
which is of second order in the weak interactions. It is
then natural to assume {\em semileptonic} decay amplitudes
to conserve {\bf CPT}~, which is fully consistent with
\mref{CPTSLDATA}, but not confirmed to the required level:
\begin{equation}
x_l-\bar x_l = 0,~~~~~~~ {\rm or}~~~~~~y_l= 0.
\end{equation}
With this
{\em assumption}, and from the data \cite{PDG}
\begin{equation}
\delta _{\rm Lept} = (3.27 \pm 0.12) \times 10^{-3} .
\end{equation}
one obtains
\begin{equation}
{\rm Im}\, \phi - {\rm Re \, cos}\,
\theta =(3.27 \pm 0.12) \times 10^{-3}
\end{equation}
and infers from \mref{cp1}
\begin{equation}
\Im\phi=(3.9\pm 0.7)\times 10^{-3},
\end{equation}
showing that {\bf T}~is violated in kaon dynamics.
This result can be stated more concisely as follows
\cite{CPLEAR2}:
\begin{equation}
A_{T} \simeq A_{{\bf CP} } =
(6.6 \pm 1.3\pm 1.0 ) \times 10^{-3}.
\mlab{1}
\end{equation}
In order to get a result independent of the assumption
that {\em direct semileptonic} kaon decays obey
{\bf CPT}~symmetry,
the CPLEAR collaboration has employed constraints from the
Bell-Steinberger relation to deduce the bound \cite{CPL}
\begin{equation}
\frac{1}{2}{\rm Re}\; (x_l - \overline x_l) - {\rm Re}\,
y_l =(-0.4\pm 0.6)\times 10^{-3} \; ,
\end{equation}
which again is fully consistent with {\bf CPT}~invariance of the
semileptonic decays. This results in establishing violation of
{\bf T}~symmetry -- provided the assumption mentioned above is valid.
\section{Phases of $\eta _{+-}$ \& $\eta _{00}$
and {\bf CPT}
\label{CPT}}
\subsection{Basic Expressions}
Manipulating \mref{ETA} we obtain
through ${\cal O}(\phi)$ and
${\cal O}$(cos$\theta$)
\begin{equation}
|\eta _{+-}|
\frac{\Delta\Phi}{{\rm sin}\phi _{SW}}=
\left( \frac{M_{\overline K} - M_{K}}{2\Delta M} +
R_{direct} \right)
\mlab{CPTTEST}
\end{equation}
\begin{eqnarray}
\Delta\Phi &\equiv&
\frac{2}{3} \phi _{+-} + \frac{1}{3}\phi _{00}
- \phi _{SW} \nonumber\\
R_{direct} &=&
\frac{1}{2} \Re~ r_A -\frac{ie^{-i\phi_{SW}}}{\sin\phi_{SW}}
\sum_{f\ne [2\pi]_0}\epsilon(f) \nonumber\\
r_A &\equiv& \frac{\overline A_0}{A_0} - 1,
\; \; \phi_{SW}\equiv
{\rm tan}^{-1}\frac{2\Delta M}{\Delta \Gamma}\nonumber\\
\epsilon(f) &=& e^{i\phi_{SW}}i\cos\phi_{SW}
\frac{\Im \Gamma_{12}(f)}
{\Delta \Gamma} \; .
\end{eqnarray}
Since {\bf CPT}~symmetry predicts $M_K=M_{\overline K}$
and $\Re~r_A={\cal O}(\xi_0^2)$, where $\xi_0=\arg~A_0$,
it implies
$|\Delta\Phi|=0$ to within the
uncertainty given by $|\sum_{f\ne [2\pi]_0}\epsilon(f)|$;
the latter sum thus represents the theoretical `noise'.
\subsection{Estimating $\sum \epsilon (f)$
}
The major kaon decay modes fall into two classes, namely
flavor-{\em non}specific or flavor-specific channels.
\begin{itemize}
\item
With
$A_f=\matel{f}{H_W}{K^0}$ and $\overline A_f=
\matel{f}{H_W}{\overline K^0}$,
we have, to first order in {\bf CP}~violation,
\begin{equation}
\Im \Gamma_{12}(f)=i\eta_f\Gamma(K\rightarrow f)\left( 1-
\eta _f
\frac{\overline A_f}{A_f}\right) \; ,
\mlab{iden}
\end{equation}
for {\bf CP}~eigenstates with eigenvalue $\eta _f$.
Im $\Gamma _{12}(f) \neq 0$ can hold only if
$\overline A_f \neq \eta _f A_f$, i.e. if there is
{\em direct} {\bf CP}~violation in the channel $f$.
Using data on $\epsilon ^{\prime}$,
Br($K_{L,S} \rightarrow 3\pi$) and
\begin{equation}
{\rm Im}\, \eta _{+-0} =
\left(-2 \pm 9~{+2\atop -1}\right)\times 10^{-3}\; ,
\; \;
\Im\eta _{000}=0.07\pm 0.16
\; \; \; \cite{3pi,BLOCH},
\mlab{Im410}
\end{equation}
where
\begin{equation}
\eta_{+-0,000} \equiv
\frac{1}{2} \left( 1 + \frac{q_1}{p_1}
\frac{\bar A(\pi ^+ \pi ^- \pi ^0, 3\pi ^0)}
{A(\pi ^+ \pi ^- \pi ^0, 3\pi ^0)}\right),
\end{equation}
we obtain
\begin{eqnarray}
|\epsilon(3\pi ^0)| &<& 1.1 \times 10^{-4}\nonumber\\
|\epsilon([2\pi ]_2)| &\simeq& 0.28 \times 10^{-6}\nonumber\\
|\epsilon((\pi ^+ \pi^- \pi ^0 )_{{\bf CP}~-[+]})| &<&
5\, [0.2]\times 10^{-6},
\end{eqnarray}
\item
Allowing for a violation of the
$\Delta Q = \Delta S$ rule in semileptonic decays
as expressed by
$
x_l \equiv \frac{\matel{l^+ \nu \pi ^-}
{{\cal H}_W}{\overline K}}
{\matel{l^+ \nu \pi ^-}{{\cal H}_W}{K}}$,
we find
\begin{equation}
|\epsilon (\pi l \nu )| \leq 4 \times 10^{-7}.
\end{equation}
\end{itemize}
\subsection{Quantifying {\bf CPT}~Tests}
With the measured values for the phases
$\phi _{+-},~\phi _{00}- \phi _{+-}$,
and $\phi _{SW}$
we arrive at a result quite consistent with zero
\cite{PDG,adler6,SCHUBERT}:
\begin{equation}
\Delta\Phi=
0.01^o \pm
0.7^o |_{exp.}
\pm 1.5^o |_{theor.} \; ,
\mlab{CPTTEST}
\end{equation}
i.e., the phases $\phi _{+-}$ and $\phi _{00}$ agree with their
{\bf CPT}~prescribed values to within $2^o$ .
{\bf CPT}~invariance is thus probed to about the
$\delta \phi /\phi _{SW} \sim
5\% $ level.
The relationship between $\phi _{+-}$, $\phi _{00}$ on one
side and $\phi _{SW}$ on the other is a truly
meaningful gauge; yet the numerical accuracy of that test is
not overwhelming. The theoretical error can be reduced significantly
by making quite reasonable assumptions on {\bf CP}~violation; however,
we refrain from doing so based on our belief that assuming
observable {\bf CPT}~breaking is not very reasonable to start with.
In \mref{CPTTEST}, the {\em theoretical} uncertainty
$\sum_f\epsilon(f)$ provides the limiting factor for
this test\cite{shabalin}; it is dominated by
$K \rightarrow 3\pi ^0$. Future experiments could reduce the uncertainty
by a factor of up to two \cite{BLOCH}.
Alternatively we can state
\begin{equation}
\frac{M_{\overline K} - M_K}{2\Delta M}
+ \frac{1}{2} {\rm Re}~ r_A
= \left( 0.06 \pm 4.0|_{exp}
\pm 9 |_{theor}
\right) \times 10^{-5} .
\end{equation}
Yet
$\Delta M$ does not provide a meaningful calibrator; for it
arises basically unchanged
even if {\bf CP}~were conserved while the latter would imply
$M_{\overline K} - M_K = 0$ and $r_A=0$ irrespective of
{\bf CPT}~breaking.
The often quoted
truly spectacular bound (for $R_{direct} = 0$)
\begin{equation}
\frac{M_{\overline K} - M_K}{ M_K} =
(0.08 \pm 5.3|_{exp}) \times 10^{-19}
\end{equation}
definitely
overstates the numerical degree to which
{\bf CPT}~invariance has been probed. $M_K$ is not generated
by weak interactions and thus cannot serve as a meaningful yardstick.
In summary: while no hint has has found
indicating a limitation to {\bf CPT}~symmetry, the {\em experimental}
evidence for it is far from overwhelming:
\begin{itemize}
\item
Comparing the phases of $\eta _{+-}$ and $\eta _{00}$ with the
superweak phase constitutes a meaningful test of
{\bf CPT}~symmetry. Yet there is a `noise' level of about
$2^o$ that cannot be reduced significantly \cite{BLOCH}.
\item
Relating the bound on the difference
$| M_{\overline K} - M_K|$ to the
kaon mass itself is extremely impressive numerically --
yet meaningless.
\item
When entertaining the idea of {\bf CPT}~violation, we should not
limit our curiosity to a single quantity like
$\Delta \Phi$ (or equivalently
$M_{\overline K} - M_K$).
\item
Finally, the reader should be reminded that {\bf CPT}~symmetry implies
$\Delta\Phi\ll\phi_{SW}$ but the converse does not follow.
\end{itemize}
\section{Consequences in a {\bf T}~Conserving World \label{TODD}}
\subsection{Reproducing $\eta _{+-}$}
Assuming nature to conserve {\bf T} ,
which implies $\phi=0$, see \mref{DS2CON}, we have:
\begin{eqnarray}
&&\frac{|\eta _{+-}|\Delta\Phi}{{\rm sin}\phi _{SW}}=
- \frac{M_{11} - M_{22}}{2\Delta M}
+\frac{1}{2} r_A ,\nonumber\\
&&
\frac{|\eta_{+-}|}{\cos\phi_{SW}}=-\frac{\Gamma_{11}-
\Gamma_{22}}{4\Delta M}{\rm tg}\phi_{SW}
-\frac{1}{2} r_A.\nonumber\\
&&\Re\cos\theta= -
\frac{M_{11}-M_{22}}{\Delta M}\sin^2\phi_{SW}
+\frac{1}{2}\frac{\Gamma_{11}-\Gamma_{22}}{\Delta M}
\sin\phi_{SW}\cos\phi_{SW}.
\mlab{3}
\end{eqnarray}
Inserting the values of $\eta_{+-}$, $\phi_{SW}$ and \mref{cp1}
we can solve for the three unknowns:
\begin{eqnarray}
\frac{M_{11}-M_{22}}{\Delta M}&\simeq& r_A\simeq
(-3.9 \pm 0.7) \times 10^{-3}
\nonumber\\
\frac{\Gamma_{11}-\Gamma_{22}}{\Delta M}
&\simeq& (- 5.0 \mp 1.4) \times 10^{-3} .
\mlab{cptviol}
\end{eqnarray}
The solution is very
{\em unnatural} -- \mref{cptviol}, for example,
requires cancellation
between {\bf CPT}~violating $\Delta S=1$ and 2 amplitudes.
Yet however unnatural they may be, we must entertain
this possibility
unless we can exclude it empirically.
As a side remark, we mention that
if we invoke the Bell-Steinberger relation in its usual
form -- meaning that kaon decays are effectively saturated
by the $K \rightarrow 2\pi , \, 3\pi , \, l \nu \pi$ channels, then
we have an additional relation \cite{bloch2}:
\begin{equation}
\frac{1}{2} r_A \approx -\frac{\Gamma_{11} - \Gamma_{22}}{4\Delta M} \; ;
\mlab{b1}
\end{equation}
i.e, \mref{3} then implies $\eta _{+-} \simeq 0$.
This is not surprising since these known modes do
not exhibit any sign of {\bf CPT}~violation.
But, as we have remarked before, in testing
{\bf CPT}~we want to stay away from invoking saturation
by the known channels.
\subsection{$K\rightarrow\pi\pi$}
Where should such a large {\bf CPT}~violation show its face?
Imposing
$r_A \neq 0$ raises the prospects of unacceptably large
direct {\bf CP}~violation in $K_L \rightarrow \pi \pi$.
\mref{ETA} can be reexpressed as follows:
\begin{equation}
\epsilon \simeq \frac{1}{\sqrt{1+(\frac{\Delta \Gamma}
{2\Delta M})^2}}e^{i\phi _{SW}}
\left( -\frac{{\rm Im}M_{12}}{\Delta M_K} + \xi _0\right)
\mlab{6.3.12}
\end{equation}
\begin{equation}
\epsilon ^{\prime} =
\frac{1}{2\sqrt{2}}\omega e^{i(\delta _2 - \delta _0)}
\frac{q_2}{p_2}
\left( \frac{\overline A_0}{A_0} - \frac{\overline A_2}{A_2}\right)
\mlab{6.107}
\end{equation}
If {\bf T}~is conserved,
$\frac{q_2}{p_2}\left( \frac{\overline A_0}{A_0} -
\frac{\overline A_2}{A_2}\right) $ is real
and \mref{6.107} then tells us \cite{PDG}
\begin{equation}
\arg\left(\frac{\epsilon'}{\epsilon}\right)
=\delta_2 - \delta_0 -\phi_{SW}\simeq - (85.5\pm 4)^\circ.
\mlab{RE0}
\end{equation}
Therefore
\begin{eqnarray}
{\rm Re} \frac{\epsilon ^{\prime}}{\epsilon} &\simeq&
\cos (\delta _2 - \delta _0 - \phi _{SW}) \cdot
\frac{|\omega|}{2\sqrt{2} |\eta _{+-}|} \cdot
|\Delta _0 - \Delta _2| \nonumber\\
&=& 0.035 \cdot \left( 0.087 ^{+0.061}_{-0.078} \right) \cdot
\left| \frac{r_A^{\prime}}{r_A} - 1 \right| =
\left( 3.0^{+2.2}_{-2.7}\right) \cdot 10^{-3}
\cdot \left| \frac{r_A^{\prime}}{r_A} - 1 \right|
\end{eqnarray}
where
\begin{equation}
r_A^{\prime} \equiv \frac{\overline A_2}{A_2} - 1
\mlab{RP}
\end{equation}
Some remarkable features can be read off from this expression:
\begin{itemize}
\item
For
\begin{equation}
\delta _2 - \delta _0 - \phi _{SW} = 90 ^{\circ}
\end{equation}
which is still allowed by the data, one obtains
\begin{equation}
{\rm Re} \frac{\epsilon ^{\prime}}{\epsilon} = 0 \; .
\end{equation}
As far as $K \rightarrow \pi \pi$ is concerned this amounts to a superweak
scenario!
\item
The empirical landscape of {\bf CP}~violation has changed
{\em qualitatively}: KTeV, confirming earlier observations of
NA 31, has conclusively established the existence of
direct {\bf CP}~violation \cite{KTEVPRL}:
\begin{equation}
{\rm Re} \frac{\epsilon ^{\prime}}{\epsilon} =
\left( 2.80 \pm 0.30 \pm 0.28 \right) \cdot 10^{-3}
\end{equation}
Including previous data and preliminary results from NA 48 one
arrives at a world average of
\begin{equation}
{\rm Re} \frac{\epsilon ^{\prime}}{\epsilon} =
\left( 2.12 \pm 0.28 \right) \cdot 10^{-3}
\end{equation}
This can be reproduced with a `canonical' $r^{\prime}_A = 0$, but only
for a very narrow slice in the phase of $\epsilon
^{\prime}/\epsilon$, namely
\begin{equation}
\delta _2 - \delta _0 - \phi _{SW} \simeq - (86.5 \pm 0.5)^{\circ} \;
.
\mlab{SLICE}
\end{equation}
\item
The dominant uncertainty here enters through the phase shifts
$\delta _{0,2}$. If $\delta _2 - \delta _0 - \phi _{SW}$ falls outside
the range of \mref{SLICE}, then $r_A^{\prime} \neq 0$ is needed to
reproduce
Re$(\epsilon ^{\prime}/\epsilon)$. As an illustration consider
$ \delta _2 - \delta _0 - \phi _{SW} = 80^{\circ}$. In that case
$1/2 \leq r_A^{\prime}/r_A \leq 5/6$ had to hold to obtain
$1 \cdot 10^{-3}\leq {\rm Re}(\epsilon ^{\prime}/\epsilon )\leq 3 \cdot
10^{-3}$. Hence $r_A^{\prime} \sim - (2 \div 4) \cdot
10^{-3}$. More generally if
\begin{equation}
\delta _2 - \delta _0 - \phi _{SW} \leq 83 ^{\circ}
\end{equation}
then the observed value of Re$(\epsilon ^{\prime}/\epsilon)$ would
imply
\begin{equation}
r_A^{\prime} \leq - 10^{-3}
\mlab{RPLB}
\end{equation}
if {\bf T}~is conserved.
\item
This would have a dramatic impact on
$K^{\pm} \rightarrow \pi ^{\pm} \pi ^0$ decays. For \mref{RPLB} implies a
sizeable {\bf CPT}~asymmetry there
\begin{equation}
\frac
{\Gamma (K^+\rightarrow \pi ^+ \pi ^0) - \Gamma (K^-\rightarrow \pi ^- \pi ^0)}
{\Gamma (K^+\rightarrow \pi ^+ \pi ^0) + \Gamma (K^-\rightarrow \pi ^- \pi ^0)}
> 10^{-3}
\end{equation}
With {\bf CPT}~symmetry we predict here
a direct {\bf CP}~asymmetry
of at most ${\cal O}(10^{-6})$ due to electromagnetic
corrections. Thirty year old data yield
$(0.8 \pm 1.2) \cdot 10^{-2}$. Upcoming experiments
will produce a much better measurement.
\end{itemize}
\subsection{$K_L \rightarrow \pi ^+\pi ^- e^+e^-$}
If the photon polarization
$\vec \epsilon _{\gamma}$ in
$
K_L \rightarrow \pi ^+ \pi ^- \gamma
$
were measured, we could form the {\bf CP}~and {\bf T}~odd
correlation $P_{\perp}^{\gamma} \equiv
\langle \vec \epsilon _{\gamma} \cdot
(\vec p_{\pi ^+} \times \vec p_{\pi ^-})\rangle$.
A more practical realization of this idea is to analyze
$K_L \rightarrow \pi ^+ \pi ^- e^+ e^-$ which proceeds like
$
K_L \rightarrow \pi ^+ \pi ^- \gamma ^* \rightarrow \pi ^+ \pi ^- e^+ e^-
$.
It allows to determine a {\bf CP}~and {\bf T}~odd moment
$\langle A\rangle$ related to
$P_{\perp}^{\gamma}$ by measuring the correlation
between the $\pi ^+ \pi ^-$ and $e^+ e^-$ planes. This effect
was predicted to be \cite{segal}
\begin{equation}
\langle A\rangle = (14.3 \pm 1.3)\%
\mlab{ATH}
\end{equation}
and observed by
KTeV \cite{ktev}:
\begin{equation}
\langle A\rangle = (13.6 \pm 2.5\pm 1.2)\%
\mlab{AEXP}
\end{equation}
It is
mainly due to the interference between the bremsstrahlung
process
$K_L \Rightarrow K_{{\bf CP} +} \rightarrow \pi ^+ \pi ^- \rightarrow
\pi ^+ \pi ^- \gamma ^*$ and a one-step M1 reaction
$K_L \rightarrow \pi ^+ \pi ^- \gamma ^*$. The former is
{\bf CP}~violating and described by $\eta _{+-}$
{\em irrespective} of
the theory underlying {\bf CP}~violation.
It is a remarkable measurement since it has revealed a
huge {\bf CP}~asymmetry in a rare channel that had not been observed
before. While {\bf T}~odd correlations have been seen before in
production processes and in nonleptonic hyperon decays, those
-- due to their sheer magnitude -- had to be blamed on
final state interactions; such an explanation turned out
to be consistent with what we know about those. The
quantity $\langle A \rangle$ on the other hand is a
{\bf T}~odd correlation {\em sui generis} since it has a chance
to be generated by microscopic {\bf T}~violation.
Yet the most intriguing question is
what does this measurement teach us about
{\bf T}~violation without reference to {\bf CPT}~symmetry?
The answer is: Nothing really! For we have just shown --
by giving a concrete example -- that if we are sufficiently determined
we can dial {\bf CPT}~violation in such a way that both the modulus and
phase of $\eta _{+-}$ are reproduced even with {\bf T}~invariant dynamics,
and it is $\eta _{+-}$ that controls $\langle A \rangle$.
\subsubsection{A Comment on the Intricacies of Final State
Interactions}
It is well-known that a non-vanishing {\bf T} -odd correlation does
not necessarily establish {\bf T}~violation since final state interactions
can induce it even if {\bf T}~is conserved. Yet even so the reader
might be surprised by our findings that a
value of $\langle A\rangle$
as large as
10\% does not establish {\bf T}~violation. For it
would be tempting to argue that in the case at hand final state
interactions could not induce an effect even within an order of
magnitude of the observed size. The argument might proceed as follows:
$\langle A \rangle$ reflects the correlation between the
$\pi ^+ -\pi ^-$ and the $e^+ - e^-$ planes; their relative
orientation can be affected by final state interactions -- but only of
the electromagnetic variety; then $\langle A \rangle \gg 1\%$ could
not arise.
If nothing else, our brute force
scenario shows that such an argument is
fallacious. This can be seen also more directly.
As stated above there
are two different contributions to
$K_L \rightarrow \pi ^+ \pi ^- e^+ e^-$, namely the M1 amplitude which is
{\bf CP}~neutral, and the bremsstrahlung one due to the presence of
{\bf CP}~violation. One should note that the presence of the this
second amplitude requires neither {\bf T}~violation nor final state
interactions!
Let us assume for the moment that arg $\eta _{+-} = 0$ were to hold.
Ignoring final state interactions both in the M1 and the
bremsstrahlung amplitudes one obtains $\langle A \rangle =0$,
since the former is imaginary and the latter real now. When the final
state interactions are switched back on, they affect the two
amplitudes differently. Interference can take place, and one finds
(with arg $\eta _{+-} = 0$) $\langle A \rangle \sim 8 \%$.
How can the orientation of the
$\pi ^+ - \pi ^-$ and the
$e^+ - e^-$ planes get shifted so much by strong final
state interactions?
The fallacy of the intuitive argument sketched above derives from its
purely classical nature. In quantum mechanics it is not surprising at
all that phase shifts between coherent amplitudes
change angular correlations.
\section{Summary
\label{SUMMARY}
}
In this note we have listed the information we
can infer
on {\bf T}~and {\bf CPT}~invariance from the data on kaon decays.
Our reasoning was guided by the conviction that once we contemplate
{\bf CPT}~breaking the notion of a reasonable or natural assumption
starts to resemble an oxymoron.
Our findings can be summarized as follows:
\begin{itemize}
\item
The presence of {\bf T}~violation in $\Delta S\neq 0$ dynamics
has been shown without invoking
{\bf CPT}~symmetry through the Kabir
test performed by CPLEAR. Yet their analysis had to
assume semileptonic kaon decays to be {\bf CPT}~symmetric or
it had to impose the Bell-Steinberger relation in its
conventional form. We do not view either assumption as
qualitatively more sacrosanct than {\bf CPT}~symmetry.
\item
$\phi _{+-,00}$ lie within $2^o$ of what is expected
from {\bf CPT}~symmetry.
\item
A meaningful yardstick for calibrating bounds on
limitations to {\bf CPT}~symmetry is provided by {\bf CP}~asymmetries.
{\bf CPT}~breaking forces could -- empirically --
still be as large as
few percent
of {\bf CP}~violating forces.
\item
It is grossly misleading to calibrate the bound on
$M_{\over K} - M_K$ inferred from $\phi _{+-}$,
$\phi _{00}$ and $\phi _{SW}$ to the kaon mass.
\item
The measured values of $\eta _{+-}$ and $\eta _{00}$ provide
us with little information on the level of
{\bf T}~versus {\bf CPT}~violation.
More specifically $\eta _{+-}$ -- both its modulus as well as
its phase -- can be reproduced with {\bf T}~invariant dynamics
(unless one imposes the Bell-Steinberger relation):
\begin{itemize}
\item
This is achieved by carefully adjusting
{\bf CPT}~violation in $\Delta S=1 \& 2$ transitions.
\item
The observed level of direct {\bf CP}~violation --
$\eta _{+-} \neq \eta _{00}$ -- is {\em not} a natural consequence of
such a
scenario. However it could arise due to a fine-tuning of
$\delta _2 - \delta _0 - \phi _{SW}$ -- which had to be viewed as
completely accidental -- or to a compensation of direct {\bf CPT}~violation
in $K_L \rightarrow [\pi \pi ]_0$ and $K_L \rightarrow [\pi \pi ]_2$.
\item
In the latter subscenario one is stuck with a {\bf CPT}~asymmetry
in $K^{\pm}\rightarrow \pi ^{\pm} \pi ^0$ that could be
up to few$\times 10^{-3}$ without upsetting any known empirical
bound.
\item
The KTeV observation of a large {\bf CP}~and {\bf T}~odd correlation
in $K_L \rightarrow \pi ^+ \pi ^- e^+ e^-$ in agreement with
theoretical predictions is highly intriguing,
yet does {\em not} constitute an unequivocal
signal for {\bf T}~violation. This has also been noted before
\cite{ELLIS} using a different line of reasoning.
\end{itemize}
\item
We are fully aware that our construction is purely ad-hoc without any
redeeming theoretical feature. Nevertheless we do not view it as
l'art pour l'art (or more appropriately non-art pour non-art):
\begin{itemize}
\item
We have shown by constructing an explicit counter-example that the
{\bf T}~odd correlation observed in $K_L \rightarrow \pi ^+ \pi ^- e^+ e^-$
does {\em not} establish {\bf T}~violation without invoking the
{\bf CPT}~theorem.
\item
As a by-product we have found that $K^{\pm} \rightarrow \pi ^{\pm}\pi ^0$
could exhibit a {\bf CPT}~asymmetry large enough to become observable
soon.
\end{itemize}
\end{itemize}
Finally we would like to add the remark that even negative
searches for {\bf CPT}~violation in kaon transitions will
{\em not} free
us from the obligation to probe for such effects in beauty
meson decays at the $B$ factories.
\vskip 3mm
{\bf Acknowledgements}
\vskip 3mm
We thank Y. Nagashima for discussions.
We are grateful to P. Bloch and his colleagues for pointing out
several relevant errors and omissions in the original version
of this paper.
The work of I.I.B. has been supported by the NSF under the grant
PHY 96-0508 and that of A.I.S. by Grant-in-Aid for Special Project
Research (Physics of CP violation).
|
\section{Introduction \label{sex1}}
Whereas the
conformational statistics of a single flexible polymer chain in dilute and
semidilute solutions are understood rather well, less is
known about the inter--molecular packing. It is well understood that a
semidilute polymeric solution builds up a temporary mesh with a
mesh size, the density screening length $\xi_\rho$, which
for macromolecular solutions can become large compared to the length scales
characterizing the individual monomers \cite{degennes,descloi}.
However, the inter--molecular packing inside the mesh but still on length
scales
large compared to the chemistry dependent local length scales is as yet
unclear. It has been the focus of recent neutron scattering experiments
\cite{ullman,jannink}, scaling considerations and field theoretic calculations
\cite{jannink,schaefer}, and of computer simulations \cite{yamakov}. Older
theories for the inter--molecular structure, which either used the random phase
approximation (RPA) \cite{benoit}
or the assumption of Gaussian inter--molecular correlations
\cite{bueche}, failed to incorporate the non--mean field like correlations on
scale $\xi_\rho$ of semidilute polymer solutions. The recent field theoretic
results lead to contradicting results as will be pointed out and resolved in
this contribution.
Integral equation theories for simple liquids directly address the problem of
inter-particle packing in dense fluids.
Starting with the work of Schweizer and Curro \cite{kcur1}, this approach has
successfully been extended to macromolecular liquids. The polymer
reference interaction site model (PRISM) integral equations have been
fruitfully applied to describe inter alia the inter--molecule correlations in
dense homopolymer systems, polymer blends and block copolymer melts
\cite{kcur3}. PRISM is a macromolecular generalization of the RISM theory of
small molecules of Chandler and Andersen \cite{chandler,chandl}.
The low density limit of PRISM theory shall be worked out in detail in
this contribution in order to discuss the density correlations on the mesh size
length scale. This works either extends \cite{thread0,thread,fractal}, or
complements \cite{threada,aviksem} previous studies.
It is a priori not related to nor required for the success of the PRISM
approach to polymer melts whether it also correctly
captures the long ranged correlations of semidilute polymer
solutions. As liquid correlations in melts generally are short
ranged, an approach like PRISM appropriate for dense systems, need not be a
useful approach to \mbox{(semi--)}
dilute solutions, where long ranged correlations
are of interest. Nevertheless, the simplification of the PRISM equations to low
polymer densities worked out here will be argued to provide a useful
description of the inter--molecular correlations building up the polymer mesh
in polymer solutions \cite{kcur3,thread0,thread}.
Criteria for the quality of the approach will be
established from comparisons with simulations, field theory
and mean field results.
The aspect of screening of the intra--molecular excluded volume shall be
neglected in this work. It would require the use of self-consistent PRISM
theory, which is considerably more demanding \cite{kcur3}. Moreover, the errors
made, when neglecting the crossover to Gaussian intra--molecular correlations
on length scales large compared to the density screening length $\xi_\rho$,
will not affect the scalings of the inter--molecular correlations for
distances smaller than $\xi_\rho$, which are the main focus of this
contribution. Thus, in the following the intra--molecular correlations shall be
characterized by a density independent polymer structure factor $\omega_q$\ , which, for
macromolecules of $N$ segments at the positions ${\bf r}_\alpha$, is defined
as follows:
\beq{1.0}
\omega_q =
\frac{1}{N} \sum_{\alpha,\beta =1}^N \langle
e^{i {\bf q} ( {\bf r}_\alpha -{\bf r}_\beta ) } \rangle \; .
\end{equation}
Its full functional form will
not be required. Knowledge of its variation for small, large and intermediate
wave vectors suffices \cite{degennes,descloi}.
For small wave vectors, the number of scattering units, the index of
polymerization $N$, where $N\gg 1$ for macromolecules,
and the global molecular size, the radius of gyration
$R_g$, can be obtained from a scattering experiment measuring $\omega_q$\ :
\beq{1.0a}
\omega_q \to N\, ( 1 - \frac 1d \; q^2 R_g^2 + \ldots \, )\;, \qquad\mbox{for }\, q R_g \ll
1\; ,
\end{equation}
where $d$ is the spatial dimension. In an intermediate wave vector range, the
macromolecule is supposed to be self similar, which leads to a power law
behavior in $\omega_q$\ determined by the fractal dimension, $d_F=1/\nu$:
\beq{1.0b}
\omega_q \to \frac{1}{(q\sigma)^{1/\nu}} \; , \qquad\mbox{for }\, \;
1/R_g \ll q \ll 1/\sigma\; .
\end{equation}
The fractal exponent $\nu$ also determines the size--mass scaling, $R_g \propto
\sigma N^\nu$, where a smooth crossover from \gl{1.0a} to \gl{1.0b} is assumed
around $qR_g\approx 1$. The assumption of an intermediate self--similar
molecular structure rules out the study of compact macromolecules, e. g. hard
sphere like colloids, but is appropriate for polymer chains in good,
$\nu=0.588\ldots$, or $\Theta$--solvents, $\nu=\frac 12$, or for rods, $\nu=1$,
which
share some properties with semi-flexible polymer molecules like actin or DNA
\cite{descloi,degennes,sackmann}.
The Kuhn'sche--segment size $\sigma$ in \gl{1.0b} is of
the order of local polymer--specific length scales where microscopic segmental
packing effects influence the complicated structure of $\omega_q$\ .
This chemistry dependent variation of $\omega_q$\ around $q\sigma\approx1$ can be
included in PRISM studies \cite{kcur3}, but shall be neglected here. Only the
self scattering contribution, $\alpha=\beta$ in \gl{1.0}, which is the only
remaining contribution for large wave vectors, $q\sigma \gg 1$, is universal
and needs to be considered.
\beq{1.0c}
\omega_q \to 1 \; , \qquad\mbox{for }\, q \sigma \gg 1\; .
\end{equation}
Thus, a generic smooth crossover from the point particle self scattering
term, \gl{1.0c}, to the self similar intra--molecular correlations, \gl{1.0b},
will be assumed. Chemistry dependent local packing will show up in all
correlation functions on microscopic length scales but will not, except for
in prefactors, affect the inter--molecular structure on global length scales
like the molecule size, $R_g$, or the mesh width, $\xi_\rho$; as will be shown
explicitly.
In order to characterize the total, including the inter--molecular, density
correlations of an interacting polymer system, further correlation functions
need to be introduced. In order to compare them with results from other
approaches it is useful to recall their definition as used in PRISM theory
\cite{hansen,kcur3}. To be specific, let us consider $n$ polymers with $N$
scattering units in a $d$--dimensional volume $V$, where in the thermodynamic
limit the number density of segments, $\varrho = \frac{nN}{V}$, is kept fixed;
i. e. $n, V \to\infty$ with $\varrho=$ const..
The local, fluctuating density is
\beq{1.1}
\varrho( {\bf r}, t ) = \sum_{i=1}^n \sum_{\alpha=1}^N \delta({\bf r} - {\bf
r}^{(i)}_\alpha(t) )\; ,
\end{equation}
with equilibrium average $\langle \varrho( {\bf r} , t ) \rangle = \varrho$.
The spatial components of the density fluctuations shall be denoted by
$\varrho_{{\bf q}}$, where:
\beq{1.2}
\varrho_{{\bf q}} = \int d^d r\; e^{i {\bf q} {\bf r}}\; \varrho({\bf r}) =
\sum_{i=1}^n \sum_{\alpha=1}^N e^{i {\bf q} {\bf r}^{(i)}_\alpha} \; .
\end{equation}
Their statistical average vanishes except for zero wave vector,
$\langle \varrho_{{\bf q}} \rangle = \varrho \delta_{{\bf q},{\bf 0}}$.
The total structure factor, $S_q$\ , as measured for example by coherent neutron
scattering \cite{descloi,hansen}
is given by the second moment of the wave vector
dependent density fluctuations, from \gl{1.2}:
\beqa{1.3}
S_q & = & \frac{1}{nN} \langle \varrho^*_q \varrho_q \rangle - nN \delta_{{\bf
q},{\bf 0}} \nonumber \\ & = &
\frac{1}{nN} \sum_{i,j =1}^n \sum_{\alpha,\beta =1}^N \langle
e^{i {\bf q} ( {\bf r}^{(i)}_\alpha -{\bf r}^{(j)}_\beta ) } \rangle - nN
\delta_{{\bf q},{\bf 0}} \; .
\end{eqnarray}
The total density fluctuations are straightforwardly separated into density
fluctuations on the identical polymer, $\omega_q$\ , the intra--molecular structure
factor, and on different polymers, $h_q$\ , the inter--molecular structure factor.
\beq{1.4}
S_q = \omega_q + \varrho \, h_q \; ,
\end{equation}
where the intra--molecular part has already been defined in \gl{1.0} above.
The inter--molecular structure factor, $h_q$\ , describes the packing of different
molecules and is given by the restricted sum $i\ne j$:
\beq{1.5}
h_q = \frac{V}{n^2N^2} \sum_{i,j =1, i\ne j}^n
\sum_{\alpha,\beta =1}^N \langle
e^{i {\bf q} ( {\bf r}^{(i)}_\alpha -{\bf r}^{(j)}_\beta ) } \rangle
- V \delta_{{\bf q},{\bf 0}} \; .
\end{equation}
The inter--molecular structure factor is the Fourier transform of the
inter--molecular pair correlation function, $g(r)$\ :
\beq{1.6}
h_q = \int d^dr e^{i{\bf q} {\bf r}} \; (\, g(r) - 1 \,) \; .
\end{equation}
The pair correlation function describes the probability averaged over all
segments of
finding at a distance $r$ from site $\alpha$ on molecule $i$ another segment
$\beta$ of a different molecule $j$. From \gls{1.5}{1.6}, one obtains:
\beq{1.7}
g(r) = \frac{V}{n^2N^2} \sum_{i,j=1;i\ne j} \sum_{\alpha,\beta =1}^N \langle
\delta( {\bf r} - ({\bf r}^{(i)}_\alpha - {\bf r}^{(j)}_\beta) ) \rangle \; .
\end{equation}
$g(r)$\ is non--negative and approaches unity for large separations $r$,
because then statistical correlations between the sites at ${\bf
r}^{(i)}_\alpha$ and ${\bf r}^{(j)}_\beta$ have vanished \cite{hansen}.
Implicit in the Eqs. (\ref{1.3}) to (\ref{1.7}) is a neglect of a special site
dependence of the density fluctuations as it might for example arise from
chain--end effects for linear polymers \cite{kcur1}. Star polymers, where
sites in the core region possibly experience very different local density
fluctuations than sites in the star--arms, also would require a more elaborate
treatment \cite{kcur3}.
Nevertheless, for arbitrary macromolecular architectures,
the above defined correlation functions
are experimentally measurable at least in principle and can also be determined
from computer simulations. They contain information about the local liquid
structure and remain meaningful in the whole accessible density range, from
dilute solutions to melts.
\section{PRISM integral equations \label{sex2}}
Whereas the effects of the excluded volume on the intra--molecular structure,
like swelling, are already taken into account in \gl{1.0}, PRISM theory
\cite{kcur1,kcur3} explicitly enforces inter--molecular excluded volume by
requiring the pair correlation function to vanish for distances smaller than
the segment size R.
\beq{2.1}
g(r) = 0 \qquad\mbox{for }\, r < R\; ,
\end{equation}
Building upon the Ornstein--Zernicke approach so successful for simple liquids
\cite{hansen}, an averaged molecular site--site Ornstein--Zernicke like
equation \cite{chandler,chandl} is formulated
and can be viewed as definition of an effective potential, the direct
correlation function $c_q$:
\beq{2.2}
h_q = \omega_q c_q \, ( \omega_q + \varrho h_q ) \; .
\end{equation}
Equations (\ref{1.4}) and (\ref{2.2}) can also be brought into a RPA--like form
which supports the interpretation of the direct correlation function as an
effective potential.
\beq{2.3}
S^{-1}_q = \omega^{-1}_q - \, \varrho c_q \; .
\end{equation}
Different from the RPA approach, $c_q$ is not considered to be given, but needs
to be found from a solution of the non--linear integral equations. Besides
Eqs. (\ref{2.1}) to (\ref{2.3}), a further equation, the ``closure''
approximation, is required to determine the solution.
The most simple and yet appropriate closure to treat
the inter--molecular steric repulsion is the Percus--Yevick (PY) approximation
which expresses the expectation that the effective potential is short ranged:
\beq{2.4}
c(r) = 0 \qquad\mbox{for }\, r > R\; ,
\end{equation}
Note, that the interaction described by $c(r)$ is localized on microscopic
length scales. Thus, the PRISM equations describe the interplay of local
inter--molecular steric interactions and long ranged intra--molecular
correlations due to macromolecular connectivity \cite{kcur1,kcur3}.
\subsection{Thread limit for (semi--) dilute solutions \label{sex2a}}
In the simplified non--self-consistent PRISM approach, the intra--molecular
structure $\omega_q$\ , from \gl{1.0}, is assumed to be given, and (mostly
numerical) techniques to solve the integral equations,
\glto{2.1}{2.4} are employed \cite{kcur3}. Note that
for a Pade approximation to $\omega_q$\ for Gaussian chains with $\nu=\frac 12$ in
$d=3$ an analytic solution of the PRISM equations on all length scales exists
\cite{threada}. The solution technique employing the Wiener--Hopf factorization
as pioneered by Baxter \cite{baxter} can straightforwardly be extended to
Gaussian chains in (low) odd dimensions, $d=5, 7, \ldots$, but a simpler
approach can also be used in order to study the low density results of PRISM
theory
analytically. For the mentioned case, $\nu=\frac 12$ in $d=3$, this was first
used in \cite{thread0,thread},
explicitly justified in \cite{threada}, and without
proof extended to exponents $\nu$ within the bounds $\frac 1d \le \nu < \frac
2d$ in \cite{fractal}. Here, general arguments
on the solutions of \glto{2.1}{2.4} allow to find the low density
limits in the more general case $\frac 1d<\nu$ and $d\ge2$.
The special low density limit, called ``thread PRISM'' model
\cite{thread0,thread},
studied in the following assumes that polymer solutions can be modeled as a
low density limit of the one--component PRISM equations for polymer
melts. Special solvent effects, are assumed to be taken into account via the
model for the intra--molecular structure, \glto{1.0a}{1.0c}.
In general, the excluded volume constraint, \gl{2.1}, and the PY closure,
\gl{2.4}, lead to a discontinuity in the pair correlation function $g(r)$\
and in the direct correlation function $c(r)$
at contact:
\beq{2.5}\begin{array}{lcll}
g_d & := & g(r\searrow R) & > 0\; ,\\
c(R-) & := & c(r \nearrow R) & \ne 0\; .
\end{array}
\end{equation}
Note, that the actual values of $g_d$ and $c(R-)$
will dependent on details of the monomer
chemistry, as the segment size $R$ obviously is a microscopic length.
To connect both quantities it is useful to express
$h_q$\ and $c_q$ as one--dimensional Fourier transforms,
\beq{2.5a}\begin{array}{lcl}
h_q & = & \int dr\; e^{i q r }\; j(r)\; ,\\
c_q & = & \int dr\; e^{i q r }\; i(r)\; ,\end{array}
\end{equation}
with the symmetric functions,
\beq{2.5b}\begin{array}{lcl}
j(r) & = & \Omega_{d-1} \int_{|r|}^\infty ds \; s\, (g(s)-1) \;
(s^2-r^2)^{\frac{d-3}{2}} \; ,\\
i(r) & = & \Theta(R-|r|)\; \Omega_{d-1} \int_{|r|}^R ds \; s\, c(s) \;
(s^2-r^2)^{\frac{d-3}{2}} \; , \end{array}
\end{equation}
where $\Omega_d=\frac{2\pi^{d/2}}{\Gamma(d/2)}$
denotes the surface of the $d$--dimensional unit sphere.
Because of \gls{2.1}{2.4}, $i(r)$ can be non--smooth
for $|r| \le R$ only, while this can happen for $j(r)$ for $|r|\ge R$.
Thus, using the large wave vector limits,
$h_q\propto -g_d \cos{qR}/q^d$ for $q\to\infty$ and similarly for
$c_q$, and the large wave vector asymptote of $\omega_q$\ , see \gl{1.0c}, \gl{2.2}
connects the discontinuities of $g(r)$\ and $c(r)$ at $r=R$, leading to:
\beq{2.6}
g_d = - \, c(R-)\; .
\end{equation}
Moreover, one concludes that $c(r)$ is finite. Thus, the scaling of the
Fourier transform of the direct correlation function with global parameters (to
be defined below) also can be connected to the contact value:
\beq{2.7}
c_q = R^d c(R-) f_c(qR) =: - g_d R^d f_c(qR)\; ,
\end{equation}
where the regular function,\newline
$f_c(x)=\int_{y<1} d^dy\, e^{i {\bf xy}}\, c(R{\bf y})/c(R)$,
has a finite value at $x=0$. For the analytically known results of the
PRISM equations these property could be shown explicitly \cite{threada}, and
they can now be used to simplify the PRISM equation for low densities.
In order to extract the large distance solution of the PRISM equations, it
proves useful to shift the microscopic length scales to zero:
\beq{2.8}
R\to0\, ,\quad \sigma\to0\qquad \mbox{with }\; \sigma/R = \mbox{const.}\; .
\end{equation}
In order to evade the trivial limit of a non--interacting ideal gas, the length
scale of the intra--molecular correlations, to be denoted by $\xi_c$, which is
proportional to the molecular size, $\xi_c\propto R_g$, is kept finite by
increasing the index of polymerization $N$.
\beq{2.9}
N\to\infty\qquad \mbox{so that }\; \xi_c \propto \sigma N^\nu =
\mbox{const.}\; .
\end{equation}
Also, in order to keep inter--molecular excluded volume active, the bare
segmental density
is increased beyond bounds:
\beq{2.10}
\varrho\to\infty\qquad\mbox{so that }\; \varrho/\varrho_* = \mbox{const.}\; .
\end{equation}
In the thread PRISM equations, there enters a typical density, the dilute
to semidilute crossover density $\varrho_*$, which is familiar from scaling
considerations
\cite{degennes}. As will be shown below, $\varrho_*$, is defined differently
for
scaling exponents $\nu$ below and above a value $\nu_c$, which denotes the
crossover to mean--field like
behavior.
\beq{2.11}
\varrho_* \propto \left\{
\begin{array}{lll}
\frac{N}{\xi_c^d} & \mbox{for} & \nu < \nu_c = \frac 2d \; ,\\
\frac{1}{N \sigma^d} & \mbox{for} & \nu > \nu_c \; .
\end{array}\right.
\end{equation}
Equations (\ref{2.8}) to (\ref{2.11}) specify the thread PRISM limit. Note,
that the divergent number density
$\varrho_*$ actually corresponds to a vanishing polymer volume fraction,$\phi$,
and thus (semi--) dilute
polymer solutions are studied as claimed.
\beq{2.12}
\phi_* = \varrho_* \sigma^d \propto \left\{
\begin{array}{lll}
1/N^{(\nu d-1)} & \mbox{for} & \nu < \nu_c \; ,\\
1/N & \mbox{for} & \nu > \nu_c\; ,
\end{array}\right.
\end{equation}
where $\phi={\cal O}(1)$ corresponds to polymer melts.
The solution of the non--linear PRISM integral equations in
the general case of course demands to find $f_c(x)$ from \gl{2.7} for all
$x$. However, a solution to \glto{2.1}{2.4} depending on $f_c(0)$ can be
constructed on large distances, $r\gg R,\sigma$, or, equivalently in the thread
limit, for finite distances, $r>0$.
This holds, because in the limit of $R\to0$, the excluded volume condition
\gl{2.1} affects a point (of measure zero) only, and thus $h_q$\ can be obtained
from the Fourier integral of $g(r)$\ outside the core, $r>0$. From the inverse
transformation and \gls{2.2}{2.7}, the contact value follows in the general
case, where $\bf R$, denotes a vector of length $R+$:
\beq{2.13}
g_d - 1 = - \frac{N A}{\varrho}\;
\int \frac{d^d q}{(2\pi)^d}\, e^{i{\bf qR}}\;
\frac{\bar{f}_c(qR)\bar{\omega}_q^2}{1+A \bar{f}_c(qR) \bar{\omega}_q}\; .
\end{equation}
The normalized functions $\bar{\omega}_q=\omega_q/N$ and
$\bar{f}_c(x)=f_c(x)/f_c(0)$ have been introduced, and the
long wavelength interaction parameter A, which
abbreviates the zero wave vector limit of the direct correlation function:
\beq{2.14}
A = - N \varrho c_{q=0} = N \varrho g_d R^d f_c(0) \; .
\end{equation}
For (semi--) dilute solutions, the limit \gl{2.8} simplifies the PRISM integral
equations,
because the core condition, \gl{2.1}, becomes irrelevant, and \gl{2.13}
assures that the connection \gl{2.6} is satisfied. The two further conditions,
\gls{2.9}{2.10}, assure that non--trivial solutions describing an interacting
polymer solution are obtained.
Because of \gl{2.9}, only the long ranged scaling form of the intra--molecular
structure factor enters:
\beqa{2.15}
\bar{\omega}_q & = & \bar{\omega}(x=q\xi_c) \nonumber \\
& = & \left\{\begin{array}{ll} 1+{\cal O}(x^2) & x\to0\; ,\\
1/x^{1/\nu} & x\to\infty \; .\end{array}\right.
\end{eqnarray}
And \gl{2.10} enforces the molecules to interact, $A\ne0$,
so that \gl{2.13} leads to a transcendental equation
determining $A$ (equivalently $g_d$), which is the only unknown parameter in
the thread inter--molecular structure factor:
\beq{2.16}
\varrho h_q = - N A \frac{\bar{\omega}_q^2}{1+ A\bar{\omega}_q}\; .
\end{equation}
\section{Thread limit results} \label{sex3}
In the dilute to semidilute concentration region, the thread PRISM result for
$h_q$\ assumes the RPA like form, \gl{2.16}, where the interaction parameter
$A$ needs to be found from \gl{2.13}. The total structure factor then also has
simple RPA--like form:
\beq{3.1}
S_q = N \frac{\bar{\omega}_q}{1+ A \bar{\omega}_q }\; ,
\end{equation}
where generally $\bar{\omega}_q$ differs from the Gaussian form and
the integrated inter--molecular interaction strength, $A$, in general
differs from the simple RPA approximation, $A\propto N\varrho R^d$.
Because of the large wave vector behavior of $\bar{\omega}$, \gl{2.15},
the limit $R\to0$ affects the integral in \gl{2.13} differently for
$\nu<\nu_c=\frac 2d$ or $\nu>\nu_c$. In the first case, the integral converges
uniformly,
and integration and limit can be interchanged, thus only $\bar{f}_c(0)=1$
enters.
In the second case, the integral converges only because of the wave vector
dependence
of $\bar{f}_c(qR)$, and leads to RPA or mean--field behavior.
\subsection{Below the mean field crossover \label{sex3a}}
In the thread equation for the interaction parameter $A$, \gl{2.13},
the limit $R\to0$ can be performed trivially
for $\nu<\nu_c=\frac 2d$, and $A$ becomes a function of $\varrho/\varrho_*$
only,
where the crossover density $\varrho_*=N\Omega_d/(2\pi\xi_c)^d$
enters.
\beq{3.2}
( 1 - g_d ) \frac{\varrho/\varrho_*}{A} =
\int_0^\infty dx \frac{x^{d-1} \bar{\omega}^2(x)}{1 + A \bar{\omega}(x)}\; .
\end{equation}
For size--mass scaling exponents $\nu$ corresponding to fractal dimensions,
$d_F=1/\nu$, equal to or exceeding the spatial dimension, i.e. for $\nu<\frac
1d$, the intra--molecular structure on long length scales determines the
thread equation (\ref{3.2}). There, screening of the intra--molecular excluded
volume has been neglected in the present approach, and the polymer segregation
effect predicted in the thread limit requires use of the self--consistent PRISM
approach \cite{fractal}. In order to avoid this complication, the exponent
$\nu$ will be restricted in the following, $\nu>\frac 1d$.
From the two limits of the integral in \gl{3.2},
constant for $A\ll 1$ and $A^{-(2-\nu d)}$ for $A\gg 1$,
the scaling form for the thread parameter can be determined.
\beqa{3.3}
A & = & (\varrho/\varrho_*) f_A(\varrho/\varrho_*) \; ,\qquad\mbox{where}
\nonumber \\
f_A(x) & \propto & \left\{\begin{array}{lll}
const. + {\cal O}(x) & \mbox{for} & x\to0\; ,\\
x^{\frac{2-\nu d}{\nu d-1}} & \mbox{for} & x\to\infty\; .
\end{array}\right.
\end{eqnarray}
In \gl{3.3}, the contact value correction was already neglected, as it is
of higher order, as can be deduced from \gl{2.14}. Actually, a scaling law
follows
from \gls{2.14}{3.3} for the contact value:
\beq{3.4}
g_d = c \frac{(\xi_c/R)^d}{N^2}\; f_A(\varrho/\varrho_*) \; ,
\end{equation}
where the identical scaling function $f_A$ from \gl{3.3} enters. The numerical
prefactor $c$ of course depends on the microscopic details of the polymer
model.
Note that for $\nu<\nu_c$ the contact value vanishes like $N^{-(2-\nu d)}$ for
$N\to\infty$ in the dilute case.
The scaling function $f_A$ also determines the density dependence of the mesh
size or density screening length, as for large reduced densities,
$\varrho\gg\varrho_*$, the width of the total
structure factor can be estimated from $S_q=(N/A)/(1+(q\xi_\rho)^{1/\nu})$,
for $q\xi_c\gg1$, which leads to:
\beq{3.5}
\xi_\rho = c' \sigma (\varrho\sigma^d)^{-\nu/(\nu d-1)}\; .
\end{equation}
For a given model of the intra--molecular structure factor, the thread
equation, \gl{3.2} with $g_d=0$, allows to
determine $f_A$ straightforwardly. Figure \ref{fig1} shows the result for the
polymer model: $\omega_q=\frac 1N \sum_{\alpha,\beta=1}^N
\exp{-\frac{q^2\bar{\sigma}^2}{6}
|\alpha-\beta|^{2\nu}}$, with the exponent given by
the Flory approximation $\nu=3/5$ corresponding to good polymer solutions
\cite{degennes,descloi}. Numerical solutions
of the microscopic PRISM equations (\ref{2.1}) to (\ref{2.4}) for this model
and for not too large degrees of polymerization, $N$, still exhibit rather
large corrections to the thread asymptote. This can be expected to be model
dependent. From Fig. \ref{fig1} one notices that the connection of the contact
value to the small wave vector interaction parameter, \gl{2.14}, already holds
for values of $N$ where the asymptotic $f_A$, \gl{3.3}, is not yet
reached. Differences appear for larger packing fractions and signal
concentrated or melt like polymer packing.
The pair correlation function, $g(r)$, in the thread limit can be obtained for
finite segment distances from the Fourier transform of \gl{2.16}. For dilute
densities, $\varrho\ll\varrho_*$ and thus $A\ll1$, it differs from the ideal
gas limit, $g(r)=1$, because of two molecule interactions. In the semidilute
concentration region, $\varrho\gg\varrho_*$ and $A\gg1$, the replacement
$A=(\xi_c/\xi_\rho)^{1/\nu}$ shows that $g(r)$ depends on the two length
scales, $\xi_c$ and $\xi_\rho$, independently.
On length scales large compared to the mesh size, $r\gg\xi_\rho$, the
inter--molecular structure factor exhibits the well known correlation hole
\cite{degennes,kcur1,kcur3,fractal}, which asymptotically for $\varrho\gg
\varrho_*$ exactly cancels off the long ranged
intra--molecular correlations:
\beq{3.5a}
h_q \to -\frac N\varrho\; \bar{\omega}(q\xi_c)\;,\qquad\mbox{for }\, q\ll 1/\xi_\rho\;;
\varrho\gg\varrho_* \; .
\end{equation}
This result is equivalent to $S_q\ll\omega_q$ for $q\xi_\rho\ll1$ in the
semidilute range \cite{degennes}. Note that the self--similar structure of the
molecule leads to the power law behavior $\varrho h_q \propto -(\sigma
q)^{-1/\nu}$ for $1/\xi_c\ll q\ll1/\xi_\rho$, which is equivalent to a power
law variation in the pair correlation function: $g(r)-1\propto (-1/\varrho)
(\sigma/r)^{d-1/\nu}$ for $\xi_\rho\ll r \ll\xi_c$ \cite{fractal}. For larger
distances, $r\gg\xi_c$, \gl{3.5a} describes how $g(r)$ decays exponentially
to its random mixing value unity.
Within the mesh size, i. e. for distances around and smaller than the density
screening length, the inter--molecular correlations do not depend on the
molecular size and $\bar{\omega}$ in \gl{2.16} can be replaced by its large
$q\xi_c$ asymptote from \gl{2.15}. This leads to
\beq{3.5b}
h_q \to \frac{-\bar{c}\;
\xi_\rho^d}{(q\xi_\rho)^{1/\nu} + (q\xi_\rho)^{2/\nu} }\; ,
\qquad\mbox{for }\, 1/\xi_c \ll q\; ; \varrho\gg \varrho_*\; ,
\end{equation}
where the limiting behaviors of $h_q$\ for $q\xi_\rho $ large or small
compared to unity can be read of immediately and $\bar{c}=N/(\varrho A
\xi_\rho^d)$ approaches a number ($\bar{c}\to 6.26\ldots$ for $\nu=3/5$).
Note, that \gl{3.5} for the density screening length ensures a smooth crossover
of \gl{3.5b} to \gl{3.5a} in the correlation hole region.
The variation of the pair correlation function, which describes the mesh
structure for $r\ll\xi_c$, can thus be obtained in closed form, if the neglect
of the cutoff of the correlation hole at $r\, \raise2.5pt\hbox{$>$}\llap{\lower2.5pt\hbox{$\approx$}}\, \xi_c$ included in
\gl{3.5a}, is kept in mind. For $r\ll\xi_c$, $g(r)$ depends on $r/\xi_\rho$
only, with:
\beqa{3.5c}
g(r) & = & 1 - \bar{c}
\int \frac{d^d y}{(2\pi)^d}\; e^{-i{\bf y r}/\xi_\rho} \;
\frac{1}{y^{1/\nu} + y^{2/\nu} } \nonumber \\
& \to & \left\{ \begin{array}{lll}
(r/\xi_\rho)^{2/\nu-d} & r\ll \xi_\rho \; ; & r\gg \sigma,R\; , \\
1 - (\xi_\rho/r)^{d-1/\nu} & r\gg \xi_\rho \; ; & r\ll \xi_c\; , \\
\end{array}\right.
\end{eqnarray}
where $\bar c$ ensures $g(0)=0$ in agreement with \gl{3.2} and constant
prefactors of order unity have been suppressed in the final two lines.
For polymer chains in good solvents, the smooth increase, $g(r\ll\xi_\rho)
\propto (r/\xi_\rho)^{1/3}$, by accident agrees with the estimate from Ref.
\cite{daoud}, $g(r\ll\xi_\rho) \propto
(r/\xi_\rho)^{(\gamma-1)/\nu}\approx(r/\xi_\rho)^{1/3}$ where
$\gamma$ is associated with the entropy of a single polymer chain
\cite{descloi}.
The depth of the correlation hole displays an intriguing dependence on the
fractal and spatial dimensionalities. The probability to find a segment of
another polymer within the considered molecule decreases strongly if $1/\nu\to
d$. From \gl{3.5c} one estimates $g(r\approx \xi_c)-1\propto - 1/N^{\nu d-1}$,
which becomes a number of order unity in the case $\nu=1/d$.
The smooth variation of $g(r)$ at short distances explains, why the
scaling of the correct contact value $g_d$, \gl{3.4}, with macroscopic
variables can be estimated from the thread solution, \gl{3.5c}, by $g_d\propto
g(\sigma)$; its dependence on the ratio of the microscopic length scales,
$\sigma/R$, however cannot generally be recovered in this way \cite{threada}.
Note that \glto{3.5}{3.5c} asymptotically apply for semidilute solutions,
$\varrho\gg\varrho_*$, whereas (\ref{2.16},\ref{3.1},\ref{3.3}) describe the
full dilute--to--semidilute crossover region.
\subsection{The mean field cases \label{sex3b}}
The condition \gl{2.13} for the contact value $g_d$, or equivalently, for
the thread
parameter $A$ becomes independent of the microscopic interaction details only
for $\nu<\nu_c$. Above the crossover exponent, $\nu>\nu_c=\frac 2d$, the
integral over the effective
potential as it enters $h_q$\ is determined by the local structure in
$f_c(qR)$. In the thread limit
\gl{2.13} becomes a linear, density--independent equation for $g_d$
with solution:
\beq{3.6}
g_d = 1 / [\, 1 + \int\frac{d^dq}{(2\pi/R)^d}\; e^{i{\bf qR}}\;
(q\sigma)^{-2/\nu} f_c(qR)\, ]\; .
\end{equation}
Thus, a finite density independent contact value follows in the mean field like
cases $\nu>\nu_c$.
Obviously, its exact value depends on the solution of the PRISM equations
considering all
microscopic details and is beyond the reach of the thread PRISM approach.
The interaction parameter $A$ thus shows the density scaling as expected within
RPA.
The reduced density $\varrho/\varrho_*$ appears,
with $\varrho_*= 1/(N\sigma^d)$, and $A$ becomes --- with a unknown
numerical constant $\tilde c$, which however may depend on the ratio $\sigma/R$
of the microscopic length scales.
\beq{3.7}
A = \tilde{c} \; \varrho/\varrho_*\; .
\end{equation}
In the semidilute density regime, the width of the total structure factor again
determines the density screening length, $S_q=(N/A)/(1+(q\xi_\rho)^{1/\nu})$
for $q\xi_c\gg1$, with the result:
\beq{3.8}
\xi_\rho = \tilde{c}' \; \sigma (\varrho\sigma^d)^{-\nu}\; .
\end{equation}
Again, the numerical prefactor, $\tilde c'$, depends on the polymer model. The
result, \gl{3.5a}, discussed for the inter--molecular structure in
the correlation hole region holds. The mesh structure factor,
$h(q\gg1/\xi_c)$, however, shows a different density scaling:
\beq{3.8b}
h_q \to \frac{-\hat{c}\,
\sigma^d \,(\xi_\rho/\sigma)^{2/\nu}}{(q\xi_\rho)^{1/\nu} +
(q\xi_\rho)^{2/\nu} }\; , \qquad\mbox{for }\, 1/\xi_c \ll q\; ,
\end{equation}
which again, with \gl{3.8}, leads to a smooth crossover for intermediate
distances, $1/\xi_c\ll q \ll 1/\xi_\rho$, but to a small distance divergence of
the thread pair correlation function, $g(r)\propto -(\sigma/r)^{(d-2/\nu)}$ for
$r\ll\xi_\rho$. This results from the neglect of
the wave vector variation of the direct correlation function, \gl{2.7}, and
reinforces that the validity of the thread $g(r)$ is restricted to
$r\gg\sigma$ for $\nu>\nu_c$, where $g(r)$ is still positive, as it must be by
definition.
\section{Discussion and comparison with other approaches \label{sex4}}
In this work, scaling limits appropriate for the dilute to semidilute
concentration regime of macromolecular solutions have been derived starting
from the microscopic
PRISM integral equations. RPA like expressions for the total density
fluctuations, the structure factor $S_q$\ , \gl{3.1}, were given, where the density
scaling of the interaction parameter, $A$, was
deduced from the local excluded volume constraint, \gl{2.1}. The thread
interaction parameter is connected to the more familiar excluded volume
parameter, $v$, via $A=\frac{N_A M \varrho}{M_0^2} v$, where $N_A$ is Avogadro's
number, $M$ the molecular and $M_0$ the monomer weight. Effective density
dependent excluded volume parameters $v(\varrho)$ have often been used in
connection with RPA expressions \cite{daoud,ullman}, and \gl{3.3}
justifies this.
The crossover of the single chain correlations to Gaussian large distance
behavior for $r\gg \xi_\rho$ due to intra--molecular excluded volume has been
neglected and would affect the model for $\omega_q$\ , \gl{1.1}, and consequently the
thread results for large distances. Use of self--consistent PRISM
\cite{kcur3} to incorporate this would be required, but \gl{3.3} for the thread
parameter $A$ indicates that no change of its density scaling can be expected.
The crossover density $\varrho_*$, \gl{2.11}, arises from the full microscopic
PRISM equations as the relevant low density scale, and importantly, the
qualitatively different definitions in the mean field, $\nu>\nu_c=\frac 2d$,
and in the non--trivial cases, $\nu<\nu_c$, are recovered. Whereas for
$\nu<\nu_c$ the molecular crossover density, $c_*=\varrho_*/N$, is defined in
terms of the molecular size only, $c_*\propto1/R_g^d$ for $\nu>\nu_c$, in the
mean field cases also a microscopic length, the segmental hard core diameter
$R$, enters, $c_*\propto1/(R_g^{2/\nu} R^{d-2/\nu})$ for $\nu>\nu_c$.
This indicates that inter--molecular steric interactions become important as
soon as the macromolecules fill space for $\nu<\nu_c$, whereas for the more
open molecules, $\nu>\nu_c$, much higher densities are required.
For chain polymers, the upper critical dimension, which
separates mean field and fluctuation dominated structures, agrees with
the renormalization group results,
$d_c=2/\nu_c=4$ \cite{degennes,descloi}. For rod polymers, $\nu=1>\nu_c$, the
mean field like behavior underlies the successful Onsager theory
of the nematic transition \cite{onsager} and is generally argued to be true
\cite{shimada}. Note that in the studied PRISM theory orientational,
``nematic'', interactions are not treated correctly \cite{kcur3}, and thus a
nematic transition for rods is missed. Very recently, PRISM has been
generalized to treat oriented polymer fluids and the isotropic--nematic liquid
crystal transition \cite{pickett}. As the (isotropic) crossover density $c_*$
for rods is of the order of the nematic transition density \cite{onsager},
a suppression of nematic order is required to study the described isotropic
semidilute rod solutions experimentally. Networks of stiff semiflexible
molecules like actin may provide good systems \cite{sackmann}.
Of course, the full PRISM integral equations, which have been introduced to
study dense polymer systems with short ranged melt--like correlations
\cite{kcur1}, incorporate wave vector dependent corrections in
e. g. the effective interaction $c_q$, see \gl{2.3}, when $1/q$ approaches
local length scales.
From the compressibility, which is connected to the zero wave vector limit of
the total structure factor, the equation of state can be obtained,
where $\Pi$ denotes the osmotic pressure \cite{hansen,kcur3}:
\beqa{4.1}
\frac{\Pi}{\varrho k_BT} & = & \frac 1N + \frac 1\varrho \int_0^\varrho
d\varrho' \frac{A(\varrho')}{N} \nonumber\\
& \propto & \left\{\begin{array}{lll}
\frac 1N\, ( 1 + {\cal O}(\varrho/\varrho_*) )
& \varrho \ll \varrho_*\; , & \\
(\varrho\sigma^d)^{\frac{1}{\nu d-1}} & \varrho\gg\varrho_*\,,&\nu<\nu_c\;,\\
\varrho\sigma^d & \varrho\gg\varrho_*\,,&\nu>\nu_c\; ,
\end{array}\right.
\end{eqnarray}
when $A$ is given by \gl{3.3}.
The non--mean field behavior for $\nu<\nu_c$ for semidilute concentrations
\cite{fractal} agrees with the exact Des Cloizeaux result \cite{descloi},
and the second virial coefficient, $1/(N\varrho_*)\propto R_g^d/N^2$,
recovers the picture of dilute polymer coils interacting like hard spheres of
radius $R_g$ \cite{aviksem}, but it does not vanish for dilute
$\Theta$--solvents, i. e. for $\nu=\frac 12$ in the present approach.
PRISM theory apparently correctly captures the leading
asymptotic behaviors, the free molecule limit $\Pi=(\varrho/N) k_BT$ for
$\varrho\ll\varrho_*$ and the $N$--independent power law for
$\varrho\gg\varrho_*$, but the next to leading terms are not described
correctly in general.
The structure of the polymer mesh in semidilute solutions, i. e. the
inter--molecular structure factor, $h_q$\ , on length scales of the order of the
density screening length, $\xi_\rho$, has not been conclusively discussed from
first principles calculations. The thread PRISM results for the non--mean field
like case of polymer chains in good solvents give explicit results,
\gls{3.5b}{3.5c}, which can be compared to results from other approaches.
\subsection{Comparison with scaling considerations \label{sex4a}}
Detailed scaling law considerations of $h_q$\ in the limit $qR_g\gg1$ have been
presented in \cite{jannink} and can be directly compared with \gl{3.5b}. The
limit, $h_q\propto \xi_\rho^d$ for $1/R_g\ll q \ll 1/\xi_\rho$ strongly differs
from the correlation hole behavior, $\varrho h_q=-(q\sigma)^{1/\nu}$ with
$\sigma$ the Kuhn'sche segment size, predicted by
the thread PRISM theory for this wave vector window. Physically, the
long ranged variation of $h_q$\ arises from the rearrangement of the polymer
mesh around a molecule on distances up to the molecule's
size. This adjustment compensates for the excess density due to the considered
molecule, leading to the small total density fluctuations expected
for concentrated systems. The correlation hole has first been predicted
and discussed for polymer melts \cite{degennes}, but PRISM theory also
predicts it for semidilute solutions \cite{kcur1}, in agreement with scaling
considerations in \cite{daoud} but in disagreement with the
mentioned scaling picture presented in \cite{jannink}. Intriguingly, PRISM
theory, \gl{3.5c}, recovers the tendency of macromolecules to
segregate for $\nu=1/d$ as discussed for ideal chains in two dimensions
\cite{degennes}.
In agreement with the thread PRISM result, a scaling law is postulated in
Ref. \cite{jannink} for the intermolecular structure factor inside the coil
radius, $h(q\gg 1/R_g) = \bar{h}(q\xi_\rho)$, which leads to the
prediction $h_q\propto q^{-d}$ for $q\to\infty$ \cite{jannink}. As a scaling
law can only hold for distances large compared to the microscopic length
scales, $q\sigma\ll 1$, this result can be compared with the thread
scaling power law, $h_q\propto q^{-2/\nu}\xi_\rho^{d-2/\nu}$
in \gl{3.5b}, and again differs.
Within thread PRISM the behavior of $h_q$ arises naturally as it matches
smoothly to the microscopic limit, $h_q\sim \frac{g_d}{q^d} \cos qR$ for $qR\gg
1$, because the contact value vanishes
asymptotically, $g_d\propto (\sigma/\xi_\rho)^{2/\nu-d}$
for $\xi_\rho \gg \sigma$.
This supports the
expectation in \cite{daoud}. Computer simulations could address this question
for polymer chain solutions as shown in section IV.C,
where corrections to the low density scaling law,
\gl{3.4}, need to be considered which will arise due to finite packing
fractions.
Accepting the existence of a scaling law for the contact value
of macromolecules in solutions, then \gl{3.4} can be used to connect the PRISM
results to field theoretic calculations for two--polymer systems.
\subsection{Comparison with field theoretic calculations \label{sex4b}}
Field theoretic calculations which employ the mapping of the polymer problem
onto the $O(n\to0)$ magnetic model lead to numerous single chain results and
have recently been extended to provide information about the inter--molecular
structure factor on short length scales \cite{jannink,schaefer}.
In \cite{jannink}, the mentioned behavior $h_q\propto q^{-d}$ for $q\to\infty$
is recovered from the field theoretic calculation and used to support the
scaling picture discussed in the previous section. The implications for $g(r)$
can be compared with
another field theoretic calculation which studies the number of intersections
of two polymer chains \cite{schaefer}. Let $\Sigma_2(R_e)$ be the number of
intersections of two random (or self--avoiding) walks whose end--to--end
distance is $R_e$:
\beq{4.3}
\frac{\Sigma_2(R_e)}{V (4\pi\sigma^2 )^{d/2}} =
\sum_{\alpha,\beta=1}^N \langle \delta({\bf r}^{(2)}_0-{\bf r}^{(1)}_0 -
{\bf R}_e )\;
\delta({\bf r}^{(2)}_\alpha-{\bf r}^{(1)}_\beta ) \rangle\; .
\end{equation}
For intermediate distances $R_e$, $\sigma\ll R_e\ll R_g$,
where the two polymers
overlap but local effects do not dominate $\Sigma_2$, the scaling
$\Sigma_2\propto
(\sigma/R_g)^{\omega_{12}(P)}$ is predicted, where the two molecule correction
to scaling exponent $\omega_{12}$ appears \cite{schaefer}.
The contact value can now be obtained from $\Sigma_2$ by integrating over all
possible distances and (trivial) factors of normalization, as can be seen from
Eqs. (\ref{1.7},\ref{2.5},\ref{2.8}).
\beq{4.4}
g_d = \int d^dR_e\; \frac{\Sigma_2(R_e)}{(4\pi\sigma^2)^{d/2}N^2}\; .
\end{equation}
Using the results for two polymers from \cite{schaefer} to obtain the scaling
of the contact value in the dilute case, one finds:
\beq{4.5}
g_d^{\rm RG}
\propto \frac{R_g^{d-\omega_{12}(P)}}{ N^2} \; ,\qquad\mbox{for }\, \; \varrho\to 0\; .
\end{equation}
The thread PRISM result, \gl{3.3}, differs from this in general, because in
PRISM theory the correction to scaling exponent is approximated to
$\omega^{\rm thread}_{12}=0$.
Its value in quadratic order in $\varepsilon=4-d$ is known, and the value
appropriate for polymer chains in good solvents turns out to
$\omega_{12}(G)=\frac 12 \varepsilon-\frac{19}{64}\varepsilon^2+\ldots\approx
0.40$ \cite{schaefer}, which can be compared to the PRISM and to the mean field
approximation, $\omega_{12}^{\rm RPA}=d-2/\nu$, which qualitatively
differs because it is negative.
The thread approximation, $\omega_{12}=0$, is correct at and above the upper
critical dimension $d_c$.
It appears difficult to envisage simple forms of $g(r)$ which reconcile
the prediction $h_q \propto q^{-d}$ for $q\to\infty$ \cite{jannink}
with the results for $\Sigma_2$ \cite{schaefer}, and the dilute limit of thread
PRISM theory qualitatively agrees with the later.
\subsection{Comparison with Monte Carlo simulations \label{sex4c}}
Monte Carlo simulations are well suited to study the inter--molecular
structure of polymer solutions but face the difficult challenge to achieve a
clear separation of the three
length scales, segment size $\sigma$ (or excluded volume size $R$), density
screening length $\xi_\rho$, and molecular size $R_g$ (or molecular
correlation length $\xi_c$); see the discussion in \cite{yamakov}. Whereas in
\cite{yamakov} in the range $1/R_g\ll q \ll 1/\xi_\rho$ a discrimination of
the
two predictions, $h_q\propto$ const. from scaling considerations \cite{jannink}
and $h_q\propto q^{-x}$ with $x\approx 1/\nu$ (as follows from the PRISM
treatment of the correlation hole)
appears possible and appears to support the latter, no clear
conclusions about the exponent in the asymptotic behavior, $h_q \to q^{-x}$
for $1/\xi_\rho\ll q \ll 1/R$, with $x=d=3$ (scaling picture),
$x=4$ (RPA), or $x=2/\nu\approx3.34$ (thread PRISM) were
possible. Even recent large scale simulations of the bond fluctuation model
(BFM) \cite{carmesin,deutsch}
do not provide a conclusive test of the large $q$
dependence if $h_q$ is considered \cite{marcus}. Figure \ref{fig3}
shows data from Ref. \cite{marcus}
for semidilute solutions and rather large chain lengths, $N=2048$,
where $\xi_c=94$, $\xi_\rho= 14$
for $c/c_*=97.7$ and
$\xi_\rho=31.1$ for $c/c_*=25.6$, and the steric
segment size is $R=2$ in units of the lattice constant of the BFM.
For a fit, the asymptotic thread PRISM prediction, \gl{3.5b}, is shifted by a
factor indicating that non--asymptotic corrections to $\bar c$ cannot be
neglected.
A clearer picture of the polymer mesh structure is provided by the pair
correlation function $g(r)$, which is the Fourier transform of $h_q$ and
asymptotically should follow \gl{3.5c} in the thread limit, $\sigma, R \ll
\xi_\rho$ and $\xi_\rho \ll \xi_c\propto R_g$. Figure \ref{fig2} shows the two
$g(r)$ for the above parameters, where $\xi_\rho$ is defined by collapsing the
simulation data onto the master curve at $g(r=\xi_\rho)=0.747$.
Note that this is an unfamiliar definition of $\xi_\rho$ which gives values
(theoretically) proportional to the standard ones.
These values of
$\xi_\rho$ also produce the collapse of the $h_q$ onto a common curve shown in
the inset of Fig. \ref{fig2} and lead to a reasonable collapse of the pair
correlation functions onto a common master curve. Finite size
corrections enter from short distances because of the finite excluded volume
segment sizes, $R/\xi_\rho$. These corrections
can also be understood as finite packing fraction corrections.
Large distance deviations from a common curve appear because of the finite
chain sizes, $\xi_c/\xi_\rho$. Nevertheless, the short distance behavior of
$g(r)$ provides a sensitive test of the various predictions.
The prediction $h_q\to 1/q^d$ would correspond to a logarithmic variation of
$g(r\to0)$, which appears to be ruled out by the data. Also the thread PRISM
prediction, $g(r\to0)\propto r^{1/3}$, appears incompatible with the data,
even if finite segment size corrections are approximated by a shift of the
$r$--origin. The RPA prediction for Gaussian polymers, $g(r\to0)-g_d\propto r$,
of a linear increase in $r$, can describe the
data over small intervals (like $0.1 < g < 0.25$)
but fails to account for the slight curvature of especially
the lower density curve. Moreover, the contact value of the RPA cannot be
expected to vanish asymptotically if parameters appropriate for a fit to $h_q$
are used. An increase in the range of a power law fit to $g(r)$
at the lower density up to an interval $0.1< g < 0.42$ is possible if the
following assumption about the pair correlation function for
semidilute solutions is made:
\beqa{final}
& g(r) \to \bar{g}(r/\xi_\rho)\; \qquad\mbox{for }\, \xi_\rho\to\infty\;\; ; \;
\xi_\rho/\xi_c
\to 0 \;,& \\
& \bar{g}(x\ll1) \propto x^{2/\nu-d+\omega_{12}(P)} \; .& \label{final2}
\end{eqnarray}
where for dilute cases the same power law with the replacement
$\xi_\rho\to\xi_c$ can be expected from scaling considerations. This power law
would match the scaling law for $g(r)$ for $r\to0$
smoothly to the calculated vanishing contact value $g_d$ from \gl{4.5}.
Note that such a matching is
predicted by PRISM. In \gl{final2} however, the exponent is corrected because
the correction to scaling exponent $\omega_{12}(P)$ is taken into account.
According to scaling arguments \cite{johner,schaefer},
there is a term of the form of \gl{final2} present
in the intra--molecular correlations also, although it is masked by
chain--end effects there.
The expected power law for good solutions, $g\propto r^{0.80}$ for $r\ll
\xi_\rho$ with $\nu=0.588$ and $\omega=0.40$ \cite{degennes,descloi,schaefer},
is compatible with the
simulation data, if a finite shift owing to a finite segment size is
anticipated. The power law $h_q\to 1/q^{2/\nu+\omega_{12}(P)}$
also is compatible with the data as shown in Fig. \ref{fig3}, but could less
be argued on data for $h_q$ only.
\section{Conclusions}
The thread PRISM results derived and discussed here justify earlier
phenomenological extensions of RPA like expressions. The density dependence of
the excluded volume parameter is derived from a microscopic incorporation of
inter--molecular excluded volume and intra--molecular connectivity.
Various comparisons with rigorous field theoretic calculations show that
leading asymptotic predictions, even for non--mean field like situations, are
captured correctly in the PRISM integral approach. The correction to scaling
exponent, which appears in the molecular mass dependence of the contact value
of two polymers, provides a typical example where thread PRISM provides a much
better description than mean field theory but fails to describe all
non--trivial correlations. PRISM theory suggests useful concepts like the pair
correlation function $g(r)$ and predicts scalings laws which provide a
framework for the interpretation of data if the exponents are corrected.
Thread PRISM thus turns out rather useful for semidilute
solutions, where it explicitly describes the inter--molecular correlations of
the polymer mesh and results from more rigorous approaches are scarce.
Moreover, as PRISM theory is successful for polymer melts, it provides
the unique possibility to approach polymer systems at all densities.
\acknowledgments{We would like to thank Profs. K. S. Schweizer, G. Jannink,
K. Binder, A. Milchev and L. Sch\"afer and Dr. J. Baschnagel
for many valuable discussions, and Dr. E. David for providing the programs to
solve the PRISM equations numerically.
This work was supported by the Deutsche
Forschungsgemeinschaft under Grants No. Fu 309/2-1 and Bi 317.}
|
\section{Introduction}
It is by now well established that extended objects play a fundamental r\^ole
in the non-perturbative completion of the superstring theories.
A few years ago this fact led to renewed interest in the
study of effective world-volume actions describing such objects.
In particular, while actions for the fundamental strings and for the M-theory
membrane~\cite{Bergshoeff:1987} have been known for a long time,
actions for the D-branes were constructed more recently
in refs~\cite{Leigh:1989,Cederwall:1996b,Cederwall:1996c,Aganagic:1996,
Bergshoeff:1996c}.
A further development was the observation that these branes can
equivalently be described by actions where the tension is generated
dynamically by a gauge-invariant world-volume $p{+}1$-form, $F_{p+1}$.
These actions have the generic form
\begin{equation}
\label{action}
S = \int{\rm d}^{p+1}\xi\sqrt{-g}\,\lambda\,[1 + \Phi(\{F_{i}\}) - ({*}F_{p+1})^2]\,,
\end{equation}
where $\{F_{i}\}$ collectively denotes the world-volume field
strengths (excluding $F_{p+1}$), and $\lambda$ is a Lagrange
multiplier for the constraint $1 +
\Phi({\{F_i\}}) - ({*}F_{p+1})^2 \approx 0$. The world-volume field strengths
(including $F_{p+1}$) have the schematic form
$F= {\rm d} A - C - ``C\kern-.1em\wedge\kern-.1em F\mbox{''}$. Here $``C\kern-.1em\wedge\kern-.1em F\mbox{''}$ denotes corrections
to the canonical, minimally coupled form determined by the requirement of
gauge invariance. The equation of motion for the tension gauge potential
$A_p$ leads to $\lambda\,{*}F_{p+1} = {\rm const}$; this constant is identified
with the tension.
An appealing feature of the formulation (\ref{action}) is that, in all known
cases, $\Phi$
is a polynomial function of its arguments. Notice also that there is no
Wess--Zumino term in the above action; this term is instead incorporated
in the $p{+}1$-form field strength $F_{p+1}$. Whenever convenient, one can
return to the formulation without the Lagrange multiplier $\lambda$ and the
tension gauge potential $A_p$ by solving their equations
of motion, thereby regaining the Wess--Zumino term in its traditional form.
Actions of the form (\ref{action}) were constructed for the M2-brane and the
fundamental strings in refs~\cite{Townsend:1992,Bergshoeff:1992} and more
recently for the D-branes in ref.~\cite{Bergshoeff:1998c}.
For the type {I\kern-.09em I}B branes it is known \cite{Schwarz:1995} that the
fundamental string (charge (1,0)) and the Dirichlet string (charge (0,1))
belong to a doublet of $(p,q)$ strings under the
non-perturbative SL(2,$\mathbb{Z}$) symmetry \cite{Hull:1995} of the
type {I\kern-.09em I}B theory, and an action for these $(p,q)$ strings which is
manifestly covariant under the SL(2,$\mathbb{Z}$) symmetry
\cite{Townsend:1997,Cederwall:1997} has been constructed. The D3-brane
on the other hand is a singlet under SL(2,$\mathbb{Z}$). In
the action which makes this property manifest, two world-volume two-form
field strengths
are required \cite{Cederwall:1998a}. In order to get the correct counting
for the degrees of freedom, an auxiliary duality relation between these
two fields has to be imposed at the level of the equations of motion. In
ref.~\cite{Cederwall:1998b} an action for the M5-brane was constructed which
circumvents the problems \cite{Witten:1996a} associated with the
self-dual world-volume three-form by implementing the self-duality
relation of this three-form as an auxiliary condition at the level of the
equations of motion. Thus, in the last two cases duality relations had to be
imposed to compensate for the fact that too many fields appear in the actions.
For the three-brane the introduction of extra fields was necessary to make a
symmetry manifest, and for the M5-brane to circumvent a topological
restriction.
In this paper auxiliary duality relations will be a recurrent theme.
A crucial requirement for supersymmetric boson-fermion matching of the
above brane actions is that of $\kappa$-symmetry, a local world-volume
symmetry for which the variation parameter $\kappa$ is a target-space spinor
which can be written as $\kappa=P_{+}\zeta=\mbox{$\frac{1}{2}$}(\hbox{1\kern-.27em l}+\Gamma)\zeta$, where $P_{\pm}$ are
mutually orthogonal projection operators, each reducing the number of
independent components of a spinor by half. These properties translate
into the requirements ${\rm tr}(\Gamma)=0$, and $\Gamma^2 = \hbox{1\kern-.27em l}$.
A long-standing problem has been the construction of a $\kappa$-symmetric
action for the type {I\kern-.09em I}B five-brane. While it is known \cite{Callan:1991}
that the theory on the world-volume is described (on-shell) by a
six-dimensional vector multiplet, there has been a debate in the literature
about whether
this multiplet should be realised in the action as a two-form or as
a four-form field strength (these are Poincar\'e dual, and hence have
the same number of degrees of freedom). Recently, a proposal based on
T-duality considerations for the bosonic part of the action for the
type {I\kern-.09em I}B NS5-brane action was put forward \cite{Eyras:1998}.
One of the main results of the present paper is the construction of a
$\kappa$-symmetric action for the type {I\kern-.09em I}B NS5-brane in a general curved
supergravity background, and with a world-volume field content which reflects
the fact that the D1-, D3- and D5-branes can end on it. The bosonic part
of our action differs from the one given in ref.~\cite{Eyras:1998}; we will
comment on this fact in later sections.
In ref.~\cite{Townsend:1996a} it was observed that in order to relate the
action for the directly dimensionally reduced M2-brane to the action for the
D2-brane, which describes the same physical object,
one has to perform a world-volume Poincar\'e dualisation of the one-form
on the world-volume of the dimensionally reduced membrane to a two-form,
which can be interpreted as the two-form field-strength
on the D2-brane world-volume,
and vice versa. Such dualisations were shown to be required also in other
interconnections between $p$-brane actions \cite{Tseytlin:1996,Aganagic:1997}.
There are however limitations on the applicability of this method.
The extension for $p\!>\!2$ to general backgrounds, where all form fields are
non-constant, was recently accomplished in ref.~\cite{Kimura:1999a}.
In the present paper we also work with general backgrounds.
The method of refs~\cite{Tseytlin:1996,Aganagic:1997} runs into more serious
problems when $p\!>\!4$, since one then encounters fifth-order algebraic
equations.
In this paper we propose a general method for constructing world-volume
actions with the requirements of $\kappa$-symmetry and gauge invariance as the
only input. To be able to apply the method one does not need to know even the
bosonic part of the action beforehand. We work within the framework where the
brane actions take the form~(\ref{action}).
The method furthermore leads to a new way of handling Poincar\'e
dualisation, which in a certain sense circumvents the above mentioned problem.
The dualisation is also more general in that it is not a
$\mathbb{Z}_2$ transformation; rather a whole set of interpolating theories
is constructed. The interpolating actions depend on
certain parameters which control the dualisation. For intermediate values of
the parameters the world-volume fields are ``doubled,'' i.e., they
come in pairs with their Poincar\'e duals.
In these formulations it can be arranged so that there is a world-volume
field corresponding to every possible brane-ending-on-another-brane
case~\cite{Strominger:1996,Townsend:1996b,Argurio:1997},
thus realising an equality among the various possible gauge
invariant world-volume field strengths on the branes. In order to obtain
the correct number of degrees of freedom for intermediate values of the
parameters auxiliary duality relations are imposed. As a byproduct of our
results we show that the form of the usual D-brane actions is determined by
super- and $\kappa$-symmetry alone (supplemented by gauge invariance).
The fundamental requirements we impose are that the world-volume fields
should be gauge invariant and satisfy Bianchi identities on the world-volume.
For the type {I\kern-.09em I}B branes we also require our actions to be invariant
under the perturbative Peccei--Quinn symmetry (this requirement is closely
related to the condition that the world-volume field strengths should satisfy
Bianchi identities). All hitherto known supersymmetric brane actions
are invariant under this symmetry.
The paper is organised as follows. In the next section we illustrate the
method for the D2-brane in the type {I\kern-.09em I}A theory.
In section~\ref{IIBbranesect}
we discuss the strings and the D3-brane in the type {I\kern-.09em I}B theory.
The construction of the action for the type {I\kern-.09em I}B NS5-brane is presented
in section~\ref{IIB5branesect}; this section also contains a discussion about
the D5-brane.
Finally, in the appendices we describe our notation and conventions, list
some properties of the ten-dimensional type {I\kern-.09em I} supergravities in
superspace relevant for our discussion, and give some details about
the method used to verify $\kappa$-symmetry of our actions.
In the latter context, we list a number of essential identities involving the
world-volume field strengths. These identities should also be useful in
applications of our results.
\setcounter{equation}{0}
\section{An introductory example: the D2/M2-brane}
In this section we will describe the method in a simple, but non-trivial,
setting---the D2-brane in type {I\kern-.09em I}A. This brane can also be described as
the dimensionally reduced M2-brane, the two actions being related via a
Poincar\'e-duality transformation of the
world-volume fields \cite{Townsend:1996a}\footnote{In order to obtain the
M2-brane action from the D2-brane action one has to perform a world-volume
Poincar\'e-duality transformation followed by an S-duality transformation,
which, given the relation between the eleventh dimension and the dilaton
\cite{Townsend:1995a,Witten:1995a}, corresponds to lifting the action to
eleven dimensions.}.
This is, in fact, the way the $\kappa$-symmetric action for the D2-brane was
first constructed~\cite{Townsend:1996a}.
The action for the dimensionally reduced M2-brane contains the world-volume
field $F^{\rm M2}_3 = {\rm d} A_2 - C_3 + B_2 \, F_1$ which generates the tension
(here $F_1 = {\rm d} A_0 - C_1$).
Similarly, an action for the D2-brane where the tension is replaced by
the world-volume field $F^{\rm D2}_3 = {\rm d} A_2 - C_3 - C_1 \, F_2$
has been constructed~\cite{Bergshoeff:1998c}
(here as usual $F_2 = {\rm d} A_1 - B_2$).
Before we continue we will briefly discuss some facts about the background
type {I\kern-.09em I}A supergravity theory (more details can be found in
appendix~\ref{sugraapp}).
The gauge-invariant field strengths in the type {I\kern-.09em I}A
theory which are relevant for the discussion in this section are
$R_2$, $H_3$ and $R_4$. These fields satisfy the Bianchi identities
\begin{equation}
{\rm d} R_2 = 0\,, \qquad {\rm d} H_3 = 0\,, \qquad {\rm d} R_4 = H_3\, R_2\,.
\end{equation}
The first two relations are ``solved'' by $R_2 = {\rm d} C_1$ and $H_3 = {\rm d} B_2$,
whereas there is an ambiguity in the definition of $R_4$;
if one requires $R_4$ to satisfy the above Bianchi identity one arrives at
the natural ``solution'' $R_4 = {\rm d} C_3 + x\,B_2 \, R_2 -
(1{-}x)\,C_1\, H_3 = {\rm d} (C_3 + x\,C_1\, B_2) - C_1\, H_3$.
{}From the last equality we see that the parameter $x$ arises from an ambiguity
in the definition of $C_3$, the most general natural field redefinition
being $C_3 \rightarrow C_3 + x\, C_1 \, B_2$. This innocent field redefinition will
play a central role in the sequel. Different values of the
parameter $x$ describe the same physics, but, as we will see later, the
corresponding descriptions of the world-volume theories can be quite
different. The background field strengths
are invariant under the gauge transformations
\begin{eqnarray}
\label{backgaug}
\delta C_1 &=& {\rm d} L_0\,, \qquad\quad \delta B_2 = {\rm d} L_1\,, \nonumber\\
\delta C_3 &=& {\rm d} L_2 - x\,R_2 \, L_1 - (1{-}x)\,H_3\, L_0\,.
\end{eqnarray}
We now turn to the construction of gauge-invariant world-volume field
strengths satisfying Bianchi identities.
The general form of these field strengths is
$F = {\rm d} A - C - ``C\wedge F{\mbox{''}}$. Here $C$ collectively denotes the
background potentials (both the RR and the NS-NS ones) and
$``C\wedge F{\mbox{''}}$ denotes possible corrections to the minimally coupled
form, required by gauge invariance.
It is easy to see that $F_1 = {\rm d} A_0 - C_1$, and $F_2 = {\rm d} A_1 - B_2$
are invariant under the gauge transformations (\ref{backgaug})
together with
$\delta A_0 = c + L_0$ and $\delta A_1 = {\rm d} l_0 + L_1$ (here $c$ is a constant),
and that they satisfy the Bianchi identities
\begin{equation}
\label{BIF12}
{\rm d} F_1 = -R_2 \,, \qquad {\rm d} F_2 = -H_3\,.
\end{equation}
Furthermore, using the above form of $R_4$, we see that the world-volume
three-form field strength
\begin{equation}
F_3 = {\rm d} A_2 - C_3 + x\, B_2\, F_1 - (1{-}x)\, C_1\, F_2
\end{equation}
obeys the Bianchi identity
\begin{equation}
\label{BIF3}
{\rm d} F_3 = -R_4 + x\,F_1 \, H_3 - (1{-}x)\,F_2 \, R_2\,,
\end{equation}
and is invariant under the above gauge transformations combined with
$\delta A_2 = {\rm d} l_1 + L_2 + x\,L_1\, F_1 + (1{-}x)\, L_0\, F_2$.
Thus, on the world-volume the field redefinition in the background has
a more profound implication: it determines which world-volume field
strengths appear. We see that for $x=1$ we obtain the correct
tension field for the description of the dimensionally reduced M2-brane,
whereas for $x=0$ we instead find the D2-brane description.
The limits $x=0$ and $x=1$ thus lead to sensible results,
but what happens for other values of $x$? The answer to this question
is given below, where we construct an action which interpolates
between the two limiting cases discussed at the beginning of the section.
The interpolation is controlled by the real parameter $x$ introduced
above. In order for the kinetic term to be positive semi-definite, it appears
desirable to impose the physical restriction $x\in [0,1]$.
We make a general Ansatz for the action of the form
\begin{equation}
\label{D2ansatz}
S = \int {\rm d}^{3}\xi\sqrt{-g}\,\lambda \left[1 +
\Phi(e^{\frac{3}{4}\phi}F_1,e^{-\frac{1}{2}\phi}F_2) -
e^{\frac{2}{4}\,\phi}({*}F_3)^2\right]\,.
\end{equation}
The dilaton dependence in this expression can be motivated by
considering the dilaton-scaling of the supergravity constraints, which,
via the basic $\kappa$-variations, directly determine the dilaton-factor
in front of each occurrence of the world-volume field strengths. We will
from now on suppress the dilaton factors; they can be reinstated at any
time by using the rules $F_1 \rightarrow e^{\frac{3}{4}\phi}F_1$,
$F_2\rightarrow e^{-\frac{1}{2}\phi}F_2$ and $F_3\rightarrow e^{\frac{1}{4}\phi}F_3$.
The equation of motion for $A_2$ is ${\rm d} [\lambda\,{*}F_{3}]=0$, whereas the ones
for $A_0$ and $A_1$ are
\begin{eqnarray}
{\rm d}\Big[\lambda\{{*}\frac{\delta\Phi}{\delta F_1} + 2 x\, B_2 \,{*}F_3\}\Big] &=& 0\,,
\nonumber \\
{\rm d}\Big[\lambda\{{*}\frac{\delta\Phi}{\delta F_2} - 2 (1{-}x)\,C_1\,{*}F_3\}\Big] &=& 0\,.
\end{eqnarray}
By using the Bianchi identities for $F_1$ and $F_2$ together with the
result ${\rm d}[\lambda\,{*}F_{3}]=0$ one can eliminate the
explicit appearance of the background field strengths:
\begin{eqnarray}
{\rm d}\Big[\lambda\{{*}\frac{\delta\Phi}{\delta F_1} - 2x\,F_2\,{*}F_3\}\Big] &=& 0\,,\nonumber \\
{\rm d}\Big[\lambda\{{*}\frac{\delta\Phi}{\delta F_2} + 2 (1{-}x)\, F_1\, {*}F_3\}\Big]
&=& 0\,.
\end{eqnarray}
It is thus consistent with the equations of motion and the Bianchi
identities to impose the ``duality'' relations
\begin{eqnarray}
\label{D2dualrel}
-2x\,{*}F_3\,{*}F_2 &=& K_1 := \frac{\delta\Phi}{\delta F_1}\,,\nonumber\\
2(1{-}x)\,{*}F_3\,{*}F_1 &=& K_2 := \frac{\delta\Phi}{\delta F_2}\,,
\end{eqnarray}
where $\Phi$ is yet to be determined.
These relations reduce the number of degrees of freedom by half, and thus
remedy the doubling of fields in the action. Auxiliary duality relations
of this kind have been used before
in the literature \cite{Bergshoeff:1996e,Cederwall:1998a,Cederwall:1998b}.
For the two limiting values $x=0$ and $x=1$ the duality
relations~(\ref{D2dualrel}) become degenerate, so that we have the
same number of degrees of freedom for all values of $x$, namely that of
a one-form $F_1$ (or, equivalently, of its Poincar\'e dual $F_2$).
Turning next to the $\kappa$-symmetry properties of the action~(\ref{D2ansatz}),
it can be shown that the $\kappa$-transformations of the world-volume fields are
\begin{eqnarray}
\delta_\kappa g_{ij} &=& 2\,{E_{(i}}^a\,{E_{j)}}^B\kappa^\alpha\,T_{\alpha Ba}\,,
\quad
\delta_\kappa\phi= \kappa^\alpha\partial_\alpha\phi \,, \nonumber\\
\delta_\kappa F_1 &=& -i_{\kappa}R_1\,,
\hspace{23.5mm}
\delta_{\kappa}F_2 = -i_\kappa H_3\,, \nonumber\\
\delta_\kappa F_3 &=& -i_{\kappa}R_4+x\,F_1\,i_{\kappa}H_3-(1{-}x)\,F_2\,i_{\kappa}R_2\,.
\end{eqnarray}
Notice the close correspondence between the variations of the world-volume
field strengths and their respective Bianchi identities given in
eqs~(\ref{BIF12}) and (\ref{BIF3}).
This correspondence holds for all cases considered in this paper.
We would now like to check whether the action~(\ref{D2ansatz}) is invariant
under these transformations. To this end, it is necessary and sufficient
to show that the variation of the constraint
$\Upsilon = 1 + \Phi - ({*}F_3)^2 \approx 0$ is zero. However, we do not
know the form of the function $\Phi$. The way out of this impasse
\cite{Cederwall:1998b,Cederwall:1998a} is to note that if we assume that the
scalar functional $\Phi$ is formed out of contractions between the
world-volume field strengths and the metric only,\footnote{We thus
exclude WZ-type terms linear in $\epsilon^{ijk}$, which is reasonable since in the
formulation we are using such terms are contained in $F_3$.}
a simple scaling argument shows that the variation of $\Upsilon$ can be
written\footnote{The part of this expression that is proportional to the
dilaton variation was derived by temporarily reinstating the suppressed
dilaton-dependence of the action~(\ref{D2ansatz}) by means of the previously
given substitution rules
$F_1\rightarrow e^{\frac{3}{4}\phi}F_1$, $F_2\rightarrow e^{-\frac{1}{2}\phi}F_2$ and
$F_3\rightarrow e^{\frac{1}{4}\phi}F_3$.}
\begin{eqnarray}
\delta_{\kappa} \Upsilon &=& K^{i}\,\delta_{\kappa}F_{i} +
\mbox{$\frac{1}{2!}$}K^{ij}\,\delta_{\kappa}F_{ij} +
\mbox{$\frac{2}{3!}$}F^{ijk}\,\delta_{\kappa}F_{ijk} - \big[\mbox{$\frac{1}{2}$} K^{(i}F^{j)} +
\mbox{$\frac{1}{2!}$}{K^{(i}}_{l}F^{j)l} \nonumber\\ && +
\mbox{$\frac{3}{2}$}\kern-.13em\cdot\kern-.13em \mbox{$\frac{2}{3!}$}{F^{(i}}_{lm}F^{j)lm}
\big]\,\delta_{\kappa}g_{ij} +
\big[ \mbox{$\frac{3}{4}$}K_1\kern-.13em\cdot\kern-.13em F_{1} -\mbox{$\frac{1}{2}$} K_2\kern-.13em\cdot\kern-.13em F_{2}
+ \mbox{$\frac{2}{4}$} F_3\kern-.13em\cdot\kern-.13em F_3\big]\,\delta_{\kappa}\phi\,.
\end{eqnarray}
This improves matters considerably since from~(\ref{D2dualrel}) we have
explicit expressions for $K_1$ and $K_2$, which are valid when the duality
relations are imposed. Hence, we are in the fortunate situation that
although we do not know the action we know its variation under a
$\kappa$-transformation. By demanding that
$\Upsilon$ be $\kappa$-invariant, we can exploit this knowledge to derive the action,
as we shall demonstrate next.
Inserting the expressions for the variations of the world-volume fields,
together with the background constraints leads to
\begin{eqnarray}
(\delta_\kappa\Upsilon)^{(1/2)} &=& -2\,{*}F_3\,{\bar{\La}}\Big[ \Xi +
2\,{*}F_1\kern-.13em\cdot\kern-.13em\gamma_2\gamma_{11} +
3\,{*}F_2\kern-.13em\cdot\kern-.13em\gamma_1 \gamma_{11} + (3{-}x)\,{*}(F_1\kern-.1em\wedge\kern-.1em F_2) + {*}F_3 \Big]
\kappa\,, \nonumber\\
(\delta_\kappa\Upsilon)^{(0)} &=& 4i\,{*}F_3\,{\bar{E}}_i
\Big[\frac{\epsilon^{ijk}}{2\sqrt{-g}}\gamma_{jk}
- {*}F_1^{ij}\,\gamma_j\gamma_{11} + {*}F_2^{i}\,\gamma_{11} +
x\,g^{ij}\,F_1\kern-.13em\cdot\kern-.13em{*}F_2\,\gamma_j + F_1^{(i}{*}F_2^{j)}\,\gamma_j \nonumber \\
&&\hspace{1.2cm} +\,\, g^{ij}\,{*}F_3\,\gamma_{j}\Big] \kappa \,.
\end{eqnarray}
These expressions should vanish when $\kappa$ is of the form $P_+\zeta = \mbox{$\frac{1}{2}$}(\hbox{1\kern-.27em l} +
\Gamma)\zeta$, which leads us to another question: how do we determine $P_+$?
This is a less serious problem since there are not that many ``natural''
terms that
can appear. In all known cases one can write ${*}P_+$ as a $p{+}1$-form
expression involving the world-volume fields and the totally
antisymmetric products of $\gamma$-matrices, $\gamma_{i_1\ldots
i_r}$, considered as forms. By making a natural Ansatz for $P_+$,
inserting it into the above variations, and then using the
formula~(\ref{general}) to expand the products of $\gamma_{i_1\ldots i_r}$'s
one obtains two expressions which are linear combinations of
$\gamma_{i_1\ldots
i_r}$'s, for various values of $r$. Requiring these expressions to vanish
forces each tensor component to vanish separately, since the $\gamma_{i_1\ldots
i_r}$'s are linearly independent (for generic embeddings).
These conditions determine expressions for the duality
relations~(\ref{D2dualrel}), which can be integrated to give $\Phi$;
for further details, see appendix~\ref{kappadetails}.
The projection operator which makes the above variation vanish is
\begin{equation}
2\,{*}F_3\,P_\pm = {*}F_3\,\hbox{1\kern-.27em l}
\mp\big[\Xi - x\,{*}F_1\kern-.13em\cdot\kern-.13em \gamma_2\gamma_{11} +
(1{-}x)\,{*}F_2\kern-.13em\cdot\kern-.13em\gamma_1 \gamma_{11} \big]\,,
\end{equation}
and the result for the action is
\begin{equation}
S = \int {\rm d}^3\xi\,\sqrt{-g}\,\lambda\,\big[1 + x\,F_1\kern-.13em\cdot\kern-.13em F_1 +
(1{-}x)\,F_2\kern-.13em\cdot\kern-.13em F_2 +
x\,(1{-}x)\,F_1\kern-.13em\cdot\kern-.13em F_1\;F_2\kern-.13em\cdot\kern-.13em F_2 - ({*}F_3)^2\big]\,,
\end{equation}
supplemented by the duality relations
\begin{eqnarray}
{*}F_3\,{*}F_1 &=& F_2 + x\,(F_1\kern-.13em\cdot\kern-.13em F_1)\,F_2 \,,\nonumber \\
-{*}F_3\,{*}F_2 &=& F_1 + (1{-}x)\,(F_2\kern-.13em\cdot\kern-.13em F_2)\,F_1 \,.
\end{eqnarray}
At this point we would like to make a few comments. Notice the
similarity between the Bianchi identity for $F_3$ and the
form of ${*}P_+$, which becomes even closer if we use the
formal rules $R_{2n}
\rightarrow -(-)^{n}\gamma_{2n-1}(\gamma_{11})^{n}$, $H_3 \rightarrow \gamma_2\gamma_{11}$.
A similar correspondence holds for all hitherto known
brane actions. If assumed to hold in all cases, it reduces the amount of
guesswork involved (basically we only had to guess the form for $P_+$)
and makes the method more algorithmic.
We will comment further on this issue in later sections.
Let us also remark that in the simple case considered above one could
alternatively have made progress by making an Ansatz for $\Phi$;
for higher-dimensional cases this approach is less feasible, however.
To summarise, in the two limits $x=0$ and $x=1$ we recover known results with
correct expressions for the projection operator $P_+$. For other values of $x$
we obtain new $\kappa$-symmetric formulations of the D2-brane action.
In particular, for the choice $x=\mbox{$\frac{1}{2}$}$ one obtains a formulation which treats
the two world-volume fields in a symmetric fashion.
We would like to stress that these actions are all equivalent
(i.e., they describe the same physical object) as is
obvious from the way the parameter $x$ was introduced.
\setcounter{equation}{0}
\section{Some further examples}
\label{IIBbranesect}
\subsection*{The type {I\kern-.09em I}B D-string}
The D-string in the type {I\kern-.09em I}B theory couples minimally to the two-form
potential $C_2$. Possible world-volume fields satisfying Bianchi
identities
are
\begin{eqnarray}
F_0 &=& \mu - C_0\,, \phantom{{\rm d} A_1 - B_2} {\rm d} F_0 = R_1\,, \nonumber \\
F_2 &=& {\rm d} A_1 - B_2\,, \phantom{\mu - C_0}{\rm d} F_2 = H_3\,,
\end{eqnarray}
where $\mu$ is a constant. In order for $F_0$ to be invariant under the
Peccei-Quinn symmetry, $\mu$ has to change to compensate for the constant shift
of $C_0$. We will comment on this fact later on. Given the above two fields,
one can construct the following expressions for the tension form $\tilde{F}_2$:
\begin{eqnarray}
\tilde{F}_2 &=& {\rm d} \tilde{A}_1 - C_2 + x\,B_2\, F_0 - (1{-}x)\,C_0 \,F_2\,, \nonumber \\
{\rm d} \tilde{F}_2 &=& -R_3 + H_3\,F_0 - (1{-}x)\,R_1\, F_2 \,.
\end{eqnarray}
Here the background field strength $R_3$ is defined as
\begin{equation}
R_3 = {\rm d} C_2 + x\,B_2 \,{\rm d} C_0 - (1{-}x)\,C_0 \,{\rm d} B_2\,,
\end{equation}
and satisfies the usual Bianchi identity, ${\rm d} R_3 = R_1\, H_3$.
The action is taken to be of the form\footnote{As for the
D2-brane we suppress the dilaton dependence; it can be reinstated by using
the rules: $F_0\rightarrow e^{-\phi}F_0$, $F_2 \rightarrow e^{-\frac{1}{2}\phi}F_2$ and
$\tilde{F}_2 \rightarrow e^{\frac{1}{2}\phi}\tilde{F}_2$.}
\begin{equation}
S = \int {\rm d} ^2 \xi \,\sqrt{-g}\, \lambda \,\big[ 1 + \Phi(F_0, F_2) -
({*}\tilde{F}_2)^2\big]\,.
\end{equation}
In order to be able to apply the method of the previous section it is
convenient to formally regard $\mu$ as the exterior derivative of a
``$-1$''-form, $A_{-1}$. This procedure will be justified later in this
section. Using the just mentioned formal trick, the duality relations can
be shown to be determined by
\begin{eqnarray}
2(1{-}x)\,{*}\tilde{F}_2\,{*}F_0 &=& K_2 := \frac{\delta \Phi}{\delta F_2}\,,\nonumber \\
-2x\,{*}\tilde{F}_2\,{*}F_2 &=& K_0 := \frac{\delta \Phi}{\delta F_0}\,.
\end{eqnarray}
As in the case of the D2-brane, the $\kappa$-variation of the constraint
$\Upsilon = 1+\Phi-({*}\tilde{F}_2{*})^2\approx 0$ can be written in terms of the $K$'s:
\begin{eqnarray}
\delta_{\kappa} \Upsilon &=& K_0\,\delta_{\kappa}F_0 + \mbox{$\frac{1}{2!}$}\,K^{ij}\,
\delta_{\kappa}F_{ij} + \mbox{$\frac{2}{2!}$}\,\tilde{F}^{ij}\,\delta_{\kappa}\tilde{F}_{ij} -
\big(\mbox{$\frac{1}{2}$}\,{K^{(i}}_{l}\,F^{j)l} + {\tilde{F}}{}^{(i}{}_{l}\,\tilde{F}^{j)l}\big)\,
\delta_{\kappa}g_{ij} \nonumber\\ && + \mbox{$\frac{1}{2!}$}\,\big(
\mbox{$\frac{1}{2}$}\tilde{K}^{ij}\,\tilde{F}_{ij} -\mbox{$\frac{1}{2}$} K^{ij}\,F_{ij}\big)\,\delta_{\kappa}\phi\,.
\end{eqnarray}
The $\kappa$-variations of the fields are $\delta_{\kappa}F_0 = -i_{\kappa}R_1$,
$\delta_{\kappa}F_2 = -i_{\kappa}H_3$ and $\delta_{\kappa}\tilde{F}_2 = -i_{\kappa}R_3 +
F_0\,i_{\kappa}H_3 - (1{-}x)\,F_2\,i_{\kappa}R_1$ (the variations of the metric and
the dilaton are the same as in the D2-brane case). Using these
expressions together with the supergravity constraints (see
appendix~\ref{sugraapp}) one then obtains the expressions
\begin{eqnarray}
(\delta_{\kappa}\Upsilon)^{(1/2)} &=& 2\,{*}\tilde{F}_2\,{\bar{\La}} \Big[-\Xi\,J + F_0\,\Xi\,K
- \,{*}F_2\,I - (1{+}x)\,F_0\,{*}F_2 - {*}\tilde{F}_2 \Big]\kappa\,, \nonumber\\
(\delta_{\kappa}\Upsilon)^{(0)} &=& 4i\,{*}\tilde{F}_2\,{\bar{E}}_i \Big[
\frac{\epsilon^{ij}}{\sqrt{-g} }\,\gamma_{j}\,J + F_0\,\frac{\epsilon^{ij}}{\sqrt{-g}}
\gamma_{j}\,K - (1{-}x)\,F_0\,{*}F_2\,g^{ij}\,\gamma_j \nonumber \\
&&\hspace{1.7cm}+\,\, g^{ij}\,{*}\tilde{F}_2\,\gamma_j \Big]\kappa\,.
\end{eqnarray}
The projection $\kappa=P_+\zeta$ which makes the above variations vanish turns out
to be
\begin{equation}
P_\pm \propto -{*}(\tilde{F}_2 - \alpha F_0\,F_2)\hbox{1\kern-.27em l} \pm [\Xi\,J + (x{+}\alpha)\,F_0\,
\Xi\,K + (1{-}(x{+}\alpha))\,{*}F_2\,I]\,,
\end{equation}
where the parameter $\alpha$ is undetermined by the requirement of
$\kappa$-symmetry.
Following the same approach as for the D2-brane one obtains the
action\footnote{Note the similarity with the result for the D2-brane in the
previous section, a fact which should be derivable from T-duality
considerations.}
\begin{equation}
S = \int {\rm d}^2 \xi\,\sqrt{-g}\,\lambda\,[1 + x\,F_0^2 + (1{-}x)\,F_2\kern-.13em\cdot\kern-.13em F_2
+ x\,(1{-}x)\,F_0^2\,F_2\kern-.13em\cdot\kern-.13em F_2]\,.
\end{equation}
The duality relations which supplement this action are
\begin{eqnarray}
{*}\tilde{F}_2\,{*}F_0 &=& F_2 + x\,F_0^2\,F_2\,, \nonumber \\
-{*}\tilde{F}_2\,{*}F_2 &=& F_0 + (1{-}x)\,F_2\kern-.13em\cdot\kern-.13em F_2\,F_0 \,.
\end{eqnarray}
It is fairly straightforward to show that the second relation follows
from the first one and the constraint $1 + \Phi = ({*}\tilde{F}_2)^2$,
thus showing that it can be consistently imposed and justifying the earlier
formal derivation.
Let us close this subsection by discussing some simple limiting cases of the
above action. When $x=0$, we recover the conventional $\kappa$-symmetric D-string
action. In the opposite limit, $x=1$, $\Phi$ equals
$1 + F_0^2 = 1 + (\mu - C_0)^2$. This result agrees with the well-known result
of refs~\cite{Tseytlin:1996,Aganagic:1997,Oda:1998},
where it was shown that $\mu$ has
the correct transformation property under the Peccei-Quinn symmetry (which in
particular transforms $C_0$ to $C_0 + 1$) to make $F_0$ invariant. It was
furthermore shown that the action could be interpreted as the action for the
fundamental string in an SL(2,$\mathbb{Z}$)-transformed background, thus
establishing the S-duality connection between the D- and F-strings at the
level of their world-volume actions.
\subsection*{The type {I\kern-.09em I}B D3-brane}
The complete $\kappa$-symmetric action for the super-D3-brane in the type
{I\kern-.09em I}B theory was constructed in ref.~\cite{Cederwall:1996b}. It is known
that the D3-brane is a singlet under the SL(2,$\mathbb{Z}$) symmetry of the type {I\kern-.09em I}B
theory. This fact was demonstrated at the level of the world-volume action in
refs~\cite{Tseytlin:1996,Green:1996a,Aganagic:1997,Kimura:1999a}, and
an action which makes the SL(2,$\mathbb{Z}$) symmetry manifest was constructed in
ref.~\cite{Cederwall:1998a}.
In this section we will apply our method to the type {I\kern-.09em I}B D3-brane, and
compare our findings with previously known results. The parameterisations for
the background field strengths which we will use are
\begin{eqnarray}
R_1 &=& {\rm d} C_0\,, \qquad\qquad\hspace{.5pt}
R_3 = {\rm d} C_2 - C_0\, H_3\,, \nonumber \\
H_3 &=& {\rm d} B_2\,, \qquad\qquad
R_5 = {\rm d} C_4 + x\,B_2 \,{\rm d} C_2 - (1{-}x)\,C_2\, H_3 \,.
\end{eqnarray}
Notice that we have not introduced a parameter in the definition for
$R_3$, in contrast to the case of the D-string. We only
introduce parameters which lead to (Poincar\'e) dual
pairs of world-volume field strengths. Although this is {\it a priori}\/ a
restriction, it is unclear whether it excludes any physically interesting
cases. The introduction of a parameter in the expression for $R_3$ would
have lead to the appearance of $F_0$ in various places, but there is
no corresponding four-form to which it can
be dual (recall that for the D-string there where two
two-forms). Given the above parameterisation the possible gauge
invariant world-volume fields and their Bianchi identities become
\begin{eqnarray}
\label{D3WVfields}
F_{2} &=& {\rm d} A_1 - B_2\,,\phantom{{\rm d}\tilde{A}_1-C_2-C_0\,F_2\,,} {\rm d} F_2 = -H_3\,,
\nonumber\\
\tilde{F}_{2} &=& {\rm d} \tilde{A}_1 - C_2 - C_0\, F_2\,,\phantom{{\rm d} A_1-B_2\,,}{\rm d}
\tilde{F}_2 = -R_3 -R_1\,F_2\,, \nonumber \\
F_{4} &=& {\rm d} A_3 - C_4 +x\,B_2\,\tilde{F}_2 - (1{-}x)\,C_2\,F_2 + x\,C_0\,B_2\,F_2 +
(x{-}\mbox{$\frac{1}{2}$})\,C_0\,F_2\,F_2\,, \nonumber\\
{\rm d} F_4 &=& -R_5 + x\,H_3\,\tilde{F}_2 - (1{-}x)\,F_2\,R_3 +
(x{-}\mbox{$\frac{1}{2}$})\,R_1\,F_2\,F_2\,.
\end{eqnarray}
There is an alternative way of looking at the parameter $x$ in $F_4$
above. Consider the expression $F^{\rm D3}_4 - x\, F_2\,\tilde{F}_2$, where
$F^{\rm D3}_4$
is the canonical D-brane expression, $F^{\rm D3}_4 = {\rm d} A_3 - C_4 -
C_2\,F_2$. By making the
field redefinitions $C_4 \rightarrow C_4 + x\,B_2\,C_2$ together with
$A_3 \rightarrow A_3 - xA_1\,d\tilde{A}_1$, we recover the expression for $F_4$
given above. Thus, at the
world-volume level the dualisation is performed by adding the term
$F_2\,\tilde{F}_2$ to $F_4$. This is analogous to the way one usually
proceeds. Looking at the dualisation this way leads to an easy
way of determining which field parameterisations are needed for the
background fields.
The variations of the world-volume form fields under a $\kappa$-symmetry
transformation are
\begin{eqnarray}
\label{D3WVvar}
\delta_{\kappa}F_2 &=& -i_{\kappa}H_3\,, \nonumber \\
\delta_{\kappa}\tilde{F}_2 &=& -i_{\kappa}R_3 - F_2\,i_{\kappa}R_1\,, \nonumber\\
\delta_{\kappa}F_4 &=& -i_{\kappa}R_5 + x\,\tilde{F}_2\,i_{\kappa}H_3 - (1{-}x)\,F_2\,i_{\kappa}R_3
- (\mbox{$\frac{1}{2}$}{-}x)\,F_2\,F_2\,i_{\kappa}R_1\,.
\end{eqnarray}
We make a general Ansatz for the action of the form\footnote{The dilaton
dependence is as usual suppressed, but can be reinstated using
the rules: $F_2 \rightarrow e^{-\frac{1}{2}\phi}F_2$,
$\tilde{F}_2 \rightarrow e^{\frac{1}{2}\phi}\tilde{F}_2$ and $F_4\rightarrow F_4$.}
\begin{equation}
\label{D3ansatz}
S = \int {\rm d}^4 \xi\, \sqrt{-g}\,\lambda \left[ 1 + \Phi(F_2,\tilde{F}_2) -
({*}F_4)^2\right]\,.
\end{equation}
It is often convenient to rewrite this action in ``form language'' as
\begin{equation}
S = - \int \lambda \left[ {*}1 + {*}\Phi(F_2,\tilde{F}_2) + F_4\,{*}F_4\right]\,.
\end{equation}
This form of the action is better suited for the derivation of the duality
relations consistent with the Bianchi identities and the equations of motion.
In the present case these duality relations must be of the form
\begin{eqnarray}
\label{D3dualrel}
2(1{-}x)\,{*}F_4\,{*}\tilde{F}_2 &=& K_2:=\frac{\delta \Phi}{\delta F_2}\,,\nonumber\\
-2x\,{*}F_4\,{*}F_2 &=& \tilde{K}_2 := \frac{\delta \Phi}{\delta \tilde{F}_2}\,.
\end{eqnarray}
In order for the action~(\ref{D3ansatz}) to be $\kappa$-invariant
it is necessary and sufficient that the constraint $\Upsilon = 1 +
\Phi(F_2,\tilde{F}_2) - ({*}F_4)^2 \approx 0$ is invariant.
By using the same argument as before we can rewrite the
$\kappa$-variation of the constraint in terms of the $K$'s as
\begin{eqnarray}
\delta_{\kappa} \Upsilon &=& \mbox{$\frac{1}{2!}$}K^{ij}\delta_{\kappa}F_{ij} +
\mbox{$\frac{1}{2!}$}\tilde{K}^{ij}\delta_{\kappa}\tilde{F}_{ij} +
\mbox{$\frac{2}{4!}$}F^{ijkl}\delta_{\kappa}F_{ijkl} - \big(\mbox{$\frac{1}{2}$}
{K^{(i}}_{l}F^{j)l} + \mbox{$\frac{1}{2}$}{\tilde{K}^{(i}}{}_{l}\tilde{F}^{j)l} \nonumber\\ && +\,
\mbox{$\frac{1}{3!}$}{F^{(i}}_{lmn}F^{j)lmn} \big)\delta_{\kappa}g_{ij} +
\mbox{$\frac{1}{2!}$}\big( \mbox{$\frac{1}{2}$}\tilde{K}^{ij}\tilde{F}_{ij} -
\mbox{$\frac{1}{2}$} K^{ij}F_{ij}\big)\delta_{\kappa}\phi\,.
\end{eqnarray}
Inserting the expressions (\ref{D3dualrel}) for the $K$'s and using the
variations (\ref{D3WVvar}) of the world-volume fields together with the
on-shell constraints for the background fields, we then obtain
\begin{eqnarray}
\label{UpsD3}
(\delta_{\kappa}\Upsilon)^{(1/2)} &=& 2\,{*}F_4\,{\bar{\La}} \Big[
\mbox{$\frac{1}{2}$}\, e^{\phi/2}\,{*}\tilde{F}^{ij}\gamma_{ij}\,K -
\mbox{$\frac{1}{2}$}\, e^{-\phi/2}\,{*}F^{ij}\gamma_{ij}\,J - e^{-\phi}\,{*}(\tilde{F} \kern-.1em\wedge\kern-.1em \tilde{F})\,I
- {*}(F \kern-.1em\wedge\kern-.1em \tilde{F}) \Big]\kappa\,, \nonumber\\
(\delta_{\kappa}\Upsilon)^{(0)} &=& 4i\,{*}F_4\,{\bar{E}}_i \Big[
\frac{\epsilon^{ijkl}}{3!\sqrt{-g} } \gamma_{jkl}\,I + e^{\phi/2}{*} \tilde{F}^{ij}
\gamma_{j}\,K + e^{-\phi/2}\, {*}F^{ij}\gamma_{j}\,J \nonumber\\
&&\hspace{1cm}-\,\, (1{-}x)g^{ij}\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em F_2)\gamma_{j} +
{*}{F^{(i}}_{l}\tilde{F}^{j)l}\gamma_j + g^{ij}\,{*}F_4\gamma_j \Big]\kappa\,.
\end{eqnarray}
The correct projection operator can be shown to be (more details can be found
in appendix~\ref{kappadetails})
\begin{eqnarray}
P_\pm &\propto& {*}\big[F_4 - \alpha\,(F_2\kern-.1em\wedge\kern-.1em \tilde{F}_2)\big]\,\hbox{1\kern-.27em l} \mp
\bigg[\Xi I + \mbox{$\frac{1}{2}$}(1 - (x+ \alpha))\,{*}F_2^{ij}
\gamma_{ij}J \,,
\nonumber \\ &&\hspace{1cm}+\,\,\mbox{$\frac{1}{2}$}(x{+}\alpha)\,{*}\tilde{F}_2^{ij}\gamma_{ij}K +
\mbox{$\frac{1}{2}$}(1-2(x{+}\alpha))\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2)I \bigg]\,.
\end{eqnarray}
Again we note the similarity with the Bianchi identity; from this
perspective the free
parameter $\alpha$ can be understood from the fact that the Bianchi
identity for $F_4$ can be rewritten as \,
${\rm d} (F_4 - \alpha F_2\,\tilde{F}_2) = -R_5 + (x+\alpha)H_3\,\tilde{F}_2 - [1-(x+\alpha)]F_2\,R_3 +
[(x+\alpha)-\mbox{$\frac{1}{2}$}]R_1\,F_2\,F_2\,$.
Inserting the above expression for $\kappa = P_+ \zeta$ into the variations
(\ref{UpsD3}) and using the expansion
formula (\ref{general}) for the product of two $\gamma$-matrices, we obtain a
set of component expressions which must each vanish in order for
$\kappa$-symmetry to hold. The final results of the analysis of these
expressions are the duality relations (for further details, see appendix
\ref{kappadetails})
\begin{eqnarray}
-{*}F_4\,{*}F_2 &=& \tilde{F}_2 - (1{-}x)\,{*}(F_2 \kern-.1em\wedge\kern-.1em \tilde{F}_2)\,{*}F_2
+ \mbox{$\frac{1}{2}$}\,(1{-}x)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2)\,{*}\tilde{F}_2 - \mbox{$\frac{1}{2}$}\,x\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em \tilde{F}_2)\,
{*}\tilde{F}_2\,, \nonumber\\
{*}F_4\,{*}\tilde{F}_2 &=& F_2 - x\,{*}(F_2 \kern-.1em\wedge\kern-.1em \tilde{F}_2)\,{*}\tilde{F}_2 +
\mbox{$\frac{1}{2}$}\,x\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em \tilde{F}_2)\,{*}F_2-\mbox{$\frac{1}{2}$}(1{-}x)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2)\,{*}F_2\,,
\end{eqnarray}
which can be derived from the $\kappa$-invariant action
\begin{eqnarray}
S &=& \int {\rm d}^4\xi\,\sqrt{-g}\,\lambda\bigg[1 + (1{-}x)\,F_2\kern-.13em\cdot\kern-.13em F_2 +
x\,\tilde{F}_2\kern-.13em\cdot\kern-.13em\tilde{F}_2
-x\,(1{-}x)\,{*}(F_2\kern-.1em\wedge\kern-.1em \tilde{F}_2)\,{*}(F_2\kern-.1em\wedge\kern-.1em\tilde{F}_2) \nonumber \\
&&\hspace{2cm} +\,\, \mbox{$\frac{1}{2}$}\,x\,(1{-}x)\,{*}(F_2\kern-.1em\wedge\kern-.1em
F_2)\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2) -\mbox{$\frac{1}{4}$}\,(1{-}x)^{2}\,{*}(F_2\kern-.1em\wedge\kern-.1em
F_2)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2) \nonumber \\ &&\hspace{2cm}-\,\,\mbox{$\frac{1}{4}$}\,x^{2}\,
{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em \tilde{F}_2)\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em \tilde{F}_2) \bigg]\,.
\label{D3xaction}
\end{eqnarray}
The value $x=0$ corresponds to the usual D3-brane action with the standard
projection operator, a result we have arrived at using only the requirements
of $\kappa$ and gauge invariance. In this limit there is no need for
the duality relations since they just define $\tilde{F}_2$ in terms $F_2$, but
$\tilde{F}_2$ appears neither in the action nor in the projection operator.
As a consequence, the duality relations can simply be dropped in this case.
Another interesting limit is $x=1$, in which the action is also of Born-Infeld
form, but with $\tilde{F}_2$ as the world-volume field. However, unlike $\tilde{F}_2$
in the limit $x=0$, $F_2$ does
not decouple completely, since the action depends implicitly on $F_2$ through
$\tilde{F}_2$ and $F_4$. The explicit dependence of $F_2$ in
$P_+$ can be removed by using the identity (\ref{D34formid}), but the implicit
dependence remains. It is nevertheless true
that the equation of motion for $A_1$ becomes non-dynamical, as can
be demonstrated in the following way.
When $x=1$, the equation of motion for $A_1$ is
\begin{equation}
{\rm d} \left[ \lambda\left\{ 2{*}F_4\,F_2- \frac{\delta {*} \Phi}{\delta \tilde{F}_2} \right\}
C_0\right] = 0 \,.
\end{equation}
Here the expression in curly brackets vanishes trivially as a consequence of
the duality relations.
The action in the limit $x=1$ differs from the dual action derived in
refs~\cite{Tseytlin:1996,Aganagic:1997}.
The reason for this discrepancy is that
the dual field of refs~\cite{Tseytlin:1996,Aganagic:1997} does not satisfy a
Bianchi identity, nor is it invariant under the Peccei-Quinn symmetry. It thus
violates two of our basic assumptions. We can reproduce the results of
refs~\cite{Tseytlin:1996,Aganagic:1997} by solving the duality relations to
eliminate $F_2$ at the level of the equations of motion and then
integrate
the result to an action; this action agrees with the one derived
earlier using conventional Poincar\'e dualisation.
Yet another interesting case is $x=\mbox{$\frac{1}{2}$}$. This is the most symmetrical choice
for the parameter $x$, and the corresponding action is, as one might suspect,
related to the manifestly SL(2,$\mathbb{Z}$)-covariant action constructed in
ref.~\cite{Cederwall:1998a}. More precisely, the action (\ref{D3xaction}) with
$x=\mbox{$\frac{1}{2}$}$ is a gauge-fixed version of the manifestly SL(2,$\mathbb{Z}$)-covariant
action constructed in ref.~\cite{Cederwall:1998a} and can easily be lifted to
the latter. In the manifestly covariant formulation the two world-volume field
strengths are combined into a single complex field,
${\mathcal F}= {\mathcal U}^r F_{2;r} $, which by construction is invariant
under SL(2,$\mathbb{Z}$) but transforms under the (local) U(1) symmetry of the
background theory.\footnote{See appendix \ref{sugraapp} for a brief
account of some relevant properties of type {I\kern-.09em I}B supergravity.}
Here $F_{2;r}={\rm d} A_{1;r}-C_{2;r}$ transforms as a doublet under
SL(2,$\mathbb{Z}$), where $C_{2;1}=B_2$, $C_{2;2}=C_2$
and similarly for $A_{1;r}$. By choosing the U(1) gauge
${\mathcal U}^{1}=-e^{\frac{1}{2}\phi} C_0+ie^{-\frac{1}{2}\phi}$
and ${\mathcal U}^2=e^{\frac{1}{2}\phi}$, one gets
${\mathcal F} = e^{\frac{1}{2}\phi}\tilde{F}_2 + i e^{-\frac{1}{2}\phi}F_2$.
Implementing this relation in the action leads to complete agreement with
the results of ref.~\cite{Cederwall:1998a} (after taking into account the
differences in conventions for $F_4$ and $R_5$). We have thus explicitly
shown in a simple way that the usual D3-brane action and the manifestly
covariant action of ref.~\cite{Cederwall:1998a} describe the same physical
object---the D3-brane---which is a singlet under SL(2,$\mathbb{Z}$).
\setcounter{equation}{0}
\section{The type {I\kern-.09em I}B 5-branes}
\label{IIB5branesect}
In type {I\kern-.09em I}B supergravity in its doubled formulation there are two six-form
gauge potentials, $C_6$ and $B_6$, which are dual to the two-form potentials
$C_2$ and $B_2$, respectively. The six-form potentials couple minimally
to two different branes: the D5-brane and the NS5-brane. Below we
discuss these two objects using the method outlined in previous sections.
We will be somewhat more elaborate in our treatment of the NS5-brane
as this is the most interesting and novel case.
\subsection*{The D5-brane}
\label{IIBD5sect}
The D5-brane couples minimally to the six-form potential $C_6$. The
standard form of the associated world-volume six-form is
\cite{Bergshoeff:1998c}
\begin{equation}
F_6 = {\rm d} A_5 - C_6 -C_4\,F_2 -
\mbox{$\frac{1}{2}$}\,C_2\,F_2\,F_2 - \mbox{$\frac{1}{6}$}\,C_0\,F_2\,F_2\,F_2\,.
\end{equation}
In addition to this tension form only $F_2$ appears in the action.
As in previous sections, this action can be
generalised by introducing parameters into the definitions of the
background field strengths. In analogy with the case of the D3-brane,
we only introduce dual pairs of world-volume fields, i.e.,
fields that are related by auxiliary duality relations. In other words,
we are dualising the two-form field
strength $F_2$ into the four-form field strength $F_4$.
In world-volume language this means that we are redefining $F_6$ as
$F_6 \rightarrow F_6 - y\,F_2\,F_4$.
We let the background fields $R_1$, $R_3$ and $R_5$ have the canonical
form $R = {\rm d} C - C\, H$, whereas $R_7$ is given by
\begin{eqnarray}
R_7 &=& {\rm d} C_6 + y\, {\rm d} C_4\, B_2 -(1{-}y)\, C_4\,H_3\,.
\end{eqnarray}
These field strengths all obey Bianchi identities of the form ${\rm d} R-R\,H=0$.
The corresponding world-volume fields are
\begin{eqnarray}
F_{2} &=& {\rm d} A_{1} - B_{2}\,, \nonumber\\
F_{4} &=& {\rm d} A_{3} - C_{4} - C_{2}\,F_{2}
-\mbox{$\frac{1}{2}$} \,C_{0}\,F_{2}\,F_{2}\,, \nonumber\\
F_{6} &=& {\rm d} A_{5} - C_{6} + y\,B_{2}\,F_{4} -
(1{-}y)\,C_{4}\,F_{2} - (\mbox{$\frac{1}{2}$}{-}y)\, C_{2}\,F_{2}\,F_{2}\,, \nonumber\\ && +
y\, B_{2}\,C_{2}\,F_{2} - \mbox{$\frac{1}{2}$} (\mbox{$\frac{1}{3}$} {-}y)\,
C_{0}\,F_{2}\,F_{2}\,F_{2}
+\mbox{$\frac{1}{2}$}\,y\,C_{0}\,B_{2}\,F_{2}\,F_{2} \,,
\end{eqnarray}
satisfying the Bianchi identities
\begin{eqnarray}
{\rm d} F_{2} &=& - H_{3}\,, \nonumber\\
{\rm d} F_{4} &=& -R_{5} - R_{3}\,F_{2} -\mbox{$\frac{1}{2}$}\, R_{1}\,F_{2}\,F_{2}\,,\nonumber \\
{\rm d} F_{6} &=& -R_7 + y\,H_3\,F_4 - (1{-}y)\,R_5\,F_2 -
(\mbox{$\frac{1}{2}$}{-}y)\,R_3\,F_2\,F_2 -
\mbox{$\frac{1}{2}$}\,(\mbox{$\frac{1}{3}$} {-}y)\,R_1\,F_2\,F_2\,F_2\,.
\end{eqnarray}
To verify that $F_4$ is an appropriate dual field in the conventional sense
one can apply the methods of refs~\cite{Tseytlin:1996,Aganagic:1997} to the
usual DBI D-brane action. One then
obtains exactly the first duality relation given in (\ref{D5dualrel}) below,
after the
identification ${*}F_6 = \sqrt{1+\Phi(F_2)}$, where
$\Phi(F_2)=\frac{\det(g+F)}{\det(g)}-1$.
Our Ansatz for the action is
\begin{equation}
S = \int \lambda\,[{*}1 + {*}\Phi(F_{2},F_{4}) + F_6\,{*}F_6]\,.
\end{equation}
The duality relations compatible with the Bianchi identities and the
equations of motion can then be shown to be
\begin{eqnarray}
\label{D5dualrel}
-2y\,{*}F_{6}\,{*}F_{2} &=& K_{4} := \frac{\delta\Phi}{\delta F_4}\,,\nonumber\\
2(1{-}y)\,{*}F_{6}\,{*}F_{4} &=& K_{2} := \frac{\delta\Phi}{\delta F_2}\,.
\end{eqnarray}
Furthermore, the variation of the constraint $\Upsilon = 1 + \Phi(F_{2},F_{4})-
({*}F_{6})^2\approx0$ is\footnote{As usual we make the assumption that $\Phi$
can be constructed from only the form fields and the metric. Moreover, the
dilaton variation is obtained using the rules
$F_2\rightarrow e^{-\frac{1}{2}\phi}F_2$, $F_4\rightarrow e^{0{\cdot}\phi}F_4$ and
$F_6\rightarrow e^{-\frac{1}{2}\phi}F_6$.}
\begin{eqnarray}
\delta_{\kappa}\Upsilon &=& \mbox{$\frac{1}{2!}$}\,K^{ij}\,\delta_{\kappa}F_{ij} +
\mbox{$\frac{1}{4!}$}\,K^{ijkl}\,\delta_{\kappa}F_{ijkl} +
\mbox{$\frac{2}{6!}$}\,F^{ijklmn}\,\delta_{\kappa}F_{ijklmn} -
\mbox{$\frac{1}{2}$}\,{K^{(i}}_{l}\,F^{j)l}\,\delta_{\kappa}g_{ij} \nonumber\\
&&- \mbox{$\frac{1}{12}$}\,{K^{(i}}_{lmn}\,F^{j)lmn}\,\delta_{\kappa}g_{ij} -
\mbox{$\frac{1}{5!}$}\,{F^{(i}}_{lmnpq}\,F^{j)lmnpq}\,\delta_{\kappa}g_{ij} -
\mbox{$\frac{1}{4}$}\,K^{ij}\,F_{ij}\,\delta_{\kappa}\phi \nonumber\\
&&- \mbox{$\frac{1}{6!}$}\,F^{ijklmn}\,F_{ijklmn}\,\delta_{\kappa}\phi\,.
\end{eqnarray}
By inserting the explicit expressions (\ref{D5dualrel}) for the $K$'s,
as well as the supergravity on-shell constraints (see appendix
\ref{sugraapp}), we obtain
\begin{eqnarray}
\label{D5variations}
(\delta_{\kappa}\Upsilon)^{(1/2)} &=& 2\,{*}F_6\,{\bar{\La}} \Big[\Xi\,J + {*}F_4\kern-.13em\cdot\kern-.13em \gamma_2\,K
- \mbox{$\frac{1}{2}$}\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2)\kern-.13em\cdot\kern-.13em \gamma_2\,J - \mbox{$\frac{1}{3}$}\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2
\kern-.1em\wedge\kern-.1em F_2)\,I \nonumber\\
&&- (1{-}y)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_4) + {*}F_6 \Big]\kappa\,, \nonumber\\
(\delta_{\kappa}\Upsilon)^{(0)} &=& 4i\,{*}F_6\,{\bar{E}}_i \Big[({*}\gamma_5)^i\,I +
\mbox{$\frac{1}{3!}$}\,{*}F_2^{ijkl} \gamma_{jkl}\,I + {*}F_4^{ij}\gamma_{j}\,K +
\mbox{$\frac{1}{2}$}\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2)^{ij}\gamma_j\,J \nonumber\\
&&+ y\, g^{ij}\,{*}(F_2\kern-.1em\wedge\kern-.1em F_4)\gamma_{j} + {*}{F_4^{(i}}_{l}F_2^{j)l}\gamma_j +
g^{ij}\,{*}F_6\,\gamma_j \Big]\kappa\,.
\end{eqnarray}
In analogy with the previously considered cases, the calculation proceeds by
inserting the projected spinor-parameter $\kappa=P_+\zeta$ using an appropriate
Ansatz for the projection operator and examining the irreducible components
of the expression obtained by application of the $\gamma$-matrix product identity
(\ref{general}); for details we refer to appendix \ref{kappadetails} quoting
here only the results.
The projection operator is found to be
\begin{eqnarray}
\label{D5projop}
P_\pm &\propto& - {*}(F_6- \alpha\,F_2\kern-.1em\wedge\kern-.1em F_4)\,\hbox{1\kern-.27em l} \pm \big[\, \Xi\,J +
(1{-}(y{+}\alpha))\,{*}F_2\kern-.13em\cdot\kern-.13em \gamma_4\,I +
(y{+}\alpha)\,{*}F_4\kern-.13em\cdot\kern-.13em \gamma_2\,K \nonumber\\ &&
+\,\, (\mbox{$\frac{1}{2}$}{-}(y{+}\alpha))\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2)\kern-.13em\cdot\kern-.13em \gamma_2\,J
+ \mbox{$\frac{1}{2}$}\,(\mbox{$\frac{1}{3}$} {-}(y{+}\alpha))\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2 \kern-.1em\wedge\kern-.1em F_2)\,I\,\big]
\,.
\end{eqnarray}
Here we note, in particular, that the correspondence with the tension field
strength holds also in this case. Moreover, the explanation for the free
parameter $\alpha$ is the same as in the D3-brane case.
The final expression for the action is
\begin{eqnarray}
\label{D5action}
S &=& \int{\rm d}^6\xi\,\sqrt{-g}\,\lambda\,\bigg[ 1 + (1{-}y)\,F_2\kern-.13em\cdot\kern-.13em F_2 +
y\,F_4\kern-.13em\cdot\kern-.13em F_4 - y\,(1{-}y)\,{*}(F_2 \kern-.1em\wedge\kern-.1em F_4)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_4) \nonumber\\
&& \qquad + \mbox{$\frac{1}{2}$}\, y\,(F_2\kern-.1em\wedge\kern-.1em
F_2)\kern-.13em\cdot\kern-.13em({*}F_4\kern-.1em\wedge\kern-.1em \,{*}F_4) + \mbox{$\frac{1}{2}$}(\mbox{$\frac{1}{2}$}{-}y)\,(F_2 \kern-.1em\wedge\kern-.1em F_2)\kern-.13em\cdot\kern-.13em
(F_2\kern-.1em\wedge\kern-.1em F_2) \nonumber\\
&& \qquad - \mbox{$\frac{1}{12}$}(\mbox{$\frac{1}{3}$} {-}y)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2 \kern-.1em\wedge\kern-.1em
F_2)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2\kern-.1em\wedge\kern-.1em F_2)- ({*}F_6)^2\bigg]\,,
\end{eqnarray}
which is to be supplemented with the duality relations
\begin{eqnarray}
\label{D5dualrels}
-{*}F_6\,{*}F_2 &=& F_4 - (1{-}y)\,{*}(F_2\kern-.1em\wedge\kern-.1em F_4)\,{*}F_2 + \mbox{$\frac{1}{2}$}\,{*}(F_2\kern-.1em\wedge\kern-.1em
F_2)\kern-.1em\wedge\kern-.1em\,{*}F_4\,, \nonumber\\
{*}F_6\,{*}F_4 &=& F_2 - y\,{*}(F_2\kern-.1em\wedge\kern-.1em F_4)\,{*}F_4
- \mbox{$\frac{1}{4}$}\,{*}[{*}(F_2\kern-.1em\wedge\kern-.1em F_2+{*}F_4\kern-.1em\wedge\kern-.1em{*}F_4)\kern-.1em\wedge\kern-.1em F_2]\nonumber \\
&& -\mbox{$\frac{1}{4}$}\mbox{$\frac{1-3y}{1-y}$}\,{*}\bigg\{
{*}[F_2\kern-.1em\wedge\kern-.1em F_2 -{*}F_4\kern-.1em\wedge\kern-.1em{*}F_4 - \mbox{$\frac{1}{3}$}\,{*}(F_2\kern-.1em\wedge\kern-.1em F_2\kern-.1em\wedge\kern-.1em
F_2)\,{*}F_2] \kern-.1em\wedge\kern-.1em F_2 \bigg\} \,.
\end{eqnarray}
In the last line of the second relation we have extracted a linear combination
of terms which can be shown to vanish identically when both duality relations
are satisfied (see appendix~\ref{kappadetails},
in particular eq.~(\ref{D5id})).
Hence, in applications of the D5-brane formulation under discussion
one may simply drop these terms and use the simplified expression.
As expected, we recover the usual D5-brane action for $y=0$.
In the opposite limit, $y=1$, we get a dual (but equivalent) description.
However, $F_2$ does not decouple completely from the action in this limit;
unlike the situation for the D3-brane, in addition to the implicit dependence,
$F_2$ also appears explicitly in the action.
In order to fully eliminate $F_2$ one needs to solve an algebraic
equation similar to the one encountered in ref.~\cite{Aganagic:1997}.
The difference compared to earlier approaches is that we have the additional
information of knowing the equations of motion and the action of the dual
theory, so we can in principle determine the dynamics of the
three-form potential.
Another interesting limit appears to be $y=\mbox{$\frac{1}{3}$} $, where the
action becomes quartic. One would like to relate the action for the
D5-brane given above to an action for the NS5-brane; we will return to
this issue once we have constructed an action for the NS5-brane.
\subsection*{The NS5-brane}
It has been known for several years \cite{Callan:1991} that the world-volume
field theory of the type {I\kern-.09em I}B NS5-brane is described (on-shell) by a
six-dimensional vector multiplet.
However, there has been a debate in the literature about whether the
action should be formulated in terms of a two-form field strength
(as for the D5-brane) or in terms of a four-form field strength (in keeping
with the prescription of accompanying a target-space S-duality transformation
with a world-volume Poincar\'e duality transformation); for discussions see
e.g.\ refs~\cite{Bergshoeff:1996a,Hull:1997}. It is
furthermore known that the D5-, D3- and D1-branes can end on the NS5-brane
(see e.g.\ ref.~\cite{Argurio:1997}). From this perspective one expects a
six-form, a four-form and a two-form to be present in the world-volume theory
of the NS5-brane. The six-form in question is different from the one
describing the tension and the need for it has been discussed in
ref.~\cite{Bergshoeff:1998a}.
Below we will construct an action whose world-volume field content reflects
the above mentioned facts.
First we need to choose a proper parameterisation of the background
field strengths. Since it is known that the fundamental string cannot
end on the NS5-brane, we use as a guiding principle the condition that
$F_2$ should not appear (not even implicitly) in the world-volume
action. This restriction leads to the following expressions for the RR
background field strengths:
\begin{eqnarray}
R_1 &=& {\rm d} C_0 \,, \nonumber\\
R_3 &=& {\rm d} C_2 + B_2\, {\rm d} C_0 \,,\nonumber\\
R_5 &=& {\rm d} C_4 + B_2\,{\rm d} C_2 +\mbox{$\frac{1}{2}$}\, B_2\, B_2 \,{\rm d} C_0\,, \nonumber\\
R_7 &=& {\rm d} C_6 + B_2 \,{\rm d} C_4 + \mbox{$\frac{1}{2}$}\, B_2\, B_2 \,{\rm d} C_2 +
\mbox{$\frac{1}{6}$}\, B_2\, B_2 \, B_2\, {\rm d} C_0 \,.
\end{eqnarray}
The corresponding world-volume field strengths become\footnote{Here,
as in the case of the D-string,
${\rm d} A_{-1}$ formally denotes an exact zero-form, i.e.\ a constant.}
\begin{eqnarray}
F_0 &=& {\rm d} A_{-1} -C_0\,, \nonumber\\
\tilde{F}_2 &=& {\rm d} \tilde{A}_1 - C_2 + F_0\, B_2 \,,\nonumber\\
F_4 &=& {\rm d} A_3 - C_4 + B_2\,\tilde{F}_2 - \mbox{$\frac{1}{2}$}\, B_2\, B_2\, F_0 \,,\nonumber\\
F_6 &=& {\rm d} A_5 - C_6 + B_2\, F_4 - \mbox{$\frac{1}{2}$}\, B_2\, B_2\, \tilde{F}_2
+ \mbox{$\frac{1}{6}$}\,B_2\, B_2\, B_2\, F_0 \,.
\end{eqnarray}
These relations can be succinctly written as $R = e^{B}\,{\rm d} C$ and $e^{-B}F =
{\rm d} A - C$,
where the last definition is iterative and $F = F_0 + \tilde{F}_2 + F_4 + F_6$.
In the same compact notation the Bianchi identities are
\begin{equation}
{\rm d} F = - R + H_3\, F\,.
\label{NS5wvBIs}
\end{equation}
The gauge-invariant field strength of the NS-NS six-form gauge potential
is chosen as
\begin{eqnarray}
H_7 &=& {\rm d} B_6 - (1{-}y)\,C_6\,{\rm d} C_0 + y\,C_0\,{\rm d} C_6 - (1{-}x)\,C_2\, {\rm d} C_4
+ x\, C_4\,{\rm d} C_2 \,,\nonumber\\
{\rm d} H_7 &=& R_7\,R_1 - R_5 \,R_3\,.
\end{eqnarray}
The associated tension form is then found to be
\begin{eqnarray}
\tilde{F}_6 &=& {\rm d} \tilde{A}_5 - B_6 - (1{-}y)\,C_6\,F_0 + y\,C_0\,F_6 - (1{-}x)\,C_2\,F_4
+ x\,C_4\,\tilde{F}_2 + (1{-}x{-}y)\,B_2\,F_0\,F_4 \nonumber\\
&&+ (\mbox{$\frac{1}{2}$}{-}x)\,B_2\,\tilde{F}_2\,\tilde{F}_2 +
(1{-}x)\,B_2\,C_2\,\tilde{F}_2 - x\,B_2\,C_4\,F_0 - y\,B_2\,C_0\,F_4 \nonumber\\
&&-(1{-}\mbox{$\frac{3}{2}$}x{-}\mbox{$\frac{1}{2}$} y)\,B_2\,B_2\,F_0\,\tilde{F}_2 -
\mbox{$\frac{1}{2}$}(1{-}x)\,B_2\,B_2\,C_2\,F_0 + \mbox{$\frac{1}{2}$}\,y\,B_2\,B_2\,C_0\,\tilde{F}_2 \nonumber\\
&& + (\mbox{$\frac{1}{3}$} {-}\mbox{$\frac{1}{2}$} x{-}\mbox{$\frac{1}{6}$} y)
\,B_2\,B_2\,B_2\,F_0^2 -\mbox{$\frac{1}{6}$}\,y\,B_2\,B_2\,B_2\,C_0\,F_0\,,
\end{eqnarray}
with Bianchi identity
\begin{eqnarray}
{\rm d} \tilde{F}_6 &=& -H_7 -(1{-}y)\,R_7\,F_0 + y\,R_1\,F_6 - (1{-}x)\,R_3\,F_4
+ x\,R_5\,\tilde{F}_2
\nonumber\\ &&+ (1{-}x{-}y)\,H_3 \,F_0\, F_4 + (\mbox{$\frac{1}{2}$}{-}x)\,H_3\,\tilde{F}_2\,\tilde{F}_2\,.
\label{NS5tF6BI}
\end{eqnarray}
The fact that one has a non-dynamical six-form in addition to the
tension form makes the situation analogous to the D-string
discussion, where two field strengths of maximal degree are present
together with a non-dynamical scalar $F_0$.
Given the above field content, the Ansatz for the action takes the by now
familiar form\footnote{Again, we suppress the dilaton dependence. When
deriving the $\kappa$-variation it is temporarily reinstated by means of
the rules $F_{2k}\rightarrow e^{(1-\frac{1}{2}k)\phi}F_{2k}$ (here $F_2=\tilde{F}_2$) and
$\tilde{F}_6\rightarrow e^{\frac{1}{2}\phi}\tilde{F}_6$.}
\begin{equation}
S= \int \lambda \left[{*}1+{*}\Phi(F_0,\tilde{F}_2,F_4,F_6)+\tilde{F}_6\,{*}\tilde{F}_6\right]\,.
\label{NS5ansatz}
\end{equation}
Compatibility between the equations of motion and the Bianchi identities
leads to the following expressions for the duality relations:
\begin{eqnarray}
\label{NS5dualrels}
2y\,{*}\tilde{F}_6\,F_0 &=& {*}K_6 := \frac{\delta{*}\Phi}{\delta F_6}\,,\nonumber\\
-2(1{-}x)\,{*}\tilde{F}_6\,F_2&=&{*}K_4:=\frac{\delta{*}\Phi}{\delta F_4}\,,\nonumber\\
2x\,{*}\tilde{F}_6\,F_4&=&{*}K_2:=\frac{\delta{*}\Phi}{\delta F_2}\,,\nonumber\\
-2(1{-}y)\,{*}\tilde{F}_6\,F_6&=&{*}K_0:=\frac{\delta{*}\Phi}{\delta F_0}\,.
\end{eqnarray}
The derivation of the last relation is at this point formal, but
will be justified by the final result.
The total $\kappa$-variation of the constraint
$\Upsilon = 1 + \Phi(F_0,\tilde{F}_2,F_4,F_6) - ({*}\tilde{F}_6)^2 \approx 0$ is
\begin{eqnarray}
\delta_{\kappa} \Upsilon &=& K_0\,\delta_\kappa F_0 +
\mbox{$\frac{1}{2!}$}\tilde{K}^{ij}\,\delta_{\kappa}\tilde{F}_{ij} +
\mbox{$\frac{1}{4!}$}K^{ijkl}\,\delta_{\kappa}F_{ijkl} +
\mbox{$\frac{1}{6!}$}K^{ijklmn}\,\delta_{\kappa}F_{ijklmn} +
\mbox{$\frac{2}{6!}$}\tilde{F}^{ijklmn}\,\delta_{\kappa}\tilde{F}_{ijklmn} \nonumber\\ &&- \big[
\mbox{$\frac{1}{2!}$}{{\tilde{K}}^{(i}}{}_{l}\,\tilde{F}^{j)l} +
\mbox{$\frac{2}{4!}$}{K^{(i}}_{lmn}\,F^{j)lmn} +
\mbox{$\frac{3}{6!}$}{K^{(i}}_{lmnpq}\,F^{j)lmnpq} \big]\,\delta_{\kappa}g_{ij} +
\mbox{$\frac{1}{2!}$}\big[ K_0\,F_0 + \mbox{$\frac{1}{2}$}\,\tilde{K}^{ij}\,\tilde{F}_{ij} \nonumber\\
&&- \mbox{$\frac{1}{2}$}\,K^{ij}\,F_{ij} -
\mbox{$\frac{1}{2}$}\kern-.13em\cdot\kern-.13em \mbox{$\frac{1}{6!}$}\,K^{ijklmn}\,F_{ijklmn} +
\mbox{$\frac{1}{2}$}\kern-.13em\cdot\kern-.13em \mbox{$\frac{2}{6!}$}\,\tilde{F}^{ijklmn}\,\tilde{F}_{ijklmn} \big]\,
\delta_{\kappa}\phi\,.
\end{eqnarray}
Inserting the duality relations (\ref{NS5dualrels}) and the
expressions for the variations of the world-volume fields with the
background constraints imposed leads to
\begin{eqnarray}
\label{NS5variations}
(\delta_{\kappa}\Upsilon)^{(1/2)} &=& 2\,{*}\tilde{F}_6\,{\bar{\La}} \Big[ \Xi\, K + F_0\,\Xi\, J
- {*}F_4\kern-.13em\cdot\kern-.13em \gamma_{2}\,J + [F_0\,{*}F_4-\mbox{$\frac{1}{2}$}{*}(F_2 \kern-.1em\wedge\kern-.1em F_2)]
\kern-.13em\cdot\kern-.13em \gamma_{2}\,K \nonumber\\ &&
+ 2 \,{*}F_6\,I + (2{-}y)F_0\,{*}F_6 - x \,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em F_4) -
\,{*}\tilde{F}_6 \Big]\kappa\,, \nonumber\\
(\delta_{\kappa}\Upsilon)^{(0)} &=& 4i \,{*}\tilde{F}_6\,{\bar{E}}_i \Big[
-({*}\gamma_5)^i\,K + F_0\,({*}\gamma_5)^i\,J
- \mbox{$\frac{1}{3!}$}\,{*}\tilde{F}_{2}^{ijkl}\,\gamma_{jkl}\,I +
\,{*}F_{4}^{ij}\,\gamma_j\, J \nonumber\\ && +
[F_0\,{*}F_4^{ij}- \mbox{$\frac{1}{2}$}\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em \tilde{F}_2)^{ij}]\,\gamma_j\,K
+ \big[{-}(1{-}x)\,g^{ij}\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em F_4) \nonumber\\ &&
+ \,{*}{F_4^{(i}}_l\, \tilde{F}_2^{j)l}
+ y\,g^{ij}\,F_0\,{*}F_6+ g^{ij}\,{*}\tilde{F}_6\big]\gamma_j \Big]\kappa\,.
\end{eqnarray}
After a rather long and in parts somewhat intricate calculation (outlined
in appendix \ref{kappadetails}) it is possible to show that the above
variations vanish when $\kappa=P_+\zeta$ and the projection operator is
\begin{eqnarray}
P_\pm &\propto& \pm \,\Big[\Xi\,K + (\mbox{$\frac{1}{2}$}{+}3\alpha)\,F_0\,\Xi\,J +
(\mbox{$\frac{1}{2}$}{+}\alpha)\,{*}\tilde{F}_2\kern-.13em\cdot\kern-.13em \gamma_4\,I - (\mbox{$\frac{1}{2}$}{-}\alpha) \,{*}F_4\kern-.13em\cdot\kern-.13em \gamma_2\,J
\nonumber\\ && -\,\, 2\,\alpha\, [F_0\,{*}F_4
-\mbox{$\frac{1}{2}$}\,{*}(\tilde{F}_2 \kern-.1em\wedge\kern-.1em \tilde{F}_2)]\kern-.13em\cdot\kern-.13em \gamma_2\,K + (\mbox{$\frac{1}{2}$}{-}3\,\alpha)\,{*}F_6\,I\Big]
\nonumber \\
&&+\,\, {*}\big[\tilde{F}_6-
(\mbox{$\frac{1}{2}$}{-}3\,\alpha{-}y)\,F_0\,F_6-(\mbox{$\frac{1}{2}$}{+}\alpha{-}x)\,\tilde{F}_2\kern-.1em\wedge\kern-.1em F_4\big]\,\hbox{1\kern-.27em l}\,,
\end{eqnarray}
where $\alpha$ is arbitrary.
Moreover, the action is given by (\ref{NS5ansatz}) with
\begin{eqnarray}
\label{NS5Phi}
{*}\Phi &=& (1{-}y)\,F_0\,{*}F_0 + x\,\tilde{F}_2\kern-.1em\wedge\kern-.1em{*}\tilde{F}_2 + (1{-}x)\,F_4\kern-.1em\wedge\kern-.1em{*}F_4
+y\,F_6\,{*}F_6 \nonumber\\ && - y\,(1{-}y)\,({*}F_6)^2\,F_0\,{*}F_0
+ x\,(1{-}x)\,(\tilde{F}_2\kern-.1em\wedge\kern-.1em F_4)\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em F_4)
\nonumber\\ &&
+ \mbox{$\frac{1}{2}$}\,(1{-}x)\,{*}({*}F_4\kern-.1em\wedge\kern-.1em{*}F_4)\kern-.1em\wedge\kern-.1em\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2
+ \mbox{$\frac{1}{2}$}\,(x{-}\mbox{$\frac{1}{2}$})\,{*}(\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2)\kern-.1em\wedge\kern-.1em\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2
\nonumber\\&& +
\mbox{$\frac{1}{3}$}\,y\,{*}F_6\,\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2
-\mbox{$\frac{1}{2}$}\,(x{-}y)
\,F_0\,{*}F_4\kern-.1em\wedge\kern-.1em\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2 + (1{-}x{-}y)\,(F_0)^2\,F_4\kern-.1em\wedge\kern-.1em{*}F_4 \nonumber\\
&&-\mbox{$\frac{1}{6}$}\,(2{-}3x{+}y)\,F_0\,{*}({*}F_4\kern-.1em\wedge\kern-.1em{*}F_4)\kern-.1em\wedge\kern-.1em F_4 -
2\,x\,y\,F_0\,{*}F_6\,\tilde{F}_2\kern-.1em\wedge\kern-.1em F_4 \nonumber \\
&& - \mbox{$\frac{1}{6}$}\,(2{-}3x{-}y)\left[{*}(F_0\,F_4 -
\mbox{$\frac{1}{2}$}\tilde{F}_2\kern-.1em\wedge\kern-.1em\tilde{F}_2)\right]^{3} \,.
\end{eqnarray}
Although this expression looks rather intimidating, it
simplifies somewhat in certain limits of parameter space.
The duality relations supplementing the action are readily obtained by
inserting the appropriate functional derivatives of this expression in
the equations~(\ref{NS5dualrels}); equivalent expressions are
listed in appendix~\ref{kappadetails}. It is also worth noting that
one has the freedom of adding to the action expressions whose variations
vanish as a consequence of the duality constraints, e.g.\ quadratic
combinations of the constraints given in eq.~(\ref{NS5idI}).
Finally, let us comment on the relation of our formulation of
the NS5-brane to the one obtained in ref.~\cite{Eyras:1998}, and to
the formulation of the D5-brane above. By inspection
of (\ref{NS5Phi}) one notices that there is no way to consistently eliminate
the four-form field
strength at the level of the action. In ref.~\cite{Eyras:1998}, T-duality
considerations led to an action for the NS5-brane which
is related to the D5-brane by a simple S-duality transformation of
the supergravity background, without the need for any Poincar\'e-duality
transformations in the world-volume. However, as for the dual D3-brane action
of refs~\cite{Tseytlin:1996,Aganagic:1997}, the world-volume field strength
of ref.~\cite{Eyras:1998} does not satisfy a Bianchi identity, nor is it
invariant under $C_0 \rightarrow C_0 + 1$; this is the primary
source for the discrepancy. However, if both actions are to describe the same
object, it must be possible to obtain the results of ref.~\cite{Eyras:1998}
from ours at the level of the equations of motion. In order to
accomplish this one must eliminate $F_4$ and $F_6$ from the
equation of motion for $\tilde{A}_1$. As a first step in this direction one
has to choose $x=1$ and $y=0$. It turns out to be rather involved (if at all
possible) to algebraically eliminate $F_4$, and we therefore temporarily
limit our considerations to backgrounds for which $C_0=0$ and
set $\mu=0$ so that $F_0=0$. Under these simplifying assumptions we
find agreement with the results of ref.~\cite{Eyras:1998}.
It would be of interest to investigate whether this is still the case when
$C_0\neq0$.
Furthermore, implementing the restriction $F_0=0$ into the expression
for $\Phi$ leads to complete agreement with the action for the D5-brane in
eq.~(\ref{D5action}), provided we let $x \rightarrow 1-y$, $\tilde{F}_2 \rightarrow -F_2$,
$F^{\rm NS}_4 \rightarrow F^{\rm D5}_4$ and $\tilde{F}_6 \rightarrow F_6$.
These transformations follow from making an SL(2,$\mathbb{Z}$) transformation of the
background; under SL(2,$\mathbb{Z}$) the combinations $(B_2, C_2)$, $(A_1, \tilde{A}_1)$,
$(B_6, C_6)$ and $(\tilde{A}_5, A_5)$ transform as doublets whereas
$C_4$ and $A_3$ are singlets.
In particular, for the SL(2,$\mathbb{Z}$) transformation
\begin{equation}
\label{S}
S =\left(\begin{array}{cc} 0 & 1 \\ -1 & 0 \end{array} \right)\,,
\end{equation}
the field strengths have the required transformation properties.
Reinstating the dilaton factors one finds that these also behave
correctly under the transformation $e^{\phi} \rightarrow e^{-\phi}$ associated
with~(\ref{S}).
\setcounter{equation}{0}
\section{Discussion}
We would like to comment briefly on some of the cases we have not
treated in this paper. For the D4-brane in the type {I\kern-.09em I}A theory, the
relevant world-volume field strengths are $F_2$ and $F_3$; since there
is no possible dual to $F_1$ we choose the parameters so that it does
not appear. The tension form is given by
\begin{equation}
F_5 = {\rm d} A_4 - C_5 + x\,B_2\,F_3 - (1{-}x)\,C_3 \, F_2 + x\,C_1\,B_2\,F_2 +
(x{-}\mbox{$\frac{1}{2}$})\,C_1\,F_2\,F_2 \,,
\end{equation}
and satisfies the Bianchi identity ${\rm d} F_5 = R_6 - x\,H_3\, F_3 -
(1{-}x)\,R_4 \, F_2 + (x{-}\mbox{$\frac{1}{2}$})\,R_2\,F_2\,F_2 = 0$. The duality relations
take the same canonical form as for the other D-branes. For $x=0$ we
recover the usual D4-brane description in terms of $F_2 = {\rm d} A_1 -
B_2$, whereas for $x=1$ we obtain a dual description in terms of
$F_3={\rm d} A_2 - C_3 -C_1\,F_2$. In the dual case the action can be
related to the one given in ref.~\cite{Aganagic:1997} by completely
eliminating $F_2$ at the expense of losing manifest gauge
invariance. For the symmetric choice $x=\mbox{$\frac{1}{2}$}$, we obtain an action
which can also be obtained from the M5-brane action given in
ref.~\cite{Cederwall:1998b} by double dimensional reduction. For the type
{I\kern-.09em I}A NS5-brane the relevant world-volume fields are $F_1$, $F_3$ and
$F_5$, while $F_2$ cannot have a dual and is thus not introduced; this
meshes nicely with the fact that the fundamental string
cannot end on the NS5-brane. The field strength $F_3$ is self-dual,
whereas there is a parameter controlling the dualisation $F_1
\leftrightarrow F_5$. In particular, for a certain choice of the
parameter the action can be related to the direct dimensional
reduction of the action in ref.~\cite{Cederwall:1998b}. By applying the
methods developed in this
paper to the D6-brane, it may be possible---for certain values of the
parameters---to lift the solution to eleven dimensions. Since it is
known that the D6-brane is obtained from the $D{=}11$
Kaluza--Klein monopole, one would in this way
obtain an action for the latter object. The known
action~\cite{Bergshoeff:1997b} for the KK-monopoles gives the standard
D6-brane action upon direct dimensional reduction.
If it is possible to lift the D6-brane action for other choices of the
parameters one would presumably obtain other formulations of the
action for the KK-monopole in $D=11$. A similar situation holds
for the {I\kern-.09em I}A D8-brane, where again it may be possible to lift the solution
to eleven dimensions for certain values of the parameters and thus make
contact with work on the M9-brane~\cite{Bergshoeff:1998d}.
In order to treat the last two cases it may be necessary to introduce more
general parameters than those considered in this paper. It would also
be of interest to extend the formalism to include the $D=10$
KK monopoles. Another issue involves the extension to massive branes
\cite{Bergshoeff:1997c}. We have seen that there is a correspondence
between the possible brane-ending-on-another-brane configurations and
the restriction to only introducing world-volume fields which have
duals. Invoking this restriction there seems to be a world-volume field
for every possible brane-ending-on-another-brane case. Conversely,
every world-volume field (except for the tension form) can ``arise'' from
the end of a brane. In addition, one also has the configurations which
arise from dualisations of the background.
For the type {I\kern-.09em I}B branes an interesting outstanding problem concerns
the construction of a manifestly SL(2,$\mathbb{Z}$)-covariant action for the type
{I\kern-.09em I}B $(p,q)$ five-branes. However, the problem of constructing
such an action turns out to be significantly more involved than the
cases treated in this section; nevertheless some progress can be made
\cite{WW}.
There exist formulations of the ten-dimensional $N=2$ supergravity theories
where all the usual bosonic fields (except for the metric) are
``doubled,'' i.e., supplemented with their Poincar\'e duals; see
e.g.\ refs~\cite{Cremmer:1998,Dall'Agata:1998}. In particular,
the dilaton is supplemented by a nine-form field strength. For each
field allowed in the doubled formulations there
appears to be an associated brane. The question therefore arises if there
exist branes which couple to the dual of the dilaton (``NS 7-branes'').
In Hull's general brane scan~\cite{Hull:1997},
which is based on considerations of the supersymmetry algebra of the
background superspace, there appears to be no place for such objects.
However, in the type {I\kern-.09em I}B theory the D7-brane has to transform under
SL(2,$\mathbb{Z}$), since the potential to which it couples does. In particular,
it is possible to transform it into a brane which couples to the dual of the
dilaton. It is furthermore known~\cite{Meessen:1998} that in a manifestly
SL(2,$\mathbb{Z}$)-covariant formulation there must appear a triplet of seven-branes
coupling to the dual eight-form potentials of the three scalars which belong
to the SL(2,$\mathbb{R}$)/U(1) coset. After gauge fixing there remain two scalars
and two dual eight-form potentials. The branes coupling to these
eight-form potentials are the usual D7-brane and an NS7-brane.
In the type {I\kern-.09em I}A theory, on the other hand, there is only one eight-form
potential, which in this case should couple to an NS7-brane.
This object appears to be different from the KK-type seven-brane of
the {I\kern-.09em I}A theory discussed in ref.~\cite{Eyras:1998b}.
It would be interesting to see how (or indeed whether) it
fits into the U-duality structure of M-theory
(see e.g.\ ref.~\cite{Obers:1998}).
It would also be desirable to have a more uniform description for the various
branes within the framework of this paper, along the lines of those available
for the D-branes both in their original formulation~\cite{Cederwall:1996b,
Cederwall:1996c,Aganagic:1996,Bergshoeff:1996c}
and in the one with dynamically generated
tension~\cite{Bergshoeff:1998c}. It should also be possible to
construct T-duality rules which relate
actions of the general form considered in this paper. Presumably it is
possible to choose the
parameters so that they are left inert under T-dualisation, as
suggested by the similarity between the D1- and D2-brane actions presented
earlier. A more thorough investigation of the correspondence between
the Bianchi identity for $F_{p+1}$ and $P_+$ which seems to hold in all cases,
might lead to a deeper understanding of $\kappa$-symmetry.
Our actions should also find applications in the study of world-volume solitons
\cite{Gibbons:1997,Callan:1997,Gutowski:1998}.
\subsection*{Acknowledgements}
The work of A.W. and N.W. was supported by the European Commission under
contracts FMBICT972021 and FMBICT983302, respectively.
|
\section{Introduction}
\defn_{\mathrm gas}{n_{\mathrm gas}}
\defn_{\mathrm gas}^{\mathrm in}{n_{\mathrm gas}^{\mathrm in}}
\defn_{\mathrm gas}^{\mathrm out}{n_{\mathrm gas}^{\mathrm out}}
\defn_{\mathrm liquid}{n_{\mathrm liquid}}
\defn_{\mathrm liquid}^{\mathrm in}{n_{\mathrm liquid}^{\mathrm in}}
\defn_{\mathrm liquid}^{\mathrm out}{n_{\mathrm liquid}^{\mathrm out}}
\defn_{\mathrm in}{n_{\mathrm in}}
\defn_{\mathrm out}{n_{\mathrm out}}
\defn_{\mathrm inside}{n_{\mathrm inside}}
\defn_{\mathrm outside}{n_{\mathrm outside}}
\def {\cal N}_{\mathrm in}{ {\cal N}_{\mathrm in}}
\def {\cal N}_{\mathrm out}{ {\cal N}_{\mathrm out}}
\def\hbox{max}{\hbox{max}}
\def\hbox{min}{\hbox{min}}
\def{\mathrm in}{{\mathrm in}}
\def{\mathrm out}{{\mathrm out}}
\def\mathop{\hbox{sinc}}{\mathop{\hbox{sinc}}}
\def{\textstyle{1\over2}}{{\textstyle{1\over2}}}
\def{\textstyle{1\over4}}{{\textstyle{1\over4}}}
\def\omega_\in{\omega_{\mathrm in}}
\def\omega_\out{\omega_{\mathrm out}}
\def {\mathrm observed} { {\mathrm observed} }
Sonoluminescence (SL) is the phenomenon of light emission by a
sound-driven gas bubble in fluid \cite{Physics-Reports}. In SL
experiments the intensity of a standing sound wave is increased until
the pulsations of a bubble of gas trapped at a velocity node have
sufficient amplitude to emit brief flashes of light having a
``quasi-thermal'' spectrum with a ``temperature'' of several tens of
thousands of Kelvin. The basic mechanism of light production in this
phenomenon is still highly controversial. We first present a brief
summary of the main experimental data (as currently understood) and
their sensitivities to external and internal conditions. For a more
detailed discussion see~\cite{Physics-Reports}.
SL experiments usually deal with bubbles of air in water, with
ambient radius $R_{\mathrm ambient} \approx 4.5 \; \mu {\rm m}$.
The bubble is driven by a sound wave of frequency of 20--30 kHz.
(Audible frequencies can also be used, at the cost of inducing
deafness in the experimental staff.) During the expansion phase,
the bubble radius reaches a maximum of order $R_{\mathrm max}\approx
\; 45 \;\mu {\rm m}$, followed by a rapid collapse down to a minimum
radius of order $R_{\mathrm min}\approx 0.5 \; \mu {\rm m}$. The
photons emitted by such a pulsating bubble have typical wavelengths of
the order of visible light. The minimum observed wavelengths range
between $200\; {\rm nm}$ and $100\; {\rm nm}$. This light appears
distributed with a broad-band spectrum. (No resonance lines, roughly a
power-law spectrum with exponent depending on the noble gas admixture
entrained in the bubble, and with a cutoff in the extreme
ultraviolet.) For a typical example, see figures~\ref{F:experiment-1}
and~\ref{F:experiment-2}. If one fits the data to a Planck black-body
spectrum the corresponding temperature is several tens of thousands of
Kelvin (typically $70,000\; {\rm K}$, though estimates varying from
$40,000 \; {\rm K}$ to $100,000 \; {\rm K}$ are common). There is
considerable doubt as to whether or not this temperature parameter
corresponds to any real physical temperature. There are about one
million photons emitted per flash, and the time-averaged total power
emitted is between $30$ and $100\;
\hbox{ mW}$.
The photons appear to be emitted by a very tiny spatio-temporal
region: Estimated flash widths vary from less than $35 \; {\rm ps}$ to
more than $380 \; {\rm ps}$ depending on the gas entrained in the
bubble \cite{Flash1,Flash2}. There are model-dependent (and
controversial) claims that the emission times and flash widths do not
depend on wavelength \cite{Flash2}. As for the spatial scale, there
are various model-dependent estimates but no direct measurement is
available \cite{Flash2}. Though it is clear that there is a frequency
cutoff at about $1 \; {\rm PHz}$ the physics behind this cutoff is
controversial. Standard explanations are (1) a Thermal cutoff
(deprecated because the observed cutoff is much sharper than
exponential) , or (2) the opacity of water in the UV (deprecated
because of the observed absence of dissociation effects).
Alternatively, Schwinger suggests that the critical issue is that (3)
the real part of the refractive index of water goes to unity in the UV
(so that there is no change in the Casimir energy during bubble
collapse). We shall add another possible contribution to the mix: (4)
a rapidly changing refractive index causes photon production with an
``adiabatic cutoff' that depends on the timescale over which the
refractive index changes. (Because the observed falloff above the
physical cutoff is super-exponential it is clear that this adiabatic
effect is at most part of the complete picture.)
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=experimental-spectrum-1.eps,height=10cm}\hfil}
\caption{%
Typical experimental spectrum: The data has been extracted from
figure 51 of reference 1, and has here been plotted as intensity
(arbitrary units) as a function of wavelength. Note that no data
has been taken at frequencies below the visible range. The spectrum
is a broad-band spectrum without significant structure. The physical
nature of the cutoff (which occurs in the far ultraviolet) is one
of the key issues under debate.}
\label{F:experiment-1}
\end{figure}
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=experimental-spectrum-2.eps,height=10cm}\hfil}
\caption{%
Typical experimental spectrum: The data has been extracted from
figure 51 of reference 1, and has here been plotted as
a number spectrum as a function of frequency.}
\label{F:experiment-2}
\end{figure}
Any truly successful theory of SL must also explain a whole series of
characteristic sensitivities to different external and internal
conditions. Among these dependencies the main one is surely the
mysterious catalytic role of noble gas admixtures. (Most often a few
percent in the air entrained in the bubble. One can obtain SL from air
bubbles with a $1\%$ content of argon, and also from pure noble gas
bubbles, but the phenomenon is practically absent in pure oxygen
bubbles.) In fact, it has been suggested that physical processes
concentrate the noble gasses inside the bubble to the extent that the
bubble consists of almost pure noble gas~\cite{Argon}, and some
experimental results seem to corroborate this
suggestion~\cite{Matula}.
Other external conditions that influence SL are: (1) Magnetic Fields
--- If the frequency of the driving sound wave is kept fixed, SL
disappears above a pressure-dependent threshold magnetic field: $H
\geq H_{0}(p)$. On the other hand, for a fixed value of the magnetic
field $H_{0}$, there are both upper and lower bounds on the applied
pressure that bracket the region of SL, and these bounds are
increasing functions of the applied magnetic field~\cite{YSK}. This
is often interpreted as suggesting that the primary effect of magnetic
fields is to alter the condition for stable bubble
oscillations. See. also~\cite{Magnetic}. (2) Temperature of the water
--- If $T_{H_{2}O}$ decreases then the emitted power $W$ increases.
The position of the peak of the spectrum depends on $T_{H_{2}O}$. It
has been suggested that the increased light emission at lower water
temperature is associated with an increased stability of the bubble,
allowing for higher driving pressures~\cite{Hilgen}.
These are only the most salient features of the SL phenomenon. In
attempting to explain such detailed and specific behaviour the
dynamical Casimir approach (QED vacuum approach) encounters the same
problems as all other approaches have. Nevertheless we shall argue
that SL explanations using a Casimir-like framework are viable, and
merit further investigation.
In this paper we shall concentrate on changes in the QED vacuum state
as a candidate explanation for SL, and try to clear up considerable
confusion as to what models based on the Casimir effect do and do not
predict. It is important to realize that changes in the static Casimir
energy in this experimental situation are big, that they have roughly
the right energy budget to drive SL, and that any purported
non-Casimir explanation for SL will have to find a way to hide the
effects of these changes in Casimir energy so as to make them
unobservable.
\section{Quantum-electrodynamic models of SL}
\subsection{Quasi-static Casimir models: Schwinger's approach}
The idea of a ``Casimir route'' to SL is due to Schwinger who several
years ago wrote a series of papers~\cite{Sc1,Sc2,Sc3,Sc4,Sc5,Sc6,Sc7}
regarding the so-called dynamical Casimir effect. Considerable
confusion has been caused by Schwinger's choice of the phrase
``dynamical Casimir effect'' to describe his particular model. In
fact, Schwinger's original model is not dynamical but is instead
quasi-static as the heart of the model lies in comparing two static
Casimir energy calculations: That for an expanded bubble with that for
a collapsed bubble. One key issue in Schwinger's model is thus simply
that of calculating Casimir energies for dielectric spheres---and
there is already considerable disagreement on this issue. A second and
in some ways more critical question is the extent to which this
difference in Casimir energies may be converted to real photons during
the collapse of the bubble---it is this issue that we shall address in
this paper. The original quasi-static incarnation of the Schwinger
model had no real way of estimating either photon production
efficiency or timing information ({\em when} does the flash occur?).
In contrast the model of Eberlein~\cite{Eberlein1,Eberlein2,Eberlein3}
(more fully discussed below) is truly dynamical but uses a much more
specific physical approximation---the adiabatic approximation. The
two models should not be confused. In this paper we shall argue that
the observed features of SL force one to make the sudden
approximation. We can then estimate the spectrum of the emitted
photons by calculating an appropriate Bogolubov coefficient relating
two states of the QED vacuum. The resulting variant of the Schwinger
model for SL is then rather tightly constrained, and should be
amenable to experimental verification (or falsification) in the near
future.
In his series of papers~\cite{Sc1,Sc2,Sc3,Sc4,Sc5,Sc6,Sc7} on SL,
Schwinger showed that the dominant bulk contribution to the Casimir
energy of a bubble (of dielectric constant $\epsilon_{\mathrm
inside}$) in a dielectric background (of dielectric constant
$\epsilon_{\mathrm outside}$) is~\cite{Sc4}
\begin{eqnarray}
E_{\mathrm cavity}
&=&
+2\frac{4\pi}{3}R^3 \; \int_0^K
{4 \pi k^2 dk\over(2\pi)^3} \; \frac{1}{2} \hbar c k
\left(
{1\over\sqrt{\epsilon_{\mathrm inside}}}-
{1\over\sqrt{\epsilon_{\mathrm outside}}}
\right) +\cdots
\nonumber\\
&=&+\frac{1}{6 \pi} \hbar c R^3 K^4
\left(
{1\over\sqrt{\epsilon_{\mathrm inside}}}-
{1\over\sqrt{\epsilon_{\mathrm outside}}}
\right) +\cdots.
\label{E:schwinger}
\end{eqnarray}
The corresponding number of emitted photons is
\begin{eqnarray}
N
&=&
+2\frac{4\pi}{3}R^3 \; \int_0^K
{4 \pi k^2 dk\over(2\pi)^3} \; \frac{1}{2}
\left(
\frac{\sqrt{\epsilon_{\mathrm outside}}}{\sqrt{\epsilon_{\mathrm
inside}}}- 1 \right) +\cdots
\nonumber\\
&=&+\frac{2}{9 \pi} (RK)^3
\left(
\frac{\sqrt{\epsilon_{\mathrm outside}}}{\sqrt{\epsilon_{\mathrm
inside}}}- 1 \right) +\cdots.
\label{N:schwinger}
\end{eqnarray}
Here we have inserted an explicit factor of two with respect to
Schwinger's papers to take into account both photon polarizations.
There are additional sub-dominant finite volume effects discussed
in~\cite{CMMV1,CMMV2,MV}. Schwinger's result can also be rephrased in
the clearer and more general form as~\cite{CMMV1,CMMV2,MV}:
\begin{equation}
E_{\mathrm cavity} = + 2 V \int \frac{d^3\vec{k}}{(2 \pi)^3} \;
\frac{1}{2}
\hbar \left[ \omega_{\mathrm inside}(k) -
\omega_{\mathrm outside}(k) \right] + \cdots
\end{equation}
\begin{equation}
N = + 2 V \int \frac{d^3\vec{k}}{(2 \pi)^3} \;
\frac{1}{2}
\left[
{\omega_{\mathrm inside}(k) \over \omega_{\mathrm outside}(k)}
- 1 \right] + \cdots
\end{equation}
Here it is evident that the Casimir energy can be interpreted as a
difference in zero point energies due to the different dispersion
relations inside and outside the bubble. The quantity $K$ appearing
above is a high-wavenumber cutoff that characterizes the wavenumber at
which the (real part of) the refractive indices ($n=\sqrt{\epsilon}$)
drop to their vacuum values. It is important to stress that this
cutoff it is not a regularization artifact to be removed at the end of
the calculation. The cutoff has a true physical meaning in its own
right.
The three main points of strength of models based on zero point
fluctuations ({\em e.g.}, Schwinger's model and its variants) are:
1) One does {\em not} need to achieve ``real'' temperatures of
thousands of Kelvin inside the bubble. As discussed in~\cite{2gamma},
quasi-thermal behaviour is generated in quantum vacuum models by the
squeezed nature of the two photon states created \cite{Eberlein2}, and
the ``temperature'' parameter is a measure of the squeezing, not a
measure of any real physical temperature\footnote{%
This ``false thermality'' must not be confused with the very specific
phenomenon of Unruh temperature. In that case, valid only for
uniformly accelerated observers in flat spacetime, the temperature is
proportional to the constant value of the acceleration. Instead, in
the case of squeezed states, the apparent temperature can be related
to the degree of squeezing of the real photon pairs generated via the
dynamical Casimir effect.}.
(Of course, one should remember that the experimental data merely
indicates an approximately power-law spectrum [$N(\omega) \propto
\omega^\alpha$] with some sort of cutoff in the ultraviolet, and with
an exponent that depends on the gases entrained in the bubble; the
much-quoted ``temperature'' of the SL radiation is merely an
indication of the scale of this cutoff $K$.)
2) There is no actual production of far ultraviolet photons (because
the refractive index goes to unity in the far ultraviolet) so one does
not expect dissociation effects in water that other models
imply. Models based on the quantum vacuum automatically provide a
cutoff in the far ultraviolet from the behaviour of the refractive
index---this observation going back to Schwinger's first papers on the
subject.
3) One naturally gets the right energy budget. For $n_{\mathrm
outside} \approx 1.3$, $n_{\mathrm inside} \approx 1$, $K$ in the
ultra-violet, and $R\approx R_{\mathrm max}$, the change in the static
Casimir energy approximately equals the energy emitted each cycle.
This last point is still the object of heated debate.
Milton~\cite{M95}, and Milton and Ng~\cite{M96,M97} strongly criticize
Schwinger's result claiming that actually the Casimir energy contains
at best a surface term, the bulk term being discarded via (what is to
our minds) a physically dubious renormalization argument. In their
more recent paper~\cite{M97} they discard even the surface term and
now claim that the Casimir energy for a dielectric bubble is of order
$E\approx \hbar c/R$. (The dispute is ongoing---see~\cite{M98}.)
These points have been discussed extensively in~\cite{CMMV1,CMMV2,MV}
where it is emphasized that one has to compare two different physical
configurations of the same system, corresponding to two different
geometrical configurations of matter, and thus must compare {\em two}
different quantum states defined on the {\em same}
spacetime\footnote{%
This point of view is also in agreement with the bag model results of
Candelas~\cite{Candelas}. It is easy to see that in the bag model one
finds a bulk contribution that happens to be zero only because of the
particular condition that $\epsilon \mu = 1$ everywhere. This
condition ensures the constancy of the speed of light (and so the
invariance of the dispersion relation) on all space while allowing the
dielectric constant to be less than one outside the vacuum bag (as the
model for quark confinement requires).}.
In a situation like Schwinger's model for SL one has to subtract from
the zero point energy (ZPE) for a vacuum bubble in water the ZPE for
water filling all space. It is clear that in this case the bulk term
is physical and {\em must} be taken into account. Surface terms are
also present, and eventually other higher order correction terms, but
they prove to not be dominant for sufficiently large
cavities~\cite{MV}.
The calculations of Brevik {\em et al}~\cite{Brevik} and Nesterenko
and Pirozhenko~\cite{Nesterenko} also fail to retain the known bulk
volume term. In this case the subtlety in the calculation arises from
neglecting the continuum part of the spectrum. They consider a
dielectric sphere in an infinite dielectric background and sum {\em
only} over the discrete part of the spectrum to calculate their
Casimir ``energy''. When the continuum modes are reintroduced the
proper volume dependence is recovered. Their conclusions regarding
the relevance of the Casimir effect to SL are then incorrect: by
completely discarding the volume (and indeed surface) contribution
they are left with a Casimir ``energy'' that can be simply estimated
by dimensional analysis to be of order $\hbar c/R$ and is strictly
proportional to the inverse radius of the bubble. This is certainly a
very small quantity insufficient to drive SL but this is also not the
correct physical quantity to calculate. For a careful discussion of
the correct identification of the physically relevant Casimir energy
see~\cite{CMMV1,CMMV2,MV}.
While we believe that the contentious issues of how to define the
Casimir energy are successfully dealt with in~\cite{CMMV1,CMMV2,MV},
one of the subsidiary aims of this paper is to side-step this whole
argument and provide an independent calculation demonstrating
efficient photon production.
In contrast to the points of strength outlined above, the main
weakness of the original quasi-static version of Schwinger's idea is
that there is no real way to calculate either timing information or
conversion efficiency. A naive estimate is to simply and directly
link power produced to the change in volume of the bubble. As pointed
out by Barber {\em et al.}~\cite{Physics-Reports}, this assumption
would imply that the main production of photons may be expected when
the the rate of change of the volume is maximum, which is
experimentally found to occur near the maximum radius. In contrast the
emission of light is experimentally found to occur near the point of
minimum radius, where the rate of change of area is maximum. All else
being equal, this would seem to indicate a surface dependence and
might be interpreted as a true weakness of the dynamical Casimir
explanation of SL. In fact we shall show that the situation is
considerably more complex than might naively be thought. We claim that
there is much more going on than a simple change in volume of the
bubble, and shall shortly focus attention on the ``bounce'' that
occurs as the contents of the bubble hit the van der Waals hard core
maximum density.
\subsection{Eberlein's dynamical model for SL}
The quantum vacuum approach to SL was developed extensively in the
work of Eberlein~\cite{Eberlein1,Eberlein2,Eberlein3}. The basic
mechanism in Eberlein's approach is a particular implementation of the
dynamical Casimir effect: Photons are assumed to be produced due to an
{\em adiabatic} change of the refractive index in the region of space
between the minimum and the maximum bubble radius (a related
discussion for time-varying but spatially-constant refractive index
can be found in the discussion by Yablonovitch\cite{Yablonovitch}).
This physical framework is actually implemented via a boundary between
two dielectric media which accelerates with respect to the rest frame
of the quantum vacuum state. The change in the zero-point modes of the
fields produces a non-zero radiation flux. Eberlein's contribution
was to take the general phenomenon of photon generation by moving
dielectric boundaries and attempt a specific implementation of these
ideas as a candidate for explaining SL.
It is important to realize that this is a second-order effect. Though
he was unable to provide a calculation to demonstrate it, Schwinger's
original discussion is posited on the direct conversion of zero point
fluctuations in the expanded bubble vacuum state into real photons
plus zero point fluctuations in the collapsed bubble vacuum
state. Eberlein's mechanism is a more subtle (and much weaker) effect
involving the response of the atoms in the dielectric medium to
acceleration through the zero-point fluctuations. The two mechanisms
are quite distinct and considerable confusion has been engendered by
conflating the two mechanisms. Criticisms of the Eberlein mechanism do
not necessarily apply to the Schwinger mechanism, and vice versa.
In the Eberlein analysis the motion of the bubble boundary is taken
into account by introducing a velocity-dependent perturbation to the
usual EM Hamiltonian:
\begin{eqnarray}
H_{\epsilon} &\!=&\!
\frac{1}{2} \int{\rm d}^3{\bf r}
\left(
{{\bf D}^2\over\epsilon} + {\bf B}^2
\right)\;,
\\
\Delta H &\!=&\!
\beta
\int{\rm d}^3{\bf r}
\frac{\epsilon -1}{\epsilon}\;
({\bf D}\wedge{\bf B})\cdot{\bf\hat r}\;.
\end{eqnarray}
This is an approximate low-velocity result coming from a power series
expansion in the speed of the bubble wall $\beta= \dot R/c$. The
bubble wall is known to collapse with supersonic velocity, values of
Mach 4 are often quoted, but this is still completely non-relativistic
with $\beta\approx 10^{-5}$. Unfortunately, when the Eberlein
formalism is used to model the observed quantity of radiation from
each SL flash the implied bubble wall velocities are superluminal,
indicating that one has moved outside the region of validity of the
approximation scheme~\cite{Eberlein-discussion}.
The Eberlein approach consists of a novel mixture of the standard
adiabatic approximation with perturbation theory. In principle, the
adiabatic approximation requires the knowledge of the complete set of
eigenfunctions of the Hamiltonian for any allowed value of the
parameter. In the present case only the eigenfunctions of part of the
Hamiltonian, namely those of $H_\epsilon$, are known. (And these
eigenfunctions are known explicitly only in the adiabatic
approximation where $\epsilon$ is treated as time-independent.) The
calculation consists of initially invoking the standard application of
the adiabatic approximation to the full Hamiltonian, then formally
calculating the transition coefficients for the vacuum to two photon
transition to first order in $\beta$, and finally in explicitly
calculating the radiated energy and spectral density. In this last
step Eberlein used an approximation valid only in the limit $k R\gg1$
which means in the limit of photon wavelengths smaller than the bubble
radius. (Compare this with our discussion of the large $R$ limit
below.) This implies that the calculation will completely miss any
resonances that are present.
Eberlein's final result for the energy radiated over one acoustic
cycle is:
\begin{equation}
{\cal W} = 1.16\:\frac{(n^2-1)^2}{n^2}\,\frac{1}{480\pi}
\left[{\hbar\over c^3}\right]
\int_{0}^{T} {\rm d}\tau\; \frac{\partial^5 R^2(\tau)}{\partial
\tau^5}\,R(\tau)\beta(\tau)\;.
\end{equation}
Eberlein approximates $n_{\mathrm inside} \approx n_{\mathrm air}
\approx 1$ and sets $n_{\mathrm outside} = n_{\mathrm water} \to
n$. The $1.16$ is the result of an integration that has to be
estimated numerically. The precise nature of the semi-analytic
approximations made as prelude to performing the numerical integration
are far from clear.
One of the interesting consequences of this result is that the
dissipative force acting on the moving dielectric interface can be
seen to behave like $R^2\beta^{(4)}(t)$, plus terms with lower
derivatives of $\beta$. This dependence tallies with results of
calculations for frictional forces on moving perfect mirrors; the
dissipative part of the radiation pressure on a moving dielectric
{\em half-space} or {\em flat} mirror is proportional to the fourth
derivative of the velocity~\cite{JR}.
By a double integration by parts the above can be re-cast as
\begin{equation}
{\cal W} = 1.16\:\frac{(n^2-1)^2}{n^2}\,\frac{1}{960\pi}
\left[{\hbar\over c^4}\right]
\int_{0}^{T} {\rm d}\tau\;
\left(\frac{\partial^3 R^2(\tau)}{\partial \tau^3}\right)^2.
\end{equation}
Then the energy radiated is also seen to be proportional to
\begin{equation}
\int_0^T (\dot R)^2 (\ddot R)^2 + ...
\end{equation}
explicitly showing that the acceleration of the interface ($\ddot
R$) and the strength of the perturbation ($\dot R$) both contribute
to the radiated energy.
The main improvement of this model over the original Schwinger
model is the ability to provide basic timing information: In this
mechanism the massive burst of photons is produced at and near the
turn-around at the minimum radius of the bubble. There the velocity
rapidly changes sign, from collapse to re-expansion. This means
that the acceleration is peaked at this moment, and so are higher
derivatives of the velocity. Other main points of strength of the
Eberlein model are the same as previously listed for the Schwinger
model. However, Eberlein's model exhibits significant weaknesses
(which do not apply to the Schwinger model):
1) The calculation is based on an adiabatic approximation which
does not seem consistent with results. The adiabatic approximation
would seem to be justified in the SL case by the fact that the
frequency $\Omega$ of the driving sound is of the order of tens of
kHz, while that of the emitted light is of the order of $10^{15}$
Hz. But if you take a timescale for bubble collapse of the order
of milliseconds, or even microseconds, then photon production is
extremely inefficient, being exponentially suppressed [as we shall
soon see] by a factor of $\exp(-\omega/\Omega)$.
In order to compare with the experimental data the model requires,
as external input, the time dependence of the bubble radius. This
is expressed as a function of a parameter $\gamma$ which describes
the time scale of the collapse and re-expansion process. In order
to be compatible with the experimental values for emitted power
$\cal W$ one has to fix $\gamma \approx 10\; {\rm fs}$. This is
far too short a time to be compatible with the {\em adiabatic}
approximation. Although one may claim that {\em the precise
numerical value of the timescale} can ultimately be modified by
the eventual inclusion of resonances it would seem reasonable
to take this ten femtosecond figure as a first self-consistent
approximation for the characteristic timescale of the driving system
(the pulsating bubble). Unfortunately, the characteristic timescale
of the collapsing bubble then comes out to be of the same order of
the characteristic period of the emitted photons, {\em violating
the adiabatic approximation used in deriving the result}. Attempts
at bootstrapping the calculation into self-consistency instead
bring it to a regime where the adiabatic approximation underlying
the scheme cannot be trusted.
2) The Eberlein calculation cannot deal with any resonances that
may be present. Eberlein does consider resonances to be a possible
important correction to her model, but she is considering ``classical''
resonances (scale of the cavity of the same order of the wavelength
of the photons) instead of what we feel is the more interesting
possibility of parametric resonances.
Finally we should mention a recent calculation that gives
qualitatively the same results as the Eberlein model although leading
to different formulae. Sch\"utzhold, Plunien, and Soff~\cite{SPS} adopt
a slightly different decomposition into unperturbed and perturbing
Hamiltonians by taking
\begin{eqnarray}
H_{0} &\!=&\!
\frac{1}{2} \int{\rm d}^3{\bf r}
\left(
{{\bf D}^2\over \epsilon_0} + {\bf B}^2
\right),
\\
\Delta H &\!=&\!
\int{\rm d}^3{\bf r}
\left(
-\frac{1}{2} {\epsilon-\epsilon_0\over\epsilon\epsilon_0} {\bf D}^2
+ {\bf \beta} \;
\frac{\epsilon -\epsilon_0}{\epsilon}\;
({\bf D}\wedge{\bf B}) \cdot {\rm \hat r}
\right),
\end{eqnarray}
Their result for the total energy emitted per cycle is given
analytically by
\begin{equation}
{\cal W} = {n^2(n^2-1)^2 \over 1890\pi}
\left[{\hbar\over c^6}\right]
\int_{0}^{T} {\rm d}\tau\;
\left(\frac{\partial^4 R^3(\tau)}{\partial \tau^4}\right)^2.
\end{equation}
The key differences are that this formula is analytic (rather than
numerical) and involves fourth derivatives of the volume of the
bubble (rather than third derivatives of the surface area). The
main reason for the discrepancy between this and Eberlein's result
can be seen as due to a different choice of the dependence on $r$
of $\beta(r,t)$. In reference~\cite{SPS} they considered the more
physical case of a localized disturbance that yields significant
contributions only over a bounded volume. (Eberlein makes the
simplifying assumption that $\beta(r,t)$ is a function of $t$ only,
which is incompatible with continuity and the essentially constant
density of water. In contrast, Sch\"utzhold\ {\em et al.} take the
radial velocity of the water outside the bubble to be $\beta(r,t)
= f(t)/r^2$.)
Putting these models aside, and before proposing new routes for
developing further research in SL, we shall give below a more detailed
discussion of some important points of Schwinger's model which seem to
us to be crucial in order to understand the possibility of a vacuum
explanation of SL.
\subsection{Timescales: The need for a sudden approximation.}
One of the key features of photon production by a space-dependent and
time-dependent refractive index is that for a change occurring on a
timescale $\tau$, the amount of photon production is exponentially
suppressed by an amount $\exp(-\omega\tau)$. Below we provide a
specific model that exhibits this behaviour, and argue that the result
is in fact generic.
The importance for SL is that the experimental spectrum is {\em not\,}
exponentially suppressed at least out to the far ultraviolet.
Therefore any mechanism of Casimir-induced photon production based on
an adiabatic approximation is destined to failure: Since the
exponential suppression is not visible out to $\omega \approx 10^{15}
\hbox{ Hz}$, it follows that {\em if\,} SL is to be attributed to
photon production from a time-dependent refractive index ({\em i.e.},
the dynamical Casimir effect), {\em then} the timescale for change in
the refractive index must be of order a {\em femtosecond\/}\footnote{
Actually, once one takes into account the refractive index of the
final state this condition can be somewhat relaxed. We ultimately find
that we can tolerate a refractive index that changes as slowly as on a
picosecond timescale, but this is still far to rapid to be associated
with physical collapse of the bubble. For the time being we focus on
the femtosecond timescale (which actually makes things more difficult
for us) to check the physical plausibility of the scenario, but keep
in mind that eventually things can be relaxed by a few orders of
magnitude.}.
Thus any Casimir--based model has to take into account that {\em
the change in the refractive index cannot be due just to the change
in the bubble radius}.
The SL flash is known to occur at or shortly after the point of
maximum compression. The light flash is emitted when the bubble is at
or near minimum radius $R_{\mathrm min} \approx 0.5\;\mu \hbox{m} =
500\;\hbox{nm}$. Note that to get an order femtosecond change in
refractive index over a distance of about $500\;{\rm nm}$, the change
in refractive index has to propagate at relativistic speeds. To
achieve this, we must adjust basic aspects of the model: We will move
away from the original Schwinger suggestion, in that it is no longer
the collapse from $R_{\mathrm max}$ to $R_{\mathrm min}$ that is
important. {\em Instead we will postulate a rapid (order femtosecond)
change in refractive index of the gas bubble when it hits the van der
Waals hard core.}
The underlying idea is that there is some physical process that gives
rise to a sudden change of the refractive index inside the bubble when
it reaches maximum compression. We have to ensure that the velocity
of mechanical perturbations, that is the sound velocity, can be a
significant fraction of the speed of light in this critical
regime\footnote{
This condition can also be slightly relaxed: One can conceive of the
change in refractive index being driven by a shockwave that appears at
the van der Waals hard core. Now a shockwave is by definition a
supersonic phenomenon. If the velocity of sound is itself already
extremely high then the shockwave velocity may be even higher. Note
that most of the viable models for the gas dynamics during the
collapse predict the formation of strong shock-waves; so we are
adapting physics already envisaged in the literature. At the same time
we are asking for less extreme conditions ({\em e.g.}, we can be much
more relaxed regarding the focussing of these shockwaves) in that we
just need a rapid change of the refractive index of the entrained gas,
and do not need to propose any overheating to ``stellar''
temperatures. It is also interesting to note that changes in the
refractive index (but that of the surrounding water) due to the huge
compression generated by shockwaves have already been considered in
the literature~\cite{HRB} ({\em cf.} page 5437).}.
We first show that the minimum radius experimentally observed is of
the same order as the van der Waals hard core radius $R_{\mathrm
hc}$. The latter can be deduced as follows: It is known that the van
der Waals excluded volume for air is $b=$0.036
l/mol~\cite{Wu-Roberts}. The minimum possible value of the volume is
then $V_{\mathrm hc} = b\cdot (\rho V_{\mathrm ambient})/m$, where
$(\rho V_{\mathrm ambient})/m$ gives the number of moles and
$V_{\mathrm ambient}$ is the ambient value of the volume. {From}
$R_{\mathrm hc} = R_{\mathrm ambient} \cdot (b
\rho/\mu)^{1/3}$ and assuming for the density of air $\rho=10^{-3}\;
{\rm gr}/{\rm cm}^3$ [$1.3 \times 10^{-3} \; {\rm gr}/{\rm cm}^3$ at
STP (standard temperature and pressure)], one gets $R_{\mathrm hc}\sim
0.48\; \mu {\mathrm m}$. This value compares favorably with the
experimentally observed value of $R_{\mathrm min}$. Moreover, the
role of the van der Waals hard core in limiting the collapse of the
bubble is suggested in~\cite{Physics-Reports} (cf. fig. 10, p.78), and
a careful hydrodynamic analysis for the case of an Argon
bubble~\cite{Fluid} reveals that for sonoluminescence it is necessary
that the bubble undergoes a so called strongly collapsing phase where
its minimum radius is indeed very near the hard core radius\footnote{
Noticeably, from~\cite{Fluid} it is easy to estimate the van der Waals
radius for the Argon bubble: $R_{\mathrm hc} \approx R_{\mathrm
ambient}/8.86$, with $R_{\mathrm ambient}=0.4 \mu {\rm m}$. This
again gives $R_{\mathrm hc}(Argon) \approx 0.45 \; \mu {\rm m}
\approx R_{\mathrm min}$.}.
It is crucial to realize that a van der Waals gas, when compressed to
near its maximum density, has a speed of sound that goes relativistic.
To see this, write the (non-relativistic) van der Waals equation of
state as
\begin{equation}
p
= {n k T \over 1-n b} - a n^2
= {\rho kT/m \over 1- \rho/\rho_{\mathrm max}}
- a {\rho^2\over m^2}.
\end{equation}
Here $n$ is the number density of molecules; $\rho$ is mass density;
$m$ is average molecular weight ($m=28.96\; \hbox{amu/molecule} =
28.96\; \hbox{gr/mol}$ for air).
Now consider the (isothermal) speed of sound for a van der Waals gas
\begin{equation}
v_{\mathrm sound}^2
= \left({\partial p \over \partial \rho}\right)_{T}
= {(kT/m) \over (1- \rho/\rho_{\mathrm max})^2}
- 2a {\rho\over m^2}.
\end{equation}
Near maximum density this is
\begin{equation}
v_{\mathrm sound}
\approx {\sqrt{kT/m} \over (1- \rho/\rho_{\mathrm max}) },
\end{equation}
and so it will go relativistic for densities close enough to maximum
density. It is {\em only} for the sake of simplicity that we have
considered the isothermal speed of sound. We do not expect the
process of bubble collapse and core bounce to be isothermal.
Nevertheless, this calculation is sufficient to demonstrate that
in general the sound velocity becomes formally infinite (and it is
reasonable that it goes relativistic) at the incompressibility
limit (where it hits the van der Waals hard core). This conclusion
is not limited to the van der Waals equation of state, and is not
limited to isothermal (or even isentropic) sound propagation.
Indeed, consider any equation of state of the form
\begin{equation}
p = \kappa({\cal V}-b,T),
\end{equation}
where ${\cal V}$ is the volume per mole occupied by the gas and
$b$ is the ``minimum molar volume'', related to the molar mass and
maximum mass density by
\begin{equation}
b = m/ \rho_{\mathrm max}.
\end{equation}
To say that $b$ is a minimum molar volume means that we want
\begin{equation}
\lim_{{\cal V} \to b} \kappa({\cal V}-b,T) = \infty.
\end{equation}
We can enforce this by demanding
\begin{equation}
p({\cal V},T) = \kappa({\cal V}-b,T) = {\kappa_1({\cal V}-b,T)\over
{\cal V}-b} + \kappa_2({\cal V}-b,T),
\end{equation}
or equivalently
\begin{equation}
p(\rho,T) =
{\kappa_3(\rho-\rho_{\mathrm max},T)\over
1-\rho/\rho_{\mathrm max}} + \kappa_4(\rho-\rho_{\mathrm max},T),
\end{equation}
with the $\kappa_i({\cal V}-b,T)$ being less singular than $p({\cal
V}-b,T)$. For typical model equations of state $\kappa_1({\cal
V}-b,T)$ and $\kappa_2({\cal V}-b,T)$ are typically differentiable and
finite. (The van der Waals, Dieterici, Bethelot, and ``modified
adiabatic'' equations are all of this form, but the Moss {\em et al}
equation of state is {\em not} of this
form~\cite{Moss-et-al}.)~\footnote{%
The Dieterici equation of state is
\[
p = {n k T \over 1 - nb} \exp\left(-{a n\over k T}\right)
= {\rho k T /m \over 1 - \rho/\rho_{\mathrm max}}
\exp\left(-{a \rho\over m k T}\right),
\]
while the Bethelot equation of state is
\[
p = {n k T \over 1 - nb} - {a' n^2 \over T}
= {\rho k T/m \over 1 - \rho/\rho_{\mathrm max}}
- {a' \rho^2 \over m^2 T},
\]
so that it is a modified van der Waals equation with a particular
temperature dependence for the $a$ parameter ($a \to a'/T$). The
``modified adiabatic'' equation of state discussed by Barber {\em
et al}~\cite{Physics-Reports} is
\[
p = p_0 \left( { 1 - n_0 b \over 1- n b} \right)^\gamma =
p_0 \left( { 1 - \rho_0/\rho_{\mathrm max}
\over 1 - \rho/\rho_{\mathrm max} } \right)^\gamma.
\]
In contrast, Moss et al~\cite{Moss-et-al} use a model equation of
state of the form
\[
p = {\rho k T\over m}(1 + \kappa)
+ {\gamma E_c \rho \over 1-\gamma}
\left[ \left({\rho\over\rho_0}\right)^\gamma
- \left({\rho\over\rho_0}\right)
\right],
\]
with $\kappa$ and $\gamma$ being adjustable parameters. This
equation of state does {\em not} exhibit a maximum hard-core
density.}
Now calculate the speed of sound, keeping some quantity ``$X$''
constant
\begin{equation}
v_{\mathrm sound}|_X
= \sqrt{ \left.\left({dp\over d\rho}\right)\right|_X }
= \sqrt{ \left({\partial p\over\partial \rho}\right)
+ \left({\partial p\over\partial T}\right)
\left.\left({dT\over d\rho}\right)\right|_X }.
\end{equation}
Then
\begin{eqnarray}
v_{\mathrm sound}^2|_X &=&
{\kappa_3(\rho-\rho_{\mathrm max},T) \over
\rho_{\mathrm max} (1-\rho/\rho_{\mathrm max})^2 }
+
\left({\partial \kappa_3\over\partial \rho}\right)
{1\over 1-\rho/\rho_{\mathrm max}}
+
\left({\partial \kappa_3\over\partial T}\right)
\left.\left({dT\over d\rho}\right)\right|_X
{1\over 1-\rho/\rho_{\mathrm max}}
\nonumber\\
&&
\qquad
+
\left({\partial \kappa_4\over\partial \rho}\right)
+
\left({\partial \kappa_4\over\partial T}\right)
\left.\left({dT\over d\rho}\right)\right|_X .
\end{eqnarray}
The net result is that as $v \to b$; {\em i.e.} $ \rho \to
\rho_{\mathrm max}$; the speed of sound becomes relativistic
(formally infinite), independent of whether or not this is constant
temperature, constant entropy, or whatever ``constant $X$'' may
be\footnote{
It should be noted that similar unphysical features also affect the
``shock wave''--based models~\cite{Physics-Reports}: Indeed, the Mach
number of the shock formally diverges as the shock implodes towards
the origin ({\em cf.} page 126 of \cite{Physics-Reports}). One way to
overcome this type of problem is the suggestion that that very near
the minimum radius of the bubble there is a breakdown of the
hydrodynamic description~\cite{Fluid}. If so, the thermodynamic
description in terms of state equations should probably be considered
to be on a heuristic footing at best.}.
There is something of a puzzle in the fact that hydrodynamic
simulations of bubble collapse do not see these relativistic effects.
Notably, the simulations by Moss {\em et al}~\cite{Moss-et-al} seem to
suggest collapse, shock wave production, and re-expansion all without
ever running into the van der Waals hard core. The fundamental reason
for this is that the model equation of state they choose does not have
a hard core for the bubble to bounce off\footnote{%
For low densities their equation of state is essentially perfect
gas, for medium densities it becomes stiffer (and the speed of
sound goes up), but for very high densities it again becomes softer,
and there is no ``maximum density''. One might think that their
$\rho_0$, the density of solid air at zero Celsius, is a maximum
density, but if you look carefully at their equation of state the
gas is still compressible at this density; the pressure and speed
of sound are both finite.}.
Instead, there are a number of free parameters in their equation of
state which are chosen in such a way as to make their equation of
state stiff at intermediate densities, even if their equation of state
is by construction always soft at van der Waals hard core
densities. If the equation of state is made sufficiently stiff at
intermediate densities then a bounce can be forced to occur long
before hard core densities are encountered. Unfortunately, as
previously mentioned, the experimental data and hydrodynamic analysis
seem to indicate that hard core densities {\em are} achieved at
maximum bubble compression.
Given all this, the use of a relativistic sound speed
is now physically justifiable, and the possibility of femtosecond
changes in the refractive index is at least physically plausible
(even though we cannot say that femtosecond changes in refractive
index are guaranteed).
So our new physical picture is this: The ``in state'' is a small
sphere of gas, radius about 500 nm, with some refractive index
$n_{\mathrm gas}^{\mathrm in}$ embedded in water of refractive index $n_{\mathrm liquid}$. There is a
sudden femtosecond change in refractive index, essentially at
constant radius, so the ``out state'' is gas with refractive index
$n_{\mathrm gas}^{\mathrm out}$ embedded in water of refractive index $n_{\mathrm liquid}\;$\footnote{%
In view of this femtosecond change of refractive index, we would
be justified in making the sudden approximation for frequencies
less than about a PHz. In this paper we do not make this
approximation except when convenient in obtaining crude analytic
estimates, but when we turn to dealing with finite-size effects in
the companion paper~\cite{Companion} the sudden approximation will
be more than just a convenience: it will be absolutely essential
in keeping the mathematical features of the analysis tractable.}.
Thus our calculations will be complementary to Eberlein's calculations.
She was driven to femtosecond timescales to fit the experimental
data, but these femtosecond timescales then unfortunately undermined
the adiabatic approximation used to derive her results. In contrast,
we shall maintain a physically consistent calculation throughout.
These arguments have now pushed dynamical Casimir effect models
for SL into a rather constrained region of parameter space. We hope
that these ideas will become experimentally testable in the near
future.
\section{Bogolubov coefficients}
As a first approach to the problem of estimating the spectrum and
efficiency of photon production we decided to study in detail the
basic mechanism of particle creation and to test the consistency
of the Casimir energy proposals previously described. With this
aim in mind we studied the effect of a changing dielectric constant
in a homogeneous medium. At this stage of development, we are not
concerned with the detailed dynamics of the bubble surface, and
confine attention to the bulk effects, deferring consideration of
finite-volume effects to the companion paper~\cite{Companion}.
We shall consider two different asymptotic configurations. An
``in'' configuration with refractive index $n_{\mathrm in}$, and an ``out''
configuration with a refractive index $n_{\mathrm out}$. These two configurations
will correspond to two different bases for the quantization of the
field. (For the sake of simplicity we take, as Schwinger did, only
the electric part of QED, reducing the problem to a scalar
electrodynamics). The two bases will be related by Bogolubov
coefficients in the usual way. Once we determine these coefficients
we easily get the number of created particles per mode and from
this the spectrum. Of course it is evident that such a model cannot
be considered a fully complete and satisfactory model for SL. This
present calculation must still be viewed as a test calculation in
which basic features of the Casimir approach to SL are investigated.
In the original version of the Schwinger model it was usual to
simplify calculations by using the fact that the dielectric constant
of air is approximately equal 1 at standard temperature and pressure
(STP), and then dealing only with the dielectric constant of water
($n_{\mathrm liquid} = \sqrt{\epsilon_{\mathrm outside}} \approx
1.3$). We wish to avoid this temptation on the grounds that the
sonoluminescent flash is known to occur within $500$ picoseconds of
the bubble achieving minimum radius. Under these conditions the gases
trapped in the bubble are close to the absolute maximum density
implied by the hard core repulsion incorporated into the van der Waals
equation of state. Gas densities are approximately one million times
atmospheric and conditions are nowhere near STP. For this reason we
shall explicitly keep track of both initial and final refractive indices.
We now describe a simple analytically tractable model for the
conversion of zero point fluctuations (Casimir energy) into real
photons. The model describes the effects of a time-dependent
refractive index in the infinite volume limit. We shall show that
for sudden changes in the refractive index the conversion of
zero-point fluctuations is highly efficient, being limited only by
phase space, whereas adiabatic changes of the refractive index lead
to exponentially suppressed photon production.
\def\nabla \times{\nabla \times}
\def\nabla \cdot{\nabla \cdot}
\def\nabla{\nabla}
\subsection{Defining the model}
Take an infinite homogeneous dielectric with a permittivity
$\epsilon(t)$ that depends only on time, not on space. The homogeneous
($dF=0$) Maxwell equations are
\begin{equation}
B = \nabla \times A;
\end{equation}
\begin{equation}
E = - \nabla \phi - {1\over c} {\partial A\over\partial t};
\end{equation}
while the source-free inhomogeneous ($*d*F=0$) Maxwell equations
become
\begin{equation}
\nabla \cdot (\epsilon E) = 0;
\end{equation}
\begin{equation}
\nabla \times {B} =
+{1\over c} {\partial\over\partial t} (\epsilon E).
\end{equation}
Substituting into this last equation
\begin{equation}
\nabla \times\left(\nabla \times A\right) =
- {1\over c} {\partial\over\partial t}
\left[ \epsilon \left(
\nabla \phi + {1\over c} {\partial A\over\partial t}
\right) \right].
\end{equation}
Suppose that $\epsilon(t)$ depends on time but not space, then
\begin{equation}
( \nabla (\nabla \cdot A) - \nabla^2 A )=
- \nabla {1\over c} {\partial\over\partial t}
(\epsilon \phi ) - {1\over c^2 } {\partial\over\partial t} \epsilon
{\partial A\over\partial t}.
\end{equation}
Now adopt a {\em generalized} Lorentz gauge
\begin{equation}
\nabla \cdot A +
{1\over c} {\partial\over\partial t} (\epsilon \phi ) = 0.
\end{equation}
Then the equations of motion reduce to
\begin{equation}
{1\over c^2 } {\partial\over\partial t} \epsilon
{\partial A\over\partial t} = \nabla^2 A.
\end{equation}
We now introduce a ``pseudo-time'' parameter by defining
\begin{equation}
{\partial\over\partial \tau} = \epsilon(t) {\partial \over \partial t}.
\end{equation}
That is
\begin{equation}
\tau(t) = \int {dt\over\epsilon(t)}.
\end{equation}
In terms of this pseudo-time parameter the equation of motion is
\begin{equation}
{\partial^2\over\partial\tau^2} A =
c^2 {\epsilon(\tau)}\nabla^2 A.
\label{eqm}
\end{equation}
Compare this with equation (3.86) of Birrell and
Davies~\cite{Birrell-Davies}. Now pick a convenient profile for
the permittivity and permeability as a function of this pseudo-time.
(This particular choice of time profile for the refractive index
is only to make the problem analytically tractable, with a little
more work it is possible to consider generic monotonic changes of
refractive index and place bounds on the Bogolubov
coefficients~\cite{Visser}.) Let us take
\begin{eqnarray}
\label{E:profile}
{\epsilon(\tau)} &=& a + b \tanh(\tau/\tau_0)
\\
&=& {\textstyle{1\over2}} (n_{\mathrm in}^2 + n_{\mathrm out}^2) + {\textstyle{1\over2}}(n_{\mathrm out}^2-n_{\mathrm in}^2)\;\tanh(\tau/\tau_0).
\end{eqnarray}
Here $\tau_0$ represents the typical timescale of the change of the
refractive index in terms of the pseudo-time we have just defined. We
are interested in computing the number of particles that can be
created passing from the ``in'' state ($t\to-\infty$, that is,
$\tau\to-\infty$) to the ``out'' state ($t\to+\infty$, that is,
$\tau\to+\infty$). This means we must determine the Bogolubov
coefficients that relate the ``in'' and ``out'' bases of the quantum
Hilbert space. Defining the inner product as:
\begin{equation}
(\phi_1,\phi_2) =
i \int_{\Sigma_\tau} \phi_1^* \;
{\stackrel{\leftrightarrow}{\partial}\over\partial\tau} \;
\phi_2\: d^3x,
\end{equation}
The Bogolubov coefficients can now be {\em defined} as\\
\begin{eqnarray}
\alpha_{ij}
&=&
({A_{i}^{\mathrm out} },{A_{j}^{\mathrm in} }),
\\
\beta_{ij}
&=&(
{A_{i}^{\mathrm out} }^{*}, {A_{j}^{\mathrm in} }).
\end{eqnarray}
Where ${A_{i}^{\mathrm in} }$ and ${A_{j}^{\mathrm out} }$ are solutions
of the wave equation (\ref{eqm}) in the remote past and remote future
respectively. We shall compute the coefficient $\beta_{ij}$ It is
this quantity that is linked to the spectrum of the ``out'' particles
present in the ``in'' vacuum, and it is this quantity that is related
to the total energy emitted. With a few minor changes of notation we
can just write down the answers directly from pages 60--62 of Birrell
and Davies~\cite{Birrell-Davies}. Birrell and Davies were interested
in the problem of particle production engendered by the expansion of
the universe in a cosmological context. Although the physical model is
radically different here the mathematical aspects of the analysis
carry over with some minor translation in the details. Equations
(3.88) of Birrell--Davies become
\begin{equation}
\omega^\tau_{\mathrm in}
= k \sqrt{a-b}
= k \sqrt{\epsilon_{\mathrm in} }
= k \; n_{\mathrm in} ;
\end{equation}
\begin{equation}
\omega^\tau_{\mathrm out}
= k \sqrt{a+b}
= k \sqrt{\epsilon_{\mathrm out} }
= k \; n_{\mathrm out} ;
\end{equation}
\begin{equation}
\omega^\tau_{\pm} =
{\textstyle{1\over2}} \; k \; \left| n_{\mathrm in} \pm n_{\mathrm out} \right| =
{\textstyle{1\over2}} | \omega^\tau_{\mathrm in} \pm \omega^\tau_{\mathrm out}|.
\end{equation}
(Here we emphasise that these frequencies are those appropriate to the
``pseudo-time'' $\tau$.) The Bogolubov $\alpha$ and $\beta$
coefficients can be easily deduced from Birrell--Davies (3.92)+(3.93)
\begin{eqnarray}
\alpha(\vec k_{\mathrm in} , \vec k_{\mathrm out} )
&=& \frac{\sqrt{\omega^\tau_{\mathrm out} \; \omega^\tau_{\mathrm in} }}{\omega^\tau_{+}} \;\;
{\Gamma(-i \omega^\tau_{\mathrm in} \tau_0) \;
\Gamma(-i \omega^\tau_{\mathrm out} \tau_0) \over
\Gamma(-i \omega^\tau_{-} \tau_0)^2 } \;
\delta^3(\vec k_{\mathrm in} - \vec k_{\mathrm out} )\\
\beta(\vec k_{\mathrm in} , \vec k_{\mathrm out} )
&=& -
\frac{\sqrt{\omega^\tau_{\mathrm out} \; \omega^\tau_{\mathrm in} }}{\omega^\tau_{-}} \;\;
{\Gamma(-i \omega^\tau_{\mathrm in} \tau_0) \;
\Gamma(i \omega^\tau_{\mathrm out} \tau_0) \over
\Gamma(i \omega^\tau_{-} \tau_0)^2 } \;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ).
\end{eqnarray}
Now square, using Birrell--Davies (3.95). We obtain\footnote{
Note that these are the Bogolubov coefficients for a scalar field
theory. For QED in the infinite volume limit the two photon
polarizations decouple into two independent scalar fields and these
Bogolubov coefficients can be applied to each polarization state
independently. Finite volume effects are a little trickier.}
\begin{eqnarray}
|\beta(\vec k_{\mathrm in} , \vec k_{\mathrm out} )|^2
&=&
{\sinh^2(\pi\omega^\tau_{-}\tau_0)\over
\sinh(\pi\omega^\tau_{\mathrm in} \tau_0) \;
\sinh(\pi\omega^\tau_{\mathrm out} \tau_0)} \;
{V\over (2\pi)^3 } \;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ).
\end{eqnarray}
We now need to translate this into physical time, noting that
asymptotically, in either the infinite past or the infinite future,
$t \approx \epsilon \tau + (constant)$, so that for physical
frequencies
\begin{equation}
\omega_{\mathrm in} =
{\omega^\tau_{\mathrm in} \over\epsilon_{\mathrm in} } =
{\omega^\tau_{\mathrm in} \over n_{\mathrm in} ^2} =
{k \sqrt{a-b}\over \epsilon_{\mathrm in} } =
k \sqrt{1\over\epsilon_{\mathrm in} }
= {k \over n_{\mathrm in} };
\end{equation}
\begin{equation}
\omega_{\mathrm out} =
{\omega^\tau_{\mathrm out} \over\epsilon_{\mathrm out} } =
{\omega^\tau_{\mathrm out} \over n_{\mathrm out} ^2} =
{k \sqrt{a+b}\over \epsilon_{\mathrm out} } =
k \sqrt{1\over\epsilon_{\mathrm out} }
= {k \over n_{\mathrm out} }.
\end{equation}
Note that there is a symmetry in the Bogolubov coefficients under
interchange of ``in'' and ``out''.
We also need to convert the timescale over which the refractive index
changes form pseudo-time to physical time. To do this we define
\begin{equation}
t_0 \equiv \tau_0 \left.{dt\over d\tau}\right|_{\tau=0}.
\end{equation}
For the particular temporal profile we have chosen for analytic
tractability this evaluates to
\begin{equation}
t_0 = {\textstyle{1\over2}} \tau_0 \left( n_{\mathrm in}^2 + n_{\mathrm out}^2 \right).
\end{equation}
After these substitutions, the (squared) Bogolubov coefficient becomes
\begin{eqnarray}
|\beta(\vec k_{\mathrm in} , \vec k_{\mathrm out} )|^2
&=&
{
\sinh^2\left(
\pi\; {\textstyle |n_{\mathrm in}^2 \omega_{\mathrm in} -n_{\mathrm out}^2 \omega_{\mathrm out}| \over\textstyle n_{\mathrm in}^2+n_{\mathrm out}^2} \; t_0
\right)
\over
\sinh\left(
2\pi \; {\textstyle n_{\mathrm in}^2 \over \textstyle n_{\mathrm in}^2+n_{\mathrm out}^2} \; \omega_{\mathrm in} t_0
\right) \;
\sinh\left(
2\pi \; {\textstyle n_{\mathrm out}^2 \over \textstyle n_{\mathrm in}^2+n_{\mathrm out}^2} \; \omega_{\mathrm out} t_0
\right)
} \;
{V\over (2\pi)^3 } \;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ).
\label{bog2}
\end{eqnarray}
We now consider two limits, the adiabatic limit and the sudden limit,
and investigate the physics.
\subsection{Sudden limit}
Take
\begin{equation}
\hbox{max} \{\omega^\tau_{\mathrm in} , \omega^\tau_{\mathrm out} , \omega^\tau_-\} \; \tau_0 \ll 1.
\end{equation}
This corresponds to a rapidly changing refractive index. In terms of
physical time this is equivalent to
\begin{equation}
2\pi\;\hbox{max}
\left\{1, {n_{\mathrm in}\overn_{\mathrm out}}, {\textstyle{1\over2}}\left|{n_{\mathrm in}\overn_{\mathrm out}}-1\right|\right\} \;
{ n_{\mathrm out}^2 \over n_{\mathrm in}^2+n_{\mathrm out}^2} \; \omega_{\mathrm out} t_0 \ll 1,
\end{equation}
which can be simplified to yield
\begin{equation}
2\pi\;\hbox{max} \{n_{\mathrm in},n_{\mathrm out}\} \;
{ n_{\mathrm out} \over n_{\mathrm in}^2+n_{\mathrm out}^2} \; \omega_{\mathrm out} t_0 \ll 1.
\end{equation}
So the sudden approximation is a good approximation for frequencies
{\em less} than $\Omega_{\mathrm sudden}$, where we define
\begin{equation}
\Omega_{\mathrm sudden} = {1\over 2\pi t_0} \;
{n_{\mathrm in}^2+n_{\mathrm out}^2\over n_{\mathrm out} \; \hbox{max}\{n_{\mathrm in},n_{\mathrm out}\}
}.
\end{equation}
The this shows that the frequency up to which the sudden approximation
holds is not just the reciprocal of the timescale of the change in the
refractive index: there is also a strong dependence on the initial and
final values of the refractive indices. This implies that we can
relax, for some ranges of values of $n_{\mathrm in}$ and $n_{\mathrm out}$, our figure of
$t_0\sim O({\rm fs})$ by up to a few orders of magnitude.
Unfortunately the precise shape of the spectrum is heavily dependent
on all the experimental parameters ($K,n_{\mathrm in},n_{\mathrm out},R$). This
discourages us from making any sharp statement regarding the exact
value of the timescale required in order to fit the data.
In the region where the sudden approximation holds the various
$\sinh(x)$ functions in equation (\ref{bog2}) can be replaced by their
arguments $x$. Then
\begin{equation}
|\beta|^2 \propto
{ (\pi [n_{\mathrm in}-n_{\mathrm out}])^2 \over
(2\pi n_{\mathrm in}) \; (2\pi n_{\mathrm out}) }.
\label{E:sb}
\end{equation}
More precisely
\begin{equation}
|\beta(\vec k_{\mathrm in}, \vec k_{\mathrm out})|^2 \approx
{1\over 4}
{ (n_{\mathrm in} - n_{\mathrm out})^2 \over
n_{\mathrm in} \; n_{\mathrm out} }\;
{V\over (2\pi)^3 } \; \delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ),
\label{E:sbsq}
\end{equation}
For completeness we also give the unsquared Bogolubov
coefficients evaluated in the sudden approximation:
\begin{eqnarray}
\alpha(\vec k_{\mathrm in}, \vec k_{\mathrm out})
&\approx&
{1\over 2}
{n_{\mathrm in} +n_{\mathrm out} \over \sqrt{n_{\mathrm in} \; n_{\mathrm out} }}\;
\delta^3(\vec k_{\mathrm in} - \vec k_{\mathrm out} ),\\
\beta(\vec k_{\mathrm in}, \vec k_{\mathrm out})
&\approx&
{1\over 2}
{|n_{\mathrm in} -n_{\mathrm out}| \over \sqrt{n_{\mathrm in} \; n_{\mathrm out} }}\;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ).
\end{eqnarray}
As expected, for $n_{\mathrm in} \rightarrow n_{\mathrm out}$, we have $\alpha \rightarrow
\delta^3(\vec k_{\mathrm in} - \vec k_{\mathrm out} )$ and $\beta \rightarrow 0$.
These result should be be compared with that obtained in the companion
paper~\cite{Companion}, where we first include finite volume
effects and then consider the large-volume limit for dielectric
bubbles in order to reproduce the original Schwinger estimate of
photon production~\cite{Sc1,Sc2,Sc3,Sc4,Sc5,Sc6,Sc7}. It should
also be compared with the discussion of Yablonovitch~\cite{Yablonovitch}
[see particularly the formulae in the paragraph between equations
(8) and (9)].
\subsection{Adiabatic limit}
Now take
\begin{equation}
\hbox{min}\{\omega^\tau_{\mathrm in} , \omega^\tau_{\mathrm out} , \omega^\tau_-\} \; \tau_0 \gg 1.
\end{equation}
This corresponds to a slowly changing refractive index. In this limit
the $\sinh(x)$ functions in the exact Bogolubov coefficient can be
replaced with exponential functions $\exp(x)$. Then
\begin{eqnarray}
|\beta|^2 &\propto&
{ \exp(2\pi \omega^\tau_- \tau_0) \over
\exp(\pi\omega^\tau_{\mathrm in} \tau_0) \;
\exp(\pi\omega^\tau_{\mathrm out} \tau_0)}
\\
&=&
{ \exp(\pi \; |\omega^\tau_{\mathrm in} - \omega^\tau_{\mathrm out}| \; \tau_0) \over
\exp(\pi\omega^\tau_{\mathrm in} \tau_0) \;
\exp(\pi\omega^\tau_{\mathrm out} \tau_0)}.
\end{eqnarray}
More precisely
\begin{equation}
|\beta(\vec k_{\mathrm in} , \vec k_{\mathrm out} )|^2
\approx
\exp\left(
-2\pi \; \hbox{min}\{\omega^\tau_{\mathrm out} ,\omega^\tau_{\mathrm in} \} \;
\tau_0
\right) \;
{V\over (2\pi)^3 } \;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ).
\end{equation}
In terms of physical time the condition defining the adiabatic limit
reads
\begin{equation}
2\pi\;\hbox{min}
\left\{1, {n_{\mathrm in}\overn_{\mathrm out}}, {\textstyle{1\over2}}\left|{n_{\mathrm in}\overn_{\mathrm out}}-1\right|\right\}
\;
{ n_{\mathrm out}^2 \over n_{\mathrm in}^2+n_{\mathrm out}^2} \; \omega_{\mathrm out} t_0 \gg 1.
\end{equation}
The Bogolubov coefficient then becomes
\begin{equation}
|\beta(\vec k_{\mathrm in} , \vec k_{\mathrm out} )|^2
\approx
\exp\left(-4\pi \; {\hbox{min}\{n_{\mathrm in},n_{\mathrm out}\} \;
n_{\mathrm out} \over n_{\mathrm in}^2 + n_{\mathrm out}^2} \; \omega_{\mathrm out} t_0
\right) \;
{V\over (2\pi)^3 } \;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ),
\end{equation}
This implies exponential suppression of photon production for
frequencies {\em large} compared to
\begin{equation}
\Omega_{\mathrm adiabatic} \equiv
{1\over 2\pi t_0}
{n_{\mathrm in}^2 + n_{\mathrm out}^2 \over n_{\mathrm out} \;\hbox{min}\{n_{\mathrm in},n_{\mathrm out},{\textstyle{1\over2}}|n_{\mathrm in}-n_{\mathrm out}|\} }.
\end{equation}
Eberlein's model~\cite{Eberlein1,Eberlein2,Eberlein3} for
sonoluminescence explicitly makes the adiabatic approximation and this
effect is the underlying reason why photon production is so small in
that model; of course the technical calculations of Eberlein's model
also include the finite volume effects due to finite bubble radius
which somewhat obscures the underlying physics of the adiabatic
approximation.
\subsection{The transition region}
Generally there will be a transition region between $\Omega_{\mathrm
sudden}$ and $\Omega_{\mathrm adiabatic}$ over which the Bogolubov
coefficient has a different structure from either of the asymptotic
limits. In this transition region the Bogolubov coefficient is well
approximated by a monomial in $\omega$ multiplied by an exponential
suppression factor, but the e-folding rate in the exponential is
different from that in the adiabatic regime. Fortunately, we will not
need any detailed information about this region, beyond the fact that
there is an exponential suppression.
\subsection{Spectrum}
The number spectrum of the emitted photons is
\begin{equation}
\label{E:spectrum0}
{dN(\vec k_{\mathrm out} )\over d^3\vec k_{\mathrm out} }
=
\int|\beta(\vec k_{\mathrm in} ,\vec k_{\mathrm out} )|^{2}
d^3\vec k_{\mathrm in} .
\end{equation}
Taking into account that $d^3\vec k_{\mathrm out} = 4\pi k_{\mathrm out}^2
\; d k_{\mathrm out} $ this easily yields
\begin{equation}
\label{E:spectrum1}
{dN(\omega_{\mathrm out} )\over d\omega_{\mathrm out} }
=
{
\sinh^2\left(
{ \textstyle\pi\; |n_{\mathrm in} -n_{\mathrm out} | \; n_{\mathrm out} \; \omega_{\mathrm out} t_0
\over
\textstyle (n_{\mathrm in}^2+n_{\mathrm out}^2)
}
\right)
\over
\sinh\left(
{\textstyle2\pi n_{\mathrm in} \; n_{\mathrm out} \omega_{\mathrm out} t_0
\over
\textstyle(n_{\mathrm in}^2+n_{\mathrm out}^2)
}
\right) \;
\sinh\left(
{\textstyle 2\pi n_{\mathrm out} ^{2}\; \omega_{\mathrm out} t_0
\over
\textstyle (n_{\mathrm in}^2+n_{\mathrm out}^2)
}
\right)
} \;
{2V\over (2\pi)^3 } \;
4\pi \omega_{\mathrm out} ^2 \; n_{\mathrm out} ^3 .
\end{equation}
(Here the factor 2 is introduced by hand by taking into account
the 2 photon polarizations). For low frequencies (where the sudden
approximation is valid) this is a phase-space limited spectrum with
a prefactor that depends only on the overall change of refractive
index. For high frequencies (where the adiabatic approximation
holds sway) the spectrum is cutoff in an exponential manner depending
on the rapidity of the change in refractive index.
A sample spectrum is plotted in figure \ref{F:toy-spectrum-1}. For
comparison figure \ref{F:toy-spectrum-2} shows a Planckian spectrum
with the same exponential falloff at high frequencies, while the two
curves are superimposed in figure \ref{F:toy-spectrum-3}.
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=sl-spectrum.eps,height=10cm,angle=270}\hfil}
\caption{%
Number spectrum (photons per unit volume) for $ n_{\mathrm in} =1$, $ n_{\mathrm out} =2$.
The horizontal axis is $\omega_{\mathrm out}$ and is expressed in PHz.
The typical timescale $t_{0}$ is set equal to one $fs$.
The vertical axis is in arbitrary units.}
\label{F:toy-spectrum-1}
\end{figure}
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=pl-spectrum.eps,height=10cm,angle=270}\hfil}
\caption{%
Number spectrum for a Planck blackbody curve with $k_B T=
(n_{\mathrm in}^2+n_{\mathrm out}^2)/ (4\pi t_0\; n_{\mathrm out} \; \hbox{min}\{n_{\mathrm in},n_{\mathrm out},{\textstyle{1\over2}}|n_{\mathrm in}-n_{\mathrm out}|\} ). $
The horizontal axis is $\omega_{\mathrm out}$ and is expressed in PHz. The
typical timescale $t_{0}$ is set equal to one fs. The vertical axis
is in arbitrary units (but with the same normalization as figure 1).}
\label{F:toy-spectrum-2}
\end{figure}
\begin{figure}
[htb]
\vbox{\hfil\psfig{figure=sl+pl-spectrum.eps,height=10cm,angle=270}\hfil}
\caption{%
Superimposed number spectra (the Planck spectrum is the dotted one).
This figure demonstrates the similar high-frequency behaviour although
low energy behaviour is different (quadratic versus linear).}
\label{F:toy-spectrum-3}
\end{figure}
\subsection{Lessons from this toy model}
{\bf Lesson 1:} This is only a toy model, but we feel it adequately
proves that efficient photon production occurs only in the sudden
approximation, and that photon production is suppressed in the
adiabatic regime. The particular choice of profile $\epsilon(\tau)$
was merely a convenience, it allowed us to get analytic exact results,
but it is not a critical part of the analysis. One might worry that
the results of this toy model are specific to the choice of profile
(\ref{E:profile}). That the results are more general can be
established by analyzing general bounds on the Bogolubov coefficients,
which is equivalent to studying general bounds on one-dimensional
potential scattering~\cite{Visser}. We shall here quote only the key
result that for any monotonic change in the dielectric constant the
sudden approximation provides a strict upper bound on the magnitude of
the Bogolubov coefficients~\cite{Visser}.
{\bf Lesson 2:} Eberlein's model for
sonoluminescence~\cite{Eberlein1,Eberlein2,Eberlein3} explicitly
uses the adiabatic approximation. For arbitrary adiabatic changes
we expect the exponential suppression to still hold with $\rho$
now being some measure of the timescale over which the refractive
index changes.
{\bf Lesson 3:} Schwinger's model for
sonoluminescence~\cite{Sc1,Sc2,Sc3,Sc4,Sc5,Sc6,Sc7} implicitly uses
the sudden approximation. It is only for the sudden approximation
that we recover Schwinger's phase-space limited spectrum. For
arbitrary changes the sudden approximation provides a rigorous
upper bound on photon production. It is only in the sudden
approximation that efficient conversion of zero-point fluctuations
to real photons takes place. Though this result is derived here
only for a particularly simple toy model we expect this part of
the analysis to be completely generic. We expect that any mechanism
for converting zero-point fluctuations to real photons will exhibit
similar effects.
\subsection{Extensions of this toy model}
The major weaknesses of the toy model are that it currently includes
neither dispersive effects nor finite volume effects. Including
dispersive effects amounts to including condensed matter physics
by letting the refractive index itself be a function of frequency.
To do this carefully requires a very detailed understanding of the
condensed matter physics, which is quite beyond the scope of the
present paper. Instead, in this section we shall content ourselves with
making order-of-magnitude estimates using Schwinger's sharp cutoff
for the refractive index and the sudden approximation.
The second issue, that of finite volume effects, is addressed more
carefully in the companion paper~\cite{Companion}. Finite volume
effects are expected to be significant but not overwhelmingly large.
{From} estimates of the available Casimir energy developed
in~\cite{MV}, the fractional change in available Casimir energy
due to finite volume effects is expected to be of order $1/(K R)
= \hbox{(cutoff wavelength)} /(2\pi \hbox{(minimum bubble radius)})$
which is approximately $(300\; \hbox{nm})/(2\pi \;500\; \hbox{nm})
\approx 10\%\;$\footnote{
Here we have estimated the cutoff wavelength from the location of
the peak in the SL spectrum. If anything, this causes us to overestimate
the finite volume effects.}.
Returning to dispersive issues: if the refractive indices
were completely non-dispersive (frequency-independent), then the sudden
approximation would imply infinite energy production. In the real
physical situation $n_{\mathrm in}$ is a function of $\omega_{\mathrm in} $
and $n_{\mathrm out}$ is a function of $\omega_{\mathrm out} $. Schwinger's
sharp momentum-space cutoff for the refractive index is equivalent,
in this formalism, to the choice
\begin{equation}
n_{\mathrm in}(k) = n_{\mathrm in} \; \Theta(K_{\mathrm in} -k) + 1 \; \Theta(k-K_{\mathrm in}),
\end{equation}
\begin{equation}
n_{\mathrm out}(k) = n_{\mathrm out} \; \Theta(K_{\mathrm out} -k) + 1 \; \Theta(k-K_{\mathrm out}),
\end{equation}
(More complicated models for the cutoff are of course possible at the
cost of obscuring the analytic properties of the model) Although in
general the two cutoff wavenumbers, $K_{\mathrm in}$ and $K_{\mathrm out}$ can be
different, it is easy to show (using the delta function and the fact
that $\beta \rightarrow 0$ when $n_{\mathrm in}=n_{\mathrm out}=1$), that this is equivalent
to a single common cutoff $K\equiv \hbox{min}\{K_{\mathrm in},K_{\mathrm out}\}$. {From}
equation (\ref{E:sbsq}), taking into account the two photon
polarizations, one obtains
\begin{eqnarray}
|\beta(\vec k_{\mathrm in} ,\vec k_{\mathrm out} )|^{2}
&\approx&
{1 \over 2} \frac{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2}{n_{\mathrm in} n_{\mathrm out}}
{V\over(2\pi)^3}\;
\Theta(K - k_{\mathrm in} ) \; \Theta(K - k_{\mathrm out} ) \;
\delta^3(\vec k_{\mathrm in} + \vec k_{\mathrm out} ).
\label{E:largeb2}
\end{eqnarray}
As a consistency check, expression (\ref{E:largeb2}) has the desirable
property that $\beta\to0$ as $n_{\mathrm out}\ton_{\mathrm in}$: That is, if there is no
change in the refractive index, there is no particle production. In
fact the computed Bogolubov coefficient is directly related to the
physical quantities we are interested in
\begin{equation}
{dN\over d^3 \vec k_{\mathrm out}}
=\int|\beta(\omega_{\mathrm in} ,\omega_{\mathrm out} )|^{2}
\; d^3 \vec k_{\mathrm in},
\end{equation}
\begin{equation}
{dN\over d k_{\mathrm out}}
=4\pi k_{\mathrm out}^2 \; \int|\beta(\omega_{\mathrm in} ,\omega_{\mathrm out} )|^{2}
\; d^3 \vec k_{\mathrm in},
\end{equation}
\begin{equation}
{dN\over d\omega_{\mathrm out}}
=4\pi \frac{n_{\mathrm out}^3 \omega_{\mathrm out}^2}{c^3} \;
\int|\beta(\omega_{\mathrm in} ,\omega_{\mathrm out} )|^{2}
\; d^3 \vec k_{\mathrm in},
\end{equation}
\begin{equation}
N=\int {dN\over d\omega_{\mathrm out}} \; d\omega_{\mathrm out} ,
\end{equation}
and
\begin{equation}
E= \hbar \int {dN(\omega_{\mathrm out} )\over d\omega_{\mathrm out} }
\; \omega_{\mathrm out} \; d\omega_{\mathrm out} .
\end{equation}
So we can now compute the spectrum, the number, and the total energy
of the emitted photons.
\begin{eqnarray}
{dN(\omega_{\mathrm out} )\over d\omega_{\mathrm out} }
=
\frac{n_{\mathrm out}}{c} \; {dN(\omega_{\mathrm out} )\over dk_{\mathrm out} }
&=&
\frac{n_{\mathrm out}}{c} \; 4\pi k^{2}_{\mathrm out} {dN(\omega_{\mathrm out} )\over d^3{\vec k}_{\mathrm out} }
\\
&\approx&
{n_{\mathrm out} \over (2\; c)}\;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm out}\,n_{\mathrm in}}
{V\over(2\pi)^3}\; 4\pi \; k_{\mathrm out} ^2 \;
\Theta(K - k_{\mathrm out} ) \;
\\
&=&
{1\over (2\; c^3)}\;{n_{\mathrm out}^2} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}}
{V\over(2\pi)^3}\; 4\pi \; \omega_{\mathrm out} ^2 \;
\Theta\left(K - \frac{n_{\mathrm out} \omega_{\mathrm out}}{c}\right)
\end{eqnarray}
The number of emitted photons is then approximately
\begin{eqnarray}
N
&\approx&
{1\over 2} \;{n_{\mathrm out}^2} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}} \;
{V\over(2\pi)^3}\; {4\pi\over3} \; (K/n_{\mathrm out})^3
\\
&=&
{1\over 12\pi^2} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}n_{\mathrm out}} \;
{V K^3}.
\end{eqnarray}
So that for a spherical bubble
\begin{eqnarray}
N
&\approx&
{1\over 9\pi} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\over\non_{\mathrm in}} \;
(R K)^3.
\end{eqnarray}
It is important to note that the wavenumber cutoff $K$ appearing in
the above formula is not equal to the observed wavenumber cutoff
$K_ {\mathrm observed} $. The observed wavenumber cutoff is in fact the upper
wavelength measured once the photons have left the bubble and entered
the ambient medium (water), so actually
\begin{equation}
K={\omega_{\mathrm max}n_{\mathrm out}\over c}={n_{\mathrm out}\overn_{\mathrm liquid}} \; K_ {\mathrm observed} .
\end{equation}
Thus
\begin{eqnarray}
N
&\approx&
{1\over 9\pi} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\over\non_{\mathrm in}} \;
\left(R \;{n_{\mathrm out}\overn_{\mathrm liquid}} \; K_ {\mathrm observed} \right)^3.
\\
&=&
{1\over 9\pi} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}} \; n_{\mathrm out}^2 \;
\left({R \omega_{\mathrm max} \over c}\right)^3.
\end{eqnarray}
The total emitted energy is approximately
\begin{eqnarray}
E&\approx&
{1\over 2} \; \frac{n_{\mathrm out}^2}{c^3} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}}
{V\over(2\pi)^3}\; 4\pi \;
\int \hbar \omega_{\mathrm out} \; \omega_{\mathrm out} ^2 \;
\Theta\left(K - \frac{n_{\mathrm out} \omega_{\mathrm out}}{c} \right) d\omega_{\mathrm out}
\\
&=&
{\hbar\over 2} \; \frac{n_{\mathrm out}^2}{c^3} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}} \;
{V\over(2\pi)^3}\; {4\pi\over4} \; (K\;c/n_{\mathrm out})^4
\\
&=&
{1\over16\pi^2} \;
{\left(n_{\mathrm out}-n_{\mathrm in}\right)^2\overn_{\mathrm in}n_{\mathrm out}^2} \;
\hbar\; c\; K \; {V K^3}
\\
&=&
\frac{3}{4}\; N\; \hbar \omega_{\mathrm max}.
\label{E:energy}
\end{eqnarray}
So the average energy per emitted photon is approximately\footnote{%
The maximum photon energy is $\hbar \; \omega_{\mathrm max} \approx 4\;
{\rm eV}$. }
\begin{equation}
\langle E \rangle =
{3\over 4} \hbar c\; K/n_{\mathrm out} = {3\over 4} \hbar \; \omega_{\mathrm
max}\sim 3 \; {\rm eV}.
\end{equation}
Taking into account this extra factor we can now consider some numerical
estimates based on our results.
\subsection{Some numerical estimates}
In Schwinger's original model he took $n_{\mathrm gas}^{\mathrm in}\approx 1$, $n_{\mathrm gas}^{\mathrm out} \approx
n_{\mathrm liquid} \approx 1.3$, $V= (4\pi/3) R^3$, with $R \approx R_{\mathrm
max} \approx 40 \; \mu {\rm m}$ and $K\approx 2\pi/(360\; {\rm
nm})$ \cite{Sc4}. Then $KR \approx 698$. Substitution of these
numbers into equation (\ref{E:schwinger}) leads to an energy budget
suitable for about {\em three} million emitted photons.
By direct substitution in equation (\ref{E:energy}) it is easy to
check that Schwinger's results can qualitatively be recovered also in
our formalism: in our case we get about {\em 1.8 million} photons for
the same numbers of Schwinger and about {\em 4 million} photons using
the updated experimental figures $R_{\mathrm max} \approx 45 \;
\mu {\rm m}$ and $K\approx 2\pi/(300\; {\rm nm})$.
A sudden change in refractive index {\em would} indeed convert the most
of the energy budget based on static Casimir energy calculations into
real photons. This may be interpreted as an independent check on Schwinger's
estimate of the Casimir energy of a dielectric sphere. Unfortunately,
the sudden (femtosecond) change in refractive index required to
get efficient photon production is also the fly in the ointment
that kills Schwinger's original choice of parameters: The collapse
from $R_{\mathrm max}$ to $R_{\mathrm min}$ is known to require
approximately $10\; {\rm ns}$, which is far too long a timescale
to allow us to adopt the sudden approximation.
In our new version of the model we have $R\approx R_{\mathrm
light-emitting-region} \approx R_{\mathrm min} \approx 500 \;
\hbox{nm} $ and take $K_ {\mathrm observed} \approx 2\pi/(200\; {\mathrm nm})$ so
that $K_ {\mathrm observed} R\approx 5 \pi \approx 15$. To get about one million
photons we now need, for instance, $n_{\mathrm in}\approx 1$ and $n_{\mathrm out}\approx 12$,
or $n_{\mathrm in}\approx 2\times 10^4$ and $n_{\mathrm out}\approx 1$, or even $n_{\mathrm out} \approx
25$ and $n_{\mathrm in}\approx 71$, though many other possibilities could be
envisaged. In particular, the first set of values could correspond to
a change of the refractive index at the van der Waals hard core due to
a sudden compression {\em e.g.}, generated by a shock wave. In this
framework it is obvious that the most favorable composition for the
gas would be a noble gas since this mechanism would be most effective
if the gas could be enormously compressed without being easily
ionizable\footnote{%
To ionize Argon requires $15 \; {\rm eV}\approx 10^5 \; {\rm K}/k_{B}$
per atom. This energy could be provided either from a heat bath at
this temperature (Bremsstrahlung) or from kinetic energy given by
atomic collisions. Both of these possibilities require very extreme
hypotheses.}.
Note that the estimated values of $n_{\mathrm gas}^{\mathrm out}$ and $n_{\mathrm gas}^{\mathrm in}$ are extremely
sensitive to the precise choice of cutoff, and the size of the light
emitting region, and that the approximations used in taking the
infinite volume limit underlying the use of our homogeneous dielectric
model are uncontrolled. (The complications attendant on any attempt at
including finite volume effects are sufficiently complex as to warrant
being relegated to a separate technical paper.~\cite{Companion}) We
should not put to much credence in the particular numerical value of
$n_{\mathrm gas}^{\mathrm out}$ estimated by these means, but should content ourselves with
this qualitative message: We need the refractive index of the contents
of the gas bubble to change dramatically and rapidly to generate the
photons. Compare this with the calculation and arguments presented by
Yablonovitch~\cite{Yablonovitch}, who points out that ionization
processes can and often do cause such sudden drops in the refractive
index.
As a final remark we stress that equation (\ref{E:schwinger}) and
equation (\ref{E:energy}) are not quite identical. The volume term for
photon production that we have just derived [equation
(\ref{E:energy})] is of second order in $(n_{\mathrm in}-n_{\mathrm out})$ and not of first
order like equation (\ref{E:schwinger}). This is ultimately due to the
fact that the interaction term responsible for converting the initial
energy in photons is a pairwise squeezing operator
(see~\cite{2gamma}). Equation (\ref{E:energy}) demonstrates that any
argument that attempts to deny the relevance of volume terms to
sonoluminescence due to their dependence on $(n_{\mathrm in}-n_{\mathrm out})$ has to be
carefully reassessed. In fact what you measure when the refractive
index in a given volume of space changes is {\em not} directly the
change in the static Casimir energy of the ``in'' state, but rather
the fraction of this static Casimir energy that is converted into
photons. We have just seen that once conversion efficiencies are taken
into account, the volume dependence is conserved, but not the power in
the difference of the refractive index. Indeed the dependence of
$|\beta|^2$ on $(n_{\mathrm in}-n_{\mathrm out})^2$, and the symmetry of the former under the
interchange of ``in'' and ``out'' states, also proves that it is the
amount of change in the refractive index and not its ``direction''
that governs particle production. This apparent paradox is easily
solved by taking into account that the main source of energy is the
acoustic field and that the amount of this energy actually converted
in photons during each cycle is a very small fraction of the total
acoustic energy.
\subsection{Estimating the number of photons}
Using the above as a guide to the appropriate starting point, we can
now systematically explore the relationship between the in and out
refractive indexes and the number of photons produced. Using
$K_ {\mathrm observed} R \approx 15$ we get
\begin{equation}
N = {119\overn_{\mathrm liquid}^3}\; (n_{\mathrm out}-n_{\mathrm in})^2 \; {n_{\mathrm out}^2 \over n_{\mathrm in}}.
\end{equation}
This equation can be algebraically solved for $n_{\mathrm in}$ as a function of
$n_{\mathrm out}$ and $N$. (It's a quadratic.) For $N=10^6$ emitted photons the
result is plotted in figure (\ref{F:photons-1}). For any specified
value of $n_{\mathrm out}$ there are exactly two values of $n_{\mathrm in}$ that lead to one
million emitted photons. To understand the qualitative features of
this diagram we consider three sub-regions.
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=photons-1.eps,height=5cm}\hfil}
\caption{%
The initial refractive index $n_{\mathrm in}$ plotted as a function of $n_{\mathrm out}$ when
one million photons are emitted in the sudden approximation.}
\label{F:photons-1}
\end{figure}
First, if $n_{\mathrm in} \ll n_{\mathrm out}$ then we can approximate
\begin{equation}
n_{\mathrm in} \approx {119 \; n_{\mathrm out}^4 \over n_{\mathrm liquid}^3 \; N}.
\end{equation}
This corresponds to the region near the origin, and we focus on this
region in figure (\ref{F:photons-2}).
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=photons-2.eps,height=5cm}\hfil}
\caption{%
The initial refractive index $n_{\mathrm in}$ plotted as a function of $n_{\mathrm out}$ when
one million photons are emitted in the sudden approximation. Here we
focus on the branch that approaches the origin.}
\label{F:photons-2}
\end{figure}
Second, if $n_{\mathrm in} \gg n_{\mathrm out}$ then we can approximate
\begin{equation}
n_{\mathrm in} \approx {n_{\mathrm liquid}^3 \; N \over 119 \; n_{\mathrm out}^2}.
\end{equation}
This corresponds to the region near the $y$ axis, and we focus on this
region in figure (\ref{F:photons-3}).
\begin{figure}[htb]
\vbox{\hfil\psfig{figure=photons-3.eps,height=5cm}\hfil}
\caption{%
The initial refractive index $n_{\mathrm in}$ plotted as a function of $n_{\mathrm out}$ when
one million photons are emitted in the sudden approximation. Here we
focus on the branch that approaches the $y$ axis.}
\label{F:photons-3}
\end{figure}
Third, if $n_{\mathrm in} \approx n_{\mathrm out}$ then we can approximate
\begin{equation}
N \approx {119\over n_{\mathrm liquid}^3} \; (n_{\mathrm in} - n_{\mathrm out})^2 \; n_{\mathrm out},
\label{nn}
\end{equation}
so that
\begin{equation}
n_{\mathrm in} \approx n_{\mathrm out} \pm \sqrt{N \; n_{\mathrm liquid}^3 \over 119 \; n_{\mathrm out}}.
\label{nnn}
\end{equation}
This corresponds to the region near the asymptote $n_{\mathrm in}=n_{\mathrm out}$.
Thus to get a million photons emitted from the van der Waals hard core
in the sudden approximation requires a significant (but not enormous)
change in refractive index. There are many possibilities consistent
with the present model and the experimental data.
\section{Experimental features and possible tests}
Our proposal shares with other proposals based on the dynamical
Casimir effect the main points of strength previously sketched. On
the other hand we feel it important to stress that the model we have
developed implies a much more complex and rich collection of physical
effects due to the fact that photon production from vacuum is no
longer due to the simple motion of the bubble boundary. The model
indicates that a viable Casimir route to SL cannot avoid a ``fierce
marriage'' with features related to condensed matter physics. As a
consequence our proposal is endowed both with general characteristics,
coming from its Casimir nature, and with particular ones coming from
the details underlying the sudden change in the refractive index.
Although the calculation presented above is just a ``probe'', we
can see that it is already able to make some general predictions
that one can expect to see confirmed in a more complete approach.
First of all the photon number spectrum the model predicts is not
a black body. It is polynomial at low frequencies ($\omega^{2}$ in
the infinite volume approximation of this paper), and in principle
this difference can be experimentally detected. (The same qualitative
prediction can be found in Sch\"utzhold\ {\em et al.} \cite{SPS}.)
Moreover the spectrum is expected to be a power law dramatically
ending at frequencies corresponding to the physical wave-number
cutoff $K$ (at which the refractive indices go to 1). This cutoff
implies the absence of hard UV photons and hence, in accordance
with experiments, the absence of dissociation phenomena in the
water surrounding the bubble.
It is the sudden approximation adopted in this paper that makes it
possible to mimic the experimentally observed spectrum. For either a
rise in refractive index from $1$ to $12$, or a drop in refractive
index from $2\times 10^4$ to $1$ one can produce approximately one
million photons with frequency mainly in the visible range. The
quasi-thermal nature of the emitted photons can be explained by the
squeezed nature of the photon pairs that are generically created via
dynamical Casimir effect (see reference \cite{2gamma}). Single-photon
measurements are then thermal but the core of the bubble is not
required to achieve the tremendous physical temperatures envisaged by
other models. The apparent temperature measured in single-photon
observables can be instead linked to the degree of squeezing of the
photon-pairs. As it will be explained in \cite{2gamma}, two-photon
observables do not exhibit the same thermal statistics, and therefore
measurements of suitable two-photon observables provide a useful
diagnostic for SL models based on the dynamical Casimir effect. This
is a general feature of all models based on photon creation from the
QED vacuum and hence it can be used as a definitive test of the
presence of a dynamical Casimir effect.
In this type of model, the flash of photons is predicted to occur at
the end of the collapse, the scale of emission zone is of the order of
$500\; \hbox{nm}$, and the timescale of emission is very short (with a
rise-time of the order of femtoseconds, though the flash duration may
conceivably be somewhat longer~\footnote{It would be far too naive to
assume that femtosecond changes in the refractive index lead to pulse
widths limited to the femtosecond range. There are many condensed
matter processes that can broaden the pulse width however rapidly it
is generated. Indeed, the very experiments that seek to measure the
pulse width \cite{Flash1,Flash2} also prove that when calibrated with
laser pulses that are known to be of femtosecond timescale, the SL
system responds with light pulses on the picosecond timescale.}.
These are points in substantial agreement with observations. In the
infinite volume limit the photons emerge in strictly back-to-back
fashion. In contrast, for a finite volume bubble we have shown
in~\cite{Companion} that the size of the emitting region constrains
the model to low angular momentum for the out states. This is a very
sharp prediction that is in principle testable with a suitable
experiment devoted to the study of the angular momentum decomposition
of the outgoing radiation.
Regarding other experimental dependencies, such as the temperature of
the water or the role of noble gases, we can give general arguments
but a truly predictive analysis can be done only after focussing on a
specific mechanism for changing the refractive index.
For instance, the presence of noble gas is likely to change
solubilities of gas in the bubble, and this can vary both bubble
dynamics and the sharpness of the boundary. Alternatively, a small
percentage of noble gas in air can be very important in the behavior
of its dielectric constant at high pressure. Indeed, while small
admixtures of noble gas will not significantly alter the
zero-frequency refractive index, from the Casimir point of view the
behaviour of the refractive index over the entire frequency range up
to the cutoff is important.
Finally, the temperature of water can instead affect the dynamics of
the bubble boundary by influencing the stability of the bubble,
changing either the solubility of air in water or the surface tension
of the latter. As observed by
Schwinger~\cite{Sc1,Sc2,Sc3,Sc4,Sc5,Sc6,Sc7}, temperature can also
affect the dielectric cutoff, and so temperature dependence of SL is
quite natural in Schwinger-like approaches.
\section{Discussion and Conclusions}
The present paper presents calculations of the Bogolubov coefficients
relating the two QED vacuum states appropriate to changes in the
refractive index of a dielectric bubble. We have verified by explicit
computation that photons are produced by rapid changes in the
refractive index, and are in agreement with Schwinger in that QED vacuum
effects remain a viable candidate for explaining SL. However, some
details of the particular model considered in the present paper are
somewhat different from that originally envisaged by Schwinger. Based
largely on the fact that efficient photon production requires
timescales of the order of femtoseconds we were led to consider rapid
changes in the refractive index as the gas bubble bounces off the van
der Waals hard core. It is important to realize that the speed of sound
in the gas bubble can become relativistic at this stage.
A key lesson learned from this paper is that in order that the
conversion of zero-point fluctuations to real photons be relevant for
sonoluminescence we would want the sudden approximation to hold for
photons all the way out to the cutoff ($200\;\hbox{nm}$; corresponding
to a period of $0.66 \times 10^{-15} \; \hbox{seconds}$). That's a
{\em femtosecond} timescale. This implies that if conversion of
zero-point fluctuations to real photons is a significant part of the
physics of sonoluminescence then the refractive index must be changing
significantly on femtosecond timescales. {\em Thus the changes in
refractive index cannot be just due to the motion of the bubble wall.}
(The bubble wall is moving at most at Mach 4~\cite{Physics-Reports},
for a $1\;\mu\hbox{m}$ bubble this gives a collapse timescale of
$10^{-10}$ seconds, about 100 picoseconds.) In this regard, the
comments of Yablonovitch~\cite{Yablonovitch} are particularly
useful. Yablonovitch points out that, for example, sudden ionization
of a gas can lead to substantial changes in the refractive index on
the sub-picosecond timescale. Nevertheless we do not necessarily
commit ourselves to ionization as being the relevant process in
sonoluminescence, and are quite content with any rapid change in
refractive index, however generated.
This suggests a slightly different physical model from Schwinger's
original suggestion: Certainly the Casimir energy changes as the
bubble collapses, but it is only in the sudden approximation that we
can justify converting almost all of the change in Casimir energy to
real photons. We thus suggest that one should not be focusing on the
actual collapse of the bubble, but rather the way in which the
refractive index changes as a function of space and time: As the
bubble collapses the gases inside are compressed, and although the
refractive index for air (plus noble gas contaminants) is 1 at STP it
should be no surprise to see the refractive index of the trapped gas
undergoing major changes during the collapse process---especially near
the moment of maximum compression when the molecules in the gas bounce
off the van der Waals' hard core repulsive potential.
Thus attempts at using the dynamical Casimir effect to explain SL are
now much more tightly constrained than previously. We have shown that
any plausible model using the dynamical Casimir effect to explain SL
must use the sudden approximation, and must have very rapid changes in
the refractive index with a timescale of femtoseconds.
If the light is being emitted only at the core bounce, at $R\approx
500 \; {\rm nm}$, then we can get the timescales we want
(femtoseconds) without superluminal effects, but we are rather limited
in the amount of angular momentum we can get out. Of course, the
present model is nowhere near a complete theory: Presently we can
relate only the asymptotic ``in'' states to the asymptotic ``out''
states via these Bogolubov coefficients. A complete theory of SL will
need to address {\em much more} specific timing information and this
will require a fully dynamical approach (from the QFT point of view)
and a deeper understanding (from the condensed matter side) of the
precise spatio-temporal dependence of the refractive index as the
bubble collapses. In absence of such a more detailed description the
present calculation is a useful first step. Moreover it allows us to
specify certain basic ``signatures'' of the effect that may be
amenable to experimental test. To this end we have developed use of
two-photon statistics as a diagnostic for the dynamical Casimir
effect~\cite{2gamma}. In this paper we have addressed the basic
physical scenario; all technical complications due to finite volume
effects are relegated to a companion paper~\cite{Companion}. We feel
that, as an explanation for sonoluminescence, we have now driven
models based on the dynamical Casimir effect into a relatively small
region of parameter space, and are hopeful of experimental
verification (or falsification) in the not too distant future.
Stripped to its fundamentals, we therefore view the Schwinger
mechanism as this: Bubble collapse leads to changes in the
spatio-temporal distribution of the refractive index, both via
physical movement of the dielectrics, and through the time-dependent
properties of the dielectrics. Changes in the refractive index drive
changes in the distribution of zero-point modes, and this change in
zero-point modes is reflected in real photon production.
In light of these observations we think that one can also derive a
general conclusion about the long-standing debate on the actual value
of the static Casimir energy and its relevance to sonoluminescence:
{\em Sonoluminescence cannot be directly related to the static Casimir
effect}. (The static Casimir effect is relevant only insofar as it
gives an approximate value for the energy budget). We hope that the
investigation of this paper will convince everyone that only models
dealing with the actual mechanism of particle creation (a mechanism
which must have the qualities we have discussed) will be able to
eventually prove or disprove the pertinence of the physics of the
quantum vacuum to Sonoluminescence. This implies that continuing
debate about the static Casimir effect can be now seen as marginal and
irrelevant with respect to the real physical problems of SL.
In conclusion the present calculation (limited though it may be)
represents an important advance: There now can be no doubt that
changes of the refractive index of the gas inside the bubble lead to
production of real photons---the controversial issues now move to
quantitative ones of precise fitting of the observed experimental
data. We are hopeful that more detailed models and data fitting will
provide better explanations of the details of the SL effect, and
specifically wish to assert that models based on the QED vacuum remain
viable.
\section*{Acknowledgments}
This research was supported by the Italian Ministry of Science (DWS,
SL, and FB), and by the US Department of Energy (MV). MV particularly
wishes to thank SISSA (Trieste, Italy) and Victoria University (Te
Whare Wananga o te Upoko o te Ika a Maui; Wellington, New Zealand) for
hospitality during various stages of this work. SL wishes to thank
Washington University for its hospitality. DWS and SL wish to thank
E.~Tosatti for useful discussions. SL wishes to thank M.~Bertola and
B.~Bassett for comments and suggestions. SL also wishes to thank
R.~Sch\"utzhold\ and G.~Plunien for illuminating discussion on the
relation between their results and Eberlein's calculations. All
authors wish to thank G.~Barton for his interest and encouragement.
Finally, the comments and interest of K.~Milton are appreciated.
|
\@ifstar{\@ssection}{\@section}{\@ifstar{\@ssection}{\@section}}
\def\@section#1
\EveryMac
\vskip-\lastskip\vskip\two
\LastMac=\Hae
\tenpoint\bls{22pt
\bf
\Raggedright
\ifAutoNumber
\advance\Sec by 1
\noindent\number\Sec\hskip 1pc \uppercase{#1}
\SecSec=0
\else
\noindent \uppercase{#1}
\fi
\nobreak
}
\def\@ssection#1
\EveryMac
\ifnum\LastMac=\Hae \vskip-\lastskip\vskip\half
\else \vskip-\lastskip\vskip\onehalf
\fi
\LastMac=\Hae
\tenpoint\bls{22pt
\bf
\Raggedright
\noindent\uppercase{#1}
}
\def\subsection#1
\EveryMac
\ifnum\LastMac=\Hae \vskip-\lastskip\vskip\half
\else \vskip-\lastskip\vskip\onehalf
\fi
\LastMac=\Hbe
\tenpoint\bls{22pt
\bf
\Raggedright
\ifAutoNumber
\advance\SecSec by 1
\noindent\number\Sec.\number\SecSec
\hskip 1pc #1
\SecSecSec=0
\else
\noindent #1
\fi
\nobreak
}
\def\subsubsection#1
\EveryMac
\ifnum\LastMac=\Hbe \vskip-\lastskip\vskip\half
\else \vskip-\lastskip\vskip\onehalf
\fi
\LastMac=\Hce
\ninepoint\bls{22pt
\it
\Raggedright
\ifAutoNumber
\advance\SecSecSec by 1
\noindent\number\Sec.\number\SecSec.\number\SecSecSec
\hskip 1pc #1
\else
\noindent #1
\fi
\nobreak
}
\def\paragraph#1
\EveryMac
\vskip-\lastskip\vskip\one
\LastMac=\Hde
\ninepoint\bls{22pt
\noindent \it #1
\rm
}
\def\tx
\EveryMac
\ifnum\LastMac=\Lie \vskip-\lastskip\vskip\half\fi
\ifnum\LastMac=\Hae \nobreak\vskip-\lastskip\vskip\half\fi
\ifnum\LastMac=\Hbe \nobreak\vskip-\lastskip\vskip\half\fi
\ifnum\LastMac=\Hce \nobreak\vskip-\lastskip\vskip\half\fi
\ifnum\LastMac=\Lie \else \noindent\fi
\LastMac=\Txe
\ninepoint\bls{22pt
\rm
}
\def\item
\par
\EveryMac
\ifnum\LastMac=\Lie
\else \vskip-\lastskip\vskip\half
\fi
\LastMac=\Lie
\ninepoint\bls{22pt
\rm
}
\def\bibitem
\par
\EveryMac
\ifnum\LastMac=\Bbe
\else \vskip-\lastskip\vskip\half
\fi
\LastMac=\Bbe
\Hang{1.5em}{1}
\eightpoint\bls{20pt
\Raggedright
\noindent \rm
}
\newtoks\CatchLine
\def\@journal{Mon.\ Not.\ R.\ Astron.\ Soc.\ }
\def\@pubyear{1994}
\def\@pagerange{000--000}
\def\@volume{000}
\def\@microfiche{} %
\def\pubyear#1{\gdef\@pubyear{#1}\@makecatchline}
\def\pagerange#1{\gdef\@pagerange{#1}\@makecatchline}
\def\volume#1{\gdef\@volume{#1}\@makecatchline}
\def\microfiche#1{\gdef\@microfiche{and Microfiche\ #1}\@makecatchline}
\def\@makecatchline{%
\global\CatchLine{%
{\rm \@journal {\bf \@volume},\ \@pagerange\ (\@pubyear)\ \@microfiche}}%
}
\@makecatchline
\newtoks\LeftHeader
\def\shortauthor#1
\global\LeftHeader{#1}
}
\newtoks\RightHeader
\def\shorttitle#1
\global\RightHeader{#1}
}
\def\PageHead
\EveryMac
\ifnum\HeaderNumber=1 \Pagehead
\else \Catchline
\fi
}
\def\Catchline{%
\vbox to 0pt{\vskip-22.5pt
\hbox to \PageWidth{\vbox to8.5pt{}\noindent
\eightpoint\the\CatchLine\hfill}\vss}
\nointerlineskip
}
\def\Pagehead{%
\ifodd\pageno
\vbox to 0pt{\vskip-22.5pt
\hbox to \PageWidth{\vbox to8.5pt{}\elevenpoint\it\noindent
\hfill\the\RightHeader\hskip1.5em\rm\folio}\vss}
\else
\vbox to 0pt{\vskip-22.5pt
\hbox to \PageWidth{\vbox to8.5pt{}\elevenpoint\rm\noindent
\folio\hskip1.5em\it\the\LeftHeader\hfill}\vss}
\fi
\nointerlineskip
}
\def\PageFoot{}
\def\authorcomment#1{%
\gdef\PageFoot{%
\nointerlineskip%
\vbox to 22pt{\vfil%
\hbox to \PageWidth{\elevenpoint\rm\noindent \hfil #1 \hfil}}%
}%
}
\everydisplay{\displaysetup}
\newif\ifeqno
\newif\ifleqno
\def\displaysetup#1$${%
\displaytest#1\eqno\eqno\displaytest
}
\def\displaytest#1\eqno#2\eqno#3\displaytest{%
\if!#3!\ldisplaytest#1\leqno\leqno\ldisplaytest
\else\eqnotrue\leqnofalse\def#2}\fi{#2}\def\eq{#1}\fi
\generaldisplay$$}
\def\ldisplaytest#1\leqno#2\leqno#3\ldisplaytest{%
\def\eq{#1}%
\if!#3!\eqnofalse\else\eqnotrue\leqnotrue
\def#2}\fi{#2}\fi}
\def\generaldisplay{%
\ifeqno \ifleqno
\hbox to \hsize{\noindent
$\displaystyle\eq$\hfil$\displaystyle#2}\fi$}
\else
\hbox to \hsize{\noindent
$\displaystyle\eq$\hfil$\displaystyle#2}\fi$}
\fi
\else
\hbox to \hsize{\vbox{\noindent
$\displaystyle\eq$\hfil}}
\fi
}
\def\@notice{%
\par\vskip-\lastskip\vskip\two%
\bls{12pt}%
\noindent\tenrm This paper has been produced using the Blackwell
Scientific Publications \TeX\ macros.%
}
\outer\def\@notice\par\vfill\supereject\end{\@notice\par\vfill\supereject\end}
\everyjob{%
\Warn{Monthly notices of the RAS journal style (\@typeface)\space
v\@version,\space \@verdate.}\Warn{}%
}
\newif\if@debug \@debugfalse
\def\Print#1{\if@debug\immediate\write16{#1}\else \fi}
\def\Warn#1{\immediate\write16{#1}}
\def\immediate\write-1#1{}
\newcount\Iteration
\newif\ifFigureBoxes
\FigureBoxestrue
\def\Single{0} \def\Double{1}
\def\Figure{0} \def\Table{1}
\def\InStack{0}
\def1{1}
\def2{2}
\def3{3}
\newcount\TEMPCOUNT
\newdimen\TEMPDIMEN
\newbox\TEMPBOX
\newbox\VOIDBOX
\newcount\LengthOfStack
\newcount\MaxItems
\newcount\StackPointer
\newcount\Point
\newcount\NextFigure
\newcount\NextTable
\newcount\NextItem
\newcount\StatusStack
\newcount\NumStack
\newcount\TypeStack
\newcount\SpanStack
\newcount\BoxStack
\newcount\ItemSTATUS
\newcount\ItemNUMBER
\newcount\ItemTYPE
\newcount\ItemSPAN
\newbox\ItemBOX
\newdimen\ItemSIZE
\newdimen\PageHeight
\newdimen\TextLeading
\newdimen\Feathering
\newcount\LinesPerPage
\newdimen\ColumnWidth
\newdimen\ColumnGap
\newdimen\PageWidth
\newdimen\BodgeHeight
\newcount\Leading
\newdimen\ZoneBSize
\newdimen\TextSize
\newbox\ZoneABOX
\newbox\ZoneBBOX
\newbox\ZoneCBOX
\newif\ifFirstSingleItem
\newif\ifFirstZoneA
\newif\ifMakePageInComplete
\newif\ifMoreFigures \MoreFiguresfalse
\newif\ifMoreTables \MoreTablesfalse
\newif\ifFigInZoneB
\newif\ifFigInZoneC
\newif\ifTabInZoneB
\newif\ifTabInZoneC
\newif\ifZoneAFullPage
\newbox\MidBOX
\newbox\LeftBOX
\newbox\RightBOX
\newbox\PageBOX
\newif\ifLeftCOL
\LeftCOLtrue
\newdimen\ZoneBAdjust
\newcount\ItemFits
\def1{1}
\def2{2}
\def\LineAdjust#1{%
\global\ZoneBAdjust=#1\TextLeading
}
\MaxItems=15
\NextFigure=0
\NextTable=1
\BodgeHeight=6pt
\TextLeading=11pt
\Leading=11
\Feathering=0pt
\LinesPerPage=61
\topskip=\TextLeading
\ColumnWidth=20pc
\ColumnGap=2pc
\def\vskip \TextLeading plus \TextLeading minus 4pt{\vskip \TextLeading plus \TextLeading minus 4pt}
\FigureBoxesfalse
\parskip=0pt
\parindent=18pt
\widowpenalty=0
\clubpenalty=10000
\tolerance=1500
\hbadness=1500
\abovedisplayskip=6pt plus 2pt minus 2pt
\belowdisplayskip=6pt plus 2pt minus 2pt
\abovedisplayshortskip=6pt plus 2pt minus 2pt
\belowdisplayshortskip=6pt plus 2pt minus 2pt
\PageHeight=\TextLeading
\multiply\PageHeight by \LinesPerPage
\advance\PageHeight by \topskip
\PageWidth=2\ColumnWidth
\advance\PageWidth by \ColumnGap
\newcount\DUMMY \StatusStack=\allocationnumber
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \NumStack=\allocationnumber
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \TypeStack=\allocationnumber
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \SpanStack=\allocationnumber
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newcount\DUMMY \newcount\DUMMY \newcount\DUMMY
\newbox\DUMMY \BoxStack=\allocationnumber
\newbox\DUMMY \newbox\DUMMY \newbox\DUMMY
\newbox\DUMMY \newbox\DUMMY \newbox\DUMMY
\newbox\DUMMY \newbox\DUMMY \newbox\DUMMY
\newbox\DUMMY \newbox\DUMMY \newbox\DUMMY
\newbox\DUMMY \newbox\DUMMY \newbox\DUMMY
\def\immediate\write-1{\immediate\write-1}
\def\GetItemAll#1{%
\GetItemSTATUS{#1}
\GetItemNUMBER{#1}
\GetItemTYPE{#1}
\GetItemSPAN{#1}
\GetItemBOX{#1}
}
\def\GetItemSTATUS#1{%
\Point=\StatusStack
\advance\Point by #1
\global\ItemSTATUS=\count\Point
}
\def\GetItemNUMBER#1{%
\Point=\NumStack
\advance\Point by #1
\global\ItemNUMBER=\count\Point
}
\def\GetItemTYPE#1{%
\Point=\TypeStack
\advance\Point by #1
\global\ItemTYPE=\count\Point
}
\def\GetItemSPAN#1{%
\Point\SpanStack
\advance\Point by #1
\global\ItemSPAN=\count\Point
}
\def\GetItemBOX#1{%
\Point=\BoxStack
\advance\Point by #1
\global\setbox\ItemBOX=\vbox{\copy\Point}
\global\ItemSIZE=\ht\ItemBOX
\global\advance\ItemSIZE by \dp\ItemBOX
\TEMPCOUNT=\ItemSIZE
\divide\TEMPCOUNT by \Leading
\divide\TEMPCOUNT by 65536
\advance\TEMPCOUNT by 1
\ItemSIZE=\TEMPCOUNT pt
\global\multiply\ItemSIZE by \Leading
}
\def\JoinStack{%
\ifnum\LengthOfStack=\MaxItems
\Warn{WARNING: Stack is full...some items will be lost!}
\else
\Point=\StatusStack
\advance\Point by \LengthOfStack
\global\count\Point=\ItemSTATUS
\Point=\NumStack
\advance\Point by \LengthOfStack
\global\count\Point=\ItemNUMBER
\Point=\TypeStack
\advance\Point by \LengthOfStack
\global\count\Point=\ItemTYPE
\Point\SpanStack
\advance\Point by \LengthOfStack
\global\count\Point=\ItemSPAN
\Point=\BoxStack
\advance\Point by \LengthOfStack
\global\setbox\Point=\vbox{\copy\ItemBOX}
\global\advance\LengthOfStack by 1
\ifnum\ItemTYPE=\Figure
\global\MoreFigurestrue
\else
\global\MoreTablestrue
\fi
\fi
}
\def\LeaveStack#1{%
{\Iteration=#1
\loop
\ifnum\Iteration<\LengthOfStack
\advance\Iteration by 1
\GetItemSTATUS{\Iteration}
\advance\Point by -1
\global\count\Point=\ItemSTATUS
\GetItemNUMBER{\Iteration}
\advance\Point by -1
\global\count\Point=\ItemNUMBER
\GetItemTYPE{\Iteration}
\advance\Point by -1
\global\count\Point=\ItemTYPE
\GetItemSPAN{\Iteration}
\advance\Point by -1
\global\count\Point=\ItemSPAN
\GetItemBOX{\Iteration}
\advance\Point by -1
\global\setbox\Point=\vbox{\copy\ItemBOX}
\repeat}
\global\advance\LengthOfStack by -1
}
\newif\ifStackNotClean
\def\CleanStack{%
\StackNotCleantrue
{\Iteration=0
\loop
\ifStackNotClean
\GetItemSTATUS{\Iteration}
\ifnum\ItemSTATUS=\InStack
\advance\Iteration by 1
\else
\LeaveStack{\Iteration}
\fi
\ifnum\LengthOfStack<\Iteration
\StackNotCleanfalse
\fi
\repeat}
}
\def\FindItem#1#2{%
\global\StackPointer=-1
{\Iteration=0
\loop
\ifnum\Iteration<\LengthOfStack
\GetItemSTATUS{\Iteration}
\ifnum\ItemSTATUS=\InStack
\GetItemTYPE{\Iteration}
\ifnum\ItemTYPE=#1
\GetItemNUMBER{\Iteration}
\ifnum\ItemNUMBER=#2
\global\StackPointer=\Iteration
\Iteration=\LengthOfStack
\fi
\fi
\fi
\advance\Iteration by 1
\repeat}
}
\def\FindNext{%
\global\StackPointer=-1
{\Iteration=0
\loop
\ifnum\Iteration<\LengthOfStack
\GetItemSTATUS{\Iteration}
\ifnum\ItemSTATUS=\InStack
\GetItemTYPE{\Iteration}
\ifnum\ItemTYPE=\Figure
\ifMoreFigures
\global\NextItem=\Figure
\global\StackPointer=\Iteration
\Iteration=\LengthOfStack
\fi
\fi
\ifnum\ItemTYPE=\Table
\ifMoreTables
\global\NextItem=\Table
\global\StackPointer=\Iteration
\Iteration=\LengthOfStack
\fi
\fi
\fi
\advance\Iteration by 1
\repeat}
}
\def\ChangeStatus#1#2{%
\Point=\StatusStack
\advance\Point by #1
\global\count\Point=#2
}
\def\InZoneB{1}
\ZoneBAdjust=0pt
\def\MakePage
\global\ZoneBSize=\PageHeight
\global\TextSize=\ZoneBSize
\global\ZoneAFullPagefalse
\global\topskip=\TextLeading
\MakePageInCompletetrue
\MoreFigurestrue
\MoreTablestrue
\FigInZoneBfalse
\FigInZoneCfalse
\TabInZoneBfalse
\TabInZoneCfalse
\global\FirstSingleItemtrue
\global\FirstZoneAtrue
\global\setbox\ZoneABOX=\box\VOIDBOX
\global\setbox\ZoneBBOX=\box\VOIDBOX
\global\setbox\ZoneCBOX=\box\VOIDBOX
\loop
\ifMakePageInComplete
\FindNext
\ifnum\StackPointer=-1
\NextItem=-1
\MoreFiguresfalse
\MoreTablesfalse
\fi
\ifnum\NextItem=\Figure
\FindItem{\Figure}{\NextFigure}
\ifnum\StackPointer=-1 \global\MoreFiguresfalse
\else
\GetItemSPAN{\StackPointer}
\ifnum\ItemSPAN=\Single \def\InZoneB{2}\relax
\ifFigInZoneC \global\MoreFiguresfalse\fi
\else
\def\InZoneB{1}
\ifFigInZoneB \def\InZoneB{3}\fi
\fi
\fi
\ifMoreFigures\Print{}\FigureItems\fi
\fi
\ifnum\NextItem=\Table
\FindItem{\Table}{\NextTable}
\ifnum\StackPointer=-1 \global\MoreTablesfalse
\else
\GetItemSPAN{\StackPointer}
\ifnum\ItemSPAN=\Single\relax
\ifTabInZoneC \global\MoreTablesfalse\fi
\else
\def\InZoneB{1}
\ifTabInZoneB \def\InZoneB{3}\fi
\fi
\fi
\ifMoreTables\Print{}\TableItems\fi
\fi
\MakePageInCompletefalse
\ifMoreFigures\MakePageInCompletetrue\fi
\ifMoreTables\MakePageInCompletetrue\fi
\repeat
\ifZoneAFullPage
\global\TextSize=0pt
\global\ZoneBSize=0pt
\global\vsize=0pt\relax
\global\topskip=0pt\relax
\vbox to 0pt{\vss}
\eject
\else
\global\advance\ZoneBSize by -\ZoneBAdjust
\global\vsize=\ZoneBSize
\global\hsize=\ColumnWidth
\global\ZoneBAdjust=0pt
\ifdim\TextSize<23pt
\Warn{}
\Warn{* Making column fall short: TextSize=\the\TextSize *}
\vskip-\lastskip\eject\fi
\fi
}
\def\MakeRightCol
\global\TextSize=\ZoneBSize
\MakePageInCompletetrue
\MoreFigurestrue
\MoreTablestrue
\global\FirstSingleItemtrue
\global\setbox\ZoneBBOX=\box\VOIDBOX
\def\InZoneB{2}
\loop
\ifMakePageInComplete
\FindNext
\ifnum\StackPointer=-1
\NextItem=-1
\MoreFiguresfalse
\MoreTablesfalse
\fi
\ifnum\NextItem=\Figure
\FindItem{\Figure}{\NextFigure}
\ifnum\StackPointer=-1 \MoreFiguresfalse
\else
\GetItemSPAN{\StackPointer}
\ifnum\ItemSPAN=\Double\relax
\MoreFiguresfalse\fi
\fi
\ifMoreFigures\Print{}\FigureItems\fi
\fi
\ifnum\NextItem=\Table
\FindItem{\Table}{\NextTable}
\ifnum\StackPointer=-1 \MoreTablesfalse
\else
\GetItemSPAN{\StackPointer}
\ifnum\ItemSPAN=\Double\relax
\MoreTablesfalse\fi
\fi
\ifMoreTables\Print{}\TableItems\fi
\fi
\MakePageInCompletefalse
\ifMoreFigures\MakePageInCompletetrue\fi
\ifMoreTables\MakePageInCompletetrue\fi
\repeat
\ifZoneAFullPage
\global\TextSize=0pt
\global\ZoneBSize=0pt
\global\vsize=0pt\relax
\global\topskip=0pt\relax
\vbox to 0pt{\vss}
\eject
\else
\global\vsize=\ZoneBSize
\global\hsize=\ColumnWidth
\ifdim\TextSize<23pt
\Warn{}
\Warn{* Making column fall short: TextSize=\the\TextSize *}
\vskip-\lastskip\eject\fi
\fi
}
\def\FigureItems
\Print{Considering...}
\ShowItem{\StackPointer}
\GetItemBOX{\StackPointer}
\GetItemSPAN{\StackPointer}
\CheckFitInZone
\ifnum\ItemFits=1
\ifnum\ItemSPAN=\Single
\ChangeStatus{\StackPointer}{2}
\global\FigInZoneBtrue
\ifFirstSingleItem
\hbox{}\vskip-\BodgeHeight
\global\advance\ItemSIZE by \TextLeading
\fi
\unvbox\ItemBOX\vskip \TextLeading plus \TextLeading minus 4pt
\global\FirstSingleItemfalse
\global\advance\TextSize by -\ItemSIZ
\global\advance\TextSize by -\TextLeading
\else
\ifFirstZoneA
\global\advance\ItemSIZE by \TextLeading
\global\FirstZoneAfalse\fi
\global\advance\TextSize by -\ItemSIZE
\global\advance\TextSize by -\TextLeading
\global\advance\ZoneBSize by -\ItemSIZE
\global\advance\ZoneBSize by -\TextLeading
\ifFigInZoneB\relax
\else
\ifdim\TextSize<3\TextLeading
\global\ZoneAFullPagetrue
\fi
\fi
\ChangeStatus{\StackPointer}{\InZoneB}
\ifnum\InZoneB=3 \global\FigInZoneCtrue\fi
\fi
\Print{TextSize=\the\TextSize}
\Print{ZoneBSize=\the\ZoneBSize}
\global\advance\NextFigure by 1
\Print{This figure has been placed.}
\else
\Print{No space available for this figure...holding over.}
\Print{}
\global\MoreFiguresfalse
\fi
}
\def\TableItems
\Print{Considering...}
\ShowItem{\StackPointer}
\GetItemBOX{\StackPointer}
\GetItemSPAN{\StackPointer}
\CheckFitInZone
\ifnum\ItemFits=1
\ifnum\ItemSPAN=\Single
\ChangeStatus{\StackPointer}{2}
\global\TabInZoneBtrue
\ifFirstSingleItem
\hbox{}\vskip-\BodgeHeight
\global\advance\ItemSIZE by \TextLeading
\fi
\unvbox\ItemBOX\vskip \TextLeading plus \TextLeading minus 4pt
\global\FirstSingleItemfalse
\global\advance\TextSize by -\ItemSIZE
\global\advance\TextSize by -\TextLeading
\else
\ifFirstZoneA
\global\advance\ItemSIZE by \TextLeading
\global\FirstZoneAfalse\fi
\global\advance\TextSize by -\ItemSIZE
\global\advance\TextSize by -\TextLeading
\global\advance\ZoneBSize by -\ItemSIZE
\global\advance\ZoneBSize by -\TextLeading
\ifFigInZoneB\relax
\else
\ifdim\TextSize<3\TextLeading
\global\ZoneAFullPagetrue
\fi
\fi
\ChangeStatus{\StackPointer}{\InZoneB}
\ifnum\InZoneB=3 \global\TabInZoneCtrue\fi
\fi
\global\advance\NextTable by 1
\Print{This table has been placed.}
\else
\Print{No space available for this table...holding over.}
\Print{}
\global\MoreTablesfalse
\fi
}
\def\CheckFitInZone{%
{\advance\TextSize by -\ItemSIZE
\advance\TextSize by -\TextLeading
\ifFirstSingleItem
\advance\TextSize by \TextLeading
\fi
\ifnum\InZoneB=1\relax
\else \advance\TextSize by -\ZoneBAdjust
\fi
\ifdim\TextSize<3\TextLeading \global\ItemFits=2
\else \global\ItemFits=1\fi}
}
\def\BF#1#2
\ItemSTATUS=\InStack
\ItemNUMBER=#1
\ItemTYPE=\Figure
\if#2S \ItemSPAN=\Single
\else \ItemSPAN=\Double
\fi
\setbox\ItemBOX=\vbox{}
}
\def\BT#1#2
\ItemSTATUS=\InStack
\ItemNUMBER=#1
\ItemTYPE=\Table
\if#2S \ItemSPAN=\Single
\else \ItemSPAN=\Double
\fi
\setbox\ItemBOX=\vbox{}
}
\def\BeginOpening{%
\hsize=\PageWidth
\global\setbox\ItemBOX=\vbox\bgroup
}
\let\begintopmatter=\BeginOpening
\def\EndOpening{%
\egroup
\ItemNUMBER=0
\ItemTYPE=\Figure
\ItemSPAN=\Double
\ItemSTATUS=\InStack
\JoinStack
}
\newbox\tmpbox
\def\FC#1#2#3#4{%
\ItemSTATUS=\InStack
\ItemNUMBER=#1
\ItemTYPE=\Figure
\if#2S
\ItemSPAN=\Single \TEMPDIMEN=\ColumnWidth
\else
\ItemSPAN=\Double \TEMPDIMEN=\PageWidth
\fi
{\hsize=\TEMPDIMEN
\global\setbox\ItemBOX=\vbox{%
\ifFigureBoxes
\B{\TEMPDIMEN}{#3}
\else
\vbox to #3{\vfil}%
\fi%
\eightpoint\rm\bls{10pt plus \Feathering}%
\vskip 5.5pt plus 6pt%
\setbox\tmpbox=\vbox{#4\par}%
\ifdim\ht\tmpbox>10pt
\noindent #4\par%
\else
\hbox to \hsize{\hfil #4\hfil}%
\fi%
}%
}%
\JoinStack%
\Print{Processing source for figure {\the\ItemNUMBER}}%
}
\let\figure=\FC
\def\figps#1#2#3#4{%
\ItemSTATUS=\InStack
\ItemNUMBER=#1
\ItemTYPE=\Figure
\if#2S
\ItemSPAN=\Single \TEMPDIMEN=\ColumnWidth
\else
\ItemSPAN=\Double \TEMPDIMEN=\PageWidth
\fi
{\hsize=\TEMPDIMEN
\global\setbox\ItemBOX=\vbox{#3
\eightpoint\rm\bls{10pt plus \Feathering}%
\vskip 5.5pt plus 6pt%
\setbox\tmpbox=\vbox{#4\par}%
\ifdim\ht\tmpbox>10pt
\noindent #4\par%
\else
\hbox to \hsize{\hfil #4\hfil}%
\fi%
}%
}%
\JoinStack%
\Print{Processing source for figure {\the\ItemNUMBER}}%
}
\def\TH#1#2#3#4{%
\ItemSTATUS=\InStack
\ItemNUMBER=#1
\ItemTYPE=\Table
\if#2S \ItemSPAN=\Single \TEMPDIMEN=\ColumnWidth
\else \ItemSPAN=\Double \TEMPDIMEN=\PageWidth
\fi
{\hsize=\TEMPDIMEN
\eightpoint\bls{10pt plus \Feathering}\rm
\global\setbox\ItemBOX=\vbox{\noindent#3\vskip 5.5pt plus5.5pt\noindent#4}}
\JoinStack
\Print{Processing source for table {\the\ItemNUMBER}}
}
\let\table=\TH
\def\UnloadZoneA{%
\FirstZoneAtrue
\Iteration=0
\loop
\ifnum\Iteration<\LengthOfStack
\GetItemSTATUS{\Iteration}
\ifnum\ItemSTATUS=1
\GetItemBOX{\Iteration}
\ifFirstZoneA \vbox to \BodgeHeight{\vfil}%
\FirstZoneAfalse\fi
\unvbox\ItemBOX\vskip \TextLeading plus \TextLeading minus 4pt
\LeaveStack{\Iteration}
\else
\advance\Iteration by 1
\fi
\repeat
}
\def\UnloadZoneC{%
\Iteration=0
\loop
\ifnum\Iteration<\LengthOfStack
\GetItemSTATUS{\Iteration}
\ifnum\ItemSTATUS=3
\GetItemBOX{\Iteration}
\vskip \TextLeading plus \TextLeading minus 4pt\unvbox\ItemBOX
\LeaveStack{\Iteration}
\else
\advance\Iteration by 1
\fi
\repeat
}
\def\ShowItem#1
{\GetItemAll{#1}
\Print{\the#1:
{TYPE=\ifnum\ItemTYPE=\Figure Figure\else Table\fi}
{NUMBER=\the\ItemNUMBER}
{SPAN=\ifnum\ItemSPAN=\Single Single\else Double\fi}
{SIZE=\the\ItemSIZE}}}
}
\def\ShowStack
\Print{}
\Print{LengthOfStack = \the\LengthOfStack}
\ifnum\LengthOfStack=0 \Print{Stack is empty}\fi
\Iteration=0
\loop
\ifnum\Iteration<\LengthOfStack
\ShowItem{\Iteration}
\advance\Iteration by 1
\repeat
}
\def\B#1#2{%
\hbox{\vrule\kern-0.4pt\vbox to #2{%
\hrule width #1\vfill\hrule}\kern-0.4pt\vrule}
}
\def\Ref#1{\begingroup\global\setbox\TEMPBOX=\vbox{\hsize=2in\noindent#1}\endgroup
\ht1=0pt\dp1=0pt\wd1=0pt\vadjust{\vtop to 0pt{\advance
\hsize0.5pc\kern-10pt\moveright\hsize\box\TEMPBOX\vss}}}
\def\MarkRef#1{\leavevmode\thinspace\hbox{\vrule\vtop
{\vbox{\hrule\kern1pt\hbox{\vphantom{\rm/}\thinspace{\rm#1}%
\thinspace}}\kern1pt\hrule}\vrule}\thinspace}%
\output{%
\ifLeftCOL
\global\setbox\LeftBOX=\vbox to \ZoneBSize{\box255\unvbox\ZoneBBOX}
\global\LeftCOLfalse
\MakeRightCol
\else
\setbox\RightBOX=\vbox to \ZoneBSize{\box255\unvbox\ZoneBBOX}
\setbox\MidBOX=\hbox{\box\LeftBOX\hskip\ColumnGap\box\RightBOX}
\setbox\PageBOX=\vbox to \PageHeight{%
\UnloadZoneA\box\MidBOX\UnloadZoneC}
\shipout\vbox{\PageHead\box\PageBOX\PageFoot}
\global\advance\pageno by 1
\global\HeaderNumber=\DefaultHeader
\global\LeftCOLtrue
\CleanStack
\MakePage
\fi
}
\catcode `\@=12
\@ifstar{\@ssection}{\@section}{Introduction}
\tx There has been significant recent progress in the study of galaxy
formation within a cosmological context, mainly due to a phenomenological
or `semi-analytical' approach to this problem. The idea is to start with
a structure formation model that describes where and when galactic
dark haloes form. A simple description of gas dynamics and star formation
provides a means to calculate the amount of stars forming in these haloes.
Stellar population synthesis models then provide the spectral evolution,
i.e.\ luminosities and colours, of these galaxies.
Many physical processes are modelled as simple functions of the circular
velocity of the galaxy halo. Therefore, the Tully-Fisher relation is
the most obvious observational relation to try and predict, as it
relates the total luminosity of a galaxy to its halo circular
velocity. However, most phenomenological galaxy formation models do
not simultaneously fit the I-band Tully-Fisher relation and the
B or K band luminosity function. When one sets the model parameters
such that the Tully-Fisher relation has the right normalization, the
luminosity functions generally overshoot (e.g.\ Kauffman, White \&
Guiderdoni 1993; Kaufmann, Colberg, Diaferio \& White 1998), certainly
for the $\Omega=1$, $H_0=50$ km s$^{-1}$ Mpc$^{-1}$
standard CDM cosmology (in the form given by Davis et al.\ 1985) that
we consider in this paper. Alternatively, when making sure that the
luminosity functions matches by changing some of the model parameters,
the Tully-Fisher relation ends up significantly shifted with respect to
the observed relation (e.g.\ Cole et al.\ 1994; Heyl et al.\ 1995).
In order to keep the modelling as analytical as possible, an extension
to the Press \& Schechter (1974) prescription for the evolution of galaxy
haloes (e.g.\ Bond et al. 1991; Bower 1991; Lacey \& Cole 1993; Kauffman
\& White 1993) has been a popular ingredient for implementations of a
phenomenological theory of galaxy formation. The extended Press-Schechter
(EPS) formalism keeps the galaxy formation model as analytical as possible,
and allows fast realizations of halo populations.
However, this prescription has the disadvantage that the large-scale
phase-space distribution of haloes is unknown, as is any information on
their internal structure and kinematics.
For many applications this information will be of great use.
Furthermore, the properties and formation histories of {\it individual}\
haloes do not match the corresponding properties and histories
as predicted by the EPS formalism (e.g.\ Lacey \& Cole 1993; White 1996).
More importantly, the EPS formalism is designed to identify collapsed
systems, irrespective of whether these contain surviving subsystems.
This `overmerging' of subhaloes into larger embedding haloes produces
a top-heavy galaxy halo mass function, and is therefore
relevant to the problem of matching both the luminosity
function and the Tully-Fisher relation.
Traditional N-body simulations suffer from a similar overmerging
problem (e.g.\ White 1976), which is of a purely numerical nature,
caused by two-body heating in dense environments (Carlberg 1994;
van Kampen 1995) when the mass resolution is too low. This is why
the statistical properties of EPS haloes and haloes found in N-body
simulations still agree quite well (e.g.\ Efstathiou et al.\ 1988;
Somerville et al.\ 1998 and references therein), despite the overmerging.
In order to circumvent these problems,
we use an N-body simulation technique that includes a built-in
recipe for galaxy halo formation, designed to prevent overmerging
(van Kampen 1995, 1997), to generate the halo population and its
formation and merger history. We show that this alone significantly changes
the predictions for the Tully-Fisher relation and the luminosity function.
In fact, it resolves most of the discrepancy sketched above, {\it and}\ allows
us to make the modelling more realistic by adding chemical evolution
and a merger-driven bursting mode of star formation to the modelling.
Star formation and its feedback to the interstellar medium are the least
understood ingredients of theories of galaxy formation. Usually one
considers a single mode of star formation, being either a continuous but
decaying rate of star formation from gas that cools within dark
haloes, or a discontinuous, bursting mode of star formation where stars
are formed at a high rate in short periods.
Here we will consider both, where the continuous mode of star formation
encompasses the case of many short bursts of star formation. We
reserve the term `starburst' for single major starburst events, lasting on
the order of 0.1 Gyr, as observed in luminous
infrared galaxies, for example (Sanders \& Mirabel 1996).
Once stars are formed, we apply the stellar population synthesis models
of Jimenez et al.\ (1998) to follow their evolution. We have enhanced
these models with a model for the evolution of the average metallicity
of the population, which depends on the starting metallicity.
Feedback to the surrounding material means that cooling properties of
that material will change with time, affecting the star formation rate,
and thus various other properties of the parent galaxy.
Several of our ingredients are thus different from earlier work:
we use numerical simulations for the formation and evolution
of the galaxy haloes, and we use newer population synthesis models,
which include chemical evolution. Nevertheless, it is still
useful to start from a model that is as similar as possible to a
published one. This allows us to assess the changes
in the model results due to the differences in modelling.
We chose use the model of Cole et al.\ (1994) as a starting point.
The lay-out of this paper is as follows: we first discuss the main problems
in Section 2, focusing on the overmerging of galaxy haloes, and ways
of resolving the problems. In Section 3 we generate
distributions of galaxies for which we calculate luminosity functions and
the Tully-Fisher relation. We compare these to those found in previous
work, and to observed ones. We discuss the interpretation of the results,
pitfalls, and possible extensions, in the final section.
\newcount\japcount
\japcount=0
\def\japitem{\advance\japcount by 1
\smallskip\noindent\rlap{(\the\japcount)}\hglue\parindent\hangindent\parindent}
\@ifstar{\@ssection}{\@section}{The phenomenological model for galaxy formation}
\tx In this section, we summarize how
galaxy formation is modelled in this work. In many
respects, we deliberately follow the approach of other
groups, so as to aid comparison of results.
Nevertheless, our methods differ in some important
respects, as outlined below.
A detailed discussion of the modelling is given
in Appendix A; this section is intended as an
overview for non-specialists.
\subsection{Overview}
\tx The key ingredients of the model are:
\japitem
{\it The merging history of dark-matter haloes.}
This is often treated by Monte-Carlo realizations of the
analytic `extended Press-Schechter' formalism. That formalism
is a successful description of collapsed systems
with densities of order 200 times the mean, and of the
history of merging that produced such systems.
However, the theory ignores substructure:
a cluster-scale halo is treated as a single system.
The alternative is to measure halo substructure directly
via N-body simulation. Special techniques are used in this
paper to prevent galaxy-scale haloes undergoing `overmerging'
owing to inadequate numerical resolution.
By allowing substructure to survive, the merger
history is simplified and the number of halo mergers is
reduced.
\japitem
{\it The merging of galaxies within dark-matter haloes.}
Each halo contains a single galaxy at formation. When haloes
merge, a criterion based on dynamical friction is used to decide how
many galaxies exist in the newly merged halo. The most massive
of those galaxies becomes the single central galaxy to which
gas can cool, while the others become its satellites. This
approximate argument is one of the more uncertain parts of
phenomenological galaxy formation schemes. We still use the
dynamical friction method, but our treatment of halo overmerging
means that part of the galaxy merging process is treated directly.
This should make our results more robust.
\japitem
{\it The history of gas within dark-matter haloes.}
When a halo first forms, it is assumed to have
an isothermal-sphere density profile. A fraction
$\Omega_b/\Omega$ of this is in the form of gas
at the virial temperature, which can cool to form
stars within a single galaxy at the centre of the halo.
Application of the standard radiative cooling
curve shows the rate at which this hot gas cools
below $10^4$~K, and is able to form stars.
Energy output from supernovae reheats some of the
cooled gas back to the hot phase. When haloes
merge, all hot gas is stripped and ends up in the new halo.
Thus, each halo maintains an internal account of the
amounts of gas being transferred between the two
phases, and consumed by the formation of stars.
\japitem
{\it Quiescent star formation.}
This is one of the more difficult areas to model.
Most authors assume a star-formation timescale that is
linearly proportional to the circular velocity
of the dark-matter halo in which the galaxy sits.
The star formation rate is equal the ratio of the amount of
cold gas available and the star-formation timescale.
The amount of cold gas available depends on the merger
history of the halo, the star formation history, and
the how much cold gas has been reheated by feedback
processes (discussed below).
\japitem
{\it Starbursts.}
We also introduce the idea that
the star-formation rate may suffer a sharp spike
following a major merger event. This is motivated
empirically by the existence of ultra-luminous
IRAS galaxies, and it allows hierarchical models
to yield behaviour resembling traditional
monolithic collapse. Our specific starburst model
is described in Appendix B.
\japitem
{\it Feedback from star-formation.}
The energy released from young stars heats cold
gas in proportion to the amount of star-formation,
returning it to the reservoir of hot gas. This
ingredient is essential to keep low-mass galaxies
and proto-galaxies from using up all their gas at
high redshift.
\table{1}{D}{\bf Table 1. \rm Parameters for the phenomenological
galaxy formation models discussed in Section 2, 3, and Appendix A.}
{\halign{%
\hskip4pt#\hfil&\hfil#\hskip4pt&\hfil#\hskip7pt&\hfil#\hskip4pt&\hfil
#\hskip4pt&\hfil#\hskip4pt&\hfil#\hskip4pt&\hfil#\hskip4pt&\hfil#\hskip2pt \cr
\multispan{9}\hrulefill\cr
\noalign{\vfilneg\vskip -0.4cm}
\multispan{9}\hrulefill\cr
\noalign{\smallskip}
Description & parameter & CAFNZ & m & n & c & s & a & b \cr
\noalign{\smallskip}
\multispan{9}\hrulefill\cr
\noalign{\smallskip}
Mass-to-light ratio: & \multispan{8} \cr
\ \ cosmological baryon fraction & $\Omega_{\rm b}$ &
0.06 & 0.06 & 0.06 & 0.06 & 0.06 & 0.06 & 0.06 \cr
\ \ $M({\rm all\ stars}) / M(m>0.1M_\odot)$ & $\Upsilon$ &
2.7 & 1.0 & 1.0 & 1.0 & 1.0 & 1.0 & 1.0 \cr
Continuous star formation: & \multispan{8} \cr
\ \ basic star formation time-scale & $\tau^0_*$ &
2 & 2 & 2 & 2 & $10^6$ & 2 & 2 \cr
\ \ power-law index & $\alpha^*$ &
$-1.5$ & $-1.5$ & $-1.5$ & $-1.5$ & $-1.5$ & $-1.5$ & $-0.5$ \cr
Bursting star formation: & \multispan{8} \cr
\ \ basic star formation time-scale & $\tau^0_{\rm b}$ &
- & - & - & - & 0.01 & 0.01 & 0.01 \cr
\ \ burst factor & $f_{\rm b}$ &
- & - & - & - & $10^6$ & 100 & 100 \cr
Feedback (reheating of cooled gas by supernovae): & \multispan{8} \cr
\ \ feedback parameter & $f_{\rm v}$ &
0.2 & 0.2 & 0.2 & 0.2 & 0.2 & 0.2 & 0.04 \cr
\ \ feedback normalization & $V_{\rm hot}$ &
140 & 140 & 140 & 140 & 140 & 140 & 100 \cr
\ \ feedback power-law index & $\alpha_{\rm hot}$ &
5.5 & 5.5 & 5.5 & 5.5 & 5.5 & 5.5 & 4.0 \cr
Galaxy merging within haloes: & \multispan{8} \cr
\ \ dynamical friction time-scale & $\tau^0_{\rm mrg}$ &
0.5$\tau_{\rm dyn}$ & 0.5$\tau_{\rm dyn}$ & 0.5$\tau_{\rm dyn}$ &
0.5$\tau_{\rm dyn}$ & 0.5$\tau_{\rm dyn}$ & 0.5$\tau_{\rm dyn}$ &
0.5$\tau_{\rm dyn}$ \cr
\ \ dynamical friction scaling law & $\alpha_{\rm mrg}$ &
0.25 & 0.25 & 0.25 & 0.25 & 0.25 & 0.25 & 0.25 \cr
Stellar populations: & \multispan{8} \cr
\ \ initial mass function & IMF &
Scalo & Salpeter & Salpeter & Salpeter & Salpeter &
Salpeter & Salpeter \cr
\ \ metallicity & $Z$ &
solar & solar & solar & evolving & evolving & evolving & evolving \cr
\noalign{\smallskip}
\multispan{9}\hrulefill\cr
\noalign{\vfilneg\vskip -0.4cm}
\multispan{9}\hrulefill\cr
\noalign{\smallskip}
}}
\japitem
{\it Stellar evolution and populations.}
Having formed stars, we wish to predict the appearance
of the galaxy that results. For this, it is necessary to
assume an IMF, and to have a spectral synthesis code.
Our work assumes the spectral models of Jimenez et al.\ (1998);
for solar metallicity, the results are not greatly
different from those of other workers.
The IMF is generally taken to be Salpeter, but any choice is possible.
Unlike other workers, we take it as established that the
population of brown dwarfs makes a negligible contribution
to the total stellar mass density, and we do not
allow an adjustable $M/L$ ratio, $\Upsilon$, for the stellar
population.
\japitem
{\it Chemical evolution.}
It would of course be unrealistic to assume that
the entire star-formation history of a galaxy can unfold
at constant metallicity. The evolution of the metals
must be followed, for two reasons: (i) the cooling of the
hot gas depends on metal content; (ii) for a given age,
a stellar population of high metallicity will be much
redder. The models of Jimenez et al.\ (1998) allow
synthetic stellar populations of any metallicity to be
constructed. Appendix C discusses how the chemical
evolution of the gas is followed.
\subsection{Model parameters}
\tx This framework is rather general, and includes a number of
parameters. Note that there are also parameters involved in the
cosmological model. In this paper, we consider only
a single variant of CDM ($\Omega=1$, $h=0.5$, $n=1$,
$\Omega_b=0.06$, $\sigma_8=0.62$), in order to illustrate
the effects of changes in the galaxy-formation model.
This cosmological model is identical to the one used by
Cole et al.\ (1994).
The equations containing the specific parameterizations of
the above assumptions are given in Appedix A.
The values assumed for these parameters in the models
investigated in this paper are given in Table 1.
\figps{1}{S}{\psfig{file=tfref.ps,width=8.5cm,silent=}}
{{\bf Figure 1.} The Tully-Fisher relation: observational data and a
simple model: single redshift collapse, with all available gas forming
stars in an 0.1 Gyr burst (solid lines), or continuously with an
exponential decay-time of 10 Gyr (dotted lines).
The squares are individual galaxies as observed by Tully et al.\ (1998),
whereas the thick straight line is an average over four fits to
observational data by four different authors (see text for details).
The thick curved line is a fit to the model of Cole et al.\
(1994, curve taken from Heyl et al.\ 1995). Note that the models of
CAFNZ do have significant star formation up to the present epoch.}
\@ifstar{\@ssection}{\@section}{The Tully-Fisher / luminosity function discrepancy}
\tx Matching both the Tully-Fisher relation, which relates the infrared
rest-frame magnitude to the circular velocity $V_{\rm c}$ of the halo,
and the B-band luminosity function has been a problem for most published
phenomenological galaxy formation scenarios. In the following we use,
as a starting point, the models of Cole et al.\ (1994, CAFNZ from here on),
which show this discrepancy as well. When CAFNZ set their free parameters
to match the B-band luminosity function, they did not find a good
match to the I-band Tully-Fisher relation: it is either a
factor of two too large in $V_{\rm c}$, or 2-3 magnitudes too faint in I.
We believe that this is largely due to an excess of bright galaxies with
rather large circular velocities, caused by overmerging within
the (extended) Press-Schechter formalism used by CAFNZ. Because of
the overabundance of massive galaxies, CAFNZ were forced to put in place
several `brakes' in order to prevent the luminosity evolution from
overshooting.
Firstly, they adopt a value for the star-formation timescale $\tau_*^0$
which is half that of the value of 4 Gyr that is found in the
numerical simulations of
Navarro \& White (1993), on which CAFNZ based their star formation
models, and which was used to set many other parameters.
Secondly, they adopted a very large mass-to-light ratio for the
stellar population: they set $\Upsilon=2.7$, where $\Upsilon$ is defined
as the ratio of all stars to that of stars above 0.1 $M_\odot$.
Finally, they assume that all hot gas has
primordial metallicity (even gas reheated by solar metallicity supernovae),
which means it cools less rapidly than enriched gas and results in less
stars being formed per unit mass. They also assumed instant enrichment
to solar metallicity for their stellar populations, which helps keep
the B-band magnitudes faint.
\subsection{Building blocks for the Tully-Fisher relation}
\tx In order to provide a frame of reference for interpreting and
understanding the Tully-Fisher relation produced by our models,
we first greatly simplify the modelling by only incorporating some
of the ingredients. The very simplest model is one in which a galaxy
turns all its gas into stars at a single epoch of formation, either
in a short burst of star formation, or as an exponentionally decaying,
but continuous star formation mode. We will refer to this type of
model as a `single redshift star formation model'. The circular velocity
is calculated using the spherical collapse model (e.g.\ Peebles 1980),
where the density at first collapse is 178 times the background value,
which gives $L \propto M_{\rm gas} \propto V_{\rm c}^3 (1+z_{\rm form})^{3/2}$.
In Fig.\ 1 we plot the resulting Tully-Fisher relation for such a model,
where the single formation redshift (for all galaxies) is annotated for
each of the model relations. The straight thick solid line is an average
of fits to four datasets (Pierce \& Tully 1992; Mould et al.\ 1993 and
references therein; Giovanelli et al.\ 1997; Mathewson, Ford \& Buchhorn
1992), whereas the squares show the data of Young (1994),
in order to give an indication of the typical spread around the
observed relation.
This average relation corresponds to $L\propto V_{\rm c}^{3.1}$,
so very close to that predicted by the spherical collapse model.
The most obvious lesson to learn from this figure is that
we can reproduce the Tully-Fisher relation simply by forming all galaxies
between a redshift of 0.5 and 4, with a peak near $z=1-2$, and turning
all available gas into stars shortly after the formation event.
Galaxies will move straight down in the diagram if only part of the gas
is used up, so late star formation from a fraction of the gas produces
the same relation as turning all gas into stars at early times.
\subsection{Observed luminosity functions}
\tx In the models of CAFNZ the B and K-band luminosity function
(LF for short) were used to set some of their free parameters,
so that they matched well to the
observed luminosity functions known at the time: the B-band LF of Loveday
et al.\ (1992) and the K-band LF of Mobasher et al.\ (1986).
Present day datasets show quite a change in the observed LF:
Zucca et al.\ (1997) observe a much steeper faint end for the B-band LF,
while Gardner et al.\ (1997) and Glazebrook et al.\ (1995) observe a
K-band LF that is fainter at the bright end, although there is some
disagreement at the faint end. It must be gratifying to CAFNZ that their models
are closer to these new results that to the data of Loveday et al.\ (1992) and
Mobasher et al.\ (1986). The significant discrepancy with the Tully-Fisher
relation remains, however, and we attempt to resolve that in the remainder
of this Section.
\subsection{The galaxy halo population}
\subsubsection{Defining haloes}
\tx In most phenomenological galaxy formation recipes published
so far the galaxy halo population and its formation and merger
history are obtained from the extended Press-Schechter (EPS)
formalism (see the introduction for references and discussion).
It is essential to realize that the EPS formalism deals only
with systems of density approximately 200 times the mean,
which are {\it assumed}\ to be virialized.
Such sets of particles are readily
identified in N-body simulations e.g. by linking particles
via the percolation (a.k.a.\ `friends-of-friends') algorithm.
Indeed, the term `halo' is not infrequently taken implicitly to
denote only systems of this sort.
However, we feel this is an unfortunate useage, and we will use
the term in this paper in a more general sense: as a virialized
local density maximum. A rich cluster is an EPS halo, but
the galaxies it contains are defined by dark-matter haloes
that constitute substructure in the cluster dark matter.
These systems collapsed before the cluster, and so have
present-day density contrasts well above 200. As the cluster forms,
the outer parts of these galaxy-scale haloes merge and
lose their identity; the high-density cores of the haloes will
nevertheless survive and remain virialized, and mark the locations
of galaxies within the cluster.
The aim of a galaxy formation model is to predict the
stellar content of these `embedded' galaxy-scale haloes.
\subsubsection{Overmerging}
\tx If one takes the galaxy haloes merger history from the EPS
formalism, all embedding haloes are retained, but embedded haloes are lost;
EPS records only when haloes are incorporated into a larger system,
but subsequent evolution within the new system is not followed.
We adopt the term `overmerging' to describe this
loss of information concerning the substructure in embedding haloes.
The use of the EPS formalism is usually motivated by the fact
that the distribution of masses of the haloes and their formation
histories are in good agreement with those found in N-body simulations
(Efstathiou et al.\ 1988; Somerville et al.\ 1998 and references therein).
However, this result depends on the choice of the group finder adopted
to identify haloes in the simulations (Suginohara \& Suto 1992).
The usual choice is percolation with a fixed linking-length, which
shares with the EPS formalism the property that embedded haloes
are not identified. It also identifies {\it any} overdense group,
not just virialized haloes.
Furthermore, overmerging also occurs in the N-body simulations to
which the EPS formalism is compared.
Small groups of particles that represent galaxy haloes
get disrupted by numerical two-body heating, especially
in large overdense systems (Carlberg 1994; van Kampen 1995).
If the numerical resolution is good enough, this problem is resolved
(Moore et al.\ 1998; Tormen, Diaferio \& Syer 1998),
but both the spatial and mass resolution need to be rather high:
Klypin, Gottl\"ober \& Kravtsov (1997) find that haloes survive for
resolutions of a few kpc and 10$^8$-10$^9$ $M_\odot$ respectively.
Unfortunately, for N-body simulations on a cosmological scale, this
requires the use of far too many particles (on the order of $10^9$)
to be practical, so a special N-body technique is needed instead.
\subsubsection{Resolving the overmerging problem}
\tx In order to prevent overmerging, we adopt a simulation technique
in which a galaxy halo formation recipe is added to an otherwise
standard N-body technique (van Kampen 1995, 1997).
The idea is that a group of particles that
has collapsed into a virialised system with a mass appropriate of a
galactic halo is replaced by a single `halo particle'. Local
density percolation, also called adaptive friends-of-friends,
is adopted for finding the groups (van Kampen 1997).
This is designed to identify the embedded haloes that the
traditional percolation group finder links up with their parent halo.
A Gaussian filter, with a length equal to the average nearest neighbour
distance for a Poisson particle distribution, is applied to calculate
a local density $n_{\rm G}$. This local density is then used to
calculate a local linking length for the percolation algorithm,
scaling as $n_{\rm G}^{-1/3}$. The linking
length also scales with the particle mass $m$ as $m^{1/2}$, in order
to account for the larger gravitational pull of the more massive halo
particles. The basic linking length (for mean density and non-halo
particles) is half the Poissonian average nearest neighbour distance.
The virial equilibrium criterion
is put into a simplified form using the half-mass radius $R_{\rm h}$,
motivated by Spitzer (1969) who found that for many equilibrium systems
\newcount\virialB
\virialB=\eqnumber
$$\langle v^2\rangle \approx 0.4 {GM\over R_{\rm h}}\ . \eqno({\rm\the\eqnumber}\step{\eqnumber})$$
A group is considered to be virialized when this criterion is satisfied
with a margin of 25 per cent.
The new halo particle is softened according to the size of the group
it replaces (thus conserving energy), and gets the position and momentum
of the centre-of-mass of the original group. Because groups of particles
can contain halo particles, merging is naturally included in this recipe as
well. Note that we do {\it not}\ assume that each halo particle hosts a
single galaxy. At formation, each halo particle may contain several
pre-existing galaxies, and the dynamical friction argument (ingredient
2 of Section 2.1) has to be invoked to see whether they merge. However,
our procedure clearly treats a good part of the general merging of
galaxies directly, which increases the robustness of the results.
If groups are sufficiently massive compared to the numerical resolution of
the code, they will avoid spurious numerical disruption in future evolution
of the density field. There is therefore no need to replace these systems
by a single particle. In fact, doing so would pose numerical problems as
supermassive simulation particles will interact violently with `normal'
N-body particles.
Therefore, an upper limit should be set to the mass of a halo particle.
However, the local density percolation group-finder we adopt here has been
designed to identify haloes in the field as well as in overdense regions.
This means that a smooth cluster containing no subhaloes will be identified
as one massive halo, while a cluster many subhaloes will not be identified
as a single halo. The centre of the cluster and the subhaloes will all
be identified as separate entities. This implies that there will
automatically be a maximum for the mass of the groups found,
depending on the environment. This is discussed in more detail in
van Kampen (1999, in preparation).
Preferrably the upper limit to the halo particle-mass should equal this
maximum group-mass, but as it depends on the group finder parameters,
it is hard to predict its value. The results should be robust
with respect to reasonable variations in this number, so the actual
choice is not too important. We set the upper mass limit to be
$7\times 10^{13} h^{-1}$ M$_\odot$, or about 1600 of the initial
simulation particles.
\figps{2}{D}{\psfig{file=nmpaper.ps,width=18.0cm,silent=}}
{{\bf Figure 2.} A comparison of galaxy halo cumulative mass distributions.
Dashed lines show the mass function for the Press-Schechter formalism for
standard CDM with two different normalizations: $\sigma_8=0.67$ (thick
dashed line), and the COBE normalization ($\sigma_8=1.19$, thin dashed line).
The thick dotted line represents the mass function for all haloes identified
in a standard dissipationless N-body simulation (for $\sigma_8=0.67$) using the
percolation algorithm, while the thin dotted line
shows the subset of those haloes that have virialized. The thickest solid line
shows the mass functions for haloes formed using the recipe of van Kampen
(1995, 1997), starting from the same initial conditions. Including all
haloes found from the percolation algorithm time at the final time
results in the mass function depicted by the thinnest solid line,
while the solid line with intermediate thickness shows the subset of
these haloes that have virialized. Note the difference between the thin solid
line and the thin dotted line, which were obtained using the same group
finder.}
\subsubsection{Our halo population}
\tx We performed two high-resolution simulations, using the methods
described above, with a (spherical) volume of $(23.2 h^{-1} {\rm Mpc})^3$,
using around $8\times10^5$ particles.
The first simulation started from initial conditions contrained
to form a galaxy cluster in the centre of the simulation volume,
while the second simulation started from unconstrained initial
conditions. We will refer to the first simulation as the
`cluster model', and to the second one as the `field model'.
The importance of overmerging is most clearly illustrated by
looking at cumulative dark halo mass functions as produced by the
Press-Schechter formalism, a standard N-body simulation, and the
present method, as described above.
In Fig.\ 2 we compare all these to each other. The thick dashed line is
the halo function from the Press-Schechter formalism for the cosmology
adopted here, with $\sigma_8=0.62$. The {\it COBE}\ normalized version
(i.e.\ $\sigma_8=1.19$) is also shown, as a thin dashed line.
We ran the field model without the
galaxy halo formation recipe switched on, i.e.\ just as a traditional
N-body simulation, and obtained the mass function using the same
group finder (see Section 3.3.3) used in the full simulation technique.
This mass function
is plotted in Fig.\ 2 as a thick dotted line. It clearly matches the
Press-Schechter mass function. However, if we now plot the mass
function for {\it virialized haloes only}, being haloes which
satisfy eq.\ (1) within 25 per cent, it falls short by a factor
of two (thin dotted line). This is because in the EPS
formalism all collapsed haloes are assumed to virialize, while
haloes as identified in numerical simulations change through merging,
secondary infall, tidal forces, etc., and are therefore not able to
remain virialized at all times. This is even true for galaxy cluster
haloes (Natarajan, Hjorth \& van Kampen 1997).
Thus, virialized haloes form a subset of all haloes at any one time.
Indeed, the haloes formed in our simulations contain about 25
per cent of all the mass in the universe, while in the EPS formalism
all matter is contained in haloes, by design.
The halo functions that are produced by the full simulations are shown
in Fig.\ 2 as solid lines. The thickest of the three lines represents
the halo function for the galaxy haloes identified by the recipe.
The thinnest solid line shows the effect of adding the virialized haloes
identified at $z=0$, including the ones above the upper mass-limit. Note that
there are only a few more haloes added, so our upper mass-limit for halo
particles is close to the maximum mass found from the halo formation
recipe employed.
The right-most solid line shows the mass function for all percolated systems
at $z=0$: galaxy halo particles, newly virialized haloes, and all other
collapsed haloes. We see that locking matter in haloes at early times
has the effect that the largest mass haloes that form in both
the Press-Schechter formalism and standard N-body simulations are now
correctly resolved into subhaloes and are not counted as single
objects. More interestingly, the maximum mass found for virialized haloes is
close to that found for observed galaxies. The core of the simulated cluster
is one of these most massive virialized haloes.
Because of our choice for the upper-limit to halo particles,
the most massive haloes are not treated as single particles. However,
they are identified in the simulation, and are incorporated into the
merger tree used for the phenomenological galaxy formation model.
\@ifstar{\@ssection}{\@section}{Resolving the Tully-Fisher / luminosity function discrepancy}
\tx Before showing our model results, a word of warning:
throughout this section we plot Tully-Fisher diagrams that contain
{\it all}\ model galaxies found from the modelling. We should therefore
bear in mind that only a subset of all these galaxies, i.e.\ the spirals,
actually belong in such a diagram.
As we do not model the morphologies of the galaxies, we aim to
at least cover the region of $V_{\rm c}-M_{\rm I}$ space occupied by the
observational data.
Galaxies outside that region may in fact be regarded as ellipticals, S0's,
and irregular galaxies, and their properties should be tested using
other observational relations, like the Faber-Jackson relation, which is
the analogue of the Tully-Fisher relation for ellipticals.
One could be tempted to merge the Tully-Fisher
and Faber-Jackson relations into one diagram, as the circular velocity
is related to the velocity dispersion through the assumption of
an isothermal density profile. However, the Faber-Jackson relation
concerns the {\it central}\ velocity dispersion, which is not equal to,
and usually larger than, the velocity dispersion of the dark matter halo.
White (1979) found that merger remnants tend towards a $r^{-3}$ density
profile, so if elliptical form through merging, they should have declining
circular velocity profiles. Observationally, it is very difficult to measure
halo properties of generally gas-poor elliptical galaxies, whereas spiral
galaxies have sufficient cold gas at large radii to act as a tracer of
the halo potential.
Thus, we need more detailed modelling before we can attempt a match to
the Faber-Jackson relation, and we therefore restrict ourselves to the
Tully-Fisher relation for now.
\subsection{Repeating CAFNZ: model {\rm m}}
\tx To see the effect of using a merger tree obtained from numerical
simulations that employ the halo formation recipe of van Kampen (1997),
instead of a merger tree generated according to the extended
Press-Schechter formalism, we applied the same prescription of gas
dynamics and star formation that CAFNZ used, with the same parameters
($\tau_*^0=2$ Gyr etc.). The only other difference with the CAFNZ model
beside the halo population is the choice of IMF and stellar population
models (see Table 1), but we believe these differences to be marginal
(see Appendix A). The spherical infall model has been adopted to
calculate the circular velocity, as in CAFNZ.
The result for the cluster simulation is plotted in Fig.\ 3a.
What is apparent from this figure is that we reproduce the CAFNZ
Tully-Fisher relation, but miss the high-$V_{\rm c}$ end of it.
Because our model galaxy haloes do not overmerge, the large-$V_{\rm c}$
haloes found by CAFNZ are divided into a number of smaller haloes
in our modelling (see Section 3.3).
The effect on the B and K band luminosity funcion is a shift to the fainter
end, as is shown in Fig.\ 4a. The thick line is the luminosity function
found by CAFNZ, while the symbols with error bars represent the
observational data described above. Our model {\rm m}
results quite clearly fit neither the observations nor the CAFNZ models.
But this is actually a very nice result, as it allows us to resolve much
of the Tully-Fisher / luminosity function discrepancy, as follows.
\figps{3}{D}{\psfig{file=tfpaper.ps,width=17.2cm,silent=}}
{{\bf Figure 3.} I-band Tully-Fisher relation from our
modelling (dots) for various choices of parameters. The first six panels
correspond to the six models discussed in the text, whereas the bottom
left panel shows model {\rm b} for a subset of the data that should
represent an observational dataset (see Section 3.4.6 for details).
We refer to the legenda and to Table 1 for the exact model specifications.
Also shown are fits to the distributions from CAFNZ (curved thin solid line),
observational data averaged over four different sets of data (straight
thick solid line, see main text for details). In the top left panel we
added observational data by Tully et al.\ (1998), shown as squares.
Note that the specific redshifts at which the galaxy formation recipe
is applied are clearly visible for model {\rm b}, and to a lesser extent
for the other models, due to the simplified modelling of starbursts.
}
\figps{4}{D}{\vskip-0.5cm\psfig{file=lfpaper.ps,width=17.5cm,silent=}}
{\vskip 0.0cm{\bf Figure 4.} B and K band luminosity functions from our
modelling (thick solid lines), corresponding to the Tully-Fisher
relations shown in Fig.\ 5 (in the same order), and those from CAFNZ
(thin solid lines). Please refer to the legenda of Fig.\ 3 and Table 1
for details on the models. Also shown are
old observational data (triangles) from Loveday et al.\ (1992) for the
B-band and Mobasher et al.\ (1993) for the K-band, and new observational
data (squares) from Zucca et al.\ (1997) for the B-band and
Gardner et al.\ (1997) and Glazebrook et al.\ (1995) for the K-band.}
\subsection{Setting $\Upsilon=1$: model {\rm n}}
\tx Considering the `brakes' that CAFNZ had to apply to their
models, it should come as a relief that our model galaxies are too
faint in B and K, since the I-band magnitudes were already
too faint for the Tully-Fisher relation to fit. We can thus apply
an {\it overall brightening}\ to the modelling in order to simultaneously
match the Tully-Fisher relation and both luminosity functions. This is
easily achieved by setting $\Upsilon=1$ instead of $\Upsilon=2.7$,
which was the value that CAFNZ needed to adopt.
This brightens the I-band Tully-Fisher relation by just over a magnitude,
with the result that the predicted distribution of points now
overlaps with the observational data (Fig.\ 3b).
If we assume that the brightest galaxies in I (for a given $V_{\rm c}$)
are those that were still forming stars up to a fairly recent epoch,
and are therefore likely to be spirals,
the predicted Tully-Fisher relation can be considered encouraging,
except for the faint end.
We discuss solutions to that at the end of this Section.
The luminosity functions are shown in Fig.\ 4b, in the corresponding panel.
The luminosity function matches the observations very well in the B-band,
but falls short in the K-band. In the next section we therefore introduce
two more ingredients that not only render the modelling more realistic,
but provides a better to the K-band luminosity function:
chemical evolution and starbursts.
\@ifstar{\@ssection}{\@section}{Chemical evolution and starbursts}
\subsection{Adding chemical evolution: model {\rm c}}
\tx Instead of adopting primordial metallicity for the cooling
function, and constant solar metallicity for the stellar
populations, we now incorporate a chemical evolution model.
The details of the method we use is described
in Appendix C. This means that we use a different set of
stellar population models, with evolving metallicities.
Once a population forms, the {\it initial}\ metallicity and the star
formation timescale $\tau_*$ determines what the metallicity will be at
later times. The enrichment of hot gas by chemically evolving stellar
populations will increase the number of stars forming at later times,
because the cooling function depends on the metallicity. We assume
enrichment to be efficient, i.e.\ the metallicity of the cooling gas
is equal to that of the populations it cools to (mass-averaged over
co-existing populations in merged galaxies).
The effect on the Tully-Fisher relation of the inclusion of chemical
evolution can be seen in Fig.\ 3c, whereas the effect on the
luminosity functions is shown in Fig.\ 4c. We see that the B-band
luminosity function ends up too bright, whereas the K-band
one is still too faint. In other words, most galaxies end up
too blue. The reason for this is that chemical enrichment boosts late
stars formation with relatively slowly decaying star formation rates.
This means that we have to find a way to form stars earlier and with much
more rapidly decaying star formation rates. This leads us to the conclusion
that we need to incoorporate a bursting mode of star formation, driven
by merging activity, which is most significant at early times (e.g.\
Carlberg 1990).
Another possibility is the addition of a dust model, which will
typically render galaxies redder. However, we leave this for
future work to explore.
\subsection{Models with starbursts only: model {\rm s}}
\tx In order to see the effect of adding a bursting mode of star formation,
we first look at a burst-only model in which continuous galaxy formation
is effectively switched off. We make the burst as strong as possible,
with $f_{\rm b}$ set to a very large value, and with a very short decay-time.
However, the duration of the starburst is set as described in Appendix B.
The Tully-Fisher relation produced by this pure starburst model is shown
Fig.\ 3d. It is quite similar to the chemical evolution model, in the
sense that more stars are formed, which brings enough galaxies on or
towards the Tully-Fisher relation.
The spread is a bit larger, as the earliest galaxies fade more rapidly if
they experience just a single starburst, therefore ending up fainter than
in model {\rm c}, whereas the late mergers are brighter because more
gas is used up at the present epoch than in model {\rm c}, which has a
much larger star formation time-scale.
However, the luminosity functions, as shown in Fig.\ 4d, are different
from the chemical evolution ones in the sense that the colours end up much
redder, providing a better match to the observational data for both bands.
\subsection{Models with both chemical evolution and starbursts: models
{\rm a} and {\rm b}}
\tx The pure starburst model produces fairly monochromatic
galaxies: all galaxies redden quickly and effectively stop forming stars,
because most mergers happen at high redshifts (e.g.\ Carlberg 1990), with a
merger rate peak at $z \approx 1$ (van Kampen 1997).
If we now switch on continuous galaxy formation again, those galaxies
that have some gas left (likely to be spirals), can form a (cosmologically)
young blue population that broadens the colour distribution {\it and}\
moves those galaxies on top of the observed Tully-Fisher relation. After all,
this relation is based on spiral galaxies in which some star formation
activity is typically found. In order to achieve this we set
$f_{\rm b}=100$ and $\tau^0_{*,\rm b}=0.01$ Gyr.
The resulting Tully-Fisher relation is plotted in Fig.\ 3e, and the
luminosity functions in Fig.\ 4e. They all match the observational data
reasonably well, except for the faint end of the Tully-Fisher relation.
However, we can easily `lift' this faint end by decreasing the amount
of feedback to $f_{\rm v}=0.04$, which comprises model {\rm b}.
The resulting Tully-Fisher relation is plotted in Fig.\ 3f, and this
is clearly shown to work. It slightly improves the B-band luminosity
function at the bright end, but at the same time worsens it at the
faint end, as shown in Figs.\ 4f.
Finally, note that the remaining deviations from the observed luminosity
functions are likely to be resolved by taking into account dust, which
will redden the colours of the galaxies.
\subsubsection{Observational selection}
\tx So far we have made no attempt at discriminating between the model
galaxies, which makes quite a difference for the Tully-Fisher relation, as
only spiral galaxies enter the relation, and only those for which a reliable
circular velocity can be measured. This means that spiral galaxies showing
any sign of interaction are typically excluded from an observational sample.
Thus, we should disregard any model galaxy that underwent a major merging
event within the last 1-2 Gyr. Furthermore, galaxies that have not been
part of a major merger event for a long time, will have a much reduced
star formation rate, and are therefore not likely to possess an obvious
disk component. So we should also disregard galaxies that have formed at
fairly high redshift and evolved passively up to the present epoch.
We select an `observational Tully-Fisher sample' from all model galaxies
by selecting only those galaxies that satisfy
1 Gyr $< t_{\rm last} <$ 6 Gyr, where $t_{\rm last}$ is the lookback time
for the last major merging (or formation) event. This corresponds to
(approximately) a redshift range of $0.05<z_{\rm last}<0.5$.
These are plotted in Fig.\ 3g. Clearly, this subset of the model data
provides a reasonable match to the observational Tully-Fisher data.
However, this attempt at observational selection is rather crude,
and will need to be refined in future work. A proper disk model is a
necessity, as is a dust model which takes the inclination of the disk
into account.
\@ifstar{\@ssection}{\@section}{Discussion and conclusions}
\tx We have argued that overmerging is an important problem for
phenomenological galaxy formation models, and is largely responsible
for the discrepancy found in earlier work between the Tully-Fisher
relation and the luminosity function. We resolved the overmerging
problem by using an N-body simulation technique that includes a galaxy
halo formation recipe. We combined this technique with simplified gas
dynamics and star formation physics, and a recent stellar population
model, in order to phenomenologically describe galaxy formation and
evolution. We included chemical evolution, and two modes of star formation:
major bursts of star formation, and a quiescent mode in which stars
form more continuously.
With this set-up we match both the B and K band
luminosity function and the I-band Tully-Fisher relation, for an
$\Omega=1$ standard CDM structure formation scenario. Resolving the
overmerging problem is the major contributor to this result,
but the inclusion of chemical evolution and starbursts are also
important ingredients.
The new ingredients we have added to the modelling of galaxy formation
are needed in order to make the models more realistic, and are not
introduced simply in order to give yet more free parameters. Nevertheless,
our resolution to the Tully-Fisher / luminosity function discrepancy
may well not be unique, and various other changes to the ingredients of the
phenomenological galaxy formation recipe might produce similar
results. For example, we have not studied the influence cosmological
parameters have on the model galaxy populations, where $\Omega$,
$\Lambda$, and $\sigma_8$ are likely to be the important parameters.
Other types of ingredients are possible as well:
Somerville \& Primack (1998) resolve some of the discrepancy using
a dust extinction model plus a halo-disk approach to feedback.
We intend to explore these issues in future work. One way of resolving
the worries about degeneracies in the cosmological/physical parameter
space will be to include data at intermediate and high redshifts.
Nevertheless, although the present work has concentrated on modelling the
properties of the low-redshift universe only, we believe that the
issues we have raised are sufficiently general that they will invariably
be part of any successful model for galaxy formation.
\@ifstar{\@ssection}{\@section}*{Acknowledgements}
\tx We thank the anonymous referee for a report that encouraged us to clarify some
parts of this paper, and Carlton Baugh, Shaun Cole, and Eduard Thommes for
useful comments.
\@ifstar{\@ssection}{\@section}*{references}
\bibitem Barnes J.E., Hut P., 1986, Nat, 324, 446
\bibitem Binney J., 1977, ApJ, 215, 483
\bibitem Bond J.R., Cole S., Efstathiou G., Kaiser N., 1991, ApJ, 379, 440
\bibitem Bower R.J., 1991, MNRAS, 248, 332
\bibitem Carlberg R.G., 1990, ApJ, 350, 505
\bibitem Carlberg R.G., 1994, ApJ, 433, 468
\bibitem Cole S., Arag\'on-Salamanca A., Frenk C.S., Navarro J.F., Zepf S.E.,
1994, MNRAS, 271, 781
\bibitem Davis M., Efstathiou G., Frenk C.S., White S.D.M., 1985, ApJ, 292, 371
\bibitem Efstathiou G., Frenk C.S., White S.D.M., Davis M., 1988, MNRAS, 235, 715
\bibitem Elmegreen B., 1999, in Beckman J., ed., Formation and evolution of
galaxies on cosmological timescales, Cambridge University Press
\bibitem Faber S.M., Gallagher J.S., 1979, ARAA, 17, 135
\bibitem Gardner J.P., Sharples R.M., Frenk C.S., Carrasco B.E., 1997, ApJ, 480, 99
\bibitem Giovanelli R., Haynes M.P., da Costa L.N., Freudling W., Salzer J.J.,
Wegner G., ApJ, 477, L1
\bibitem Glazebrook K., Peacock J.A., Miller L., Collins C., 1995, MNRAS, 275, 169
\bibitem Heavens A.F., Jimenez R., 1999, submitted to MNRAS
\bibitem Heyl J.S., Cole S., Frenk C.S., Navarro J.F., 1995, MNRAS, 274, 755
\bibitem Jimenez R., Padoan P., Matteucci F., Heavens A., 1998, MNRAS, 299, 123
\bibitem Jimenez R., Dunlop J., Peacock J., MacDonald J., J{\o}rgenson U.G., 1999,
MNRAS, in press
\bibitem Kauffman G., White S.D.M., 1993, MNRAS, 261, 921
\bibitem Kauffman G., White S.D.M., Guiderdoni, 1993, MNRAS, 264, 201
\bibitem Kauffman G., Colberg J.M., Diaferio A., White S.D.M., 1998,
astro-ph/9805283
\bibitem Klypin A., Gottl\"ober S., Kravtsov A.V., 1997, astro-ph/9708191
\bibitem Kurucz R., 1992, ATLAS9 Stellar Atmosphere
Programs and 2 km/s Grid CDROM Vol. 13
\bibitem Lacey C.G., Cole S., 1993, MNRAS, 262, 627
\bibitem Longair M.S., 1998, Galaxy formation, Springer Verlag
\bibitem Loveday J., Peterson B.A., Efstathiou G., Maddox S.J., 1992,
ApJ, 90, 338
\bibitem Mathewson D.S., Ford V.L., Buchhorn M., 1992, ApJS, 81, 413
\bibitem Matteucci F., Francois P., 1989, MNRAS, 239, 885
\bibitem Mihos J.C., Hernquist L., 1994, ApJ, 431, L9
\bibitem Mihos J.C., Hernquist L., 1996, ApJ, 464, 641
\bibitem Mobasher B., Ellis R.S., Sharples R.M., 1986, MNRAS, 223, 11
\bibitem Moore B., Governato F., Quinn T., Stadel J., Lake G., 1998, ApJ, 499, L5
\bibitem Mould J.R., Akeson R.L. Bothun G.D., Han M-S., Huchra J.P.,
Roth J., Schommer R.A., 1993, ApJ, 409, 14
\bibitem Natarajan P., Hjorth J., van Kampen E., 1997, MNRAS, 286, 329
\bibitem Navarro J.F., White S.D.M., 1993, MNRAS, 265, 271
\bibitem Peebles P.J.M., 1980, Large-scale structure in the Universe, Princeton
\bibitem Pierce M.J., Tully R.B., 1992, ApJ, 387, 47
\bibitem Press W.H., Schechter P., 1974, ApJ, 187, 425
\bibitem Rees M.J., Ostriker J.P., 1977, MNRAS, 179, 541
\bibitem Roberts M.S., Haynes M.P., 1994, ARAA, 32, 115
\bibitem Sanders D.B., Mirabel I.F., 1996, ARAA, 34, 749
\bibitem Silk J., 1977, ApJ, 211, 638
\bibitem Somerville R.S., Lemson G., Kolatt T.S., Dekel A., 1998, astro-ph/9807277
\bibitem Somerville R.S., Primack J.R., 1998, astro-ph/9802268
\bibitem Spitzer Jr.\ L., 1969, ApJ, 158, L139
\bibitem Suginohara T., Suto Y., 1992, ApJ, 396, 395
\bibitem Sutherland R., Dopita M.A., 1993, ApJS, 88, 253
\bibitem Tormen G., Diaferio A., Syer D., 1998, 299, 728
\bibitem Tully R.B., Pierce M., Huang J.-S., Saunders W., Verheijen M.,
Witchalls P., 1998, AJ, 115, 2264
\bibitem van Kampen E., 1995, MNRAS, 273, 295
\bibitem van Kampen E., 1997, in Clarke D.A., West M.J., eds., Proc. 12th
`Kingston meeting' on Theoretical Astrophysics: Computational Astrophysics,
ASP Conf. Ser. Vol. 123. Astron. Soc. Pac., San Francisco, p. 231,
astro-ph/9904270
\bibitem White S.D.M., 1976, MNRAS, 177, 717
\bibitem White S.D.M., 1979, MNRAS, 189, 831
\bibitem White S.D.M., 1996, in Schaeffer et al., eds., Cosmology and
Large-scale structure, Proc. 60th Les Houches School, Elsevier, p.349
\bibitem Zucca E., Zamorani G., Vettolani G., Cappi A.S., Merighi R., Mignoli M.,
Stirpe G.M., MacGillivray H., et al., 1997, A\&A, 326, 477
\figps{5}{S}{\psfig{file=vcpaper.ps,width=8.5cm,silent=}}
{{\bf Figure A1.} Comparison of circular velocity from the spherical
infall model and from N-body simulations (top panel), and a similar
comparison for the half-mass radii (bottom panel). The solid lines
simply indicate equality.}
\@ifstar{\@ssection}{\@section}*{Appendix A: details of the galaxy formation recipe}
\eqnumber=1
\tx This Appendix provides some further details of the phenomenological modelling
of galaxy formation adopted for this paper. We chose to use the modelling
of CAFNZ as a starting point, but note that many authors
use similar models with differerences in the details or in the choice
of parameters. Significant changes to the CAFNZ models are:
the halo merger history and mass function, the inclusion of star bursts,
and the modelling of chemical evolution.
All parameters encountered in the various ingredients of the model
are listed in Table 1, along with their values for the specific models
discussed in the main paper.
\@ifstar{\@ssection}{\@section}*{\it Halo population and its merger history}
\tx In most phenomenological galaxy formation recipes published
so far the galaxy halo population and its formation and merger
history are obtained from the extended Press-Schechter (EPS) model
(see the introduction for references). Instead, we use the haloes
found an N-body simulations which includes a recipe for the formation
of galaxy haloes (van Kampen 1995, 1997), as described in Section 3.3.3.
The actual N-body code used is the Barnes \& Hut (1986) treecode.
Haloes are assumed to form according to the spherical collapse model
(e.g.\ Peebles 1980), and are assumed to settle as isothermal spheres,
with a constant circular velocity $V_{\rm c}$.
The spherical collapse model provides a relation between the mass
of the halo, its circular velocity, and its formation redshift
(e.g.\ White 1996) :
$$V_{\rm c} = \Bigl({M_{\rm halo}\over 2.35\times 10^5 h^{-1}
{\rm M}_\odot}\Bigr)^{1\over 3}
\Bigl(1+z_{\rm form}\Bigr)^{1\over 2}\
{\rm km s}^{-1}\ , \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
where a truncated isothermal density profile has been assumed.
One is forced to adopt this crude model as the extended
Press-Schechter formalism that is typically used does not
provide information on the kinematics of the haloes.
The N-body simulations do provide us with that information, so we can
obtain $V_{\rm c}$ directly from the velocity dispersion of the halo,
which is in virial equilibrium by definition. Again assuming an
isothermal density profile, we have
$$V_{\rm c}=\Bigl({2\over 3}\Bigr)^{1\over 2} \langle v^2\rangle^{1\over 2}
\approx \Bigl(0.33 {GM_{\rm halo}\over R_{\rm h}}\Bigr)^{1\over 2},
\eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
where $\langle v^2\rangle$ is obtained from the numerical simulation.
Basically, we allow dark haloes with the same mass and $z_{\rm form}$ to
have a range of values for $R_{\rm h}$, and therefore different $V_{\rm c}$,
while CAFNZ assume the fixed value
$$\Bigl({R_{\rm h}\over 100 h^{-1}{\rm kpc}}\Bigr) =
\Bigl({M_{\rm halo}\over 6.37\times 10^{12} h^{-1} {\rm M}_\odot}
\Bigr)^{1\over 3} {1\over 1+z_{\rm form}}. \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
In order to access the significance of the difference between using
$V_{\rm c}(\langle v^2\rangle)$ and $V_{\rm c}(M,z)$, we plotted
$V_{\rm c}(M,z)$ versus $V_{\rm c}(\langle v^2\rangle)$ (Fig.\ A1).
Clearly, on {\it average} the spherical infall model is a good
approximation. The scatter represents the spread in half-mass radii
and velocity dispersions for a given mass found in the simulations.
An upper limit to the mass of a galaxy halo particle needs to be set,
as discussed in detail in Section 3.3.4.
In practice, we set the upper mass limit to be
$7\times 10^{13} h^{-1}$ M$_\odot$, or about 1600 of the initial
simulation particles. Although justifyable in numerical terms,
this mass is also close to the one that results from the cooling
criterion $t_{\rm cool} < t_{\rm coll}$, where $t_{\rm coll}$ is
the halo collapse time for the spherical collapse model
(Rees \& Ostriker 1977; Binney 1977; Silk 1977).
For primordial abundances, this limit is about
$1.5\times10^{13} \Omega_{\rm b} h^{-1}$M$_\odot$ (White 1996),
i.e.\ $\approx 1.0 h^{-1}\times10^{12}$M$_\odot$ for our choice
of $\Omega_{\rm b}$.
For enriched gas, with abundances near solar, the limit goes up to
$5-10\times 10^{13} h^{-1}$M$_\odot$.
\@ifstar{\@ssection}{\@section}*{\it Formation and merging of galaxies within dark haloes}
\tx For the modelling of the galaxies that populate the dark matter haloes
we follow the approach of CAFNZ. In their scheme, galaxies
form in the centre of dark haloes. When dark haloes merge, their galaxies
merge with a delay giving by the dynamical friction timescale.
During this dynamical friction phase a dark halo can thus contain several
galaxies at the same time. One of these galaxies will be designated as the
galaxy to which halo hot gas can cool, while the others are considered
satelites. It is important to realize that this part of the CAFNZ
prescription is kept identical, whereas the formation and evolution
of the dark halo population is different.
Thus, dynamical friction is assumed to delay the merging of galaxies
within merged haloes, with the delay given by
$$\tau_{\rm mrg}=\tau^0_{\rm mrg}(M_{\rm halo}/M_{\rm gal})^{\alpha_{\rm mrg}}
\ . \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
In the following, all gas dynamics and star formation happens in between
`major events', which are either the formation of a halo (plus galaxy),
the merger of one or more haloes, or the merger of one or more galaxies
within an already merged halo. These three events define lifetimes for
haloes and galaxies over which gas can cool, stars can form, etc.,
according to processes we describe next.
\@ifstar{\@ssection}{\@section}*{\it Stars and gas}
\tx Gas dynamics and star formation are modelled using analytical
relations that are power-law functions of the halo circular
velocity, $V_{\rm c}$, which is a constant for a given halo as
we assume an isothermal density profile.
Each galaxy has a reservoir of stars, cold gas, and hot gas.
A newly formed galaxy has all its baryons, with a total mass of
$\Omega_b M_{\rm halo}$, in the form of hot gas. The gas temperature
is assumed to quickly settle to the virial temperature, which is a
function of $V_{\rm c}$ only:
$$ T_{\rm gas} = T_{\rm vir} =
{\mu m_{\rm p}\over 2 k} V_{\rm c}^2. \ \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
The three constants $\mu$, $m_{\rm p}$ and $k$ are the mean
molecular weight, proton mass, and Boltzmann's constant, respectively.
Hot gas will cool radiatively at a rate which
depends on the density $\rho$, temperature $T$, and
metallicity $Z$ of the hot gas. The cooling time-scale is given by
$$\tau_{\rm cool}={3\over 2}{\rho_{\rm gas}(r)\over \mu m_{\rm p}}
{k T_{\rm gas}\over n_{\rm e}^2(r)\Lambda(T_{\rm gas},Z_{\rm gas})},
\ \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
where $n_{\rm e}$ is the electron density, and $\Lambda(T,Z)$ the
cooling function. For the latter we take the cooling functions for
primordial and solar metallicities as given by Sutherland \& Dopita (1993),
and obtain the cooling function for any given metallicity by linear
interpolation or extrapolation of $\log \Lambda(T,0.0002)$ and
$\log \Lambda(T,Z_\odot)$.
For gas at a constant temperature with a homogeneous chemical abundance,
the cooling rate depends only on the density, and is therefore a function
of radius. Thus, at a large enough radius $r_{\rm cool}$ the cooling
time will be larger than the time passed between two major halo events
(formation or merger).
The mass encompassed by this cooling radius is transferred from the hot
to the cold gas reservoir between these two events.
Stars can form from cold gas, thus transferring mass from the cold
gas reservoir to the stellar population of the galaxy. In the present
modelling, mass ejected by dying stars is {\it not}\ transferred back
to any of the gas reservoirs. However, cold gas can be reheated and thus
transferred back to the hot gas reservoir by the energy output of
supernovae. This process is called `feedback' for short.
The star formation rate is proportional to the amount of cold gas
actually available:
$$d M_*(t,V_{\rm c}) / d t = M_{\rm cold}(t,V_{\rm c}) / \tau_*(V_{\rm c})
\ , \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
with the star formation time-scale given by
$$\tau_*(V_{\rm c})=\tau_*^0 (V_{\rm c}/ 300\ {\rm km\ s}^{-1})^{\alpha_*}
\ . \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
It is assumed that the reheating of cold gas by supernovae can be modelled as
$$d M_{\rm hot}(t,V_{\rm c}) / d t = \beta(V_{\rm c}) d M_*(t,V_{\rm c}) / d t
\ , \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
where the feedback proportionality parameter $\beta$ is given by
$$\beta(V_{\rm c})=(V_{\rm c}/V_{\rm hot})^{-\alpha_{\rm hot}}
\ . \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
The overall feedback parameter $f_v$, defined as the fraction of energy
released by supernovae that is dumped into the cold gas reservoir as
kinetic energy, determines the values of the constants
$\alpha_*$, $\alpha_{\rm hot}$, and $V_{\rm hot}$.
In this paper we mostly use $f_v=0.2$, for which CAFNZ found the following
values from fits to simulation data by Navarro \& White (1993):
$\alpha_*=-1.5$, $\alpha_{\rm hot}=5.5$, and $V_{\rm hot}=140$ km s$^{-1}$.
For a detailed description of the feedback process and its parameters
we refer to CAFNZ.
The merging of haloes and galaxies also determine where stars, cold,
and hot gas end up. A merger of haloes brings together all amounts of hot
gas of the merging haloes, and reheat the new single hot gas reservoir to
the virial temperature of the newly formed halo. It also brings together
all galaxies that were part of the original haloes. The most massive of these
will now be the one to which hot gas can cool and form stars. CAFNZ set
the circular velocity of this halo to be that of the newly formed dark
halo, which can be rather large, unless its merger time
$\tau_{\rm mrg}$, as given by eq.\ (A4), is larger than the time available
until the next merger event (or the present epoch), in which case the
original $V_{\rm c}$ is retained.
All other galaxies retain their identity (their stars and cold gas
reservoir) if they do not merge with the most massive one, i.e.\ if
$\tau_{\rm mrg}$ is smaller than the time available until the next
merger event (or the present epoch).
In that case stars keep forming from the cold gas left in each of
those galaxies, whereas feedback not only reheats part of the gas,
but also expells it to the hot gas reservoir of the common dark halo.
\@ifstar{\@ssection}{\@section}*{\it Stellar population models}
\tx To predict the spectra and photometric properties of galaxies we have used
the extensive library of synthetic stellar population models computed by
Jimenez et al.\ (1998; 1999), where a detailed description of the models can
be found. Here we briefly review the main ingredients of the models. We
computed simple synthetic stellar populations (SSPs), i.e. stellar populations
with fixed metallicity and formed in a burst of infinitesimal duration, for
metallicities between 1/100 Z$_{\odot}$ and 2 Z$_{\odot}$ and ages between 0.01
and 14 Gyr. The library uses new interior models (Jimenez et al.\ 1998) and a
new set of stellar photospheres for the coolest models (T$_{\rm eff} < 6000$),
while Kurucz (1992) models are used for T$_{\rm eff} > 6000 $K.
Two important new features of the models
are the inclusion of an accurate treatment of all post-main sequence
evolutionary stages and the use of better theoretical photospheric models
for low temperatures.
Simple stellar populations (SSP's) are the building blocks of any
arbitrarily complicated population since
the latter can be computed as a sum of SSP's, once the star formation rate is
provided. In other words, the luminosity of a stellar population of age $t_0$
(since the beginning of star formation) can be written as:
$$L_{\lambda}(t_0)=\int_{0}^{t_0} \int_{Z_i}^{Z_f}
L^{\rm SSP}_{\lambda}(Z,t_0-t)\, dZ\, dt \ \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$
where the luminosity of the SSP is:
$$L^{\rm SSP}_{\lambda}(Z,t_0-t)= \int_{m_{\rm d}}^{m_{\rm u}} \dot M_*(Z,m,t)\,
l_{\lambda}(Z,m,t_0-t)\, dm\ \eqno(A{\rm\the\eqnumber}\step{\eqnumber})$$ and $l_{\lambda}(Z,m,t_0-t)$ is
the luminosity of a star of mass $m$, metallicity $Z$ and age $t_0-t$, $Z_i$
and $Z_f$ are the initial and final metallicities, $m_{\rm d}$ and $m_{\rm u}$
are the smallest and largest stellar mass in the population and
$d M_*(Z,m,t) / d t$ is the star formation rate at the time $t$ when the
SSP is formed.
We assume that the initial mass function (IMF) is universal and is given by
a Salpeter IMF ($x=1.35$). In order to take account for the chemical evolution
we have computed detailed models that link $d M_* / d t$ and $Z$, i.e.\
$M_*(Z,t)$ since we have assumed that the $m$ dependence is constant
(a constant IMF). This is described in Appendix C.
\@ifstar{\@ssection}{\@section}*{Appendix B: starbursts}
\eqnumber=1
\tx Besides a quiescent mode of star formation, we consider a bursting
mode of star formation associated with a major formation or merging event.
This star formation mode can be the dominant one, especially at early times
when encounters and mergers are frequent, and most galaxies form.
The strength of the starburst will depend on the clumpiness
of the matter distribution that is induced by an encounter or merging
event, on the duration of the starburst phase, and on the amount of cold
gas available to form stars, which can be substantially enhanced by strong
inflows of gas into the central region of the galaxy during the final stages
of a merger event (Mihos \& Hernquist 1996).
Here we adopt a simplified approach, by assuming that the SFR during
the starburst phase can be modelled using the same scaling relations
as for the quiescent star formation mode, but with different
star formation and cooling rates. The feedback strength is assumed to be the
same for both star formation modes. Also, we adopt the same Salpeter
IMF for both modes, a choice that is supported by recent observational
data (see Elmegreen 1999).
The duration of the starburst, $t_{\rm b}$, is assumed to be equal to
the dynamical timescale of the new dark matter halo:
$t_{\rm b}=\tau_{\rm dyn}\equiv R_{\rm h}/\sigma_{\rm v}$. Using the
spherical infall model (eq.\ A1), the virial theorem (eq.\ \the\virialB),
and the isothermal profile (eq.\ A2), we find
\newcount\bursttime
\bursttime=\eqnumber
$$t_{\rm b}\approx 0.024 (1+z_{\rm form})^{-3/2} h^{-1}{\rm Gyr}\ .
\eqno(B{\rm\the\eqnumber}\step{\eqnumber})$$
Therefore, a single starburst at low redshift lasts longer than one
at high redshift, but as the frequency is much lower, it will still
be at high redshifts that starbursts are more important.
Note that the actual starburst might last for a much shorter time
than the dynamical timescale, depending on the amount of material
available in the bulges of the merging galaxies (Mihos \& Hernquist 1994),
so $t_{\rm b}$ should be taken as a maximum.
The bursting star formation timescale $\tau_{*,\rm b}$ is assumed to
scale with the halo circular velocity $V_{\rm c}$ in the same manner as
the continuous star formation timescale $\tau_*$, i.e.\
$$\tau_{*,\rm b}(V_{\rm c})= \tau^0_{*,\rm b}
(V_{\rm c}/ 300\ {\rm km\ s}^{-1})^{\alpha_*}\eqno(B{\rm\the\eqnumber}\step{\eqnumber})$$
(analogous to eq.\ (A8) in Appendix A).
It is also necessary to introduce a `burst factor' $f_{\rm b}$ which
models the enhanced
cooling of hot gas and stronger inflow of cold gas during the
star bursting phase. It is expressed as a density enhancement over and
above the mean isothermal profile that will form after the bursting event.
If $f_{\rm b}$ is larger than unity, cooling will be more efficient,
so that star formation will be enhanced during the bursting phase.
Thus, during a starburst of duration $t_{\rm b}$, star formation is
enhanced by having more cold gas available to form stars, and by having
those stars formed at a higher rate. The shorter star formation timescale
will result in a different stellar population as compared to the continuous
mode for two reasons: firstly and simply due to the different formation
timescale, but secondly because the chemical evolution of the population
will be different.
\figps{6}{S}{\psfig{file=plotmetal.ps,width=8.5cm,silent=}}
{{\bf Figure C1.} Metallicity evolution for various star formation
time-scales $\tau_*$.}
\figps{7}{S}{\psfig{file=chem_shade.ps,width=9.0cm,silent=}}
{{\bf Figure C2.} B-K colour for a stellar population with an initial
metallicity of 0.01, as a function of age and star formation
time-scale $\tau_*$.}
\@ifstar{\@ssection}{\@section}*{Appendix C: chemical evolution}
\eqnumber=1
\tx Most of the phenomenological models published so far make the
simplifiying assumption that hot gas cools with primordial metallicity,
while the stellar populations form and evolve at a constant solar
metallicity. This `instant' enrichment scheme helps to produce galaxies
that are red enough in order to match observations, but is only realistic
for star bursting galaxies, not for galaxies that form stars more
continuously. Furthermore, the hot gas around metal-ejecting galaxies
is enriched, which will boost cooling at intermediate and low redshifts.
We therefore include chemical enrichment in our modelling,
in the following way.
The chemical evolution of a stellar population can be divided into two
parts: the chemical enrichment rate at which the gas is polluted
by dying stars and the formation rate at which gas is transformed into stars.
In reality these two aspects are related, since the more metals are injected
into the interstellar medium, the more efficient star formation should be,
and also vice versa (e.g.\ low surface brightness galaxies have low
metallicity and low star formation rates). In this paper, we assume that
the star formation rate depends solely on the circular velocity of the halo
(see Appendix A).
We then compute the enrichment of the interstellar medium using accurate
nucleosynthesis prescriptions. We refer to Jimenez et al.\ (1999) for a
detailed description of the chemical evolution equations used and the stellar
yields.
In principle, one should compute the hydrodynamical evolution of the baryons
for each halo and calculate the surface density of the baryonic disc in order
to determine the star formation rate as well as the infall rate and the
outflow from supernovae. Since our simulations do not have a hydrodynamical
description of the baryonic component, we proceed in a different way.
We use the chemical evolution models of Matteucci \& Francois (1989),
but with a star formation rate given by eq.\ (A7), thus implicitely
assuming a specific disk model. As our star formation rate is similar
to that of e.g. Heavens \& Jimenez (1999) and Jimenez et al.\ (1998),
this disk model will be close to the disk model of these authors.
With this set-up it is possible
to obtain a surface that describes $\tau_*$ vs. $Z$(t), where $\tau_*$
is given in terms of $V_{\rm c}$ by eq.\ (A8). We show $Z$(t) for four
different values of $\tau_*$ in Fig.\ C1. For a given $\tau_*$ and initial $Z$,
a population will evolve with time as described in Appendix A.
An example of the B-K colour evolution is shown in Fig.\ C2.
\@ifstar{\@ssection}{\@section}*{Appendix D: star formation below the N-body resolution limit}
\tx The N-body simulation technique has a lower limit on the mass of the
haloes identified. In the hierarchical picture these must have formed
through merging of haloes with masses below that limit. We thus need to
approximate the stellar and gas mass resulting from the evolution along
the merger tree below the numerical resolution limit.
Both modes of star formation need to be accomodated, with the complication
that merging of small-scale structure is relatively frequent at early
times. Although both modes will occur intermittently, we approximate the
average of all these periodic contributions taken together, by assuming
a basic star formation timescale of 0.5 Gyr.
We estimate the amount of gas cooled in the sub-tree by considering
a single population which has two thirds of the circular velocity of
the halo in which the sub-tree terminates, and half its age at the
time of formation.
This approximation will be improved upon in future work
by linking analytical merger subtrees, obtained using the extended
Press-Schechter formalism, to the leaves of the merger tree
extracted from the numerical simulation.
For the present paper, these subtrees below the N-body resolution
limit are not important, as they represent old stellar populations
that have faded over many Gyr, and are likely to end up in bulges and
centres of elliptical galaxies. They therefore contribute little
to the present day luminosities that enter the Tully-Fisher relation
and luminosity functions.
\end
|
\section{Introduction}
The study of electric processes in semiconductor materials plays an
important role in understanding the physics of semiconductor devices as
well as in their design and development [1,2]. One of the most difficult
problems is the consideration of strongly nonequilibrium effects in
essentially nonuniform semiconductors. At the same time the latter can
display quite interesting specific features caused by nonuniform
distributions of charge carriers [3--6].
For instance, electric current through a semiconductor device can display
rather abnormal behaviour, with transient fluctuations corresponding to
the flow of the current against the applied voltage [5,6]. In the short
communications [5,6] a model case of a unipolar semiconductor was considered,
with numerical analysis not including the relaxation parameters and diffusion
coefficients.
The aim of the present paper is to study nonequilibrium processes, with a
strongly nonuniform initial distribution of charge carriers, for realistic
semiconductor materials. We consider the general case of a semiconductor
with two kinds of charge carriers, positive and negative. Numerical
analysis takes account of relaxation and diffusion effects. The total
current through a semiconductor device is considered, together with the
currents across the left and right surfaces of this device. And also the
role of the generation--recombination noise is analysed.
In Sec.2 we collect the main equations defining the problem and needed for
the following analysis. An approximate analytical solution of these equations
is presented in Sec.3, which permits to show explicitly the motion of charge
carriers under an applied voltage. The conditions for the occurrence of the
negative electric current, directed against this applied voltage, are derived
in Sec.4. This anomalous transient effect is illustrated by numerical
solutions. Sec.5 contains conclusion and discussion with several
suggestions for the possible practical usage of the considered effect.
Appendix contains the proof that the approximate regular solution
asymptotically coincides with the regular stable solution of the exact
equations.
\section{Basic Equations}
The charge carriers are characterized by the densities $\;\rho_1>0\;$ and
$\;\rho_2<0\;$ that are functions of the space position
$\;\stackrel{\rightarrow}{r}\;$ and time $\;t\;$, i.e.
$\;\rho_i=\rho_i(\stackrel{\rightarrow}{r},t)\;$ with $\;i=1,2\;$. Carrier
transport can be described in terms of a semiclassical approach called the
drift--diffusion approximation which is the basis of the majority of
semiconductor device models [1,2]. This approach is, of course,
phenomenological since it does not concern the microscopic derivation of
the used parameters, such as mobilities or relaxation widths. The values
of these parameters depend on a number of different underlying causes.
For instance, the lattice structure of the considered semiconductor is
important for defining the values of these parameters. Thus, the
scattering of carriers on phonons influences both mobilities as well as
damping parameters. However, the calculation of such parameters is a
separate problem that is not the aim of the present paper. In the
semiclassical approach all parameters are assumed to be given a priori.
With the given phenomenological parameters, the drift--diffusion
approximation is known to give a very good description of realistic
semiconductor devices, which explains why this approximation is so widely
used [1,2]. The first set of equations in this approach consists of the set
of continuity equations
\begin{equation}
\frac{\partial\rho_i}{\partial t} +\stackrel{\rightarrow}{\nabla}
\cdot\stackrel{\rightarrow}{j}_i + \frac{\rho_i}{\tau_i} =\xi_i
\end{equation}
for each kind of carriers. The relaxation term, with the relaxation time
$\;\tau_i\;$, is taken in the simplest form since, as will be clear from
what follows, it does not play an essential role in transient processes
occurring at times $\;t\ll\tau_i\;$. The right--hand side of (1) is the
generation--recombination noise [7] which is always present in semiconductor
devices. Not to overcomplicate the problem we do not include other types
of noise [7--9] assuming that they are of less importance.
Another equation is a Maxwell equation
\begin{equation}
\varepsilon\stackrel{\rightarrow}{\nabla} \cdot \stackrel{\rightarrow}{E}
= 4\pi(\rho_1 +\rho_2) \; ,
\end{equation}
with the dielectric permittivity $\;\varepsilon\;$.
The electric--current density in (1) is
\begin{equation}
\stackrel{\rightarrow}{j}_i =
\mu_i\rho_i\stackrel{\rightarrow}{E} -
D_i\stackrel{\rightarrow}{\nabla}\rho_i ,
\end{equation}
where $\;\mu_1>0\;$ and $\;\mu_2<0\;$ are the carrier mobilities and
$\;D_i\equiv(\mu_i/e_i)k_BT\;$ are the diffusion coefficients in which
$\;e_1>0\;$ and $\;e_2<0\;$ are the carrier charges, $\;k_B\;$ is the
Boltzmann constant, and $\;T\;$ is temperature. The first and second
terms in (3) are the drift and diffusion current densities, respectively.
Adding here the displacement current, one has the total current density
\begin{equation}
\stackrel{\rightarrow}{j}_{tot} =
\stackrel{\rightarrow}{j}_1 + \stackrel{\rightarrow}{j}_2 +
\frac{\varepsilon}{4\pi}
\frac{\partial\stackrel{\rightarrow}{E}}{\partial t}\; .
\end{equation}
If the device is biased with an externally applied voltage $\;V_0\;$
along a path $\;\stackrel{\rightarrow}{l}\;$, then
\begin{equation}
\int_{\stackrel{\rightarrow}{l}}\stackrel{\rightarrow}{E}
(\stackrel{\rightarrow}{r},t)d \stackrel{\rightarrow}{l} = V_0 .
\end{equation}
For concreteness, we consider a positive bias, that is, $\;V_0>0\;$. It is
assumed that metal contacts supplying the external voltage are nondamaging,
that is, do not induce in their vicinity incubation effects.
Consider a plane device of the width $\;L\;$ and area $\;A\;$. Then instead of
$\;\stackrel{\rightarrow}{r}\;$ we are to deal with one space variable
$\;x\in[0,L]\;$.
An important characteristic is the transit time
\begin{equation}
\tau_0\equiv\frac{L^2}{\mu V_0} ,
\end{equation}
where $\;\mu=\min\{\mu_1,|\mu_2|\}\;$. Usually, $\;\mu_1<|\mu_2|\;$.
This time is to be compared with the relaxation times $\;\tau_i\;$. It
is often more convenient to deal with inverse times called widths. For
instance,
\begin{equation}
\gamma_i\equiv\frac{1}{\tau_i} ,
\end{equation}
with $\;i=1,2\;$, define the relaxation widths.
Another convenience is to deal with dimensionless quantities, which will
be done in what follows. To return to dimensional quantities, we shall
imply that $\;x\;$ is measured in units of $\;L\;$; $\;t\;$ and $\;\tau_i\;$,
in units of $\;\tau_0\;$; and other physical quantities, in the
corresponding units listed below:
$$ D_0\equiv \mu V_0 , \qquad E_0\equiv\frac{V_0}{L} , \qquad Q_0\equiv
\varepsilon AE_0 , $$
\begin{equation}
\rho_0\equiv\frac{Q_0}{AL} , \qquad \xi_0\equiv\frac{\rho_0}{\tau_0},
\qquad J_0\equiv\frac{Q_0L}{\tau_0} .
\end{equation}
For the case considered, Eq.(1) reduces to
\begin{equation}
\frac{\partial\rho_i}{\partial t} +\mu_i\frac{\partial}{\partial x}
\left (\rho_iE\right ) - D_i\frac{\partial^2\rho_i}{\partial x^2} +
\gamma_i\rho_i =\xi_i ,
\end{equation}
and Eq.(2) becomes
\begin{equation}
\frac{\partial E}{\partial x} =4\pi(\rho_1 +\rho_2) ,
\end{equation}
where
\begin{equation}
0 < x < 1 , \qquad t > 0 .
\end{equation}
Condition (5) for an applied voltage reads
\begin{equation}
\int_0^1E(x,t)dx = 1 .
\end{equation}
These equations are to be supplemented by initial conditions
\begin{equation}
\rho_i(x,0) =f_i(x) \qquad (i=1,2) .
\end{equation}
Electric field can be expressed, from Eq.(10), as a functional
\begin{equation}
E(x,t) = 1 +4\pi\left [ Q(x,t) -\int_0^1Q(x,t)dx\right ]
\end{equation}
of the charge densities, so that
\begin{equation}
Q(x,t) =\int_0^x\left [ \rho_1(x',t) +\rho_2(x',t)\right ]dx' .
\end{equation}
And the total density of current (4) can be written as
\begin{equation}
j_{tot} =\mu_1\rho_1 E - D_1\frac{\partial\rho_1}{\partial x} +
\mu_2\rho_2 E - D_2\frac{\partial\rho_2}{\partial x} +
\frac{1}{4\pi}\frac{\partial E}{\partial t} \; ,
\end{equation}
where $\;j_{tot}=j_{tot}(x,t)\;$.
The quantities that can be measured and that we are going to study in what
follows are the current across the left surface
\begin{equation}
J(0,t)\equiv j_{tot}(0,t) \; ,
\end{equation}
the current across the right surface
\begin{equation}
J(1,t) \equiv j_{tot}(1,t) \; ,
\end{equation}
and the total current through the device
\begin{equation}
J(t) = \int_0^1 j_{tot}(x,t) dx \; .
\end{equation}
The latter, using Eqs. (16) and (12), can be presented as
$$
J(t) = \int_0^1
\left [ \mu_1\rho_1(x,t) + \mu_2\rho_2(x,t)\right ] E(x,t)dx +
$$
\begin{equation}
+ D_1 \left [ \rho_1(0,t) - \rho_1(1,t)\right ] +
D_2 \left [ \rho_2(0,t) - \rho_2(1,t)\right ] \; .
\end{equation}
Our aim is to study the peculiarities in the time dependence of the
electric current when the initial conditions (13) correspond to a strongly
nonuniform charge distribution. Such nonuniform distributions can be
prepared in different ways. For example, one can organize a nonuniform
distribution in the process of growing of a semiconductor sample. Another
way is to irradiate semiconductor by narrow laser beams [10]. One more
possibility is by forming heavily doped layers by ion irradiation [11].
This method makes it possible to form narrow layers of positive carriers
with a density of $\;10^{20}cm^{-3}\;$.
\section{Carrier Densities}
To understand better the physics of processes resulting from a nonuniform
initial distribution of charge carriers, it would be useful to find
an analytical, though approximate, solution to the system of equations
(9) and (10). This can be done by means of the method of scale
separation [12,13], whose mathematical foundation is based on the
Krylov--Bogolubov averaging method [14].
The first step in the method of scale separation [12,13] is to classify
the solutions onto fast and slow. In our case this can be done as follows.
The electric field, as is seen from Eq. (14), is the functional of
the charge densities, averaging the latter over the space variable $\;x\;$.
This results in that $\;E\;$ varies in space slower than $\;\rho_i\;$. On
the other hand, the voltage integral (12) shows that the electric field,
being averaged over space, does not depend on time. This means that
$\;E\;$ can be treated as a slow function in time. Therefore, the electric
field $\;E\;$ can be regarded as a slow solution, as compared to the
charge densities $\;\rho_i\;$, with respect to both space and time. This
permits us to consider the equation (9) for a fast solution $\;\rho_i\;$
keeping there $\;E\;$ as a space--time quasi--integral. After solving
Eq. (9), the found $\;\rho_i\;$ is to be substituted into Eq. (14) giving
an equation for $\;E\;$ which can be solved iteratively.
In solving Eq. (9), it is convenient to continue $\;\rho_i\;$ outside the
region of $\;x\in(0,1)\;$ by defining $\;\rho_i\;$ as zero for $\;x <
0\;$ and $\;x > 1\;$. Then we may invoke the Fourier transforms with
respect to $\;x\;$. Finally, we obtain
\begin{equation}
\rho_i(x,t) =\rho_i^{reg}(x,t) +\rho_i^{ran}(x,t) ;
\end{equation}
the first term being the regular solution
\begin{equation}
\rho_i^{reg}(x,t) =\int_{-\infty}^{+\infty}G_i(x-x',t)f_i(x')dx'
\end{equation}
induced by the initial condition (13), while the second term being the
random solution
\begin{equation}
\rho_i^{ran}(x,t) =\int_0^t\int_{-\infty}^{+\infty}
G_i(x-x',t-t')\xi_i(x',t')dx'dt'
\end{equation}
generated by the noise. The Green function in Eqs. (22) and (23) is
\begin{equation}
G_i(x,t) =\frac{1}{2\pi}\int_{-\infty}^{+\infty}\exp\{ ikx
-i\omega_i(k)t\} dk
\end{equation}
with the spectrum
\begin{equation}
\omega_i(k) =\mu_iEk- iD_ik^2 -i\gamma_i .
\end{equation}
Function (24) has the properties
$$ G_i(x,0) =\delta(x) , \qquad \int_{-\infty}^{+\infty}G_i(x,t)dx =
e^{-\gamma_it} . $$
In the case considered, the integration in (24) can be realized
explicitly resulting in
\begin{equation}
G_i(x,t) =\frac{1}{2\sqrt{\pi D_it}}\exp\left\{
-\frac{(x-\mu_iEt)^2}{4D_it} -\gamma_it\right\} .
\end{equation}
As the initial condition in Eq. (13) it is reasonable to accept the
physically realistic case of the Gaussian distribution
\begin{equation}
f_i(x) =\frac{Q_i}{Z_i}\exp\left\{-\frac{(x-a_i)^2}{2b_i}\right\} ,
\end{equation}
in which $\;0<a_i<1\;$ and
$$ Q_i=\int_0^1f_i(x)dx , \qquad
Z_i=\int_0^1\exp\left\{-\frac{(x-a_i)^2}{2b_i}\right\} dx . $$
With the initial condition (27), the regular solution (22) becomes
\begin{equation}
\rho^{reg}_i(x,t)=\frac{Q_ib_i}{Z_i\sqrt{b_i^2+2D_it}}\exp\left\{
-\frac{(x-\mu_iEt-a_i)^2}{2b_i^2+4D_it}-\gamma_it\right\} .
\end{equation}
The regular and random solutions satisfy the initial conditions
\begin{equation}
\rho_i^{reg}(x,0) = f_i(x), \qquad \rho_i^{ran}(x,0) = 0 .
\end{equation}
The regular solution, as time increases, moves with the velocity
$\;\mu_iE\;$, becomes wider and smaller, so that
\begin{equation}
\lim_{t\rightarrow\infty}\rho_i^{reg}(x,t)=0 .
\end{equation}
The behaviour of the random solution depends on that of the noise. It is
customary to treat the latter as the white noise with the averaging
properties
\begin{equation}
\langle\xi_i(x,t)\rangle = 0 , \qquad
\langle\xi_i(x,t)\xi_j(x',t')\rangle =\gamma_{ij}\delta(x-x')\delta(t-t') .
\end{equation}
Accepting (31), one has
\begin{equation}
\langle\rho_i^{ran}(x,t)\rangle = 0 ,
\end{equation}
and, consequently,
\begin{equation}
\lim_{t\rightarrow\infty}\langle\rho_i(x,t)\rangle = 0 .
\end{equation}
Since the electric field, according to (14), is a linear functional in
$\;\rho_i\;$, we find
\begin{equation}
\lim_{t\rightarrow\infty}\langle E(x,t)\rangle = 1 .
\end{equation}
Although the limiting values (33) and (34) have been obtained by analysing
the approximate solutions, it is possible to show (see Appendix) that the
limits (33) and (34) are stable stationary solutions of the exact equations.
The random solution (23) influences the electric current (2) through the
correlator
\begin{equation}
\langle\rho_i^{ran}(x,t)\rho_j^{ran}(x',t)\rangle =
\gamma_{ij}\int_0^tG_{ij}(x-x',t')dt' ,
\end{equation}
in which
\begin{equation}
G_{ij}(x-x',t)\equiv\int_{-\infty}^{+\infty}G_i(x-x'',t)G_j(x'-x'',t)dx'' .
\end{equation}
With the Green function (26), this gives
\begin{equation}
G_{ij}(x,t) =\frac{1}{2\sqrt{\pi(D_i+D_j)t}}\exp\left\{
-\frac{(x-\mu_iEt+\mu_jEt)^2}{4(D_i+D_j)t} -(\gamma_i+\gamma_j)t\right\} .
\end{equation}
At large time, Eq.(37) decays by the law
\begin{equation}
G_{ij}(x,t)\simeq\frac{1}{2\sqrt{\pi(D_i+D_j)t}}\exp\left
(-\gamma_{eff}t\right ) ,
\end{equation}
as $\;t\rightarrow\infty\;$, with the effective attenuation
$$ \gamma_{eff}\equiv\frac{(\mu_i-\mu_j)^2E^2}{4(D_i+D_j)} +\gamma_i
+\gamma_j . $$
Consequently, the correlator (35) tends to a time constant.
At small time, one has
\begin{equation}
G_{ij}(x,t)\simeq\frac{1}{2\sqrt{\pi(D_i+D_j)t}}\exp\left\{
-\frac{x^2}{4(D_i+D_j)t}\right\} ,
\end{equation}
as $\;t\rightarrow 0\;$. Therefore, the correlator (35) behaves as
\begin{equation}
\langle\rho_i^{ran}(x,t)\rho_j^{ran}(x',t)\rangle\simeq
\frac{\gamma_{ij}\sqrt{t}}{2\sqrt{\pi(D_i+D_j)}}\exp\left\{
-\frac{(x-x')^2}{4(D_i+D_j)t}\right\} ,
\end{equation}
when $\;t\rightarrow 0\;$. Equation (40) shows that the influence of
noise at small times is exponentially suppressed. This conclusion is of
high importance for the following analysis.
\section{Electric Current}
We have now enough information about the behaviour of the system in order
to answer the question: Is it possible that a negative electric current
could appear, directed against the applied voltage?
The first evident necessary condition for such a possibility is the space
nonuniformity of the charge densities. Really, if $\;\rho_i(x,t)\;$ is
uniform in $\;x\;$, then from Eq. (20) it follows immediately that the
electric current is the positively defined quantity
$\;\mu_1\rho_1+\mu_2\rho_2>0\;$.
The properties of the carrier densities, studied in the previous section,
are such that, even, if at the initial time $\;\rho_i(x,t)\;$ is
nonuniform in space, it tends to become uniform with time. Consequently,
if a negative electric current would appear, this could happen only at
the initial stage of the process, when $\;t\ll 1\;$.
In this way, the occurrence of negative electric current, if any, can
arise only as a principally transient effect, when the charge densities
are yet nonuniform. The processes of diffusion and relaxation need some
time to make $\;\rho_i\;$ uniform. Therefore, there always can be found
such a time $\;t\ll 1\;$ when diffusion and relaxation are yet not important.
But these processes shorten the time of a negative--current fluctuation,
if it appears. Thus, the conditions favoring the longer lifetime of such
a fluctuation would be $\;D_i\ll 1\;$ and $\;\gamma_i\ll 1\;$.
The influence of noise, according to Eq. (40), is exponentially small at the
initial stage. Thus, even a strong noise would not kill the effect,
although it, of course, would shorten the negative--fluctuation lifetime.
So, the condition favoring the longer lifetime is weak noise, when
$\;\gamma_{ij}\ll 1\;$.
After understanding the necessary and favoring conditions for the
transient effect of a negative--current fluctuation, let us elucidate
sufficient conditions for the inequality
\begin{equation}
J(t) < 0
\end{equation}
occurring at $\;t\ll 1\;$. As a limiting case we may take $\;t=0\;$ and
the maximally nonuniform initial density
\begin{equation}
\rho_i(x,0) = f_i(x) = Q_i\delta(x-a_i)
\end{equation}
following from condition (27) under $\;b_i\rightarrow 0\;$. Then from
expression (20) we readily get \begin{equation}
J(0)=\mu_1Q_1E(a_1,0)+\mu_2Q_2E(a_2,0) .
\end{equation}
This emphasizes once again that for such a nonuniform initial charge
density the diffusion, relaxation, and noise do not influence much the
value
of the electric current $\;J(0)\;$. The corresponding electric field,
defined by Eq. (14), is
\begin{equation}
E(x,0) = 1+4\pi Q_1 [ a_1 -\Theta(a_1 -x)] +4\pi Q_2 [ a_2 -\Theta(a_2 -x) ] ,
\end{equation}
where $\;\Theta(x)\;$ is the unit step function.
Combining Eqs. (41), (43), and (44), we have the inequality
$$ \mu_1Q_1\left\{ 1 +4\pi Q_1\left ( a_1 -\frac{1}{2}\right ) +
4\pi Q_2\left [ a_2 -\Theta(a_2- a_1)\right ]\right\} + $$
\begin{equation}
+ \mu_2Q_2\left\{ 1 +4\pi Q_2\left ( a_2 -\frac{1}{2}\right ) +
4\pi Q_1\left [ a_1 -\Theta(a_1- a_2)\right ]\right\} < 0
\end{equation}
for the parameters of the system allowing the appearance of a negative
current.
There can be a number of different cases satisfying inequality (45). To show
that such situations do really exist, consider a particular example when
$\;a_1=a_2\equiv a\;$. Then Eq.(41) reduces to
\begin{equation}
(\mu_1Q_1 +\mu_2Q_2)E(a,0) < 0 .
\end{equation}
From equality (44) we get
$$ E(a,0) = 1+4\pi Q \left ( a -\frac{1}{2}\right ) \qquad
(Q \equiv Q_1 +Q_2 ) . $$
Recall that $\;\mu_1\;$ and $\;Q_1\;$ are positive, while $\;\mu_2\;$ and
$\;Q_2\;$ are negative; so that $\;\mu_iQ_i > 0\;$. Thence, inequality
(46) can be hold only if
\begin{equation}
E(a,0) < 0 .
\end{equation}
As follows from solution (28), the quantity $\;\mu_iE\;$ plays the role of
the effective velocity of motion for the corresponding charge packet. In the
case of inequality (47), we have $\;\mu_1E < 0\;$ and $\;\mu_2E > 0\;$. This
means that the positive carriers effectively move against the applied voltage;
and the negative carriers, along the latter; that is, they move oppositely to
what one would expect. Hence, the negative electric current is related to the
anomalous drift of charge carriers.
Substituting into Eq. (47) the electric field, we find
\begin{equation}
4\pi Q \left ( \frac{1}{2} - a\right ) > 1 .
\end{equation}
Depending on whether $\;Q\;$ is positive or negative, inequality (48) yields
$$ a < \frac{1}{2} -\frac{1}{4\pi Q} \qquad ( Q > 0) , $$
\begin{equation}
a > \frac{1}{2} +\frac{1}{4\pi|Q|} \qquad (Q < 0) .
\end{equation}
Taking also into account that $\;0<a<1\;$, we obtain from inequalities (49)
the condition
\begin{equation}
|Q|> \frac{1}{2\pi} .
\end{equation}
Equations (49) and (50) are sufficient conditions for the appearance of a
negative electric current at the initial stage of the process. Similarly,
it is easy to show that the currents (17) and (18) can also become
negative in the transient regime.
To study in more detail the behaviour of the electric current as a
function of time, we have solved Eqs.(9) and (10) numerically. In
accordance with the above analysis, the case favoring the
negative--current fluctuation is accepted, when $\;\gamma_{ij}\ll 1\;$.
The initial conditions are given by the Gaussian form (27). The voltage
integral (12) plays the role of the boundary condition for the electric
field. For the charge densities one may take the Neumann or Dirichlet
boundary conditions [1]. We have tried both and found that the general
picture does not change much, with the only difference that the
calculational procedure is less stable for the Dirichlet conditions. To
achieve the best stability, we opted for the Neumann boundary conditions.
For the characteristic parameters we accept the values typical of
semiconductors [1,2], such as $\;Si\;$. Then the diffusion coefficients
are $\;D_1\sim 10\; cm^2/s,\; D_2\sim 30\; cm^2/s$. The mobility of
positive carriers $\mu_1\sim 500\; cm^2/Vs$ for the average concentration
$\;10^{13}-10^{14}cm^{-3}\;$ and $\;\mu_1\sim 200\; cm^2/Vs\;$ for the
concentration $\;10^{17}-10^{18}cm^{-3}\;$. The mobility of electrons
$\mu_2\sim 1500\;$ for the average concentration $\;10^{13}-10^{14}\;$ and
$\;\mu_2\sim 400\;$ for the concentration $\;10^{16}-10^{18}\;$. The
recombination time $\;\tau_1\sim\tau_2\sim 10^{-12}-10^{-10}s\;$, hence
the relaxation width $\;\gamma_1\sim\gamma_2\sim 10^{10}-10^{12}s^{-1}\;$. We
consider a plane device of the size $\;A\sim 1\;cm^2,\; L\sim0.1-1\;cm\;$,
with the applied voltage $\;V_0\sim 10^3-10^5\; V\;$. For the calibration
parameters in Eq.(8) we get $\mu\sim 10^3cm^2/Vs,\;V_0\sim 10^3-10^5V,\;
D_0\sim 10^6-10^8cm^2/s,\; E_0\sim 10^3-10^5V/cm,\; Q_0\sim 10^3-10^6V\;cm,
\; \rho_0\sim 10^3-10^7V/cm^2,\; J_0\sim 10^8-10^{15}V\; cm^2/s$. The
transit time (6) is $\tau_0\sim 10^{-6}-10^{-9}s$.
The results of our numerical calculations are presented in the Figs.1 to
6, where we show the time dependence of the electric current across the
boundaries as well as the behaviour of the total current through the
device. These figures demonstrate that the appearance of negative electric
current is really a transient effect occurring at dimensionless times
$t\ll 1$, which in dimensional units means that $t\ll\tau_0$. All values
in the figure captions are given in dimensionless units employing the
calibration parameters from Eq. (8). Also, for shortness, we write
$a_1\equiv a$ and $b_1=b$.
Fig.1 shows that the electric current through the left boundary of the
semiconductor sample, through its right boundary, and the total electric
current are different. This difference is not merely quantative but can
be qualitative, so that the negative current may happen at the right
surface and on average through the sample, but may be absent on the left
surface. Such a difference depends on semiconductor characteristics as
mobilities and relaxation widths. This suggests the possibility of
employing the principal difference in the behaviour of the currents for
extracting information on the semiconductor characteristics. For example,
in Fig. 2 it is seen that changing the electron mobility mainly
influences the current across the left surface. The transient negative
current becomes more pronounced when increasing the absolute value of the
electron mobility, as is seen in Fig. 3. The lifetime of the
negative--current fluctuation strongly depends on the relaxation width,
which is illustrated in Fig. 4. Increasing the relaxation width shortens
the fluctuation lifetime. Figure 5 demonstrates the role of the total
initial charge on the occurrence of the negative current, and Fig. 6 shows
the role of the initial distribution of charge carriers. The importance
of special conditions for the initial charge and its location has been
discussed in detail above. The amplitude of the transient negative--current
fluctuation becomes smaller when the charge layer at initial time is
shifted farther from the left surface of the semiconductor sample.
\section{Discussion}
We have considered electric processes in nonequilibrium nonuniform
semiconductors. The transport equations are taken in the standard
drift--diffusion approximation that is widely used for describing realistic
semiconductor devices. Both analytical and numerical solutions of these
equations are accomplished.
It is shown that under special circumstances an unusual transient
phenomenon appears displaying negative electric current. The
necessary condition for such an anomalous current is nonuniformity of
carrier densities at the initial stage. A general sufficient condition
(45) is derived and its particular forms (49) and (50) are analysed in
detail. We studied the influence of diffusion, relaxation, and of
generation--recombination noise and showed that these processes, even
being strong, do not destroy the effect although may shorten the lifetime
of a negative--current fluctuation. Therefore such an anomalous electric
current can really be observed in semiconductor devices.
An important physical question is how one could use the considered effect
for practical applications. Several possibilities of using this effect
can be suggested:
\vspace{2mm}
(i) The appearance of the transient negative--current fluctuation is
rather sensitive to the characteristic parameters of semiconductor, such
as the carrier mobilities and relaxation widths. Therefore one could use
the observation of the current fluctuation for defining these parameters.
This could be done in the following way. Assume that we know all
parameters except one, say a mobility or a relaxation coefficient.
Comparing the time--dependence of the measured current with that of the
calculated one, we may try to find such a value of the sought parameter
that the measured and the calculated behaviour of the electric currents
be as close as possible to each other, in the optimal case, be almost
coinciding. Then the so fitted quantity would give the value of the
sought parameter.
\vspace{2mm}
(ii) When the layer of charge carriers is formed by irradiating
semiconductor with an ion beam, the stuck ions are distributed
approximately in the Gaussian law centered at the mean free path of the
ions. A necessary condition for the occurrence of the negative current is
that the charge layer is located at a particular distance from the
semiconductor surface, as e.g. in Eq. (49). Thence, this effect is very
sensitive to the initial location of charge carriers and, thus, could be
used for measuring the mean free path of ions in specific semiconductor
materials.
\vspace{2mm}
(iii) The value of the total charge in the initial nonuniform layer is
also crucially important for the occurrence of the negative current, as
is seen from Eqs. (48)--(50). Hence, studying this current, we could
measure the initial charge. The latter may be unknown when the initial
distribution of charge carriers is formed by irradiating semiconductor
with narrow laser beams whose influence on the generation of carriers in
not precisely known.
\vspace{2mm}
(iv) Semiconductor devices often work in the close vicinity of radiation
sources, such as atomic reactors, or under the influence of other strong
radiation, as cosmic rays. In the presence of radiation, the functioning
of semiconductor devices can be drastically disturbed because of the
arising carrier nonuniformities. This can lead not only to the
malfunctioning of semiconductor devices but even to dramatic accidents.
In order to prevent from these, one could employ controlling schemes
reacting to the appearance of the negative current, signaling by this
that the level of the carrier nonuniformity induced by irradiation has
become dangerous.
\vspace{2mm}
(v) As we have shown, the generation--recombination noise does not
destroy the effect of the negative--current fluctuation. However, there
exist other types of noise [7--9] whose influence on the supression of
this effect can be different. Therefore, analysing the peculiarity
of the electric negative--current fluctuation, one could judge what kind
of noise dominates the process in the studied semiconductor.
It is certainly not possible to enumerate all feasible applications of
the considered effect. But we hope that the examples listed above do
demonstrate that the specific unusual features of the negative--current
transient effect could provide us several interesting physical
applications. Three types of such applications are, generally,
admissible. One type is for investigating the characteristics of
semiconductor materials. Another type can be used for studying the
properties of irradiating beams. And, finally, this effect can be
employed for the practical purpose of creating special controlling
instruments.
\vspace{5mm}
{\bf Acknowledgement}
\vspace{3mm}
We appreciate a grant from the University of Western Ontario, London,
Canada. One of the authors (M.S.) is grateful to NSERC of Canada for
financial support in the form of a research grant.
\newpage
{\Large{\bf Appendix. Stationary Solution}}
\vspace{5mm}
Here we prove that Eqs.(33) and (34) are the stable stationary solutions
of the exact equations (9) and (10) under the voltage condition (12). To
this end, because of equality (32), it is sufficient to prove that
$$ \lim_{t\rightarrow\infty}\rho_i^{reg}(x,t)= 0 , \qquad
\lim_{t\rightarrow\infty}E^{reg}(x,t) = 1 , $$
where $\;E^{reg}\;$ implies the functional (14) including the dependence
on only $\;\rho_i^{reg}\;$. In what follows we shall write for brevity
$\;\rho_i\;$ and $\;E\;$ keeping in mind $\;\rho_i^{reg}\;$ and
$\;E^{reg}\;$. The proof will be based on the method of multipliers [15].
Write Eq.(9) in the form
$$ \frac{\partial}{\partial t}\rho_i(x,t) =v_i(x,\rho,t) , $$
with the velocity field
$$ v_i(x,\rho,t) = D_i\frac{\partial^2\rho_i}{\partial x^2}
-\mu_i\frac{\partial}{\partial x}(\rho_iE) -\gamma_i\rho_i . $$
Define the multiplier matrix
$$ M_{ij}(x,x',t) =\frac{\delta\rho_i(x,t)}{\delta\rho_j(x',0)} $$
and the Jacobian matrix
$$ L_{ij}(x,x',\rho,t) =\frac{\delta v_i(x,\rho,t)}{\delta\rho_j(x',t)} . $$
The latter, with the given velocity field, consists of the elements
$$ L_{ii}(x,x',\rho,t) =\left ( D_i\frac{\partial^2}{\partial x^2}
-\mu_iE\frac{\partial}{\partial x} -\gamma_i\right )\delta(x-x') - $$
$$ - 4\pi\mu_i(2\rho_i+\rho_j)\delta(x-x') -
\mu_i\frac{\partial\rho_i}{\partial x}\frac{\delta E(x,t)}{\delta\rho_i(x',t)} , $$
$$ L_{ij}(x,x',\rho,t) = -4\pi\mu_i\rho_i\delta(x-x') -
\mu_i\frac{\partial\rho_i}{\partial x}\frac{\delta E(x,t)}{\delta\rho_j(x',t)} , $$
where $\;i\neq j\;$. Here, the variational derivative of the electric
field can be found from expression (14) giving
$$ \frac{\delta E(x,t)}{\delta\rho_i(x',t)} = 4\pi\left [
\frac{\delta Q(x,t)}{\delta\rho_i(x',t)} + x' - 1\right ] , \qquad
\frac{\delta Q(x,t)}{\delta\rho_i(x',t)} =\Theta(x-x') . $$
Varying the evolution equation for $\;\rho_i\;$, we obtain the equation
$$ \frac{\partial}{\partial t}M_{ij}(x,x',t) =\sum_k\int_0^1
L_{ik}(x,x'',\rho,t)M_{kj}(x'',x',t)dx'' $$
for the multiplier matrix, with the initial condition
$$ M_{ij}(x,x',0) =\delta_{ij}\delta(x-x') , $$
following from the variation of condition (13).
For the case $\;\rho_i=0\;$ and $\;E=1\;$, the Jacobian matrix is
$$ L_{ij}(x,x',0,t) =\delta_{ij}\left (
D_i\frac{\partial^2}{\partial x^2} -\mu_i\frac{\partial}{\partial x} -
\gamma_i\right ) \delta(x-x') . $$
This leads to the equation
$$ \frac{\partial M_{ij}}{\partial t} =
\left ( D_i\frac{\partial^2}{\partial x^2}
-\mu_i\frac{\partial}{\partial x} -\gamma_i\right ) M_{ij} $$
for the multiplier matrix. From the latter equation, invoking the Fourier
transform
$$ M_{ij}(x,x',t)
=\frac{1}{2\pi}\int_{-\infty}^{+\infty}M_{ij}(k,t)e^{ik(x-x')}dk , $$
we find
$$ M_{ij}(k,t) =\delta_{ij}\exp\left\{ -(i\mu_ik + D_ik^2 +\gamma_i)t
\right\} $$
As is seen,
$$ | M_{ij}(k,t)| < 1 $$
for all $\;k\in(-\infty,+\infty)\;$ and $\;t>0\;$. Hence the motion in
the vicinity of the stationary solutions $\;\rho_i=0\;$ and $\;E=1\;$ is
stable. It is also asymptotically stable since
$$ \lim_{t\rightarrow\infty} M_{ij}(k,t) = 0 $$
for all $\;k\in(-\infty,+\infty)\;$. This completes the proof.
|
\section{Introduction}
Functionally useful proteins are sequences of amino acids that fold rapidly
under appropriate conditions (temperature range, acidity of the water
solution etc) into their native states commonly assumed to be their ground
state configurations\cite{Dill}. The dynamics of folding is akin to
motion in a rugged free energy landscape\cite{Bryngelson} and it crucially
depends on two factors: the interactions between the amino acids
and the target conformations.
In this paper, we focus on the role of non-native contact energies
in the folding process.
We study this issue in a simple model which is an extension of the
so called Go model proposed by Go and Abe\cite{Go}
Specifically, we consider the standard
two-dimensional lattice model of 16 monomers.
Its Hamiltonian is as follows
\begin{equation}
H \; \; = \; \; \sum_{i<j} \, \alpha_{ij} \Delta_{ij} \; ,
\end{equation}
where $\Delta_{ij}=1$ if monomers $i$ and $j$ are in contact and
$\Delta_{ij}=0$ otherwise (monomers $i$ and $j$ are considered
to be in contact if they are seperated by one lattice bond and
$|i-j| \neq 1$). The quantity $\alpha_{ij}=-1$ if monomers $i$ and $j$
are in contact in the native conformation and $\alpha_{ij}=\alpha$
otherwise. The sequence is thus defined by the native conformation.
We allow $\alpha$ to be attractive
($\alpha < 0$) or {\em repulsive}
($\alpha > 0$).
In what follows $\alpha$ is assumed to be larger
than -1.0 so that the native state is guaranteed to
be a maximally compact conformation.
In the original Go model\cite{Go} $\alpha=0$.
It should be noted that using model (1) one can monitor the effect of
non-native contact energies by varying only
one parameter $\alpha$. Furthermore,
the two-dimensional model is simple enough to study the effect of the
target conformations on the folding dynamics.
It should be noted that there have already been several studies
of models with repulsion in the non-native contacts:
a Gaussian model \cite{Shriva}, a designed model \cite{Gutin},
and the so called HP+ model
\cite{HP+}. The generalized Go model that we study here allows one to
vary the strength, and sign,
of the non-native contacts relative to the native ones.
We focus on 4 target conformations shown in Fig. 1. There are 37 compact
conformations (for the two-dimensional 16-monomer chain one has 69 compact
$4\times 4$ conformations but for the Go-like model only 38
of them remain different due to the end-to-end reversal symmetry
and one is not accessible kinetically)
which may act as the native conformations. Among
these $S_1$ and $S_4$ shown in Fig. 1 are the two fastest folders at
$T \neq 0$, whereas $S_2$ and $S_3$, also shown in Fig. 1, have intermediate
folding properties.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig1.eps}}
\caption{Four sequences studied in this paper
. $S_1$ and $S_3$ have $\delta E=0$.}
\end{figure}
It should be noted that the repulsive non-native contact interaction are
expected to improve the foldicity\cite{Shriva,Gutin,HP+} because they
restrict the
size of the relevant phase space. In our case, however, this effect becomes
so dramatic that for $\alpha > 0.1$
sequences $S_1$ and $S_3$ can fold with a finite $t_{fold}$ even at $T=0$.
Such an exotic phenomenon has been also observed in the HP model for some
13-monomer chains \cite{Chan1} so it is not restricted to the Go-like
sequences that we study. Here, we study how this arises as a function
of $\alpha$ and show how do folding characteristics, i.e. characteristic
temperatures and folding times, depend on $\alpha$.
The ability of $S_1$ and $S_3$ to fold at $T=0$
may be partly understood by the fact that for these sequences
the repulsive non-native interactions reduce the number of local minima
by two orders of magnitude compared to the case of $\alpha < 0$.
Furthermore, repulsive interactions dramatically affect partitioning
of the phase space into regions associated with the local energy minima.
We demonstrate this by
using the disconnectivity graph technique \cite{Karplus}
and show, in particular, that connectivities to the folding funnel
become simplified significantly.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig2.eps}}
\caption{Temperature dependence of $t_{fold}$ for $S_1$
and $S_2$ and for 3 selected values of $\alpha$,
$\alpha = -0.4, 0$ and 1. The results are based on 2 - 4 batches,
each containing 200 trajectories from initial random conformations.}
\end{figure}
We have found that the folding temperature, $T_f$, increases with $\alpha$.
This result agrees with that of Camacho\cite{Camacho} for an
effectively zero-dimensional model\cite{Shakhnovich1}.
We obtain it, however,
not only by the numerical calculations for the
lattice model but also by the analytical argument.
The minimum folding time, $t_{min}$, defined at the temperature
where the folding is fastest has found to decrease with $\alpha$ and
it gets saturated for $\alpha \rightarrow \infty$.
The folding dynamics of a chain is studied by a Monte Carlo procedure
that satisfies the detailed balance condition\cite{Malte}, and was
motivated by the studies presented in Ref.\cite{Chan1,Chan}. The dynamics
allows for single and two-monomer (crankshaft) moves. For each
conformation of the chain one has $A$ possible moves and the maximum
value of $A$, $A_{max}$, is equal to $A_{max}=N+2$. In our 16-monomer case
$A_{max}=18$. For a conformation with $A$ possible moves, probability
to attempt any move is taken to be $A/A_{max}$ and probability not to do
any attempt is 1-$A/A_{max}$ \cite{Chan1,Chan}. In addition, probability
to do a single move is reduced by the factor of 0.2 and to do
the double move - by 0.8 \cite{Chan,Chan1}.
The attempts are rejected or accepted as in the standard Metropolis
method.
The folding time is equal to the total number of Monte Carlo attemps
divided by $A_{max}$.
We have carried out the Monte Carlo simulations to determine the dependence
of the median folding time,
$t_{fold}$, on $T$ and $\alpha$. The results for $S_1$ and $S_2$
are shown in Figure 2.
For each temperature, $t_{fold}$ is obtained based on 200 independent runs
starting from random configurations. The results are averaged over 2 - 4
batches, of 200 trajectories each.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig3.eps}}
\caption{Dependence of the folded fraction on $\alpha$ at $T=0$ for
4 sequences shown in Figure 1. The results are averaged
over 4 - 6 simulations, each
corresponding to 200 trajectories.}
\end{figure}
For sequence $S_2$ the standard U-shape\cite{Socci} for the temperature
dependence of $t_{fold}$ is observed for all values of $\alpha$. In
other words, no qualitative change occurs if the non-native contact energies
change from attractive to repulsive.
In the case of sequence $S_1$, however, for $\alpha=1$
the standard U-shape disappears suggesting that the glass
transition temperature $T_g$ which is operationally
defined as the value of the temperature at which the median time
is equal to some cut-off value (usually this cut-off value is chosen to
be 300000 Monte Carlo steps for the two-dimensional 16-monomer
chain\cite{Socci}) becomes zero. In the standard scenario,
at low temperatures the system may get trapped in some local minima
and the folding process becomes extremly
slow. The folding time is then governed by
the Arrhenius law, $t_{fold} \sim \exp(\delta E/T)$\cite{Shakhnovich,Malte},
where $\delta E$ is the energy barrier energy (at $T=0$ one has
$t_{fold} \rightarrow \infty$).
Thus, the absence of the $U$-shape dependence
suggests that the energy barrier $\delta E=0$.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig4.eps}}
\caption{Dependence of $T_f$, $T_{min}$ and $t_{min}$ for
sequence $S_1$
on $\alpha$. The results are averaged over 4 batches, each containing 200
trajectories.}
\end{figure}
In order to know whether $\delta E$ is exactly zero or not one has to study
the folding at $T=0$. At zero temperature the system gets trapped in some
local minimum or in the native state. If the fraction of the trajectories
from random conformations
fold into the native state is bigger than 50$\%$, then the chain
is said to be folded and $\delta E=0$.
The fraction of folded trajectories is shown
in Fig. 3 for $S_1$, $S_2$, $S_3$ and $S_4$.
For $S_1$ and $S_3$ this fraction becomes bigger
than $50\%$ for $\alpha > \alpha _c$, where $\alpha _c \approx 0.1$.
Sequences $S_2$ and $S_4$ have $\delta E \neq 0$ for any value of $\alpha$.
It is interesting to mention that $S_4$ folds even faster than $S_1$ at
$T \neq 0$ but its foldicity becomes much worse at $T=0$.
Furthermore, the folding rate of $S_3$ at $T \neq 0$ is slower than for the
three other sequences and yet $S_3$ can fold at $T=0$. Thus
the geometry of the native targets has a dramatic effect
on the folding at $T=0$. Among the 37 maximally compact $4\times4$ structures
it is only $S_1$ and $S_3$ that do not obey the Arrhenius law.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig5.eps}}
\caption{The same as in Figure 4 but for
sequence $S_2$.}
\end{figure}
We study the dependence of $T_f$ on $\alpha$ for
two typical sequences $S_1$ and $S_2$ shown in Fig. 1.
The total number of conformations of the 16-monomer chain is
only 802075\cite{Lau,Dinner} and
is amenable to exact enumeration. This allows for an exact evaluation of the
equilibrium parameters such as the folding temperature $T_f$. The latter
is defined as a temperature at which the probability of occupancy of
the native state is $1/2$. The results for sequences $S_1$ and $S_2$
are shown in Fig. 4 and 5.
Obviously, $T_f$ increases with $\alpha$ but this dependence
gets weaker for larger values of $\alpha$.
The increase of $T_f$ with $\alpha$ may be understood in the following
simple way. Let $P_{\Gamma _0}(\alpha)$ be the probability of occupying
the native state of conformation $\Gamma _0$. Then
\begin{equation}
P_{\Gamma _0}(\alpha) \; \; = \; \;
\frac{\exp^{-\beta E_{\Gamma _0}}}{\sum_{\Gamma}
\exp^{-\beta\alpha n - \beta E'_{\Gamma}}} \; \; ,
\end{equation}
where $E_0$ is the energy of the native state, $n$ is the number of
non-native contacts and $E'_{\Gamma}$ is the part of energy in conformation
$\Gamma$ which corresponds to the native contacts.
Then
\begin{equation}
\frac{\partial P_{\Gamma _0}}{\partial \alpha} \; \; \geq
\; \; \beta <n>_T \; \geq 0 \; \; ,
\end{equation}
where $<n>_T$ is the average number of non-native bonds at temperature $T$.
So the probability of being in the native state cannot decrease with $\alpha$.
$T_f$, therefore, should increase with increasing $\alpha$ and
then become $\alpha$-independent.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig6.eps}}
\caption{Dependence of $N_{lm}$ on $\alpha$ for
$S_1$ and $S_2$.}
\end{figure}
It should be noted that
in Camacho's model \cite{Camacho} $T_f$
was found to increase with
$\alpha$ ($\alpha<0$) linearly. Our results presented in Figures 4 and
5 show that the region of $\alpha$ where one can observe the linear
dependence is rather narrow. Such region becomes much wider, for example,
in the case of the 27-monomer chain in three dimensions (the results are
not shown). Overall, the results for $T_f$, shown in Fig. 4 and 5,
demonstrate that the non-native contact repulsive energies improve
both the thermodynamic stability
and dynamical characteristics of folding.
We now focus on the $\alpha$ dependence of
the temperature at which the folding time is minimal, $T_{min}$.
The results for $S_1$ and $S_2$ are shown
in Fig. 4 and Fig. 5.
For positive values of $\alpha$, $T_f$ and $T_{min}$ are
comparable for both
sequences and they should be good folders\cite{Socci}.
For large negative values of $\alpha$, $T_{min}$ is bigger than $T_{f}$
but $S_1$ and $S_2$ remain good folders by the
Thirumalai-Camacho criterion\cite{Thirumalai,Camacho1} -
the peaks of the structural susceptibility and
the specific heat coincide for this interval of $\alpha$.
(For $\alpha=0$, the coincidence of the peaks
is, in fact, a general feature of the Go models because the proximity
to the native state and occurence of rapid changes in an average energy as a
function of $T$ are both controlled by establishment of the same
native contacts).
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig7.eps}}
\caption{The disconnectivity graph for $S_1$ and $\alpha =0$.
$N_c$ is a symbolic notation for the index labelling the local energy minima.}
\end{figure}
Fig. 4 and 5 also show the dependence of the minimal folding time,
$t_{min}$, on $\alpha$.
The dependence seems to saturate at large values of $\alpha$.
If one extends the Camacho result to positive $\alpha$
then $t_{min}$ should decrease with $\alpha$ exponentially\cite{Camacho}.
Our results suggest that such conclusion is valid only for $\alpha < 0$
but not for $\alpha>0$.
Thus, the repulsive non-native contact energies make
the polypeptide chain to fold faster which is similar
to what has been observed in Ref.\cite{Shriva,HP+}.
In order to understand why the repulsion improves the folding so much
we study the dependence of number of local minima, $N_{lm}$, on $\alpha$.
The results for $S_1$ and $S_2$ are shown in Fig.6.
$N_{lm}$ of $S_1$ is found to be smaller than for $S_2$.
Clearly, the number
of local minima strongly depends on $\alpha$ and for
$\alpha >0$ it become by 2 orders
of magnitude smaller than that for $\alpha <0$.
So, the positive values of $\alpha$ make the energy landscape less rugged
and the folding dynamics get faster.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig8.eps}}
\caption{The same as in Fig. 7 but for $\alpha =1$.
}
\end{figure}
In order to get more insight into the nature of the energy landscapes
in the models studied here, we use the
disconnectivity graph technique which maps the potential energy
surface onto the set of local minima \cite{Karplus,Wales}.
The technique involves checking what local energy minima are
connected by pathways (sets of moves that are allowed kinetically)
that do not exceed a given total threshold energy.
For each value of this energy the minima
are divided into disconnected sets of mutually accessible minima
separated by barriers.
The local minima which share
the lowest
energy threshold are joined at a node by lines and are called a basin
corresponding to the threshold.
The procedure of construction is stoped when one gets only
one basin for all of the minima.
For $S_1$ the number of local minima is equal to 152 and 81
for $\alpha =0$ and $\alpha =1$, respectively. The construction
of the graphs, therefore,
may be done exactly.
The disconnectivity
graphs obtained for $S_1$ and $\alpha =0$ and $\alpha =1$
are shown in Fig. 7 and Fig. 8, respectively. For both values
of $\alpha$ sequence $S_1$ is a good folder and consequently,
the structure corresponding to a folding funnel is clearly visible.
However, the funnel for the repulsive case
of $\alpha =1$ has a significantly less complex pattern than for
$\alpha =0$. Moreover, the repulsive non-native interactions
reduce the number of the local energy minima that have
links to the native state below any predetermined energy threshold.
Thus, the comparison of the
disconnectivity graphs also shows that the repulsion may
facilitate the folding substantially.
\begin{figure}
\epsfxsize=3.2in
\centerline{\epsffile{fig9.eps}}
\caption{The density of the ($V$-shape) local minima,
$P_{LM}$, in which the system
gets trapped at $T=0$ is plotted versus the number of native contacts
for $S_1$ and $S_2$.
We choose $\alpha = 0.1$ and $\alpha = -0.1$. The results are obtained
for the batch of 200 trajectories. In the case of $S_1$ the local minima are
different for $\alpha =-0.1$ and $\alpha =0.1$.
For $S_2$ there are 19 common local minima which are marked by dotted lines.}
\end{figure}
We now address the question of what happens with trapped local minima
when $\alpha$ changes from negative to positive values. Fig. 9 shows the
histogram of the local minima for $\alpha=-0.1$
and 0.1 for sequence $S_1$ and $S_2$. Interestingly, for $S_1$
none of the local
minima obtained
for $\alpha=-0.1$ appears for $\alpha=0.1$. The situation
changes dramatically for $S_2$ for which 19
local minima are common for both
$\alpha=-0.1$ and $\alpha=0.1$.
Our results suggest that for the
sequence with $\delta E =0$ the local
minima in which the chain is trapped at the negative values of $\alpha$
are effectively avoided if $\alpha$ is changed to positive values.
In conclusion, we state that for
the simple extended Go model
$T_f$ ($t_{min}$) increases (decreases) with the non-native contact
energy and it gets saturated for large values of $\alpha$.
The complexity of the disconnectivity trees becomes reduced on making
$\alpha$ more and more repulsive so that some two dimensional
sequences may even lose the Arrhenius like behavior of $t_{fold}$
at low temperatures.
It would be interesting to determine whether there are
any three-dimensional Go-like sequences that fold even at $T=0$.
Another important question is
what kind of an effective parameter $\alpha$,
or its conceptual equivalent, characterizes real proteins.
We thank Jayanth R. Banavar, P. Garstecki
and T. X. Hoang for many useful discussions.
This work was supported by KBN (grant No. 2P03B-025-13).
\vspace{0.5cm}
|
\section{Introduction}
\label{sec:intro}
The phenomena resulting from the deflection of electromagnetic radiation in
a gravitational field are referred to as {\em gravitational lensing} (GL)
and an object causing a detectable deflection is known as a {\em gravitational
lens}. The basic theory of GL was developed by Liebes\cite{Lie64},
Refsdal\cite{Ref64}, and Bourossa and Kantowski\cite{BK75}.
For detailed discussions on GL see the monograph by Schneider {\it et al.}
\cite{Schetal92} and reviews by Blandford and Narayan\cite{BN92},
Refsdal and Surdej\cite{RS94}, Narayan and Bartelmann\cite{NB96} and
Wambsganss\cite{Wam98}.
The discovery of quasars in 1963 paved the way for observing point
source GL. Walsh, Carswell and Weymann\cite{WCW79} discovered the first
example of GL. They observed twin images QSO 0957+561 A,B separated by $5.7$
arcseconds at the same redshift $z_s = 1.405$ and mag $\approx 17$.
Following this remarkable discovery more than a dozen convincing multiple-imaged
quasars are known.
The vision of Zwicky that galaxies can be lensed was crystallized when
Lynds and Petrosian\cite{LP86} and Soucail {\it et al.}\cite{Souetal87}
independently observed
giant blue luminous {\em arcs} of about $20$ arcseconds long in the rich
clusters of galaxies. Paczy\'{n}ski\cite{Pac87} interpreted these
giant arcs to be distorted images of distant galaxies located behind the
clusters. About $20$ giant arcs have been observed in the rich clusters.
Apart from the giant arcs, there have been also observed weakly distorted
{\em arclets} which are images of other faint background
galaxies\cite{Tys88}.
Hewitt {\it et al.}\cite{Hew88} observed the first Einstein ring MG1131+0456 at
redshift $z_s = 1.13$. With high resolution radio observations, they
found the extended radio source to actually be a ring of diameter
about $1.75$ {\em arcseconds}. There are about half a dozen observed rings
of diameters between $0.33$ to $2$ arcseconds and all of them are found in
the radio waveband; some have optical and infrared counterparts as
well\cite{Wam98}.
The general theory of relativity has passed experimental tests
in a weak gravitational field with flying colors; however, the theory
has not been tested in a strong gravitational field. Testing the
gravitational field in the vicinity of a compact massive object, such as
a black hole or a neutron star, could be a possible avenue for such
investigations. Dynamical
observations of several galaxies show that their centres contain
massive dark objects. Though there is no iron-clad evidence, indirect
arguments suggest that these are supermassive black holes; at least, the
case for black holes in the Galaxy as well as in NGC4258 appears to be strong
\cite{Ricetal98}. These could be possible observational targets to test
the Einstein theory of relativity in a strong gravitational field through GL.
Immediately after the advent of the general theory of relativity,
Schwarzschild obtained a static spherically symmetric asymptotically
flat vacuum solution to the Einstein equations, which was later found to
have an event horizon when maximally extended; thus this solution represents
the gravitational field of a spherically symmetric black hole (see in Hawking
and Ellis\cite{HE73}).
Schwarzschild GL in the weak gravitational field region (for which the
deflection angle is small) is well-known\cite{Schetal92}.
Recently Kling {\it et al.}\cite{KNP99} developed an iterative approach to GL
theory based on approximate solutions of the null geodesics equations,
and to illustrate their method they constructed the iterative lens
equations and time of arrival equation for a single Schwarzschild lens.
In this paper we obtain a lens equation that allows for the large
bending of light near a black hole, model the Galactic supermassive
``black hole'' as a Schwarzschild lens and study point source lensing
in the strong gravitational field region, when the bending angle can be
very large. Apart from a primary image and a secondary image
(which are observed
due to small bending of light in a weak gravitational field)
we get a theoretically infinite sequence of images on both sides
close to the optic axis; we term them {\em
relativistic images}. The relativistic images are formed due to large
bending of light in a strong gravitational field in the vicinity of $3M$,
and are usually greatly demagnified (the magnification decreases very fast
with an increase in
the angular position of the source from the optic axis). Though the
observation of relativistic images is a very
difficult task (it is very unlikely that they will be observed in near
future), if it ever were accomplished it would support the general
theory of relativity in a strong gravitational field inaccessible to test
the theory in any other known way and would also give an upper bound to the
compactness of the lens. This is the subject of study in this paper. We use
geometrized units (the gravitational constant $G = 1$ and the speed of light
in vacuum $c = 1$ so that $ M \equiv M G / c^2$).
\section{ Lens equation, magnification and critical curves}
\label{sec:lenseqn}
In this section we derive a lens equation that allows for the large
bending of light near a black hole.
The lens diagram is given in Fig.1. The line joining the observer $O$ and
the lens $L$ is taken as the reference (optic) axis. The spacetime under
consideration, with the lens (deflector) causing strong curvature, is
asymptotically flat; the observer as well as the source are situated in the
flat spacetime region (which can be embedded in an expanding Robertson-Walker
universe).
\begin{figure*}
\epsfxsize 6cm
\epsfbox{fig1.eps}
\caption[ ]
{ The lens diagram:
$O, L$ and $S$ are respectively the positions of the observer, deflector
(lens) and source. $OL$ is the reference (optic) axis. $\angle LOS$ and
$\angle LOI$ are the angular separations of the source and the image
from the optic axis. $SQ$ and $OI$ are respectively tangents to the
null geodesic
at the source and observer positions; $LN$ and $LT$, the perpendiculars
to these tangents from $L$, are the impact parameter $J$. $\angle OCQ$,
is the Einstein bending angle. $D_{s}$
represents the observer-source distance, $D_{ds}$ the lens-source distance
and $D_{d}$ the observer-lens distance.
}
\label{fig1}
\end{figure*}
$SQ$ and $OI$ are tangents to the null geodesic at the source and image
positions, respectively; $C$ is where their point of intersection would be
if there were no lensing object present. The angular
positions of the source and the image are measured from the optic axis $OL$.
$\angle LOI$ (denoted by $\theta$) is the image position
and $\angle LOS$ (denoted by $\beta$) is the source position if there were no
lensing object. $\hat{\alpha}$ (i.e. $\angle OCQ$) is the Einstein deflection
angle.
The null geodesic and the background broken geodesic path $OCS$ will be almost
identical, except near the lens where most of bending will take place. Given
the vast distances from observer to lens and from lens to source,
this will be a good approximation, even if the light goes round and round the
lens before reaching the observer.
We assume that the line joining the point $C$ and the location of the lens $L$
is perpendicular to the optic axis. This is a good approximation for small
values of $\beta$. We draw perpendiculars $LT$ and $LN$ from $L$ on the
tangents $SQ$ and $OI$ repectively and these represent the impact parameter
$J $. $D_s$ and $D_d$ stand for the distances of the source
and the lens from the observer, and $D_{ds}$ represents the lens-source
distance, as shown in the Fig. 1. Thus, the lens equation may be expressed as
\begin{equation}
\tan\beta = \tan\theta - \alpha ,
\label{LensEqn}
\end{equation}
where
\begin{equation}
\alpha \equiv
\frac{D_{ds}}{D_s} \left[\tan\theta + \tan\left(\hat{\alpha} - \theta\right)\right].
\label{Alpha}
\end{equation}
The lens diagram gives
\begin{equation}
\sin\theta = \frac{J}{D_d}.
\end{equation}
A gravitational field deflects a light ray and causes a change in the
cross-section of a bundle of rays. The magnification of an image is defined
as the ratio of the flux of the image to the flux of the unlensed source.
According to Liouville's theorem the surface brightness is preserved
in gravitational light deflection. Thus, the magnification of an image
turns out to be the ratio of the solid angles of the image and of the unlensed
source (at the observer). Therefore, for a circularly symmetric
GL, the magnification of an image is given by
\begin{equation}
\mu = \left( \frac{\sin{\beta}}{\sin{\theta}} \ \frac{d\beta}{d\theta} \right)^{-1}.
\label{Mu}
\end{equation}
The sign of the magnification of an image gives the parity of the image. The
singularities in the magnification in the lens plane are known as {\em critical
curves} (CCs) and the corresponding values in the source plane are known as
{\em caustics}. Critical images are defined as images of $0$-parity.
The tangential and radial magnifications are expressed by
\begin{equation}
\mu_t \equiv \left(\frac{\sin{\beta}}{\sin{\theta}}\right)^{-1}, ~ ~ ~
\mu_r \equiv \left(\frac{d\beta}{d\theta}\right)^{-1}
\label{MutMur}
\end{equation}
and singularities in these give {\em tangential critical curves} (TCCs)
and {\em radial critical curves} (RCCs), respectively; the corresponding
values in the source plane are known as {\em tangential caustic} (TC)
and {\em radial caustics} (RCs), respectively. Obviously, $\beta = 0$
gives the TC and the corresponding values of $\theta$ are the TCCs.
For small values of angles $\beta$, $\theta$ and $\hat{\alpha}$
equations $(\ref{LensEqn})$ and $(\ref{Mu})$ yield the
approximate lens equation and magnification, respectively,
which have been widely used in studying lensing in a weak gravitational field
\cite{Schetal92}.
\section{Schwarzschild spacetime and the deflection angle}
\label{sec:DefAngle}
The Schwarzschild spacetime is expressed by the line element
\begin{equation}
ds^2=\left(1-\frac{2M}{r}\right)dt^2- \left(1-\frac{2M}{r}\right)^{-1} dr^2
-r^2\left(d\vartheta^2+\sin^2 \vartheta d\phi^2\right),
\label{SchMetric}
\end{equation}
where $M$ is the Schwarzschild mass. When this solution is maximally extended
it has an event horizon at the Schwarzschild radius $R_s = 2 M$.
The deflection angle $\hat{\alpha}$ for a light ray with
closest distance of approach $r_o$ is (Chapters $8.4$ and $ 8.5$
in \cite{Wei72})
\begin{equation}
\hat{\alpha}\left(r_o\right) = 2 \ {\int_{r_o}}^{\infty}
\frac{dr}{r \ \sqrt{\left(\frac{r}{r_o}\right)^2 \left(1-\frac{2M}{r_o}\right)
-\left(1-\frac{2M}{r}\right)} } - \pi
\label{AlphaHatR0}
\end{equation}
and the impact parameter $J$ is
\begin{equation}
J = r_o \left(1-\frac{2M}{r_o}\right)^{-\frac{1}{2}} .
\label{ImpParaR0}
\end{equation}
A timelike hypersurface $\{r = r_0\}$ in a spacetime is defined as a photon
sphere if the Einstein bending angle of a light ray with the closest
distance of
approach $r_0$ becomes unboundedly large. For the Schwarzschild metric
$r_0 = 3M$ is the photon sphere and thus the deflection angle $\hat{\alpha}$
is finite for $r_0 > 3M$.
The Einstein deflection angle for large $r_o$ is\cite{Viretal98}
\begin{equation}
\hat{\alpha}\left(r_o\right) =
\frac{4M}{r_o} + \frac{4M^2}{{r_o}^2}\left(\frac{15\pi}{16}-1\right)
+ . . . . . \ \ \ .
\label{AlphaHatWkField}
\end{equation}
We mentioned the above result only for completeness as it is not much
known in the literature.
As we are interested to study GL due to light deflection in a strong
graviational field we will use Eq. $(\ref{AlphaHatR0})$ for any
further calculations.
Introducing radial distance defined in terms of the Schwarzschild radius,
\begin{equation}
x = \frac{r}{2M} , ~ ~ ~
x_o = \frac{r_o}{2M} ,
\label{XX0}
\end{equation}
the deflection angle $\hat{\alpha}$ and the impact paprameter $J$
take the form
\begin{equation}
\hat{\alpha}\left(x_o\right) = 2 \ {\int_{x_o}}^{\infty}
\frac{dx}{x \ \sqrt{\left(\frac{x}{x_o}\right)^2
\left(1-\frac{1}{x_o}\right)
-\left(1-\frac{1}{x}\right)}} - \pi
\label{AlphaHatX0}
\end{equation}
and
\begin{equation}
J = 2M x_o \left(1-\frac{1}{x_o}\right)^{-\frac{1}{2}}.
\label{ImpParaX0}
\end{equation}
In the computations in the following section we require the first derivative
of the deflection angle $\hat{\alpha}$ with respect to $\theta$. This is
given by (see in \cite{Viretal98})
\begin{equation}
\frac{d\hat{\alpha}}{d\theta} = \hat{\alpha}'\left(x_o\right)
\frac{dx_o}{d\theta} ,
\label{DAlphaByDTheta}
\end{equation}
where
\begin{equation}
\frac{dx_o}{d\theta} =
\frac{
x_o \left(1-\frac{1}{x_o}\right)^{\frac{3}{2}}
\sqrt{1-\left(\frac{2M}{D_{d}}\right)^2 {x_o}^2 \left(1-\frac{1}{x_o}\right)^{-1}}
}
{\frac{M}{D_{d}} \left(2x_o-3\right)}
\label{DX0ByDTheta}
\end{equation}
and the first derivative of $\hat{\alpha}$ with respect to $x_o$ is
\begin{equation}
\hat{\alpha}'\left(x_o\right) = \frac{3-2x_o}{{x_o}^2\left(1-\frac{1}{x_o}\right)}
{\int_{x_o}}^{\infty}
\frac{\left(4 x - 3\right) dx}
{\left(3 - 2 x\right)^2 \ x \ \sqrt{\left(\frac{x}{x_o}\right)^2
\left(1-\frac{1}{x_o}\right)
-\left(1-\frac{1}{x}\right)}} .
\label{DAlphaHatByX0}
\end{equation}
\section{Lensing with the Galactic supermassive ``black hole''}
It is known that the Schwarzschild GL in a weak gravitational field gives
rise to an Einstein ring when the source, lens and observer are aligned,
and a pair of images (primary and secondary) of opposite parities when
the lens components are misaligned.
However, when the lens
is a massive compact object a strong gravitational field is
``available'' for investigation. A light ray can pass close to the
photon sphere and go around the lens once, twice, thrice, or many times
(depending on the impact parameter $J$ but for $J>3\sqrt{3} M$) before
reaching the observer. Thus, a massive
compact lens gives rise, in addition to the primary and secondary images, to a
large number (indeed, theoretically an infinite sequence) of
images on both sides of the optic axis. We call these images (which are formed
due to the bending of light through more than $3\pi /2$) {\em relativistic
images}, as the light rays giving rise to them pass through a strong
gravitational field before reaching the observer.
We call the rings which are
formed by bending of light rays more than $2 \pi$, {\em relativistic
Einstein rings}.
We model the Galactic supermassive ``black hole'' as a Schwarzschild lens.
This has mass $M = 2.8 \times 10^6 M_{\odot}$ and the distance $D_d =
8.5 kpc$\cite{Ricetal98}; therefore, the ratio of the mass to the
distance $M/D_d \approx 1.57 \times 10^{-11}$. We consider a point
source, with the lens situated half way between the source and the
observer, i.e. $D_{ds}/D_s = 1/2$. We allow the angular
position of the source to change keeping $D_{ds}$ fixed.
We compute positions and magnifications of two pairs of outermost relativistic
images as well as the primary and secondary images
for different values of the angular positions of the source. These
are shown in figures 2 and 3 and Table 1 (for relativistic images)
and in Fig. 4 and Table 2 (for primary and secondary images). The angular
positions of the primary and secondary images as well as the critical curves
are given in arcseconds; those for relativistic images as well as relativistic
critical curves are expressed in microarcseconds.
In Fig.2 we show how the positions of outer two relativistic images
on each side of the optic axis change as the
source position changes. To find the angular positions of images on the
same side of the source we plot $\alpha$ (represented by
continuous curves on right side of the figure) and
$\tan\theta-\tan\beta$ (represented by dashed curves)
against $\theta$ for a given
value of the source position $\beta$; the points of intersection give the
image positions (see the right side of the Fig. $3$).
\begin{figure*}
\hbox{\epsfxsize=8 cm\epsfbox{fig2a.eps}\hskip 0.1 cm
\epsfxsize=9 cm\epsfbox{fig2b.eps}}
\caption[ ]{ This gives relativistic image positions for a given source
position. $\alpha$ and $\tan\theta-\tan\beta$ are plotted against
the angular position $\theta$ of the image; these are represented by
the continuous and the dashed curves, respectively. For a given position
of the source, the points of intersections of the continuous curves
(the two outermost ones on each side being shown)
with the dashed curves give the angular positions of relativistic images.
The Galactic ``black hole'' (mass $M= 2.8 \times 10^6 M_{\odot}$
and the distance $D_d = 8.5 kpc$ so that $M/D_d \approx 1.57
\times 10^{-11}$) serves as the lens, $D_{ds}/D_s = 1/2,$
and $\beta = \mp 0.075$ radian ($\approx \mp 4.29718\mathop{\rm {{}^\circ}} $). $\theta$
is expressed in microarcseconds. The angular position of a relativistic
image changes very slowly with respect to a change in the source
position.
}
\label{fig2}
\end{figure*}
\begin{table*}
\caption[ ]{ Magnifications and positions$^{\rm c}$ of relativistic images }
\begin{flushleft}
\begin{tabular}{ccccc}
\multicolumn{2}{c}{Images on the opposite side of the source }&
Source position $\beta$ &
\multicolumn{2}{c}{ Images on the same side of the source} \\
\qquad$\mu^{outer}$&$\mu^{inner}$&
$ $
&\qquad$\mu^{inner}$&$\mu^{outer}$ \\
\hline\noalign{\smallskip}
\qquad\phantom{11}$-3.5 \times 10^{-12}$ &\phantom{11} $-6.5 \times 10^{-15}$
&\qquad $1$
&\qquad $6.5 \times 10^{-15}$ &\phantom{11} $3.5 \times 10^{-12}$ \\
\qquad\phantom{11}$-3.5 \times 10^{-13}$ &\phantom{11} $-6.5 \times 10^{-16}$
&\qquad $10$
&\qquad $6.5 \times 10^{-16}$ &\phantom{11} $3.5 \times 10^{-13}$ \\
\qquad\phantom{11}$-3.5 \times 10^{-14}$ &\phantom{11} $-6.5 \times 10^{-17}$
&\qquad $10^2$
&\qquad $6.5 \times 10^{-17}$ &\phantom{11} $3.5 \times 10^{-14}$ \\
\qquad\phantom{11}$-3.5 \times 10^{-15}$ &\phantom{11} $-6.5 \times 10^{-18}$
&\qquad $10^3$
&\qquad $6.5 \times 10^{-18}$ &\phantom{11} $3.5 \times 10^{-15}$ \\
\qquad\phantom{11}$-3.5 \times 10^{-16}$ &\phantom{11} $-6.5 \times 10^{-19}$
&\qquad $10^4$
&\qquad $6.5 \times 10^{-19}$ &\phantom{11} $3.5 \times 10^{-16}$ \\
\qquad\phantom{11}$-3.5 \times 10^{-17}$ &\phantom{11} $-6.5 \times 10^{-20}$
&\qquad $10^5$
&\qquad $6.5 \times 10^{-20}$ &\phantom{11} $3.5 \times 10^{-17}$ \\
\qquad\phantom{11}$-3.5 \times 10^{-18}$ &\phantom{11} $-6.5 \times 10^{-21}$
&\qquad $10^6$
&\qquad $6.5 \times 10^{-21}$ &\phantom{11} $3.5 \times 10^{-18}$ \\
\noalign{\smallskip}
\noalign{\smallskip}
\end{tabular}
\end{flushleft}
\begin{list}{}{}
\item[$^{\rm a}$]
The lens is the Galactic ``black hole'' (mass $M= 2.8 \times 10^6
M_{\odot} $ and the distance $D_d = 8.5$ kpc so that $M/D_d
\approx 1.57 \times 10^{-11} $). The ratio of the lens-source
distance $D_{ds}$ to the observer-source distance $D_s$ is taken to be
$1/2$. Angles are given in microarcseconds.
\item[$^{\rm b}$]
$\mu$ is the magnification and the sign on this refers to the parity of
the image.
\item[$^{\rm c}$]
For the source positions considered here, the angular positions of two
pairs of outermost relativistic images are $\approx \pm 16.898$ and
$\approx \pm 16.877$ microarcseconds ( $+$ sign refers to images on the
same side of the source and $-$ sign refers to images on opposite side of
the source).
\end{list}
\end{table*}
Similarly, we plot $- \alpha$ and $-\tan\theta-\tan\beta$ vs. $-\theta$ and
points of intersection give the image positions on the opposite side of
the source (see left side of the Fig. $2$). We have taken $\beta = \mp
0.075$ radian ($\approx \mp 4.29718\mathop{\rm {{}^\circ}}$). In fact there are a sequence
of theoretically an infinite number of
continuous curves which intersect with a given dashed curve giving rise
to a sequence of an infinite number of images on both sides of the optic
axis. We have plotted only two sets of such curves
(note that the third set of continuous curves comes to be very close to the
second set and therefore it is not possible to show them in the same figure)
demonstrating appearance of two relativistic images on both sides of the
optic axis. For $\beta=0$
the points of intersection of the continuous curves with the dashed
curve give a sequence of infinite number of relativistic tangential critical
curves (relativistic Einstein rings). As $\beta$ increases any image on
the same side of source moves away from the optic axis, whereas any image
on the opposite side of the source moves towards the optic axis. The
displacement of relativistic images with respect to a change in the
source position is very small (see Fig. $2$). The two sets of
outermost relativistic images
are formed at about $17$ microarcseconds from the optic axis.
In Fig. 3 we plot the tangential magnification $\mu_t$ as well as the
total magnification $\mu$ vs. the image position $\theta$ near the two
outermost relativistic tangential critical curves. The singularities in
$\mu_t$ give the angular radii of the two relativistic Einstein rings.
In Fig. 4 we plot the same for the primary-secondary images; the singularity
in $\mu_t$ gives the angular position of the Einstein ring.
The magnification for relativistic images falls extremely fast (as compared
with the case of primary and secondary images) as the source position increases
from perfect alignment.
The tangential parity (sign of $\mu_t$) as well as the total parity (sign of
$\mu$) are positive for all images on the same side of the source and
negative for all images on the opposite side of the source. The radial
parity (sign of $\mu_r$) is positive for all the images in Schwarzschild
lensing.
\begin{figure*}
\epsfxsize 17cm
\epsfbox{fig3.eps}
\caption[ ]{
The {\em tangential magnification} $\mu_t$ denoted by
dotted curves and the {\em total magnification} $\mu$ denoted by
continuous curves are plotted against the image position
(expressed in microarcseconds) near relativistic tangential critical
curves. The figures on the right side give the magnification for the
outermost relativistic image, whereas those on left side are for
a relativistic image adjacent to the previous one. The lens is the
Galactic ``black hole'' ($M/D_d \approx 1.57 \times 10^{-11}$) and
$D_{ds}/D_s = 1/2$. The singularities in magnifications show
the angular positions of the relativistic tangential critical curves.
The origin of the $\theta$-axis for the figures on the left side is $16.87715$
and for each on right side is $16.89825$.
}
\label{fig3.eps}
\end{figure*}
\begin{table*}
\caption[ ]{Positions and magnifications of primary and secondary images }
\begin{flushleft}
\begin{tabular}{crccr}
\multicolumn{2}{c}{Secondary images }&
Source position $\beta$ &
\multicolumn{2}{c}{ Primary images } \\
\qquad$\theta$ & $\mu$ &
$$&
\qquad$\theta$ & $\mu$ \\
\hline\noalign{\smallskip}
\qquad\phantom{11}$1.157494$ &\phantom{11} $-5787.20 $
&\qquad $10^{-4}$
&\qquad $1.157594$ &\phantom{11} $5788.21 $ \\
\qquad\phantom{11}$1.157045$ &\phantom{11} $-578.27 $
&\qquad $10^{-3}$
&\qquad $1.158045$ &\phantom{11} $579.27 $ \\
\qquad\phantom{11}$1.152555$ &\phantom{11} $-57.38 $
&\qquad $10^{-2}$
&\qquad $1.162555$ &\phantom{11} $58.38 $ \\
\qquad\phantom{11}$1.108619$ &\phantom{11} $-5.30 $
&\qquad $10^{-1}$
&\qquad $1.208624$ &\phantom{11} $6.30 $ \\
\qquad\phantom{11}$0.760918$ &\phantom{11} $-0.23 $
&\qquad $1$
&\qquad $1.760914$ &\phantom{11} $1.23 $ \\
\qquad\phantom{11}$0.529680$ &\phantom{11} $-0.05 $
&\qquad $2$
&\qquad $2.529674$ &\phantom{11} $1.05 $ \\
\qquad\phantom{11}$0.394711$ &\phantom{11} $-0.01 $
&\qquad $3$
&\qquad $3.394704$ &\phantom{11} $1.01$ \\
\qquad\phantom{11}$0.310831$ &\phantom{11} $-0.005 $
&\qquad $4$
&\qquad $4.310823$ &\phantom{11} $1.005$ \\
\qquad\phantom{11}$0.254986$ &\phantom{11} $-0.002 $
&\qquad $5$
&\qquad $5.254977$ &\phantom{11} $1.002$ \\
\end{tabular}
\end{flushleft}
\begin{list}{}{}
\item[$^{\rm a}$]
The same as in Table 1, except angles are given here in arcseconds.
\end{list}
\end{table*}
\begin{figure*}
\epsfxsize 10cm
\epsfbox{fig4.eps}
\caption[ ]{
The {\em tangential magnification} $\mu_t$ (represented by dotted curves)
and the {\em total magnification} $\mu$ (represented by continuous curves)
are plotted, near the outermost tangential critical curve, as a function
of the image position (in arcseconds). The singularity gives the position
of the outermost tangential critical curve (angular radius of the Einstein ring).
The Galactic ``black hole'' is the lens and $D_{ds}/D_s$ is taken to be
$1/2$.
}
\label{fig4.eps}
\end{figure*}
\begin{table*}
\caption[ ]{Einstein and relativistic Einstein rings}
\begin{flushleft}
\begin{tabular}{lllll}
Rings & $\theta_E$ & $\hat{\alpha}$ &$\frac{r_o}{2M}$ \\
\noalign{\smallskip}
\noalign{\smallskip}
\hline\noalign{\smallskip}
Einstein ring & $1.157544$ $arcsec$ & $2.315089$ $arcsec$ & $178193$ \\
Relativistic Einstein ring I & $16.898$ $\mu$$as$ & $2\pi+33.80$ $\mu$$as$
& $1.545115$\\
Relativistic Einstein ring II & $16.877$ $\mu$$as$ & $4\pi+33.75$ $\mu$$as$
& $1.501875$\\
\noalign{\smallskip}
\end{tabular}
\end{flushleft}
\begin{list}{}{}
\item[$^{\rm a}$]
The same as ({\rm a}) of Table 1, except $arcsec$ and $\mu as$
used here refer to arcseconds and microarcseconds, respectively.
$\theta_E$ stands for the angular positions of tangential critical
curves.
\end{list}
\end{table*}
In Table 3 we give the angular radii $\theta_E$ of the Einstein and two
relativistic Einstein rings. We also give the corresponding values for the
deflection angle $\hat{\alpha}$ and the closest distance of approach
$x_o$ for the light rays giving rise to these rings. We define an ``effective
deflection angle'' $\hat{\alpha} - 2 \pi $ times the number of revolution
the light ray has made before reaching the observer. Table 3 shows that the
effective deflection angle for a ring decreases with the decrease in its
angular radius, which is expected from the geometry of the lens diagram. The
same is true for images on the same side of the optic axis, i.e. the effective
deflection angle is less for images closer to the optic axis.
\newpage
The supermassive ``black holes'' at the centres of $NGC3115$
and $NGC4486$ have $M/D_d \approx 1.14 \times 10^{-11}$ and $1.03
\times 10^{-11}$, resepctively\cite{Ricetal98}, which are very close to the
case of the Galactic ``black hole'' we
have studied. Therefore, if we study lensing with these ``black holes''
keeping $D_{ds}/D_s = 1/2$, we will get approximately the
same results. The
angular radius of the Einstein ring in the Schwarzschild lensing
is expressed by $\theta_E = \{4M D_{ds}/(D_d D_s)\}^{1/2}$.
For a source with $D_{ds} < D_s$ one has $0 < \left( D_{ds}/D_s\right) <1$.
If we consider $D_{ds}/D_s$ different than $1/2$
the magnitude of the Einstein ring can easily be estimated.
As relativistic images are formed due to light deflection
close to $r_o = 3M$, their angular positions will be very
much less sensitive to a change in the value of $D_{ds}/D_s$.
We have considered the sources for $D_d < D_s$; however, sources with
$D_d > D_s$ will also be lensed and will also give rise to relativistic
images. Thus, all the sources of the universe will be mapped as relativistic
images in the vicinity of the black hole (albeit as very faint images).
Gravitational lensing with stellar-mass black holes
will also give rise to relativistic images; however, unlike in the
case of supermassive ``black hole'' lensing, these images
will not be resolved from their primary and secondary images with present
observational facilities.
\section{Relativistic images as test for general relativity in strong
gravitational field}
For the Galactic ``black hole'' lens, Fig. 2 and Table 1 (see caption
{\rm (c)} ) show the angular positions of the two outermost sets of
relativistic images (two images on each side of the optic axis)
when a source position is given.
In fact, there is a sequence of a large number of relativistic Einstein rings
when the source, lens and observer are perfectly aligned, and when the
alignment is ``broken'' there is a sequence of large number of relativistic
images on both sides of the optic axis. However, for a
given source position their magnifications decrease very fast as the
angular position $\theta$ decreases (see Table 1), and therefore the outermost
set of images, one on each side of the optic axis, is observationally
the most significant. The angular separations among relativistic images are
too small to be resolved with presently available instruments and therefore
all these images would be at the same position; however, these
relativistic images will be resolved from the primary and secondary images
and thus resolution is not a problem for observation of relativistic
images.
If we observe a full or ``broken'' Einstein ring near
the centre of a massive dark object at the centre of a galaxy with a faint
relativistic image of the same source at the centre of the ring, we would
expect that the central (relativistic) image would
disappear after a short period of time. If seen, this would be a great success
of the general theory of relativity in a strong gravitational field.
Observation of relativistic images would also give an upper bound on the
compactness of the lens. To get a relativistic image a light ray has to
suffer a deflection by an angle $\hat{\alpha} > 3 \pi/2$. For the closest
distance of approach $r_o = 3.208532 M$ the deflection angle $\hat{\alpha}
= 269.9999 \mathop{\rm {{}^\circ}} $ and therefore $r_o/M = 3.208532$ can be considered
as an upper bound to the compactness of the lens.
The fact that the magnification of a relativistic image decreases very fast
as the source position increases from its perfect alignment with the lens and
observer can be exploited to give a better estimate of the compactness of the
lens. For the lens system considered in section four, the outermost
relativistic Einstein ring has angular radius about $16.898$ microarcseconds
and this is formed due to light rays bending at the closest distance of
approach $r_o \approx 3.09023 M_{\odot}$ (see Table $3$). As a relativistic
image can be observed only very close to a relativistic TCC, the above value
of the $r_o/M$ gives an estimate of compactness of the massive dark object.
There are some serious difficulties hindering the observation of
the primary-secondary image pair near a galactic centre; the observation
of relativistic images is even much more difficult. The
extinction of electromagentic radiation near the line of sight to galactic
nuclei would be appreciable; the smaller the wavelength, the larger the
extinction. The interstellar scattering and radiation at several frequencies
from the material accreting on the ``black hole'' would make these observations
more difficult. Due to these obstacles no lensing event near a galactic centre
has been observed till now, but it seems this is a very worthwhile project.
There are some additional difficulties for observing relativistic images.
First, these images are very much demagnified unless the source, lens
and observer are highly aligned.
When the source position $\beta$
decreases the magnification increases rapidly and therefore one may possibly
get observable relativistic images, but only if the source, lens and observer
are highly aligned ($\beta << 1$ microarcsecond) and the source has a large
surface brightness. Quasars and supernovae would be ideal sources for
observations of relativistic images. The number of observed quasars is low
(about $10^4$, see in \cite{Wam98}) and therefore the probability that a
quasar will be highly aligned along the direction of any galactic centre
of observed galaxies is extremely small. Similarly, there is a very small
probability that a supernova will be strongly aligned with any galactic
centre. We considered a normal star in the Galaxy to be a point
source (note that we took $D_{ds}/D_s = 1/2$). We cannot
use the point source approximation when such a source is very close to the
caustic ($\beta = 0$) and therefore studies of extended source lensing are
needed. Second, if relativistic images were observed it would be for a short
period of time because the magnification decreases very fast with increase
in the source position; however, the time scale for observation of
relativistic images will be greater for lensing of more distant sources.
It is highly improbable that the relativistic images
would in fact be observed in a short observing period and a long term
project to search for such images would not have reasonable probability
of success. Nevertheless the
possibility remains that such images might be detected through lucky
observations in the vicinity of galactic centers.
\section{Summary}
We obtained a lens equation which allows an arbitrary large
values of the deflection angle and used the deflection angle expression
for the Schwarzschild metric obtained by Weinberg\cite{Wei72}. This gives
the bending angle of a light
ray passing through the Schwarzschild gravitational field for a closest
distance of approach $r_o$ in the range $3M < r_o < \infty$. Using this we
studied GL due to the Galactic ``black hole'' in a strong
gravitational field.
Apart from a pair of images (primary and secondary) which are observed due to
light deflection in a weak gravitational field, we find a sequence of large
number of relativistic images on both sides of the optic axis due to large
deflections of light in a strong gravitational field near the photon sphere
$r_o = 3M$. Among these relativistic images, the outermost pair is
observationally the most important. Though these relativistic images are
resolved from the primary and secondary images, there are serious difficulties
in observing them. However, if it were to succeed it would be a great triumph
of the general theory of relativity and would also provide valuable
information about the nature of massive dark objects.
Observations of relativistic images would confirm the Schwarzschild geometry
close to the event horizon; therefore these would strongly support the black
hole interpretation of the lensing object.
In the investigations in this paper we modelled the massive compact
objects as Schwarzschild lens. However, it is worth investigating Kerr
lensing to see the effect of rotation on lensing in strong gravitational
field, especially when the lens has large intrinsic angular momentum to
the mass ratio. There have been some studies of Kerr weak field lensing
(see Rauch and Blandford\cite{RB94} and references therein). In passing,
it is worth mentioning that any spacetime endowed with a photon sphere
(as defined in Section $\ref{sec:DefAngle}$) and acting as a gravitational
lens would give rise to relativistic images.
\begin{acknowledgements}
Thanks are due to H. M. Antia, M. Dominik, J. Kormendy, J. Lehar, and
D. Narasimha for helpful correspondence, and J. Menzies and P. Whitelock
for helpful discussions on the visibility of images. This research was
supported by FRD, S. Africa.
\end{acknowledgements}
|
\section{Introduction}
The properties of vector meson in nuclear medium have been the focus of
current interest due to their potential role to provide one with a direct
observable of the nuclear medium effects, associated with chiral
symmetry restoration, via dileptons in p-A or A-A reactions\cite{L98}.
Indeed dileptons from Relativistic Heavy
Ion Collisions (RHIC)\cite{CERES} seemed to suggest a non-trivial change of
the
vector meson spectral density in a hot/dense environment, which can be
understood in terms of model calculations\cite{LKB95} based on
decreasing vector meson masses in hot/dense medium\cite{BR91,HL92}.
However, some model calculations\cite{RCW97,KW98} based on changes of the
vector meson spectral densities obtained using effective hadronic model
calculations with no decrease in the mass also seem to
explain the main features of the CERES data.
In all of the approaches, the central question is, how the
spectral density changes in hot/dense matter\cite{BR98}.
In this talk, I will discuss the results for the scalar mass
shift and dispersion effects (three momentum dependence) of the peak position
of the spectral density for the light quark system ($\rho, \omega$),
the strange quark system ($\phi$) and the heavy quark system ($J/\psi$)
in nuclear medium.
\subsection{Vector mesons in vacuum}
The distinctive features appearing in the dilepton spectrum in p-A or A-A
reactions are
the vector meson peaks; the $\rho$, $\phi$ and the $J/\psi$, whose
(mass, width) in the vacuum are
(770, 150), (1020, 4.4), (3100, 0.086) MeV. It is interesting to compare
the phase spaces of their decay into their corresponding two pseudo
particles. For the $\rho$, its two pion decay, which accounts for
most of the total width, has a large phase space.
This is so because the pion is a Goldstone boson and has a small mass.
For the $J/\psi$, its decay
into D-mesons are forbidden, because the D mesons are not Goldstone bosons
and two times its mass is greater than the mass of $J/\psi$.
For the $\phi$, the situation is something
in between and
its decay into two kaons has very little phase space.
Hence, chiral symmetry breaking is partly
responsible for the large difference in their width.
As for the masses of the vector mesons, among other models, QCD sum rules
provide an indirect relation to QCD condensates. For the light quark system,
$\langle \bar{q} q \rangle$ is dominantly responsible for its mass,
for the strange quark system, $\langle \bar{s}s \rangle$ is responsible,
and for the $J/\psi$, the charmed quark mass and a small non
perturbative contribution from the gluon condensate $\langle
\frac{\alpha_s}{\pi} G^2 \rangle$ are responsible.
Therefore, if chiral symmetry gets restored at finite density or temperature,
non-trivial changes will occur to the masses and widths of the vector
mesons.
\subsection{Quark condensate at finite density}
The temperature dependence of the quark condensates have been calculated long
ago in the lattice\cite{K83,FU86}.
The result is that the light quark condensate will
go to zero above the critical temperature. For heavier quark masses,
the changes become smaller.
At present, due to technical reasons, no lattice result exist at finite
density. In all what follows, we will use linear density approximation
\begin{eqnarray}
\langle O \rangle_{\rho_n} =\langle O \rangle_0+ \rho_n
\times \langle O \rangle_N
\end{eqnarray}
where $ O$ is any operator, $\rho_n$ the nucleon density, and the subscripts
$\rho_n, 0 ,N$ denotes
the nuclear, the vacuum and the nucleon expectation values.
Then, we have the following model
independent result\cite{DL91,CFG92} for the quark condensate
\begin{eqnarray}
\langle \bar{q} q \rangle_{\rho_n}= \langle \bar{q} q \rangle_0
+ { \Sigma_{\pi N} \over 2 \hat{m} } \rho_n
\sim \langle \bar{q} q \rangle_0 \cdot \biggl(1-0.2 \, \frac{\rho_n}{\rho_0}
\biggr),
\end{eqnarray}
and for the strange part,
\begin{eqnarray}
\langle \bar{s} s \rangle_{\rho_n}= \langle \bar{s} s \rangle_0
+ y { \Sigma_{\pi N} \over 2 \hat{m} } \rho_n
\sim \langle \bar{s} s \rangle_0 \cdot \biggl(1-0.05 \, \frac{\rho_n}{\rho_0}
\biggr),
\end{eqnarray}
where, $y=2 \langle \bar{s} s \rangle_N /( \langle \bar{u} u \rangle_N
+ \langle \bar{d} d \rangle_N )$ and $\rho_0$ is the nuclear matter
saturation density.
One notes that already at nuclear matter, we have partial restoration of
chiral symmetry; namely, the chiral order parameter is reduced
by 20\%. Nuclear matter provides a stable environment with non trivial
vacuum changes. Hence, if anything happens to the vector mesons at high
temperature or density, the tendencies will already be apparent at nuclear
matter.
\subsection{Gluon condensate at finite density}
The temperature dependence of the gluon condensates has also
been calculated on the lattice\cite{CD87,Lee89,AHZ91,KB93}.
The result is that it will stay almost constant up to the critical temperature
and then reduce to about 60\% of its vacuum value above the critical
temperature.
The leading density behavior can be obtained from the trace anomaly relation
to leading order in $\alpha_s$\cite{DL91}
\begin{eqnarray}
T_\mu^\mu=-\frac{9}{8} \frac{\alpha_s}{\pi} G^2+ \sum m_q \bar{q} q
\end{eqnarray}
Taking the nucleon expectation value of it and using the most recent
determination of the nucleon mass in the chiral limit\cite{BM96}, we have
\begin{eqnarray}
\langle \frac{\alpha_s}{\pi} G^2 \rangle_{\rho_n}=
\langle \frac{\alpha_s}{\pi} G^2 \rangle_0 \cdot
\biggl(1-0.05 \, \frac{\rho_n}{\rho_0}
\biggr)
\label{gluon}
\end{eqnarray}
Hence, we have a non-trivial change in the gluon condensate, although its
relative change is smaller than the case of the quark condensate.
As we have seen, the condensates have non-trivial change in the nuclear
medium, which will have a non-trivial effect on the vector meson properties
in nuclear medium. Experimentally, there are attempts to produce and observe
the decay of vector mesons inside a heavy nuclei and also produce
meson bound nuclei.
Hence, the theoretical study and the subsequent experimental verification of
changes of vector meson properties in nuclear medium will provide a solid
basis for understanding the hadronic effects in RHIC and medium effects in
general.
\section{QCD constraints}
There have been many model calculations to study vector meson properties
in nuclear medium. Here, we will avoid any model calculation and derive
a constraint equation that any model calculation should satisfy.
The foundation of this approach was laid in ref.\cite{HL92}
Consider the correlation function of the vector current
$J_\mu=\bar{q} \gamma_\mu q$ at finite density;
\begin{eqnarray}
\Pi_{\mu\nu} (\omega^2, {\bf q}^2 ) &=& i \int d^4x e^{iqx}\langle
T [ J_\mu(x) J_\nu(0) ] \rangle_{\rho_n}.
\label{ope1}
\end{eqnarray}
Here $q=(\omega,{\bf q})$.
In what follows, when we give result for explicit vector meson, we will
use the currents $J_\mu^{\rho,\omega}=\frac{1}{2} ( {\bar u} \gamma_\mu
u \mp {\bar d} \gamma_\mu d )$ for the $\rho,\omega$ mesons,
$J_\mu^{\phi}={\bar s} \gamma_\mu s $ for the $\phi$ and
$J_\mu^{J/\psi}={\bar c} \gamma_\mu c $ for the $J/\psi$.
In general, because the vector current is conserved,
the polarization tensor in eq.(\ref{ope1}) will have only two invariant
functions\cite{tensors}.
\begin{eqnarray}
\Pi_{\mu\nu}(\omega^2,{\bf q}^2)=\Pi_T q^2 {\rm P}^T_{\mu\nu}+ \Pi_L q^2
{\rm P}^L_{\mu\nu},
\label{ope2}
\end{eqnarray}
where we assume the ground state to be at rest, such that, $
{\rm P}^T_{00} = {\rm P}^T_{0i}={\rm P}^T_{i0}=0$,
${\rm P}^T_{ij} = \delta_{ij}-{\bf q}_i {\bf q}_j/{\bf q}^2 $ and $
{\rm P}^L_{\mu\nu} = (q_\mu q_\nu/q^2-g_{\mu\nu}- {\rm P}^T_{\mu\nu})$.
When ${\bf q} \rightarrow 0$,
$\Pi_L=\Pi_T$, as in the vacuum.
We will
make a small ${\bf q}$ expansion of the correlation function and look
at its energy dispersion relation at fixed ${\bf q}$,
\begin{eqnarray}
{\rm Re} \Pi_{L,T}(\omega^2,{\bf q}^2) = {\rm Re} \left( \Pi^0(\omega^2,0)+
\Pi_{L,T}^1(\omega^2,0) ~ {\bf q}^2 + \cdot \cdot \right) \nonumber \\
= \int_0^\infty du^2 \left(
{\rho^0(u^2,0) \over (u^2-\omega^2)} + {\rho^1_{L,T}(u^2,0)
\over (u^2-\omega^2)} ~ {\bf q}^2 + \cdot \cdot \right),
\label{ope3}
\end{eqnarray}
where $\rho(u^2,{\bf q}^2)=1/\pi {\rm Im} \Pi^R(u^2 ,{\bf q}^2)$, and
$R$ denotes the retarded correlation function.
We will construct a constraint equation for $\Pi^0,\Pi_L^1$ and $\Pi_T^1$.
For $\Pi_L^1,\Pi_T^1$, we will only look at the ``non-trivial'' ${\bf q}$
dependence\cite{L98c}. A simple method to extract the
``non-trivial'' ${\bf q}$ dependence is to express the polarization function
in terms of $Q^2, {\bf q}^2$ ($\Pi(Q^2, {\bf q}^2)$)
and extract the linear ${\bf q}^2$ term.
In general for each polarization functions ($\Pi^0,\Pi_L^1,\Pi_T^1$),
the OPE \cite{Wilson69,muta} looks as follows,
\begin{eqnarray}
\label{eq2}
\Pi (\omega^2)=\sum_n C_{n} \,
\langle O_{n} \rangle.
\end{eqnarray}
Here the $O_n$ are operators of (mass) dimension $n$, renormalized at
a scale $\mu^2$, and $C_{n}$ are the perturbative Wilson
coefficients, which for the light quark system can be written as
$C_{n}=\frac{c_n}{(-\omega^2)^{n/2}}$ and for heavy quark system as
$C_{n}=\frac{c_n}{m_h^n}$.
Let us first discuss the light quark system.
After the Borel transformation, the dispersion relation for any one of the
polarization function($\Pi^0,\Pi_L^1,\Pi_T^1$) becomes
\begin{eqnarray}
{\rm B.T. \,}{\rm Re}\Pi(M^2)=
\int ds \rho(s)e^{-s/M^2},
\label{constraint}
\end{eqnarray}
The left hand side, which is the Borel transform (B.T.) of the OPE,
looks like
the following when including operators up to dimension 6,
\begin{eqnarray}
{\rm B.T. \,}{\rm Re}\Pi(M^2)={\rm B.T. \,}\Pi(M^2)_{pert}
+\frac{c_4}{1! M^2}\langle O_{4} \rangle
+\frac{c_6}{2! M^4}\langle O_{6} \rangle
\end{eqnarray}
The truncation is valid as long as $M^2$ is sufficiently large. The
minimum $M^2_{min}$ is usually determined by requiring the correction from
higher dimensional operators to be less than 30\% of the perturbative
contribution.
Now the constraints for the spectral density would be eq.(\ref{constraint}),
applied above $M^2> M^2_{min}$.
As can be seen in eq.(\ref{constraint}), the Borel transformation also
changes the
weighting factor of the spectral
density to an $exp(-s/M^2)$. This has the following advantage
for practical applications of our constraint. For small values of the
Borel mass, the contribution of the spectral density at larger energy is
exponentially suppressed. Consequently, in a model calculation, one can
concentrate on the changes of the spectral density near the vector meson mass
region and below and model the higher energy part with a simple pole like
contribution.
The constraint equation in the vacuum are well satisfied by the
spectral density in the vacuum. As we will see, in most cases the changes
in the operators $\langle O_{n} \rangle$ are known. Hence, starting from the
vacuum form of the spectral density (Im$\Pi$), we can study what
changes are consistent with the constraint equation. This provides
model independent QCD constraints that any model calculation should satisfy.
One can go one step further and try to parameterize the spectral density
with a simple delta function type of pole and a continuum and determine the
changes in the parameters.
\section{Light quark system $\rho,\omega$}
\subsection{$\Pi_0$}
The operators that dominate the change in the OPE in eq.(\ref{constraint})
for the light quark system ($\rho,\omega$) are the quark operators\cite{HL92}
\begin{eqnarray}
\langle O_{4} \rangle & \rightarrow & \langle \bar{q} \gamma_\mu
D_\nu q \rangle \propto \int dx x [q(x)+\bar{q}(x) ] \nonumber \\
\langle O_{6} \rangle & \rightarrow & \langle ( \bar{q} q )^2
\rangle \propto
\langle ( \bar{q} q )^2 \rangle_0 + 2 \rho_n \langle \bar{q} q \rangle_0
\langle \bar{q} q \rangle_N
\label{lpi0}
\end{eqnarray}
The first equation of eq.(\ref{lpi0}) dominates the changes in the dimension
4 operators and is related to the well known second moment of the quark
distribution function. The second equation of eq.(\ref{lpi0}), for which
we have used the ground state saturation hypothesis\cite{HL92},
dominates the changes in the dimension 6 operators.
Using a delta function assumption for the spectral density in the constraint
equation gives the
following result for the scalar mass shift at ${\bf q}=0$\cite{HL92},
\begin{eqnarray}
{ m_V(\rho_n) \over m_V(\rho_n=0) }= 1- (0.16 \pm 0.06) {\rho_n \over
\rho_0}
\label{lpi0m}
\end{eqnarray}
This result is also consistent with other model
calculations\cite{Walecka,ST95,FS96} or the Brown-Rho scaling
argument \cite{BR91}.
A detailed comparison of the constraint equation in eq.(\ref{constraint})
with a hadronic calculation, based on chiral SU(3) dynamics with explicit
vector mesons were performed in \cite{KKW97}. The result shows a very
good agreement between the OPE and the phenomenological spectral density
put into the constraint equation in eq.(\ref{constraint}). However,
the hadronic calculation gave a large increase in width with a small decrease
in mass for the $\rho$ and a large decrease in mass with a small
increase in width for the $\omega$. Hence only the result for the
$\omega$ is consistent
with the result in eq.(\ref{lpi0m}). Later it was found that this was
a general result, given the uncertainty in the ground state hypothesis for the
four quark condensate in the medium\cite{LPM98};
namely, that there exists a band
in the mass vs. width plane that satisfies the constraint equation
in eq.(\ref{constraint}).
\subsection{$\Pi_T$}
The constraint for the nontrivial ${\bf q}$ dependence in
eq.(\ref{constraint}) has no $\Pi(M^2)_{pert}$ and has contributions from
operators with explicit spin index\cite{L98c}. The contributions from
dimension 4 operators are related to the twist-2 matrix elements and are well
known. The dimension 6 operators are dominated also by the twist-2 matrix
elements. Hence the constraint equation has little uncertainty coming
from the OPE. The operators that dominate are
\begin{eqnarray}
\langle O_{4} \rangle & \rightarrow & \langle \bar{q} \gamma_\mu
D_\nu q \rangle_N \propto \int dx x [q(x)+\bar{q}(x)] \nonumber \\
\langle O_{6} \rangle & \rightarrow & \langle \bar{q} \gamma_\mu
D_\nu D_\alpha D_\beta
q \rangle_N \propto \int dx x^3 [q(x)+\bar{q}(x) ]
\label{lpit}
\end{eqnarray}
Using a delta function assumption for the spectral density and allowing the
parameters to change to leading order in density and in ${\bf q}^2$,
we find the following non-trivial momentum dependence in the peak
position\cite{L98c},
\begin{eqnarray}
{ m_\rho(\rho_n) \over m_\rho(\rho_n=0) }= 1- (0.023 \pm 0.007) \left(
{ {\bf q} \over 0.5 } \right)^2 {\rho_n \over \rho_0} \nonumber \\
{ m_\omega(\rho_n) \over m_\omega(\rho_n=0) }= 1- (0.016 \pm 0.005) \left(
{ {\bf q} \over 0.5 } \right)^2 {\rho_n \over \rho_0},
\label{lpitm}
\end{eqnarray}
where ${\bf q}$ is in GeV/c unit.
This shows a very small momentum dependence compared to the expected scalar
mass shift in eq.(\ref{lpi0m}).
A detailed comparison of the constraint equation in eq.(\ref{constraint})
for the momentum dependence for the transverse direction has been
made\cite{FLK99} to the hadronic calculation, where the vector-meson nucleon
scattering amplitude is obtained by resonance saturation
in the s-channel. The result shows that the existing model
calculations tend to overestimate the constraint. This is due to the
large $\rho- N-\Delta(1232)$ coupling, which is obtained from the Bonn
potential\cite{Bonn}.
However, the existing calculations used a non-covariant
monopole form factor\cite{RCW97,PPLLM98} normalized off shell
\begin{eqnarray}
F( {\bf q})= { \Lambda^2 \over \Lambda^2 +{\bf q}^2 }.
\end{eqnarray}
On the other hand, the large $\rho- N-\Delta(1232)$ coupling in the Bonn
potential is defined with a dipole form factor normalized at the on shell
point of the vector meson,
\begin{eqnarray}
F_\rho(q^2)= \biggl( { \Lambda_\rho^2-m_\rho^2 \over
\Lambda_\rho^2- q^2 } \biggl)^{2}.
\label{ff}
\end{eqnarray}
This reduces the Delta contribution to the rho meson self energy at
the invariant mass around $m_\Delta-m_N$ by approximately a factor of 4.
After this correction, we find that the model calculations give very
good agreement with the constraint equation\cite{FLK99}.
\subsection{$\Pi_L$}
The constraint for the longitudinal direction is dominated by
twist-2 quark and gluon operators
$ \langle \bar{q} \gamma_{\mu_1}
D_{\mu_2} .. D_{\mu_n} q \rangle_N \propto \int dx x^{n-1}
[q(x)+\bar{q}(x) ],
\langle G_{\mu_1}^\alpha
D_{\mu_2} .. G_{\mu_n \alpha} \rangle_N \propto \int dx x^{n-1} g(x)$,
Using a delta function assumption for the spectral density and allowing the
parameters to change to leading order in density and in ${\bf q}^2$,
we find\cite{L98c},
\begin{eqnarray}
{ m_V(\rho_n) \over m_V(\rho_n=0) }= 1- (0.004 \pm 0.002) \left(
{ {\bf q} \over 0.5 } \right)^2 {\rho_n \over \rho_0},
\label{lpilm}
\end{eqnarray}
for both the $\rho$ and $\omega$. This is a very small effect and
no detailed comparison with any hadronic calculations exits yet.
\section{Strange quark system $\phi$}
The OPE in the constraint equation for the $\phi$ meson is dominated by
$ \langle m_s \bar{s} s \rangle $ for the scalar mass shift and
$\langle \bar{s} \gamma_\mu D_\nu s \rangle $ for the momentum dependence.
Assuming a delta function ansatz for the pole, we find
\begin{eqnarray}
{ m_\phi(\rho_n) \over m_\phi(\rho_n=0) }= 1- (0.03 \pm 0.015) {\rho_n \over
\rho_0} + (0.0005 \pm 0.0002) \left(
{ {\bf q} \over 0.5 } \right)^2 {\rho_n \over \rho_0},
\end{eqnarray}
for the transverse vector meson and the momentum dependence for the
longitudinal $\phi$ is about a factor of two larger.
\section{Heavy quark system $J/\psi$}
For the heavy quark system, we look at the constraints from the
moments\cite{SVZ,RRY},
\begin{eqnarray}
\label{dispersion}
M_n &\equiv& \left. { 1 \over n!} \left( {d \over d \omega^2} \right)^n
{\rm Re} \Pi(\omega^2) \right|_{\omega^2=-Q_0^2} \nonumber \\ &&=
\int_{4m_{c}^2}^{\infty}\frac{\rho(s)}{(s+Q_0^2)^{n+1}}ds.
\end{eqnarray}
The changes in OPE in nuclear medium for $M_n$ is dominated by the
change in the gluon condensate in eq.(\ref{gluon}).
To study the $J/\psi$ at rest in the nuclear matter, we will approximate
the spectral density with a delta function for the lowest state. This is
valid even in nuclear matter, because for a $J/\psi$ at rest, inelastic
interactions with nucleons such
as $J/\psi+N \rightarrow \bar{D}+\Lambda_c$ do not occur.
With this assumption, the constraint equation allows for the determination
of mass shift of the $J/\psi$ in nuclear matter. We find\cite{KKLMW98} for
the mass shift
\begin{eqnarray}
\Delta m_{J/\psi} \simeq -7 \, {\rm MeV}.
\end{eqnarray}
This corresponds
to small $J/\psi$- and $\eta_c$-nucleon scattering lengths $a=-\mu_r \Delta m/(2 \pi
\rho_N) \simeq (0.1-0.2)\, $fm ($\mu_r$ is the meson-nucleon reduced mass).
Our results for the mass shifts of the lowest $\bar{c}c$ states are
surprisingly close to those reported in ref.
\cite{Luke+92,Brodsky+97,Teramond+98,Haya}.
Although the expected mass shift is small, the result has little uncertainty
coming from OPE and puts reliable
constraint on charmonium mass shift which should be met by further studies of
heavy quark systems in dense matter.
\section{Conclusion}
We have derived and explored the consequences of
model independent constraints for the
vector meson polarization at ${\bf q}=0$ and ${\bf q} \neq 0$ for all the
vector mesons $\rho,\omega, \phi, J/\psi$.
Most of them have very little uncertainty in the OPE side of the constraint
equation and can be used as reliable constraints on all model calculation of
the vector meson properties in medium.
\section{Acknowledgments}
I would like to thank B. Friman, T. Hatsuda, H. Kim, S. Kim, F. Klingl,
P. Morath and W. Weise for the collaboration, which this talk is
based upon. This work was supported in part by
KOSEF through grant no. 971-0204-017-2 and 976-0200-002-2 and the Korean
Ministry of Education through grant no. 98-015-D00061.
|
\section*{Figure Captions}
\newcounter{FIG}
\begin{list}{{\bf FIG. \arabic{FIG}}}{\usecounter{FIG}}
\item
The tree--level Feynman diagram for the process $t\rightarrow b\nu_{\tau}
\bar{\tau}$.
\item
The SUSY induced CP--violating one--loop diagrams for the process
$t\rightarrow b\nu_{\tau}\bar{\tau}$.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of $\arg(A_t)$ for
$\tan\beta=1.2$, $m_2=150GeV$, $\mu=-40GeV$, ${M}=200GeV$, c=0.2.
When $m_{\tilde{t}_1}=150GeV$, $A_{CP}$ reaches its maximum.
\item
The quantity $f_{CP}$ defined in Eq.~\ref{rati} is plotted as
function of $\arg(A_t)$. All the parameters are the same as that of
Fig.3.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of SUSY parameter $\mu$,
for several values of $m_2$, for $\tan\beta=1.2$, ${M}=160GeV$, c=0.15,
$\arg(A_t)=0.5\pi$.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of SUSY parameter $\mu$,
for several values of $m_2$, $\tan\beta=5$. All the other parameters are
the same as that of Fig.3.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of SUSY parameter $\mu$,
for several values of $m_2$, $\tan\beta=15$. All the other parameters are
the same as that of Fig.3.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of $\tan\beta$, for several
values of $m_2$, $\mu=-70GeV$, ${M}=160GeV$, c=0.15, $\arg(A_t)=0.5\pi$.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of $\tan\beta$, for several
values of $m_2$, $\mu=-50GeV$, ${M}=160GeV$, c=0.15, $\arg(A_t)=0.5\pi$.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of ${m_2}$ for
${M}=160GeV$, c=0.15, $\arg(A_t)=0.5\pi$, $\mu=-50GeV$ for $\tan\beta=2$
and $\mu=-60Gev$ for $\tan\beta=5,10$.
\item
The CP asymmetry $A_{CP}$ is plotted as a function of the mass of the
light top squark, for different values of $m_2$, $\tan\beta=2.5$,
$\mu=-50GeV$, $\arg(A_t)=0.5\pi$.
\item
The tree--level and one--loop Feynman diagrams for the process
$t\rightarrow b\chi^0\chi^+$.
\end{list}
\end{document} |
\section{Introduction}
\hspace{5mm}
Duality between string theory in Anti-de Sitter (AdS) background space-time
and conformal field theory was conjectured in \cite{Mal}. This correspondence
was further elaborated in \cite{GKPW}. Especially near-horizon geometry of
nearly coincident D3-branes\cite{Dbrane} is $AdS_5 (\times S^5)$ and its
isometry is the conformal symmetry. On the other hand four-dimensional
${\cal N}=4$ supersymmetric Yang-Mills (SYM) theory is conformally invariant.
It was shown in \cite{JKY1} that there is a correspondence between the
conformal symmetries on both sides.
Near-horizon geometry of other Dp-branes is not AdS space-time and
this conjecture does not simply apply. The corresponding SYM theory is
not apparently conformally invariant because the coupling constant is
dimensionful. Recently, however, it was pointed out by Jevicki et
al\cite{JY}\cite{JKY2} that the near-horizon geometry of Dp-branes can
also be interpreted as conformally invariant, if one varies the string
coupling constant together with other backgrounds. Then they claimed
that by regarding the coupling constant of SYM theory as a background
field and transforming this background appropriately with other fields,
they can also make SYM theory confomally invariant. They also argued
that the 2 conformal transformations on the SYM and supergravity sides
are related by some coordinate transformations.\cite{JKY2}
We will not adopt this proposal to regard the coupling constant as a
background, because a coordinate-dependent coupling constant
breaks the special conformal symmetry as well as supersymmetry (SUSY).
(See sec 4.) The coupling constant must be kept constant. In this paper
we will specialize to the D-particle system and show that this system has
conformal symmetry. For this purpose we will use the SUSY
transformation to induce a variation of the functional measure which is
equivalent to effectively changing the coupling constant. Additional BRS
transformation needs to be also performed. The ordinary conformal
transformation combined with the SUSY and BRS transformation yields the
desired, complete conformal transformation.
In sec 2 we will show that by introducing an auxiliarly field the SUSY
transformation can be extended to an off-shell-closed symmetry, with respect
to whose generator the D-particle action can be rewritten as an exact form.
This is the T-dual version of the nil-potent symmetry studied in \cite{MNS}.
In sec 3 this symmetry is used to effectively change the string coupling
constant in the action by transformation of the field variables.
The parameter of the transformation is chosen to be field-dependent
and non-local. In order to cancel the variations of the gauge-fixing and
ghost action BRS transformation with a field-dependent parameter needs to
be also performed. In sec 4 these results are used to show that D-particle
effective theory has conformal symmetry. Discussions will be given in sec 5.
\section{Q-symmetry of the D-particle effective action}
\hspace{5mm}
The action which describes the low-energy dynamics of
D-particles\cite{Dbrane} is given \cite{Eaction} by
\beq
S = \int dt \frac{1}{g} \mbox{Tr} \left( \frac{1}{2} \sum_{i=1}^9(D X^i)^2
+ \frac{1}{4} \sum_{i,j=1}^9 [X^i,X^j]^2 -\frac{i}{2} \psi^T D \psi
- \frac{1}{2} \sum_{i=1}^9 \psi^T \gamma^i [X^i,\psi] \right),
\label{DPaction}
\eeq
where $D= \partial_t -i A^0$ is a gauge covariant derivative and $X^i$,
$A^0$ and $\psi$ are N $\times$ N hermitian matrices. $\psi$ is also a
sixteen-component spinor. $g$ is the string coupling constant.
$T$ on $\psi$ stands for a transposition. Some conventions for $\gamma$
matrices and the spinor $\psi$ are given in Appendix.
In what follows summation symbols will be omitted. Indices $i,j, \ldots$
run from 1 to 9, and $a,b,\ldots$ from 1 to 8.
Action (\ref{DPaction}) is invariant under the extended SUSY
transformation.\cite{BFSS}
\beqa
\delta X^i & = &-i \epsilon^T \gamma^i \psi, \qquad
\delta A^0 = i \epsilon^T \psi, \nonumber \\
\delta \psi & = & - (DX^i) \gamma^i \epsilon - \frac{i}{2} [X^i,X^j]
\gamma^{ij} \epsilon + \epsilon'
\label{Nsusy}
\eeqa
Let us pick up a particular transformation with $\epsilon'=0$,
$\epsilon = \varepsilon \zeta$ and $\zeta= (\frac{1}{\sqrt{2}},0,0,
\frac{1}{\sqrt{2}},0,\cdots,0)^T$. $\varepsilon$ is a Grassmann odd constant.
Generality will not be lost by this choice of $\epsilon$, because any
$\epsilon$ can be put in this form by SO(9) rotation.
By substituting the representation of $\gamma$'s (\ref{A2}), (\ref{A3})
and that of $\psi$ (\ref{B1}), (\ref{B2}) into (\ref{Nsusy}), we explicitly
obtain the SUSY transformation of the component fields.
\beqa
\delta X^a & = & -i \varepsilon \psi_a \quad (a=1,\cdots,8), \qquad
\qquad
\delta X^9 = -i \varepsilon \eta, \nonumber \\
\delta A^0 & = & i \varepsilon \eta, \qquad \qquad \qquad
\delta \psi_a = i \varepsilon ([X^9,X^a] +i DX^a) \quad (a=1,\cdots,8),
\nonumber \\
\delta \vec{\chi} & = & -\frac{1}{2} \varepsilon \vec{ E}, \qquad
\qquad
\delta \eta = - \varepsilon D X^9
\label{susy}
\eeqa
Here $\psi_a$, $\vec{\chi}=(\chi^1,\cdots,\chi^7)$ and $\eta$ are components
of $\psi$ and $\vec{E}$ a seven-vector function
of $X^a$. These are defined in Appendix.
We could achieve the goal of this paper by working directly with
transformation (\ref{susy}). It is, however, more instructive to relate it
to the nil-potent transformation of \cite{MNS}.
Let us introduce an auxiliarly seven-vector
$\vec{H}$ by adding a term $\int dt (2g)^{-1} \mbox{Tr} (\vec{H}+\frac{1}{2}
\vec{E})^2$ to (\ref{DPaction}). $\vec{H}$ will coincide with $-\frac{1}{2}
\vec{E}$ by the equations of motion. The transformation rule of $\vec{\chi}$
is modified to $\delta \vec{\chi} = \varepsilon \vec{H}$. $\delta \vec{H} $ is
defined to be $\varepsilon ([X^9,\vec{\chi}]+ iD \vec{\chi})$.
Now the square of this new
transformation turns out to be equivalent to a time translation up to a gauge
transformation: $ [Q^2, \vec{\chi} ] = i\partial_t \vec{\chi}
+ [X^9+A^0, \vec{\chi} ]$, {\it etc}.\footnote{
The algebra of transformation (\ref{Nsusy}) closes only on shell, {\it i.e.}
when equations of motion are used. By the introduction of $\vec{H}$,
Q part of the SUSY algebra closes off shell.}
Here $Q$ is the generator of this
transformation.($\delta=\varepsilon Q$) We will hereafter call this
a Q transformation. Importantly, action (\ref{DPaction}) can be
rewritten into a Q-exact form.
\beq
S_{coh} = \int dt \frac{1}{g} \left\{ Q, \mbox{Tr} \left(
-\frac{1}{2} \eta DX^9 +
\frac{1}{2} \vec{\chi} \cdot \vec{E} + \frac{1}{2} \vec{\chi} \cdot
\vec{H} + \frac{i}{2} \psi_a [X^a,X^9] - \frac{1}{2} \psi_a DX^a
\right) \right\}
\label{Cohaction}
\eeq
This action is the T-dual version of the cohomological action for D-instantons
considered in \cite{MNS}.
To quantize this model we will choose the background field gauge.
We will decompose $X^i$
into a background field $B^i$ and a quantum fluctuation $Y^i$
\beq
X^i = B^i + Y^i
\eeq
and choose a gauge function\cite{JKY2}
\beq
G= \partial_t A^0 + i [ B^i, Y^i ] .
\eeq
The gauge fixing and ghost actions are given by
\beqa
S_{gf} & = & \int dt \ \frac{-1}{2g} \ \mbox{Tr} G^2, \\
S_{gh} & = & \int dt \ i \ \mbox{Tr} \left\{ \bar{C} \left( \partial_t D C
+ [ B^i,[X^i,C ] \ ] \right) \right\} .
\eeqa
The total action $S_{tot} = S_{coh}+S_{gf}+S_{gh}$ is invariant under the
BRS transformation
\beqa
\delta_B X^i & = & -\lambda \ [C,X^i], \qquad \qquad
\delta_B A^0 = i \ \lambda \ DC, \nonumber \\
\delta_B \psi & = & - \lambda \ \{ C, \psi \}, \qquad \qquad
\delta_B \vec{H} = - \lambda \ [C,\vec{H}], \nonumber \\
\delta_B C & = & - \lambda \ C^2, \qquad \qquad
\delta_B \bar{C} = \lambda \ \frac{1}{g} \ G.
\label{BRS}
\eeqa
Here $\lambda$ is a Grassmann odd parameter.
\section{Transformation with a field-dependent parameter}
\hspace{5mm}
Because of the Q-exact nature of action (\ref{Cohaction}) the
string coupling constant $g$ in (\ref{Cohaction}) can be effectively
changed by a Q transformation of fields without varying $g$ explicitly.
The variation $\delta g$ may even be a
function of $t$. We will carry out this program in two steps.
We first perform a Q transformation
\beqa
\delta_Q X^a & = & -i \varepsilon \psi_a, \qquad
\qquad
\delta_Q X^9 = -i \varepsilon \eta, \nonumber \\
\delta_Q A^0 & = & i \varepsilon \eta, \qquad \qquad \qquad
\delta_Q \psi_a = i \varepsilon ([X^9,X^a] +i DX^a),
\nonumber \\
\delta_Q \vec{\chi} & = & \varepsilon \vec{H}, \qquad
\qquad
\delta_Q \eta = - \varepsilon D X^9, \qquad
\delta_Q \vec{H} = \varepsilon ([X^9, \vec{\chi}] + i D \vec{\chi}),
\nonumber \\
\delta_Q C & = & 0, \qquad \delta_Q \bar{C} =0,
\label{Qtransf}
\eeqa
with the following parameter.
\beq
\varepsilon =
\int dt \frac{i}{g^2} \delta g(t) \mbox{Tr} \left(
-\frac{1}{2} \eta DX^9 - \frac{1}{2} \psi_a DX^a +
\frac{1}{2} \vec{\chi} \cdot \vec{E} + \frac{1}{2} \vec{\chi} \cdot
\vec{H} + \frac{i}{2} \psi_a [X^a,X^9]
\right)
\label{epsilon}
\eeq
This is a field-dependent and non-local transformation.
Action(\ref{Cohaction}) is invariant,
while the gauge fixing and ghost actions are changed.
\beqa
\delta_Q S_{gf} & = & -\varepsilon \int dt \frac{1}{g} \mbox{Tr} \{ G
\left(i \partial_t \eta +[B^a,\psi_a ] +[B^9,\eta]
\right) \},
\label{varSgf} \\
\delta_Q S_{gh} & = & -\varepsilon \int dt \mbox{Tr} \{ \bar{C} \left(
i \partial_t \{ \eta,C \} +[B^a, \{ \psi_a,C \} \ ] + [ B^9, \{ \eta,C
\} \ ]
\right) \} \qquad \qquad
\label{varSgh}
\eeqa
The functional measure is not invariant, either, because the transformation
parameter depends on the fields. Calculation of the superjacobian is
straightforward and we obtain\footnote{
Similar calculation of the superjacobian for 4d bosonic gauge theory
in a different context was performed in \cite{FradPal}. See also \cite{JKY2}.}
\vspace{.3cm}
$(1+ \delta_Q) {\cal D} (\mbox{Fields}) =$ \\
\vspace{.1cm}
$\displaystyle \ \ \ {\cal D} (\mbox{Fields})
\exp \left\{ i \int dt \frac{-\delta g(t)}{g^2} \mbox{Tr} \left( -\frac{i}{2}
\psi_a D \psi_a + \frac{1}{2} \psi_a [X^9,\psi_a] - \frac{1}{2} \vec{\chi}
\cdot [Q, \vec{E}]
\right. \right. $
\beqa
\qquad \qquad & & - \frac{1}{2} \vec{\chi} \cdot [X^9, \vec{\chi}]
- \frac{i}{2} \vec{\chi} \cdot D \vec{\chi} + \frac{1}{2} [X^9, X^a]^2
+ \frac{1}{2} (DX^a)^2 + \frac{1}{2} (DX^9)^2 \nonumber \\
\qquad \qquad & & \left. \left. -\frac{i}{2} \eta D \eta
- \eta [X^a,\psi_a] - \frac{1}{2} \eta [X^9,\eta]
+ \frac{1}{2} \vec{H}^2 + \frac{1}{2} \vec{H} \cdot \vec{E}
\right) \right\} .
\label{varMeasure}
\eeqa
The exponent on RHS is equal to the difference $i \{ S_{coh}(g+\delta g)
-S_{coh}(g)\}$. Because $S_{gh}$ does not depend on $g$, and the variation of
$S_{gf}$ with respect to $g$ is BRS exact and does not contribute to the path
integral, the effect of the transformation (\ref{Qtransf}) is just to change
$g$ to $g+\delta g(t)$ in $S_{tot}$.
Secondly, to cancel the variations (\ref{varSgf}) and (\ref{varSgh}) we
perform the BRS transformation (\ref{BRS}) with the parameter
\beq
\lambda = \varepsilon \int dt \mbox{Tr} \left\{ \left( \partial_t \eta -i
[B^a,\psi_a] -i [B^9,\eta] \right) \ \bar{C} \right\} .
\label{lambda}
\eeq
While the total action is invariant, the functional measure changes as
in (\ref{varMeasure}). It can be checked that this change cancels out
(\ref{varSgf}) and (\ref{varSgh}) exactly.
To summarize, the combination
\beq
\delta=\delta_Q+\delta_B
\label{varQB}
\eeq
has the same effect as changing $g$ into $g+\delta g(t)$. It is important to
notice that this infinitesimal transformation cannot be repeated to generate
a non-constant $g(t)$, because Q symmetry is broken if $g$ is not a constant.
It is straightforward to eliminate $\vec{H}$ from (\ref{varQB}) by using
$\langle \vec{H}\rangle = -\frac{1}{2} \vec{E}$ and $\langle H_{\alpha
\beta}^A(t) H_{\gamma \delta}
^B(t') \rangle = ig \delta^{AB} \delta_{\alpha \delta} \delta_{\beta \gamma}
\delta (t-t') + \frac{1}{4} E_{\alpha \beta}^A(t) E_{\gamma \delta}^B(t')$.
Here $\alpha,\beta,\gamma,\delta=1,\cdots,$N and $A, B=1,\cdots,7$ are
U(N) and vector indices, respectively.
The results are the same as (\ref{varQB}) except for replacement of
$\varepsilon$ (\ref{epsilon}) and $\lambda$ (\ref{lambda}) by
\beqa
\tilde{\varepsilon} & = &
\int dt \frac{i}{g^2} \delta g(t) \mbox{Tr} \left(
-\frac{1}{2} \eta DX^9 - \frac{1}{2} \psi_a DX^a
+\frac{1}{4} \vec{\chi} \cdot \vec{E} + \frac{i}{2} \psi_a [X^a,X^9]
\right) \nonumber \\
& = &\int dt \frac{i}{g^2} \delta g(t) \mbox{Tr} \left\{ \zeta^T \left(
-\frac{1}{2} \gamma^i DX^i - \frac{i}{4} \gamma^{ij} [X^i,X^j] \right) \psi
\right\} \label{tilep}
\eeqa
and
\beqa
\tilde{\lambda} & = &
\tilde{\varepsilon} \int dt \mbox{Tr} \left\{ \left( \partial_t \eta -i
[B^a,\psi_a] -i [B^9,\eta] \right) \, \bar{C} \right\} \nonumber \\
& = & \tilde{\varepsilon} \int dt \mbox{Tr} \left\{ \zeta^T \left( \partial_t
\psi - i \gamma^i [B^i,\psi] \right) \, \bar{C} \right\},
\label{tillam}
\eeqa
respectively, and
\beq
\delta \vec{\chi} = -\frac{1}{2} \tilde{\varepsilon} \vec{E} - \tilde{\lambda}
\{C,\vec{\chi} \} -\frac{\delta g(t)}{2g} \vec{\chi},
\eeq
where the last term comes from the contraction of 2 $\vec{H}$'s.
\section{Conformal symmetry}
\hspace{5mm}
It was claimed in \cite{JKY2} that Dp-brane action has conformal symmetry
provided one regards $g$ as a background field $g(x)$ and makes it
transformed appropriately together with other fields. They called this
symmetry a generalized conformal one.
The conformal group is generated by translation, dilatation and
special conformal transformation (SCT). We will specialize to the
D-particle case. These 3 transformations on $t$ are defined by
\beqa
\delta t & = & -a \quad (\mbox{translation}), \\
\delta t & = & -a t \quad (\mbox{dilatation}), \\
\delta t & = & -a t^2 \quad (\mbox{SCT}).
\eeqa
Here $a$ is an infinitesimal parameter. A field $F(t)$ of a scale dimension
$w$ is then transformed as
\beq
\delta_n F(t) = a (nwt^{n-1} + t^n \partial_t) F(t)
\eeq
for translation ($n=0$), dilatation ($n=1$) and SCT ($n=2$).
Actions $S$ and $S_{coh}$ are invariant under these transformations only if
the string coupling constant $g$ is also transformed like
\beq
\delta_n g = 3 a n t^{n-1} g.
\label{varg}
\eeq
The scale dimensions $w$ of the fields are 1, 1, $\frac{3}{2}$, 2
for $X^i$, $A^0$, $\psi$ and $\vec{H}$, respectively.
Because (\ref{varg}) for SCT ($n=2$) depends on $t$ explicitly, the authors of
\cite{JY} \cite{JKY2} proposed to regard $g$ itself as a function of
the coordinate $t$ and assumed a transformation rule
\beq
\delta_n g(t) = a(3nt^{n-1}+ t^n \partial_t) g(t).
\eeq
As for $S_{gf}$ and $S_{gh}$ these are not invariant under SCT
even if $C$ and $\bar{C}$ are assigned scale dimensions 0 and $-1$.
By using the formalism of \cite{FradPal} the authors of \cite{JKY2} then
showed that the variations of these actions
can be cancelled by a non-local BRS transformation (\ref{BRS})
with $\lambda = \nu_n$.\footnote{Here $n=2$ for SCT. An index $n$ is
introduced for later convenience. }
\beq
\nu_n = -n(n-1)ia \int dt \mbox{Tr} \bar{C} A^0
\label{BRSnu}
\eeq
We do not adopt their proporsal to regard $g$ as a $t$-dependent background
field $g(t)$ because of the following reasons.
\begin{itemize}
\item Once $g$ is promoted to a function of $t$, then action (\ref{DPaction})
is {\em no longer} invariant under SCT because the integration by parts
is used in the proof of invariance.
After a simple calculation we indeed obtain
\beq
\delta_2 S = a \int dt \frac{1}{g(t)} \partial_t \mbox{Tr} (X^i)^2.
\label{vS}
\eeq
\item If $g$ is a function of $t$, $S$ is {\em not} invariant under
SUSY transformation (\ref{Nsusy}) due to the same reason as above.
This symmetry is important and cannot be abandoned.
\end{itemize}
For these reasons we will leave $g$ a constant and will not change it.
In the previous sections we showed that the same effect as changing $g$
can be accomplished by transforming the field variables.
By combining (\ref{varQB}) and the BRS transformation (\ref{BRS}) with
the parameter (\ref{BRSnu}), we obtain the 3 conformal transformations
($n=0,1,2$)\footnote{In fact we do not change $g$. Instead we use
transformation (\ref{varQB}) to cancel the effective variation of $g$ due to
the conformal transformation. Thus $g$ remains constant. This also enables
us to repeat infinitesimal transformations to obtain a finite one.},
\beqa
\Delta_n X^i & = & a(nt^{n-1}+t^n \partial_t)X^i -i \tilde{\varepsilon}_n
\zeta^T \gamma^i \psi-(\tilde{\lambda}_n+\nu_n) [C,X^i], \nonumber \\\Delta_n A^0 & = & a(nt^{n-1}+t^n \partial_t)A^0 + i \tilde{\varepsilon}_n
\zeta^T \psi+ i (\tilde{\lambda}_n+\nu_n)DC, \nonumber \\
\Delta_n \psi_a & = & a \left(\frac{3}{2}nt^{n-1} +t^n \partial_t \right)
\psi_a +i \tilde{\varepsilon}_n ([X^9,X^a]+iDX^a) -(\tilde{\lambda}_n+\nu_n)
\{C,\psi_a \}, \nonumber \\
\Delta_n \vec{\chi} & = & a \left(\frac{3}{2}nt^{n-1} +t^n \partial_t \right)
\vec{\chi}-\frac{1}{2} \tilde{\varepsilon}_n \vec{E} -
\frac{\delta_ng}{2g} \vec{\chi}-(\tilde{\lambda}_n+\nu_n) \{C,\vec{\chi} \},
\nonumber \\
\Delta_n \eta & = & a \left(\frac{3}{2}nt^{n-1} +t^n \partial_t \right) \eta
-\tilde{\varepsilon}_n DX^9-(\tilde{\lambda}_n+\nu_n) \{C,\eta \},
\nonumber \\
\Delta_n C & = & at^n \partial_t C -(\tilde{\lambda}_n+\nu_n)C^2,
\nonumber \\
\Delta_n \bar{C} & = & a(-nt^{n-1} +t^n \partial_t)\bar{C}
+(\tilde{\lambda}_n+\nu_n) \frac{1}{g} G.
\label{Conformal}
\eeqa
Here the first 2 lines are simplified by using $\zeta$.
$\tilde{\varepsilon}_n$ and $\tilde{\lambda}_n$ are (\ref{tilep}) and
(\ref{tillam}), respectively,
with $\delta g$ replaced by $\delta_ng$
(\ref{varg}).
In the rest of this section we will compute the expectation value
$\langle \Delta_n X^i \rangle$, which will become a symmetry
transformation $\Delta_n B^i$ of the effective action $\Gamma [B^i]$.
This may be performed in perturbative expansions in $g$.
Because $\tilde{\varepsilon}_n$ (\ref{tilep}) and $\tilde{\lambda}_n$
(\ref{tillam}) are of order ${\cal O} (g^{-1})$, it turns out that
in order to obtain ${\cal O} (g^m)$ result we have to perform $(m+1)$-loop
calculation. Here we will present only the result of one-loop calculation
( ${\cal O} (g^0)$). The result and details of two-loop calculation will be
reported in a forthcoming paper \cite{HN}.
The background field $B^i(t)$ is diagonal and linear in $t$.
We found up to first order in $\dot{B}=\partial_t B$ that
$\langle -\tilde{\lambda}_n [C,X^i] \rangle$ is of order ${\cal O} (g^1)$ and
\beq
\langle-i \tilde{\varepsilon}_n \zeta^T \gamma^i \psi \rangle_{\alpha \beta}
= -\frac{1}{2} \delta_{\alpha \beta} \int dt' \frac{\delta_n g(t')}{g} \theta
(t-t') \dot{B}_{\alpha}^i(t').
\eeq
Here $B^i_{\alpha}$ is the $\alpha$-th diagonal component of $B^i$ and
$\theta (t-t')$ is a step function. $\langle-\nu_n [C,X^i] \rangle$
was obtained in \cite{JKY2} and is also of order
${\cal O} (g^1)$. We thus obtain the \lq quantum' transformation rule
\beq
\Delta_1 B^i_{\alpha}(t) = a(1+t \partial_t) B^i_{\alpha}(t)
-\frac{3}{2} a B^i_{\alpha}(t) = a(-\frac{1}{2} +t \partial_t) B^i_{\alpha}(t)
\label{dil}
\eeq
for dilatation and
\beqa
\Delta_2 B^i_{\alpha}(t) & = & a(2t + t^2 \partial_t) B^i_{\alpha}(t)
- 3a \int^t dt' \ t' \ \dot{B}^i_{\alpha}(t') \nonumber \\
& = & a(-t + t^2 \partial_t) B^i_{\alpha}(t) + 3a \int^t dt' B^i_{\alpha}(t')
\label{sct}
\eeqa
for SCT. It is easy to verify that the tree-level effective action
\beq
\Gamma_0 [B] = \int dt \frac{1}{2g} (\dot{B}^i)^2
\label{Eaction}
\eeq
is left invariant under (\ref{dil}) and (\ref{sct}).
We note that the scale dimension $w$ of the backgound $B^i_{\alpha}$
has shifted from 1 to $-\frac{1}{2}$ due to quantum effects.
We also find that SCT rule (\ref{sct}) acquired an extra non-local term.
Nonetheless we can check that (\ref{dil}), (\ref{sct}) and
$\Delta_0 B^i_{\alpha} =a \partial_t B^i_{\alpha}$ generate the conformal
algebra.
\section{Discussion}
\hspace{5mm}
We found that the low-energy effective theory of D-particles is
invariant under conformal transformation (\ref{Conformal}).
The transformation rule is obtained explicitly, although the parameters
depend on the field variables in a non-local way.
Here we stress the point that $g$ is {\em not} changed under the
transformation. We also found that the scale dimension of $X^i$ changed from
1 to $-\frac{1}{2}$ due to quantum effects.
Because the actions of Dp-branes are related by T-dual transformation
\cite{Taylor}, the procedure adopted in this paper is easily generalized
to all Dp-branes.
It can be verified that all Dp-brane theories have similar
conformal symmetry. This conformal symmetry will put strong constraints on
the correlation functions. Some lower-point functions may be obtained by
solving conformal Ward-Takahashi identities.\cite{FradPal}
Action (\ref{DPaction}) also defines the Matrix
theory \cite{BFSS}. Study of the conformal W-T identities may also shed some
light on the understanding of M theory.
Another important issue is the AdS/CFT correspondence.
Because the near-horizon geometry of Dp-branes ($p \neq 3$) is
not an AdS space-time and the conformal group is not its isometry,
this correspondence does not simply apply. Because Dp-brane
system has turned out to have conformal symmetry, however, it is expected that
its near-horizon geometry may also have the corresponding \lq conformal
symmetry'. If such a symmetry is found, Dp-brane effective theory may also be
interpreted as a boundary \lq conformal field theory' in the classical
background of the Dp-branes. It will play an important r\^{o}le in
determining the effective action for the radial distance of a probe Dp-brane
in the background field of N coincident Dp-branes placed at the
origin.\cite{Mal} The non-local nature of transformation (\ref{sct}),
however, makes the problem difficult and the realization of such symmetry
on the supergravity side is not yet clear.
The calculation of $\langle \Delta_n X^i \rangle$ to higher orders may
elucidate this point. The result will be reported elsewhere\cite{HN}.
Finally, generalization of the present analysis to the superconformal
transformation will be straightforward and may be useful in putting further
constraints on the Dp-brane dynamics.
\section*{Note added}
\hspace{5mm}
If $g(t)$ is set a constant after SCT, the variation of $S$ (\ref{vS})
vanishes. We were informed by T.~Yoneya that this is the generalized
conformal symmetry of \cite{JY}\cite{JKY2}. We were also informed that he
could extend this conformal symmetry to that for a non-constant $g$ by
using the SO(2,1) orbit of $g$. We thank T.~Yoneya for comments and
discussions.
\section*{Appendix}
\hspace{5mm}
$\gamma^i, \quad i=1, \cdots, 9$ are 16 $\times$ 16 real
symmetric matrices and satisfy Clifford algebra
\beq
\{\gamma^i, \gamma^j \}= 2 \delta^{ij}.
\eeq
We choose the following special representation.
\beqa
\gamma^i & = & i \sigma_2 \otimes \mu^i = \left( \begin{array}{cc}
0 & \mu^i \\
- \mu^i & 0 \end{array}
\right) \quad (i=1,\cdots,7), \nonumber \\
\gamma^8 & = & \sigma_1 \otimes {\bm 1}_8 = \left( \begin{array}{cc}
0 & {\bm 1}_8 \\
{\bm 1}_8 & 0 \end{array} \right), \nonumber \\
\gamma^9 & = & \sigma_3 \otimes {\bm 1}_8 = \left( \begin{array}{cc}
{\bm 1}_8 & 0 \\
0 & - {\bm 1}_8 \end{array} \right)
\label{A2}
\eeqa
8 $\times$ 8 matrices $\mu^i$ are given by
\beqa
\mu^1 & = & i \sigma_2 \otimes i \sigma_2 \otimes i \sigma_2, \qquad
\mu^2 = {\bm 1}_2 \otimes \sigma_1 \otimes i \sigma_2, \qquad
\mu^3 = {\bm 1}_2 \otimes \sigma_3 \otimes i \sigma_2, \nonumber \\
\mu^4 & = & \sigma_1 \otimes i \sigma_2 \otimes {\bm 1}_2, \qquad
\mu^5 = \sigma_3 \otimes i \sigma_2 \otimes {\bm 1}_2, \qquad
\mu^6 = i \sigma_2 \otimes {\bm 1}_2 \otimes \sigma_1, \nonumber \\
\mu^7 & = & i \sigma_2 \otimes {\bm 1}_2 \otimes \sigma_3.
\label{A3}
\eeqa
The spinor $\psi$ is decomposed into 2 eight-component spinors $\psi^{(i)}$:
\beq
\psi = \left( \begin{array}{c}
\psi^{(1)} \\
\psi^{(2)} \end{array} \right)
\label{B1}
\eeq
Their components are given by
\beqa
\psi^{(1)} & = & \frac{1}{\sqrt{2}} (\eta+ \chi^7,
-\chi^2-\chi^4,\chi^2-\chi^4,\eta-\chi^7,\chi^1-\chi^6,-\chi^3-\chi^5,
-\chi^3+\chi^5,-\chi^1-\chi^6)^T, \nonumber \\
\psi^{(2)} & = & \frac{1}{\sqrt{2}} (-\psi_2+\psi_8, \psi_3-\psi_5,
\psi_3+\psi_5,\psi_2+\psi_8,\psi_1+\psi_7,-\psi_4+\psi_6,\psi_4+\psi_6,
\psi_1-\psi_7)^T \nonumber \\
& &
\label{B2}
\eeqa
The seven-vector $\vec{E}=(E^1,\cdots,E^7)$ is defined by
\beqa
E^1 & = & 2i \{ -[X^1,X^2]-[X^3,X^4]-[X^5,X^6]-[X^7,X^8] \}, \nonumber \\
E^2 & = & 2i \{ [X^1,X^4]+[X^2,X^3]-[X^5,X^8]-[X^6,X^7] \}, \nonumber \\
E^3 & = & 2i \{ -[X^1,X^3]+[X^2,X^4]-[X^5,X^7]+[X^6,X^8] \}, \nonumber \\
E^4 & = & 2i \{ [X^1,X^6]+[X^2,X^5]+[X^3,X^8]+[X^4,X^7] \}, \nonumber \\
E^5 & = & 2i \{ -[X^1,X^5]+[X^2,X^6]+[X^3,X^7]-[X^4,X^8] \}, \nonumber \\
E^6 & = & 2i \{ [X^1,X^8]+[X^2,X^7]-[X^3,X^6]-[X^4,X^5] \}, \nonumber \\
E^7 & = & 2i \{ -[X^1,X^7]+[X^2,X^8]-[X^3,X^5]+[X^4,X^6] \}.
\label{C1}
\eeqa
\newpage
|
\section{Introduction}
{}It was proposed in \cite{TeV} that the fundamental Planck scale
can be around a TeV. In string theory, this implies that the string scale
$M_s$ is lowered all the way down to TeV scale\cite{lyk,anto,ST}.
In this picture, the Standard Model (SM) fields reside inside of $p\leq 9$
spatial dimensional $p$-branes (or the intersection
of different sets of branes, as in \cite{ST}),
while gravity lives in the higher (10 or 11)
dimensional bulk of spacetime. For $3<p<9$, this ``brane world'' scenario
appears to be flexible enough so that various properties such as
gauge and
gravitational coupling unification, dilaton stabilization, and the weakness of
the SM gauge couplings can be satisfied within
this framework \cite{BW}. The weakness of the four dimensional
gravitational coupling is due to the presence of at least two large
($\gg 1/M_s$) compact directions transverse to the $p$-branes on which the
SM fields are localized. Issues such as coupling
unification \cite{dien}, proton stability \cite{TeVphen},
neutrino masses \cite{neutrino},
fermion masses and mixing \cite{flavor},
and astrophysical/cosmological implications \cite{TeVphen,astro}
have also been addressed \cite{related,zurab,quiros}.
It has been argued that the
possibility of TeV scale gravity appears viable.
If the fundamental Planck scale were around a TeV,
clear signals due to strong gravitational interactions would appear at the
LHC, extending existing constraints \cite{collider}.
Furthermore, light Kaluza-Klein (KK) modes living in
the bulk can be produced.
{}In general, the conventional Planck mass $M_P \sim 10^{19}$ GeV is
related
to the string scale by $(V_{p-3} V_{9-p} M_s^6)=(M_P/M_s)^2$, where
$V_{p-3}$ and $V_{9-p}$ are the compactified volumes inside and
transverse to the $p$-branes respectively.
Specifically, it is convenient to discuss the brane world picture in
Type I string models, where brane fields are open string modes.
The scenario given in Ref\cite{ST} is of this type.
Here the Planck mass is given by
\begin{equation}
M_P \sim M_s (M_s R_1) (M_s R_2) (M_s R_3) \\
\sim M_{Pl} (M_{Pl} R_1)
\end{equation}
where $R_i$ are the three radii characterizing the 3 compactified tori and
$M_{Pl}$ is the 6-dimensional Planck mass scale \cite{TeV},
in the case with $n=2$ large dimensions. To stabilize the dilaton
expectation value (and maybe even to induce SUSY breaking), the
string coupling $g_s$ is likely to be strong (or, more
precisely,
$g_s {\ \lower-1.2pt\vbox{\hbox{\rlap{$>$}\lower5pt\vbox{\hbox{$\sim$}}}}\ } 1$, and the theory appears not to have a dual weak coupling description).
For $p=5$, a typical squared gauge coupling $\alpha=g^2/(4 \pi)$ is given
by $\alpha \sim g_s/(M_s R_i)^2$ for $i=$2 or 3.
To obtain the weak SM gauge couplings from
a generic strong string coupling
requires that
$(M_s R_2) (M_s R_3) \sim 10-100$.
For $n=2$, a constraint from supernova
cooling requires $M_{Pl} > 30-50$
TeV \cite{TeVphen,SN1987A}, implying that $M_s > 3-20$ TeV.
This value for $M_s$ is compatible with the SUSY
models \cite{zurab} and the brane world picture \cite{BW}.
In this case, neither the Tevatron nor LHC energies may
be high enough to detect
either the strong gravity effect or the gravity KK modes.
Are there other signatures at these energies that may be used to test the
brane world scenario? The answer is yes, as we show here.
{}The collider signatures that we discuss here
do not arise from strong gravity or bulk KK modes (as in
Refs. \cite{collider,nath}) but come from
two generic features of Type I string theory, which should also
be generic in
the brane world picture: \\
\noindent $\bullet$
(1) In contrast to perturbative heterotic string theory, in some
Type I string models (which can be realized as Type II orientifolds and
are dual to non-perturbative heterotic strings),
there can be more than one anomalous $U(1)$ gauge fields.
They must be massive, though some of them can
be relatively light and may be produced in high energy colliders
as $Z^{\prime}$ vector bosons. \\
\noindent $\bullet$
(2) There are closed string modes which are not bulk modes, that is,
they do not propagate freely in the bulk. Some of these modes are
localized
away from the branes so their couplings to the brane modes may be
exponentially suppressed.
However, some of them may sit on the branes. For $p>3$, some of
these modes may even be localized inside the branes,
{\em i.e.}, in the dimensions which are compactified.
In contrast to the KK modes (from either the branes or the bulk),
these localized modes can be produced as a resonance
from the scattering of two brane modes which have no momentum in
the compactified dimensions, with a significant
({\em i.e.}, not gravitational strength) coupling.
Some of these localized modes are the pseudo-Goldstone
bosons eaten by the anomalous $U(1)$ gauge bosons
mentioned above. Experimental constraints require the remaining ones to pick up
relatively large masses, but one of them may behave like an axion.
{}For example, in an orbifold string model, these fields are the twisted
sector moduli which are stuck at the orbifold fixed points (or fixed lines).
Unlike the untwisted sector
states which are free to propagate in the bulk,
the twisted sector states have no momentum or winding in
the compactified dimensions, and hence are localized in the internal space.
While the couplings of the brane fields with the bulk fields
are suppressed by a factor of
$(V_{p-3} V_{9-p} M_s^6)^{-1}=(M_s/M_P)^2$
due to wavefunction normalization in the compactified
dimensions (and hence the couplings are of
gravitational strength), this suppression factor (or part of it)
is absent in the corresponding coupling with these twisted fields.
If these fields happen to sit on the branes, the
coupling with the brane fields is simply $\kappa \sim g_s$ and is
therefore independent of $M_s$, $V_{p-3}$ and $V_{9-p}$.
Before SUSY breaking, these fields, being moduli, are massless.
However, to be compatible with experiment, these states must pick up masses.
Assuming that dynamical SUSY breaking takes place
on the branes we expect these twisted fields to obtain masses
somewhat comparable to the SUSY breaking scale, $M_{SUSY} \sim O(1)$ TeV.
In the scattering of two brane modes with no momentum in the compactified
dimensions, the brane KK modes cannot be singly produced, since this would
violate momentum conservation along the branes.
In contrast, the momenta
orthogonal to the branes need not be conserved since the branes break
translation invariance in these orthogonal directions. Therefore, the bulk KK
modes can be singly produced by the brane modes, with a coupling in the
amplitude $\propto 1/M_P$.
After the summation over the large multiplicity $\sim (s^{1/2}R)^n$ of bulk KK
modes (where $n$ denotes the number of large compactification dimensions
transverse to the branes with size $\sim R$, and $s$ denotes the CM mass
energy squared), the cross section for processes
involving real bulk KK emission is $\propto 1/M_{Pl}^{n+2}$. For processes
involving the exchange of virtual bulk KK modes, summing over the exchanges
that do not have strong propagator suppression,
the amplitude is $\sim (1/M_P^2)(s^{1/2}R)^n
\propto 1/M_{Pl}^{n+2}$, and the resultant cross sections
are $\propto 1/M_{Pl}^{2(n+2)}$.
Given that $M_{Pl} \mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}} 30-50$ TeV ($n=2$), these effects may be too small to
be observable at Tevatron or even LHC energies. In contrast to the KK modes,
the twisted modes which have no momentum or winding in the compactified
dimensions can be singly produced by the brane modes with a coupling
$\propto 1/M_s$. They can thus be produced as resonances with strengths that
are not suppressed by the size of the extra large
dimensions. Despite the fact that the
couplings of these twisted fields with the Standard Model fields do not appear
at the renomalizable level, and the cross sections for processes involving the
exchange of these twisted fields are hence suppressed by the factor $1/M_s^4$,
their experimental signatures may well be larger than those of the gravity KK
modes.
In a typical Type I string model,
there are abelian gauge fields (brane modes) with field-theoretic triangle
anomalies. They become massive generically, with some of the above-mentioned
twisted RR scalars being the would-be-Goldstone bosons.
If the mass of an abelian gauge field happens to be in the LHC energy
regime,
it may be seen as a resonance in the hadron scattering experiment
and in the Drell-Yan process.
Since these massive gauge bosons are
brane fields, they couple to the
quark-antiquark and lepton-antilepton pair with the strength comparable
to that of the SM interactions.
In contrast to perturbative heterotic string theory,
there can be more than one anomalous $U(1)$ gauge fields for Type I strings.
Moreover, these $U(1)$ gauge fields can be relatively light
and are possible candidates of the $Z^{\prime}$ bosons.
\section{Some Generic Features of the Brane World}
{}For illustrative purposes, let us first consider compactification of
Type I string theory to 4D on a ${\mathbb Z}_3$ orbifold \cite{z3}.
The ${\mathbb Z}_3$
generator $g$ acts on the complex
coordinates $z_1$, $z_2$, $z_3$ of the compactified dimensions
$T^6=T^2 \times T^2 \times T^2$ as follows:
$g z_i = \omega z_i$ for $i=1,2,3$
where $\omega=e^{2i \pi/3}$. The resulting model has ${\cal N}=1$
supersymmetry. The twisted sector states are located at points
in the compactified dimensions
that are invariant under the orbifold action.
Here there are $27$ fixed points at which the
${\mathbb Z}_3$ twisted sector states are located:
$(z_1,z_2,z_3)=(1/\sqrt{3})e^{\pi i/6}( k_1 R_1,k_2 R_2 ,k_3 R_3)$,
where $k_i=0,1,2$ and $R_i$ are the radii of the three $T^2$'s
(in each $T^2$, the two radii are the same because of the ${\mathbb Z}_3$
orbifold symmetry).
The twisted sector states at each fixed point form
an ${\cal N}=1$ neutral chiral supermultiplet.
In particular, the scalar component is $\phi_k + i a_k$ where
$\phi_k$ ($a_k$) comes from the NS-NS (R-R) sector.
Consistency ({\em i.e.},
tadpole cancellation) requires introducing open string and
$32$ $D9$-branes.
In the T-dual picture where
two of the complex dimensions $z_1$ and $z_2$ are dualized,
the $D9$-branes become $D5$-branes.
The gauge group is maximal when
the $D5$-branes are sitting on top on each other, for instance, when they
are all located at $z_1=z_2=0$.
In this case, the model has $SO(8) \times SU(12) \times U(1)$ gauge
symmetry. The open string sector gives rise to
three families of charged chiral multiplets in the
$({\bf 8},\overline{\bf 12})(-1)$ and $({\bf 1},{\bf 66})(+2)$
representations.
In this brane configuration, not all the twisted sector states
are located on the branes.
In particular, out of the $27$ twisted sector chiral multiplets,
only $3$ of them (the ones with $k_1=k_2=0$)
are located on the branes. The remaining twisted sector states
are located at a non-zero distance from the branes.
{}The strength of the coupling of a twisted field with the brane
fields depends on the
number of dimensions in which the twisted field lives, and its
location from the branes.
In the above example, the ${\mathbb Z}_3$ twisted sector states
live in four dimensions and so for the corresponding fixed points
which lie on the branes, the couplings with
the brane modes are unsuppressed, {\em i.e.}, $\kappa \sim g_s$.
For the remaining twisted moduli located
at non-zero distances
$|X|$ from the branes, the corresponding couplings with the brane fields
are exponentially suppressed as $\exp(-(M_s |X|)^2)$.
\begin{figure}
\centering
\epsfxsize=3 in
\hspace*{0in}\vspace*{.2in}
\epsffile{twisted1.eps}
\caption{\footnotesize{A schematic diagram of various types of twisted fields.
The complex coordinate $z_1$ is suppressed in this diagram.
The ${\mathbb Z}_3$ and ${\mathbb Z}_6$
twisted fields are localized at fixed points indicated by
crosses and a circle respectively.
The ${\mathbb Z}_2$ twisted field are
localized at fixed loci indicated
by dotted lines.
The two solid lines which are separated by $R_3$ (the size of the
compactified dimensions in the $z_3$ direction) are identified.
Our brane world is the dotted line with $z_1=z_2=0$
and so the ${\mathbb Z}_6$ fixed point is localized on our brane.}}
\label{twistedfig}
\end{figure}
{}In more realistic string models, there can be more
than one type of twisted fields. In particular, there can be
some twisted fields which are located
at fixed loci rather than fixed points
in the internal space. For example, in the ${\mathbb Z}_6$ orbifold \cite{z6}
generated by the above ${\mathbb Z}_3$ twist and an
additional ${\mathbb Z}_2$ twist where the
${\mathbb Z}_2$ generator $R$ acts on the coordinates as follows:
$R z_1 = - z_1$ , $R z_2 = - z_2$ , and $R z_3= z_3$.
The ${\mathbb Z}_2$ twisted sector states are located
at $16$ fixed loci in the internal space with $z_1$ and
$z_2$ given by
$(z_1,z_2)= {1 \over 2} \left( (a_1+ib_1)R_1, (a_2+ib_2)R_2 \right)$
where $a_i,b_i=0,1$.
While the ${\mathbb Z}_2$ twisted fields have no momentum or winding
in the
complex dimensions $z_1$ and $z_2$,
they are free to propagate in the complex dimension not twisted by the
${\mathbb Z}_2$ generator, {\em i.e.}, $z_3$.
Therefore, the ${\mathbb Z}_2$ twisted sector states
live in six dimensions. Even for the ${\mathbb Z}_2$ twisted
states that sit on the branes (the ones with $a_2=b_2=0$),
the corresponding couplings are suppressed
by the volume of the compactified dimensions in which the
${\mathbb Z}_2$ twisted sector states are free to propagate.
In other words, $\kappa \sim g_s / (V_3 M_s^2)$
where $V_3$ is the volume of the compactified dimensions not twisted by
${\mathbb Z}_2$.
In addition to the
${\mathbb Z}_3$ and ${\mathbb Z}_2$ twisted fields described above,
there are also
${\mathbb Z}_6$ twisted sector states.
A schematic diagram of various types of twisted fields
is given in Figure \ref{twistedfig}.
{}The gauge group of this model is
$[SU(6)\otimes SU(6) \otimes SU(4) \otimes U(1)^3]^2$.
Although this contains the SM, the
residual gauge symmetry is too large for the model to be phenomenologically
interesting. The rank of the gauge group can be reduced by turning on the
untwisted NS-NS sector B-field background \cite{bij}. In particular,
a three-family Pati-Salam like model was constructed in this
framework \cite{ST}. Some other four-dimensional ${\cal N}=1$
Type I string models were presented in Ref. \cite{zk}.
As noted above, the twisted fields are massless before SUSY is
broken, but would pick up masses
after SUSY breaking and dilaton stabilization.
The coupling of the twisted R-R fields $a_k$ to a $U(1)$ gauge field is
proportional to ${\mbox{Tr}} (\gamma_k \lambda_i) ~\partial_{\mu} a_k
A_i^{\mu}$,
where $\lambda_i$ is the Chan-Paton wavefunction of the abelian gauge field
$A_i^{\mu}$,
and $\gamma_k$ defines the action of the
orbifold on the D-branes.
The coupling of $a_k$ with a pair of
gauge fields (abelian or non-abelian) is $\propto
{\mbox Tr} (\gamma_k^{-1} \lambda_G^2) ~ a_k F \tilde{F}$
where $\lambda_G$ is the Chan-Paton wavefunction of the gauge fields
with gauge group $G$. Here, $F$ ($\tilde F$) is the
field strength (dual field strength) for the abelian or non-abelian gauge
fields. If $G$ is strongly coupled, the twisted
R-R fields $a_k$ can play the role of axions.
In contrast to perturbative heterotic string theory, a generic Type I
string model can have more than one anomalous $U(1)$ \cite{ibanez}.
Their triangle anomalies are cancelled by the generalized Green-Schwarz
mechanism \cite{gs} which involves the exchange of the R-R axions.
It is well known that the coupling which mixes a R-R axion with a $U(1)$ gauge
field is part of the $U(1)$ gauge-invariant combination
${1\over 2} {M_s^2} (\partial_{\mu} a_i - g_i e_i A_i^{\mu})^2$
where $A_i^{\mu}$, $i=1,2,...N$ are the anomalous $U(1)$'s, $g_i$
are the gauge couplings and $e_i$
are the corresponding charges.
The $N$ $U(1)$ gauge bosons become massive, eating up $N$ of the
twisted R-R fields (more precisely, $N$ linear combinations
of the twisted R-R fields $a_k$ described above).
Of the remaining twisted RR fields, it is likely that all except
one are relatively heavy. Although we do not expect continuous global
symmetry in a string model, it is plausible that some approximate global
symmetries are present in the low energy effective field theory,
{\em i.e.}, such symmetries are broken only by high-dimensional operators in
the effective theory. If an approximate Peccei-Quinn symmetry is
present, effectively, we have an axion field.
In this situation, strong interaction chiral symmetry dictates its
properties, that is, its mass and coupling follow from its mixing with the
neutral pion, with the mixing given by $f_{\pi}/M_s$. For $M_s \sim 10$ TeV,
this is clearly ruled out by experiments\cite{kim}. However, typically,
this axion field will also couple to other non-abelian gauge fields in
the hidden sector, and their gauge couplings are expected to be strong
at relatively high scales (compared to the QCD scale). In this situation,
the axion will pick up a relatively large mass and so can avoid
detection \cite{tye}.
{}Let us now focus on the couplings of the twisted states with the SM
fields. The bosonic states from the twisted sectors have spin-$0$, and
so can couple to a pair of gauge fields. In particular, the above mentioned
couplings
of the R-R axions with a pair of gauge fields
belong to this type. Similarly, the
spin-$0$ fields from the NS-NS sectors also couple to the gauge fields,
in the form
\begin{equation}
{\mbox Tr} (\gamma_k^{-1} \lambda_G^2) ~ \phi_k F^2 ~.
\end{equation}
{}The spin-$0$ fields from the twisted sectors
can also couple to
a pair of fermions ${\psi}_L$ and $\chi_R$ with opposite chirality.
Since the twisted fields (denoted collectively as $T$)
are gauge singlets, the three-point coupling to the Standard Model fermions
$T \overline{\psi}_L \chi_R$ is absent, since it is not a Standard Model
singlet.
However, the lowest order non-renormalizable couplings, such as
\begin{equation}
{\cal C}_{Qd} {g_s^{\footnotesize{3/2}} \over M_s} \overline{Q}_L d_R H_d T, \ \
{\cal C}_{Qu} {g_s^{\footnotesize{3/2}} \over M_s} \overline{Q}_L u_R H_u T, \ \
{\cal C}_{L\ell} {g_s^{\footnotesize{3/2}} \over M_s} \overline{L}_L \ell_R H_d T, \ \
\label{ffht}
\end{equation}
are presumably allowed \footnote{The powers of $g_s$ can be seen as follows.
The coupling between three open string states is proportional to
$g_s^{1/2}$ whereas the coupling of one closed and two open string states
is proportional to $g_s$. The above
$4$-point terms can be factorized into a product of these two
types of three-point couplings.},
are presumably allowed, where, in standard notation, $Q={u \choose d}$,
$L= {\nu_\ell \choose \ell}$, $H_{d,u}$ are the $Y=1,-1$ Higgs in the
MSSM, and we suppress generation labels
(the ${\cal C}$'s are dimensionless coupling matrices in generation space).
There are two types of cubic couplings that result from these quartic
couplings. First, nonperturbative effects can generate a nonzero
potential for the $T$ fields and hence they can acquire non-zero vevs.
These produce the cubic couplings
\begin{equation}
{\cal C}_{Qd}
{g_s^{\footnotesize{3/2}} \over M_s} \overline{Q}_L d_R H_d \langle T \rangle, \ \
{\cal C}_{Qu}
{g_s^{\footnotesize{3/2}} \over M_s} \overline{Q}_L u_R H_u \langle T \rangle, \ \
{\cal C}_{L\ell}
{g_s^{\footnotesize{3/2}} \over M_s} \overline{L}_L \ell_R H_d \langle T
\rangle
\label{yukawa}
\end{equation}
These would combine with the original dimension-$4$ Yukawa operators, and the
generational/flavor structure of the matrices ${\cal C}$ may help to yield
the observed hierarchical fermion masses and quark mixing.
Second, when $H_u$ and $H_d$ pick up vevs,
the quartic operators in (\ref{ffht}) will yield the cubic operators
\begin{equation}
{\cal C}_{Qd}
{g_s^{\footnotesize{3/2}} \over M_s} \overline{Q}_L d_R \langle H_d \rangle T, \ \
{\cal C}_{Qu}
{g_s^{\footnotesize{3/2}} \over M_s} \overline{Q}_L u_R \langle H_u \rangle T, \ \
{\cal C}_{L\ell}
{g_s^{\footnotesize{3/2}} \over M_s} \overline{L}_L \ell_R \langle H_d
\rangle T
\label{3point}
\end{equation}
For simplicity, let us neglect mixing effects. If
the ${\cal C}$ have the structure of the quark and lepton mass hierarchies,
$\sim m_f/M_{ew} << 1$ for all fermions $f$ other than the top quark, then the
induced 3-point couplings in (\ref{3point}) are presumably rather small.
Note that in the Pati-Salam case in Ref \cite{ST}, the right-handed fermion
components $f_R$ are $SU(2)_R$ doublets, which generically gives rise to a
further suppression factor $\propto M_R/M_s$, where $M_R$ denotes the
mass of the $W_R$ vector boson.
Thus, some linear combinations of the twisted RR scalars are eaten by
the anomalous $U(1)$ gauge bosons. The remaining ones $a_i$ together with
the twisted NS-NS scalars $\phi_i$ can appear in the low
energy effective action. Some of the
lowest order terms in the effective Lagrangian involving
the couplings of these twisted modes or the anomalous $U(1)$
gauge fields with the SM fields are:
\begin{eqnarray}\label{leff}
{\cal L} =&&
{g_s \over M_s} \left(
\sum_{i,j} {\cal C}^{(1)}_{ij} \phi_i F_j^2
+ \sum_{i,j} {\cal C}^{(2)}_{ij} a_i F_j \tilde{F}_j \right)
+ {g_s^{3/2} \over M_s} \left(
\overline{f}_L f_R H \sum_{i,f} {\cal C}^{(3)}_i a_i
+ \overline{f}_L f_R H \sum_{i,f} {\cal C}^{(4)}_i \phi_i
\right) \nonumber \\
+ && \left(
\sum_{i,f} g_i e_{Li} \overline{f}_L \gamma_{\mu} f_L A^{\mu}_i
+ \sum_{i,f} g_i e_{Ri} \overline{f}_R \gamma_{\mu} f_R A^{\mu}_i
\right) + \dots
\end{eqnarray}
where $\overline{f}_L f_R H=\overline{Q}_Ld_RH_d$, $\overline{Q}_Lu_RH_u$,
$\overline{L}_L\ell_RH_d$, generation
indices are suppressed, ${\cal C}^{(k)}_{ij}$ and ${\cal C}^{(k)}_i$ are
model-dependent coefficients,
$A_i^{\mu}$ denote the anomalous $U(1)$ gauge fields, with gauge
couplings $g_i$, and $e^L_i$ and $e_i^R$ are the corresponding charges.
$F_j^2$ and $F_j \tilde F_j$ sum over abelian and
nonabelian gauge fields.
The nonperturbative effects that generate a nonzero potential for the moduli
(as is necessary for dilaton stablization and is related to SUSY breaking)
will, in general, produce various quadratic, cubic, and quartic couplings
involving these moduli. These are subject to obvious constraints; for example,
they must be invariant under the orbifold twists.
In this letter, we discuss Type I strings on orbifolds, and
more generally,
when the orbifold singularities are smoothed out (``blown-up'') to a Calabi-Yau
manifold as the twisted NS-NS scalars $\phi_k$ acquire non-zero vevs. The
coupling of the twisted modes with the brane fields is
$\kappa \sim g_s f(\langle \phi_k \rangle)$ where $f(\langle \phi_k \rangle)$
is a function of $\langle \phi_k \rangle$.
Even in the generic case where $\langle \phi_k \rangle$ is comparable with the
string scale, it is likely that the suppression factor for the bulk fields,
{\em i.e.}, $(V_{p-3} V_{9-p} M_s^6)^{-1}$ is still much smaller than
$f(\langle \phi_k \rangle)$ and so there are closed string fields which couple
to the brane fields more strongly than the bulk fields. Moreover, we expect
that the features we discuss here would still apply for other realizations of
the brane world (which may be viewed as dual to the Type I description), such
as non-perturbative heterotic string/M theory models with solitonic fivebranes
\cite{ovrut} as well as F-theory \cite{vafa} models.
\section{Experimental Signatures and Constraints}
There are several experimental implications of eq. (\ref{leff}).
In general, there could be couplings such as $g_s\mu_{ijk} T_iT_jT_k$, where
the coefficient $\mu_{ijk}$ would presumably be of order the electroweak scale.
These would allow a subset of the $\xi = \phi_k,a_k$ to decay rapidly, with
widths $\Gamma \sim g_s^2\mu^2/m_\xi$. We concentrate here on the light $\xi$
fields for which these decay channels are kinematically forbidden, as
well as the subset of the heavier $\xi$ fields whose analogous decays do not
occur because the corresponding cubic or quartic operators are absent
(as a consequence, for example, of noninvariance under the orbifold twists).
In the following, we refer collectively to the various
mass eigenstates of $\phi$ and $a$ and suppress their multiplicity; similarly,
we use ${\cal C}$ to refer to the appropriate linear combinations of
the original ${\cal C}_{ij}^{(1)}$ and ${\cal C}_{ij}^{(2)}$ for these
mass eigenstates.
The masses
$m_\xi$, $\xi=\phi,a$, are expected to be of roughly of order the SUSY breaking
scale, which is comparable to the electroweak symmetry breaking scale. The
couplings in (\ref{leff}) give rise to the decays $\xi \to gg$ as well as
$\xi \to \gamma\gamma$, and (if kinematically allowed) $\xi
\to ZZ, \ W^+W^-$. From the lowest-order graphs, in terms of
\begin{equation}
\Gamma_0 = \frac{g_s^2 {\cal C}^2}{64 \pi} \Bigl
(\frac{m_\xi}{M_s}\Bigr )^2 m_\xi
\label{gammat}
\end{equation}
we find $\Gamma(\xi \to gg)= (N_c^2-1)\Gamma_0$ where $N_c=3$,
$\Gamma(\xi \to \gamma\gamma)=\Gamma_0$,
$\Gamma(\xi\to ZZ)=\Gamma_0(1-4m_Z^2/m_\xi^2)^{1/2}$, and
$\Gamma(\xi \to W^+W^-)=2\Gamma_0(1-4m_W^2/m_\xi^2)^{1/2}$,
with $\xi=\phi,a$.
Since one expects $g_s \sim O(1)$, these lowest-order calculations are only
rough estimates. Summing the partial widths and taking $g_s \sim O(1)$,
${\cal C} \sim O(1)$, and $m_{W,Z}^2/m_\xi^2 <<1$, we get
\begin{equation}
\Gamma_\xi \sim 0.1(m_\xi/M_s)^2m_\xi
\label{gammatotphi}
\end{equation}
For $m_\xi=1$ TeV, $M_s=10$ TeV, this gives $\Gamma_\xi \sim
10^{-3}m_\xi \sim 1$ GeV. The dominant decays of the $\phi,a$, namely
$(\phi,a) \to gg$, should not involve large missing energy and should in
principle allow one to reconstruct the mass.
The couplings $\phi F^2$ and $a F \tilde F$ with the gluon field
in (\ref{leff}) contribute to single jet inclusive, dijet, and multijet
production in $\bar p p$ (and, equally, in $pp$) collisions. The $\xi$
can be produced in the process $gg \to g \xi$ involving the usual triple
Yang-Mills vertex multiplied by $g_s{\cal C}/M_s$. The resultant decay of the
$\xi$ yields 3-jet events. The rate, relative to regular QCD 3-jet events is
roughly $(g_s{\cal C}/\alpha_3)^2({\hat s}/M_s)^2(1-m_\xi^2/{\hat s})^{1/2}$
(where $\hat s$ is the $gg$ center of mass energy squared) which could be
$O(10^{-2})$ for $\sqrt{s}=1.8$ TeV and our illustrative values
of $g_s, M_s, m_\xi$.
In $gg \to gg$ scattering, there are new contributions from graphs involving
the coupling of $gg$ to $\xi=\phi,a$ in the $s,t$ and $u$ channels. From the
tree-level graphs, we compute the new contribution to $gg \to gg$ to be
$d\sigma/d\hat t = |{\cal M}|^2/(16 \pi \hat s^2)$ where
\begin{eqnarray}
& & \overline{\sum}|{\cal M}|^2 =
\frac{(N_c^2-1)}{16}\Bigl (\frac{g_s{\cal C}}{M_s} \Bigr )^4
\sum_{m=m_\phi,m_a} \biggl [
\frac{\hat s^4}{(\hat s-m^2)^2} +
\frac{\hat t^4}{(\hat t-m^2)^2} +
\frac{\hat u^4}{(\hat u-m^2)^2} + \cr\cr
& & \frac{\hat s^4+\hat t^4+\hat u^4-2\hat u^2(\hat s^2+\hat t^2)}
{2(\hat s-m^2)(\hat t-m^2)} +
\frac{\hat s^4+\hat t^4+\hat u^4-2\hat s^2(\hat t^2+\hat u^2)}
{2(\hat t-m^2)(\hat u-m^2)} +
\frac{\hat s^4+\hat t^4+\hat u^4-2\hat t^2(\hat s^2+\hat u^2)}
{2(\hat s-m^2)(\hat u-m^2)} \biggr ]
\label{gg}
\end{eqnarray}
where integration over final state phase space involves a (1/2) factor for
identical particles. (As indicated, the $\phi F^2$ and $a F \tilde F$
contributions are of the same form, with $m_\phi \to m_a$.)
Since $g_s \sim O(1)$, higher order corrections can be substantial; however,
(\ref{gg}) gives a rough estimate. For $\hat s
\simeq m^2$, the $s$-channel term has a strong
resonant enhancement (since the $s$-channel propagator is actually
$\propto 1/(m \Gamma)$ at $\hat s=m^2$),
and, in particular, the $M_s^{-4}$ factor from the
coupling is cancelled by the $M_s^4$ factor in $\Gamma_\xi^{-2}$ arising from
the square of the $s$-channel propagator. Inserting (\ref{gammatotphi}), we
find that $\overline{\sum}|{\cal M}|^2 \simeq 50(g_s{\cal C})^4$. This is
comparable to the regular QCD contribution, which, e.g., for $gg$ CM angle
$\hat \theta=\pi/2$, gives $\overline{\sum}|{\cal M}|^2 \simeq 30 g_3^4$,
where $g_3$ is the SU(3) coupling. The most striking effect is that there
would be peaks in the dijet invariant mass at $M_{JJ}=m_{\phi,a}$.
We recall
that there are, in general, several $\phi_k$ and $a_k$, so there could be
several such resonances.
Let us compare the size of this effect with
the conventional scenario in which $M_s \sim M_P$.
In that case, the moduli fields $\phi,a$ could also be of order the SUSY
breaking scale, and, given that $M_s$ cancels out at the resonance for
$\hat s = m_\xi^2$, the existence of this resonance, {\it per se}, would not
constitute evidence of low-string scale, as opposed to conventional $M_s \sim
M_P$ models; however, the contribution of (\ref{gg}) to
$d^2\sigma/dM_{JJ}d\cos \hat \theta$ involves the convolution
over the momentum fractions $x_1,x_2$ of the gluons, and the off-resonant
contributions to this convolution would be very different for low- and
high-string scale theories because of the quite different $M_s$ scales.
Current data on 2-jet and inclusive jet production in $\bar p p$
collisions from D0 is in excellent agreement with QCD predictions
\cite{d0jets}. This is also true of the CDF data in the region where the $gg
\to gg$ subprocess makes its main contribution, for $E_T \mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle<}{\sim}$}} 300$ GeV
\cite{jetrev} (the latter resulting from the rapid rise of the gluon
distribution functions $g(x)$ for small $x$ and the fact that $E_T^2 = (\hat
s/4)\sin^2 \hat \theta < x_1 x_2 s/4$). Although the new contributions from
$\phi,a$ exchange can only roughly be estimated, in view of the expected large
$g_s \sim O(1)$ and the model-dependent factor ${\cal C}$, we infer that if
${\cal C} \sim 1$, then a safe bound in order to avoid
conflict with this data is
$m_{\phi,a} \mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}} O(1)$ TeV. Furthermore, if $m_{\phi,a}$ are sufficiently
large so that there is no significant resonant contribution, the data would
still constrain $M_s \mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle>}{\sim}$}} O(1)$ TeV, given that the corresponding couplings
$g_s {\cal C} \sim O(1)$. In models with a low string scale and large
compactification radii, there will also be a new contributions to $\bar p p$
scattering from processes involving the exchange or emission of gravitational
KK modes; however, in a global data analysis, one could still distinguish
between the effects of these KK modes and of the $\phi,a$.
Besides the contribution of the $\phi_k F_j^2$ term to gauge couplings,
via $\langle \phi_k \rangle$,
these terms also contribute to various loop effects which could
be significant.
Another important implication concerns neutral vector bosons.
In contrast to perturbative heterotic string theory,
there can be several anomalous $U(1)$ gauge symmetries in Type I string
models.
Moreover, in perturbative heterotic string theory where
$M_s \sim M_P$, the associated anomalous $U(1)$ vector bosons are
too heavy to be detected.
However, the
anomalous $U(1)$ vector bosons in the present scenario can be relatively
light ($\mathrel{\raisebox{-.6ex}{$\stackrel{\textstyle<}{\sim}$}}$ few TeV), and are possible candidates for
$Z^{\prime}$ bosons. These vector bosons gain masses by ``eating'' some of the
above axions. The resultant $Z^\prime$'s couple to the brane fields with
strength comparable with SM interactions. Current lower bounds on such
$Z^{\prime}$ bosons are of order 800 GeV, depending on their couplings
\cite{rpp}.
In addition to the scalar component fields $\phi,a$, the twisted moduli chiral
superfield ${\cal T}$ also has modulino component fields $\tilde \phi,
\tilde a$. We denote these collectively as $\tilde T$. Clearly these would
play the role of a electroweak singlet (``sterile'') neutrinos. Once a nonflat
superpotential is generated (nonperturbatively) for the twisted moduli, there
would be terms of the form ${\cal T}{\cal T}$ yielding the bilinears $\tilde
T^T C \tilde T$, with mass coefficients presumably of order the SUSY breaking
scale and hence comparable to the electroweak scale. If allowed by the matter
parities resulting from the underlying string theory, there could also be cubic
${\cal L}{\cal T}{\cal H}$ terms in the superpotential, which would give rise
to Dirac neutrino mass terms of the form $\bar \nu_L \tilde T_R$. Note that
the $\tilde T$ fields couple to brane fields, in particular,
neutrinos, without the volume suppression factors affecting the couplings of
the bulk fields to brane fields. We shall discuss this further elsewhere.
This contrasts with the situation for
candidates for sterile neutrinos coming from the bulk \cite{neutrino}.
The role of a generation dependent
anomalous $U(1)$ gauge symmetry in the
observed hierarchical fermion mass structure has been studied in the
context of perturbative heterotic string theory \cite{u1a}.
An important difference here is that
there can be
several anomalous $U(1)$'s (for example, the $Z_6$ model discussed
earlier has $4$ anomalous $U(1)$'s), and different families have different
$U(1)$ charge assignments, as in the case in Ref \cite{ST}.
Hence the possibilities in the brane world are more
intricate.
The numerous testable implications of models with low
string scale are clearly of great interest and
deserve further theoretical and experimental investigation.
\acknowledgments
{}We thank Philip Argyres, Zurab Kakushadze and Piljin Yi for
discussions. The research of G.S. and R.S. is partially supported by the NSF
grant PHY-97-22101. The research of S.-H.H.T. is partially supported by
the NSF.
|
\section{Introduction}
Let $X$ be a projective variety over ${\mathbb C}.$ Let $X_{an}$ be the
analytic space associated to $X.$ Let $c_1 : Pic(X) \to H^{2}(X_{an},
{\mathbb Z})$ be the map which associates to a line bundle (or equivalently
a Cartier divisor) on $X$ its cohomology class. We may identify the
N\'eron-Severi group $NS(X)$ with the image of $Pic(X)$ in
$H^{2}(X_{an},{\mathbb Z})$ under the above map.
If $X$ is smooth, then by the Hodge decomposition theorem, we know that
$$H^{2}(X_{an}, {\mathbb C}) = H^{2,0}(X_{an}) \oplus H^{1,1}(X_{an}) \oplus
H^{0,2}(X_{an}).$$
Let $F^{1}H^{2}(X_{an},{\mathbb C}) = H^{2,0}(X_{an}) \oplus H^{1,1}(X_{an}).$
The {\em Lefschetz theorem on $(1,1)$ classes} ([GH], [L]) states that if
$X$ is a smooth, projective variety, then
$$NS(X) = \{\alpha \in H^{2}(X_{an},{\mathbb Z}) | \alpha_{{\mathbb C}} \in
F^{1}H^{2}(X_{an},{\mathbb C})\}.$$
If $X$ is an arbitrary singular variety then by [D],~Theorem~8.2.2 the
cohomology groups of $X$ with ${\mathbb Z}$-coefficients carry mixed Hodge
structures. Hence it makes sense to talk of $F^{1}H^{2}(X_{an},{\mathbb C})$ for
such a variety $X$. Spencer Bloch, in a letter to Jannsen [J, appendix A],
asks whether the ``obvious'' extension of the Lefschetz $(1,1)$
theorem is true for singular projective varieties, i.e., is it true that
$$NS(X) = \{ \alpha \in H^{2}(X_{an}, {\mathbb Z}) | \alpha_{{\mathbb C}} \in
F^{1}H^{2}(X_{an},{\mathbb C})\} ?$$
Barbieri-Viale and Srinivas [BS1] gave a counterexample to this question.
Let $X$ be a surface defined by the homogenous equation $w(x^3 - y^2z) +
f(x,y,z) = 0$ in $\P^3_{{\mathbb C}},$ where $x,y,z,w$ are homogenous coordinates
in $\P^{3}_{{\mathbb C}}$ and $f$ is a ``general'' homogenous polynomial over ${\mathbb C}$
of degree $4.$ They showed that for such an $X$,
\[NS(X)\propsubset\{\alpha \in H^{2}(X_{an},{\mathbb Z}) |
\alpha_{{\mathbb C}} \in F^{1}H^{2}(X_{an}, {\mathbb C})\}.\]
In the same paper [BS1] the authors ask the following question.
Let $X$ be a complete variety over ${\mathbb C}$. Let $H^{1}(X, {\mathcal H}^{1}_{X})$
be the subgroup of $H^{2}(X_{an},{\mathbb Z})$ consisting of Zariski-locally
trivial cohomology classes, i.e., $\eta \in H^{2}(X_{an},{\mathbb Z})$ lies in
$H^{1}(X,{\mathcal H}^{1}_{X})$ if and only if there exists a finite open
cover $\{U_{i}\}$ of $X$ by Zariski open sets such that $\eta \mapsto 0$
under the restriction maps $H^{2}(X_{an}, {\mathbb Z}) \to H^{2}((U_i)_{an}, {\mathbb Z})$
for all $i$. Is
$$NS(X) = \{\alpha \in H^{1}(X,{\mathcal H}_{X}^{1}) | \alpha_{{\mathbb C}} \in
F^{1}H^{2}(X_{an},{\mathbb C})\} ?$$
We remark that for a smooth, projective variety $X$, if a cohomology class
$\eta \in H^{2}(X_{an},{\mathbb Z})$ is zero when restricted to a nonempty Zariski
open set $U \subset X,$ then $\eta$ is the class of a divisor. So
$H^{1}(X,{\mathcal H}^{1}_{X}) = NS(X)$ and the above question has a positive
answer for a smooth, projective variety $X$. For any
projective variety $X$, there is an inclusion $NS(X)\subset H^{1}(X,{\mathcal
H}_{X}^{1})$; Barbieri-Viale and Srinivas also give an example in [BS1]
of a singular variety for which this inclusion is strict.
In general, for any projective variety $X$ over ${\mathbb C},$
$$NS(X) \subset \{\alpha \in H^{1}(X,{\mathcal H}^{1}_{X}) | \alpha_{{\mathbb C}}
\in F^1 H^2 (X_{an},{\mathbb C})\}.$$
This follows from the inclusion $NS(X)\subset H^{1}(X,{\mathcal
H}_{X}^{1})$, combined with
$$Ker(H^{2}(X_{an},{\mathbb C}) \to H^{2}(X_{an}, {\mathcal O}_{X_{an}})) \subset F^1
H^2 (X_{an},{\mathbb C}),$$
which is a consequence of results of Du Bois [DB] (and is also implicit
in [D]). When $X$ is normal, we prove the reverse inclusion, thereby
answering the question in the affirmative, for the normal case.
The statement of our Main Theorem is
\begin{thm}\label{(1,1)} Let $X$ be a normal, projective variety over
${\mathbb C}.$ Then
$$NS(X) = \{ \alpha \in H^{1}(X, {\mathcal H}^{1}_{X}) | \alpha_{{\mathbb C}} \in F^1
H^2 (X_{an},{\mathbb C})\}.$$
\end{thm}
We also describe a counterexample, of a non-normal irreducible projective
3-fold with smooth normalization (isomorphic to
$\P^1\times\P^1\times\P^1$) for which the question has a {\em negative}
answer. However, it seems likely that the conclusion of the Theorem holds
for any {\it semi-normal} projective variety $X$ over ${\mathbb C}$.
Our proof of Theorem~\ref{(1,1)} is in two steps: first we show that
$H^{1}(X, {\mathcal H}^{1}_{X})\subset H^2(X_{an},{\mathbb Z})$ is a sub-MHS of
level~1; hence by [D], it determines a 1-motive, which we show to be
an extension of a direct sum of Tate structures ${\mathbb Z}(-1)$
by that of ($H^1$ of) an abelian variety. In the second part of the proof,
we give a direct construction of a certain 1-motive, using the Zariski
topology on $X$, and show that it is isogenous to the earlier one. The
Theorem will be an immediate corollary.
In a future work, we hope to use the second, algebraically defined
1-motive to also obtain the analogue of the Tate conjecture in our
situation, which would similarly characterize the ${\mathbb Z}_{\ell}$-span of the
classes of Cartier divisors in $H^2_{\rm et\/}(X_{\bar{K}},{\mathbb Z}_{\ell}(1))$,
for a normal projective variety $X$ over a number field $K$.
In another direction, our result suggests a question analogous to the
Hodge conjecture. Let $X$ be a normal projective variety over ${\mathbb C}$, and
$\alpha:X_{an}\to X$ be the obvious continuous map, leading to a Leray
spectral sequence
\[E_2^{p,q}=H^p(X,R^q\alpha_*{\mathbb Q}_{X_{an}})\implies H^{p+q}(X_{an},{\mathbb Q}),\]
with an induced decreasing {\em Leray filtration}
$\{L^pH^n(X_{an},{\mathbb Q})\}_{p\geq 0}$ on each cohomology group
$H^n(X_{an},{\mathbb Q})$. Let
\[{\rm Hg}^p(X)=L^pH^{2p}(X_{an},{\mathbb Q})\cap F^pH^{2p}(X_{an},{\mathbb C}).\]
Is ${\rm Hg}^p(X)$ the image of the $p$-th Chern class
map $c_p:K_0(X)\otimes{\mathbb Q}\to H^{2p}(X,{\mathbb Q})$? Note that this does not hold
without some hypothesis like (at least) normality; for example, Bloch's
letter to Jannsen [J,Appendix A] gives a counterexample. On the other
hand, results of Collino [Co] imply it when $X$ has a unique singular
point. Note also that, unlike the standard Hodge conjecture, the positive
answer (our Theorem above) for divisors does not automatically imply a
positive answer for the case of 1-cycles, since we do not have Poincar\'e
duality.
This work formed part of the first author's Ph.D. thesis, written at
the Tata Institute of Fundamental Research, Mumbai, submitted to the
Mumbai University in March, 1997.
\section{Some preliminaries}
\subsection{Constructible sheaves}
We will need below some technical results on constructible sheaves on
complex algebraic varieties. We begin by recalling the appropriate
definitions, the first from [V] and the second from [BS2].
\begin{defn} Let $X$ be a complex algebraic variety. We say that a sheaf
${\mathcal F}$ of abelian groups on the analytic space $X_{an}$ is
(algebraically) {\em ${\mathbb Z}$-constructible} if there is a finite
decomposition $X=\cup_{i\in I} X_i$, where each $X_i$ is irreducible and
Zariski closed in $X$, such that if $U_i = X_i - \cup_{X_j \propsubset
X_i}X_{j}$, then each $U_i$ is non-singular, $X$ is the disjoint union of
the $U_{i}$, and ${\mathcal F}|_{(U_i)_{an}}$ is a locally constant sheaf
whose fibre is a finitely generated group. We call any such collection of
subsets $\{X_i\}_{i\in I}$ an {\em admissible family} of subsets for
$\sF$. \end{defn}
\begin{defn} A sheaf ${\mathcal G}$ on a scheme $X$ (over an algebraically
closed field $k$, say) is said to be {\em ${\mathbb Z}$-constructible for the
Zariski topology} if we can express $X$ as a finite union $X=\cup X_i$,
where $X_i \subset X$ are Zariski closed, such that if $U_{i}= X_{i} -
\cup_{X_j \propsubset X_i} X_{j}$, then each $U_i$ is non-singular, $X$ is
the disjoint union of the $U_{i}$, and ${\mathcal G}|_{U_i}$ is a {\em
constant sheaf} associated to a finitely generated abelian group. We call
any such collection of subsets $\{X_i\}_{i\in I}$ an {\em admissible
family} for $\sG$.
\end{defn}
\begin{rmk}
We note that in the cited works, it is not required that the ``open
strata'' $U_i$ are non-singular, but this may clearly be assumed as well
without loss of generality, by refining any given stratification which has
all the remaining properties.
\end{rmk}
Note that if $\{X_i\}_{i\in I}$ is an admissible family of subsets for a
${\mathbb Z}$-constructible sheaf in either of the senses above, then there is a
natural partial order on the index set $I$ given by $j\leq i\iff X_j\subset
X_i$. Then we clearly have $U_i=X_i-\cup_{j<i}X_j$. Note that $U_i=X_i$
precisely when $i$ is a minimal element of $I$ with respect to the partial
order.
We recall the following basic result from [V], which is made use of
below.
\begin{thm}\label{Verdier} If $f:Y \to X$ is a
morphism of ${\mathbb C}$-varieties and ${\mathcal F}$ is a ${\mathbb Z}$-constructible sheaf
on $Y_{an},$ then $R^{i}f_{*}{\mathcal F}$ is a ${\mathbb Z}$-constructible sheaf on
$X_{an}$.
\end{thm}
We also need a certain general sheaf-theoretic result, which is presumably
well-known, but for which we do not know a reference. Let $X$ be a
topological space, $\{U_i\}_{i\in I}$ any finite collection of locally
closed subsets of $X$ which stratify $X$ ({\it i.e.\/},\ $X$ is the disjoint union of
the $U_i$, and for each $i$, the closure $X_i:=\bar{U_i}$ is again a union
of some $U_j$). Let $\leq $ denote the obvious partial order on $I$,
given by $i\leq j$ \iff $X_i\subset X_j$.
Let $f_i:U_i\to X$ be the inclusion. If $\sF$ is a sheaf of abelian groups
on $X$, and $\{i_0\leq i_1\leq\cdots\leq i_p\}$ is a $p$-chain in $I$, let
\[\sF_{i_0i_1\cdots
i_p}:=(f_{i_0})_*f_{i_0}^{-1}(f_{i_1})_*f_{i_1}^{-1}\cdots
(f_{i_p})_*f_{i_p}^{-1}\sF.\]
Note that the sheaves $\sF_{i_0\cdots i_p}$ define a cosimplicial sheaf on
the simplicial space $X\times N(I)$, where $N(I)$ denotes the nerve of $I$
(regarded as a discrete simplicial space). The augmentation $X\times
N(I)\to X$ gives rise to a complex of sheaves on $X$
\[0\to \sF\to \bigoplus_{i\in I}\sF_i\to\bigoplus_{\{i_0\leq
i_1\}\in N_1(I)}\sF_{i_0i_1}\to\cdots\to \bigoplus_{\{i_0\leq\cdots\leq
i_p\}\in N_p(I)}\sF_{i_0\cdots i_p}\to\cdots\hspace{1cm}\cdots\;\;(*)\]
\begin{lemma}\label{resol}
The above complex $(*)$ is a resolution of $\sF$.
\end{lemma}
\begin{proof}
If $x\in U_i$, then taking the stalks at $x$, we have an associated
cosimplicial abelian group $(\sF_{i_0i_1\cdots i_p})_x$, and
a corresponding augmented complex. Clearly $(\sF_{i_0i_1\cdots i_p})_x=0$
unless $i\leq i_0$. Since the partially ordered subset $(I\geq i)=\{j\in
I\mid i\leq j\}$ has a minimal element, one sees easily that
the stalk complex at $x$ is contractible (note that if $x\in U_i$, and
$\sigma=\{i_0\leq\cdots\leq i_p\}$ is a $p$-simplex in the nerve of
$(I\geq
i$), the stalks at $x$ of $\sF_{i_0\cdots i_p}$ and $\sF_{ii_0\cdots
i_p}$ are naturally isomorphic, where $\{i\leq i_0\leq\cdots\leq i_p\}$
is the cone over $\sigma$ with vertex $i$).
\end{proof}
\begin{rmk}
In case $\sF$ is ${\mathbb Z}$-constructible for the Zariski
topology on a scheme $X$, and $\{X_i\}$ is an admissible family for $\sF$,
such that $\sF\mid_{U_i}$ is the constant sheaf associated to $A_i$, then
$\sF_{i_0i_1\cdots i_p}$ is just the constant sheaf $(A_{i_p})_{X_{i_0}}$.
In particular, for a ${\mathbb Z}$-constructible sheaf in the Zariski topology, we
obtain a {\em flasque resolution}.
\end{rmk}
The key technical result of this section is the following.
\begin{lemma}\label{const1}
Let $\sA=A_{X_{an}}$ be a constant sheaf on a complex algebraic variety $X$,
and let $\sG$ be a ${\mathbb Z}$-constructible sheaf on $X_{an}$. Let $f:\sA\to
\sG$ be a sheaf homomorphism, and take $\sF={\rm image}\, f$. Let
$a:X_{an}\to X$ be the natural continuous map from the analytic space
$X_{an}$ to $X$, which is the identity on points. Then we have the
following.
\begin{points}
\item $a_*\sF$ is a constructible sheaf on $X$ for the Zariski topology.
\item The natural map $a^{-1}a_*\sF\to\sF$ is an isomorphism, and the
natural map $a_*\sA\to a_*\sF$ is surjective, {\it i.e.\/},\ $a_*\sF$ is a quotient
of the constant sheaf on $X$ associated to the abelian group $A$.
\item Let $\{X_i\}_{i\in I}$ be an admissible family of subsets for $\sG$.
Then it is also admissible for $\sF$, and for $a_*\sF$.
There is an exact sequence
\[0\to H^0(X_{an},\sF)\to \bigoplus_{i\in
I}H^0((U_i)_{an},\sF\mid_{(U_i)_{an}})\to
\bigoplus_{\begin{array}{c}i\leq j\\
i,j\in I\end{array}}H^0((U_{j})_{an},\sF\mid_{(U_{j})_{an}})
\]
\end{points}
\end{lemma}
\begin{proof} We first claim that if $U\subset X$ is an irreducible
(Zariski) locally closed subset such that $\sG\mid_{U_{an}}$ is locally
free, then $\sF\mid_{U_{an}}$ is a constant sheaf associated to a finitely
generated abelian group, which is a quotient of $A$. Indeed,
$\sG\mid_{U_{an}}$ corresponds to a representation of the fundamental
group of $U_{an}$ (with respect to any convenient base point), while
$A_{U_{an}}$ corresponds to the trivial representation. The sheaf map
$f\mid_{U_{an}}$ is then a morphism of local systems, whose image
$\sF\mid_{U_{an}}$ is clearly a trivial ({\it i.e.\/},\ constant) local subsystem of
$\sG\mid_{U_{an}}$.
Now let $\{X_i\}_{i\in I}$ be admissible for $\sG$. As observed above,
$\sF\mid_{(U_i)_{an}}$ is constant for each $i$, and so $\{X_i\}_{i\in I}$
is also admissible for $\sF$. From lemma~2 of [BS2], it follows that
$a_*\sF$ is ${\mathbb Z}$-constructible for the Zariski topology.
>From the beginning of the exact sequence $(*)$ of lemma~\ref{resol} (for
$\sF$ on $X_{an}$) we have inclusions
\[\sF\hookrightarrow\bigoplus_{i\in I}\sF_i,\;\;a_*\sF\hookrightarrow\bigoplus_{i\in
I}a_*\sF_i,\;\;a^{-1}a_*\sF\hookrightarrow\bigoplus_{i\in I}a^{-1}a_*\sF_i. \]
We see at once from the definitions that $a_*\sF_i$ is (the direct
image on $X$ of) a constant sheaf on $X_i$, for
each $i$, and the natural sheaf map $a^{-1}a_*\sF_i\to\sF_i$ is injective.
Since $\sA$ is a constant sheaf, we also have that $a^{-1}a_*\sA\to \sA$
is an isomorphism. Now from the commutative diagram
\[\diagram
a^{-1}a_*\sA \dto_{\cong}\rto & a^{-1}a_*\sF \rto|<<\ahook \dto|<<\ahook
& \bigoplus_{i\in I} a^{-1}a_*\sF_i \dto|<<\ahook \\
\sA\rto|>>\tip & \sF \rto|<<\ahook & \bigoplus_{i\in I}\sF_i
\enddiagram
\]
we deduce that $a^{-1}a_*\sF\to \sF$ is an isomorphism, and that the
natural map $a^{-1}a_*\sA\to a^{-1}a_*\sF$ is surjective. This implies
that $a_*\sA\to a_*\sF$ is surjective as well, and that $\{X_i\}$ is
admissible for $a_*\sF$. The exact sequence in (iii) of the lemma is
obtained from the resolution of lemma~\ref{resol} for $a_*\sF$.
\end{proof}
\subsection{A homological lemma}
We prove here an abstract homological lemma (lemma~\ref{[PS]}) which is a
variant of a lemma in [PS], which we will need later. The lemma is
formulated and proved with abelian groups, but a similar argument yields
it in an arbitrary abelian category.
Suppose we have the following $9$-diagram, in the category
of complexes of abelian groups, with exact rows and columns.
$$\diagram & 0 \dto & 0 \dto & 0 \dto \\
0 \rto & C_{11}^{\d} \rto \dto & C_{12}^{\d} \rto
\dto & C_{13}^{\d} \rto \dto & 0 \\
0 \rto & C_{21}^{\d} \rto \dto & C_{22}^{\d} \rto
\dto & C_{23}^{\d} \rto \dto & 0 \\
0 \rto & C_{31}^{\d} \rto \dto & C_{32}^{\d} \rto \dto &
C_{33}^{\d} \rto \dto & 0 \\
& 0 & 0 & 0 \enddiagram$$
Applying the cohomology functor we get an infinite double sequence with
exact rows and columns as shown below:
$$\diagram H^{i-2}(C_{22}^{\d}) \rto \dto & H^{i-2}(C_{23}^{\d})
\rto
\dto & H^{i-1}(C_{21}^{\d}) \rto \dto & H^{i-1}(C_{22}^{\d}) \rto
\dto & H^{i-1}(C_{23}^{\d}) \dto \rto & \\
H^{i-2}(C_{32}^{\d}) \rto \dto & H^{i-2}(C_{33}^{\d}) \rto
\dto & H^{i-1}(C_{31}^{\d}) \rto \dto & H^{i-1}(C_{32}^{\d}) \rto
\dto & H^{i-1}(C_{33}^{\d}) \dto \rto & \\
H^{i-1}(C_{12}^{\d}) \rto \dto & H^{i-1}(C_{13}^{\d}) \rto
\dto & H^{i}(C_{11}^{\d}) \rto \dto & H^{i}(C_{12}^{\d}) \rto \dto &
H^{i}(C_{13}^{\d}) \dto \rto & \\
H^{i-1}(C_{22}^{\d}) \rto \dto & H^{i-1}(C_{23}^{\d}) \rto
\dto & H^{i}(C_{21}^{\d}) \rto \dto & H^{i}(C_{22}^{\d}) \rto \dto &
H^{i}(C_{23}^{\d}) \dto \rto & \\
H^{i-1}(C_{32}^{\d}) \dto \rto & H^{i-1}(C_{33}^{\d})
\dto \rto & H^{i}(C_{31}^{\d}) \dto \rto & H^{i}(C_{32}^{\d}) \dto
\rto & H^{i}(C_{33}^{\d}) \rto \dto & \\
& & & & \enddiagram$$
Suppose now that we have an element $\alpha \in H^{i}(C^{\d}_{rs})$
(say for example $\alpha \in H^{i}(C^{\d}_{22})$ in the above
diagram) such that $\alpha \mapsto 0$ under both the maps with domain
$H^i(C^{\d}_{rs})$. We can then do a diagram chase in the above
cohomology diagram in the following way. Suppose $\alpha\in
H^{i}(C^{\d}_{22})$; arbitrarily choose lifts
$\beta_{1} \in H^{i}(C^{\d}_{21})$ and
$\beta_{2} \in H^{i}(C^{\d}_{12})$ lifting $\alpha.$
Let $\beta_{1} \mapsto \gamma_{1} \in H^{i}(C^{\d}_{31})$ and let
$\beta_{2} \mapsto \gamma_{2} \in H^{i}(C^{\d}_{13}).$ Then since
$\gamma_{1} \mapsto 0 \in H^{i}(C^{\d}_{32})$ and $\gamma_{2}
\mapsto 0 \in H^{i}(C^{\d}_{23})$, there exist $\delta_{1}$ and
$\delta_{2}$, both in $H^{i-1}(C^{\d}_{33})$, lifting
$\gamma_{1}$ and $\gamma_{2}$ respectively.
We can do a similar diagram chase beginning with an element $\alpha\in
H^i(C^{\d}_{rs})$, for arbitrary $i,r,s$, and end up with two elements
$\delta_1,\delta_2$ in the same group $H^j(C^{\d}_{r+1\;s+1})$, where we
read the subscripts modulo 3, and $j$ is either $i-1$, $i$ or $i+1$,
depending on $(r,s)$ (we end up at the two places in the diagram which
have the same entry, and are each 1 `knight's move' away from the starting point).
Let $\bar{H^j}(C^{\d}_{r+1\;s+1})$ denote the quotient of
$H^j(C^{\d}_{r+1\;s+1})$ by the subgroup generated by the images of the
two maps in the large commutative cohomology diagram with range
$H^j(C^{\d}_{r+1\;s+1})$. For example,
\[\bar{H^{i-1}}(C^{\d}_{3,3})=\frac{H^{i-1}(C^{\d}_{33})}{{\rm
image}\,H^{i-1}(C^{\d}_{23})+{\rm image}\,H^{i-1}(C^{\d}_{32})}.\]
\begin{lemma} \label{[PS]} With the notation as above, we have
$$(\delta_{1} - \delta_{2})\mapsto 0 \in \bar{H^{j}}(C^{\d}_{r+1\;s+1}).$$
\end{lemma}
\begin{proof} We first note that, by an argument with mapping cones and
cylinders (rotating the distinguished triangles in the 9-diagram), we may
assume that $\alpha \in H^{i}(C^{\d}_{22})$ without loss
of generality. For such an $\alpha$ the analogous result for the
cohomology diagram arising from a 9-diagram in the category of sheaves
has been proved by Parimala and Srinivas [PS, Sec 3]. The proof of this
lemma is entirely analogous: regarding the given 9-diagram as a (bounded)
double complex of complexes, one considers the total complex, which is a
5-term exact sequence of complexes, say
\[0\to {\mathcal C}_0\to{\mathcal C}_1\to{\mathcal C}_2\to{\mathcal C}_3\to{\mathcal C}_4\to 0.\]
Regarding this again as a double complex, there is a spectral sequence
\[E_1^{r,s}=H^s({\mathcal C}_r)\implies H^{r+s}(Tot({\mathcal C}_{\d}))\]
(the limit is in fact 0). Then the conclusion of the lemma is
interpreted as giving two (equivalent) ways of computing the differential
$E_2^{2,i}\to E_2^{4,i-1}$.
\end{proof}
\begin{rmk} An analogue of lemma~\ref{[PS]} can be formulated for a
9-diagram in the derived category of abelian groups
$$\diagram
C_{11} \rto \dto & C_{12} \rto
\dto & C_{13} \rto \dto & C_{11}[1]\dto \\
C_{21} \rto \dto & C_{22} \rto
\dto & C_{23} \rto \dto & C_{21}[1] \dto\\
C_{31} \rto \dto & C_{32} \rto \dto &
C_{33} \rto \dto & C_{31}[1]\dto \\
C_{11}[1]\rto& C_{12}[1]\rto & C_{13}[1]\rto & C_{11}[2] \enddiagram$$
where the rows and columns are distinguished triangles, and where the
cohomology diagram considered earlier is replaced by the diagram obtained
by applying any abelian group valued cohomological functor (of course a
still more general formulation is also possible). This is {\em false};
O.~Gabber has kindly shown us a counterexample.
\end{rmk}
\begin{rmk} A version of the above lemma~\ref{[PS]} also appears in a
letter from U. Jannsen to B. Gross.
\end{rmk}
\section{A short exact sequence of mixed Hodge structures}
In this section we make an analysis of the mixed Hodge structure on
\[H^1(X,\sH^1_X)=\mbox{ subgroup of Zariski locally trivial elements in
$H^2(X_{an},{\mathbb Z})$}.\]
Our goal is to describe it as an extension of a direct sum of Tate Hodge
structures ${\mathbb Z}(-1)$ by a polarizable pure Hodge structure of weight 1.
Let $X$ be our given normal projective variety over ${\mathbb C}$. Let $Y$ be a
resolution of singularities of $X$, and let $Y_{an}$ be the
associated analytic space of $Y$. We have the following commutative
diagram
$$\diagram Y_{an} \rto^{a^{Y}} \dto^{\pi^{an}} & Y \dto^{\pi} \\
X_{an} \rto^{a^{X}} & X \enddiagram$$
The Leray spectral sequence for the constant sheaf ${\mathbb Z}={\mathbb Z}_{Y_{an}}$ and
the map $\pi^{an}: Y_{an} \to X_{an}$ leads to an exact sequence:
\begin{equation}\label{1} 0 \to H^1 (X_{an}, {\mathbb Z}) \to H^1 (Y_{an},{\mathbb Z}) \to
H^0(X_{an}, R^1 \pi^{an}_{*}{\mathbb Z}) \to H^2 (X_{an},{\mathbb Z}) \to H^{2}(Y_{an}, {\mathbb Z})
\end{equation}
Note that since $X$ is normal, we have $\pi^{an}_{*}{\mathbb Z} \cong {\mathbb Z}$.
Define a new sheaf ${\mathcal F}_{{\mathbb Z}}$ on $X_{an}$ by
\begin{equation}\label{FZ}
{\mathcal F}_{{\mathbb Z}} = {\rm image}\,(H^1 (Y_{an}, {\mathbb Z})_{X_{an}} \to
R^{1}\pi^{an}_{*}{\mathbb Z}).
\end{equation}
Here by $H^{1}(Y_{an},{\mathbb Z})_{X_{an}}$ we mean the constant sheaf on $X_{an}$
associated to the group $H^1 (Y_{an}, {\mathbb Z}),$ and the map on sheaves is
induced at the level of presheaves by the restriction map on cohomology
$H^{1}(Y_{an},{\mathbb Z}) \to H^{1}((\pi^{an})^{-1}(U_{an}),{\mathbb Z})$ where $U_{an}
\subset X_{an}$ is open. By taking global sections we have the following
commutative diagram,
$$\diagram H^{1}(Y_{an},{\mathbb Z}) \rto \drto & H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})
\dto|<<\ahook \\ & H^{0}(X_{an},R^{1}\pi^{an}_{*}{\mathbb Z}) \enddiagram$$
Hence we have an inclusion,
$$0 \to \frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})}{{\rm Im}(H^{1}(Y_{an},{\mathbb Z}))}
\to \frac{H^{0}(X_{an},R^{1}\pi^{an}_{*}{\mathbb Z})}{{\rm Im}(H^{1}(Y_{an},{\mathbb Z}))}
= Ker(H^{2}(X_{an},{\mathbb Z}) \to H^{2}(Y_{an},{\mathbb Z}))$$
where the last equality is due to the above exact sequence (\ref{1}) of
low degree terms of the Leray spectral sequence.
Note that $\sF_{{\mathbb Z}}$ satisfies the hypotheses of lemma~\ref{const1}, with
$A=H^1(Y_{an},{\mathbb Z})$ and $\sG=R^1\pi^{an}_*{\mathbb Z}$ (the latter is algebraically
${\mathbb Z}$-constructible by theorem~\ref{Verdier}). Hence the following
properties hold.
\begin{points}
\item $\sF_{{\mathbb Z}}$ is algebraically ${\mathbb Z}$-constructible.
\item $a^X_*\sF_{{\mathbb Z}}:=\sG_{{\mathbb Z}}$ is ${\mathbb Z}$-constructible for the Zariski
topology, and $(a^X)^*\sG_{{\mathbb Z}}\cong\sF_{{\mathbb Z}}$.
\item The natural sheaf map
\begin{equation}\label{eqsur}
H^1(Y_{an},{\mathbb Z})_X\to a^X_*\sF_{{\mathbb Z}}
\end{equation}
is surjective.
\end{points}
\begin{lemma}\label{incl}
$\displaystyle{\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})}{Im(H^{1}(Y_{an},{\mathbb Z}))}
\subset H^{1}(X,{\mathcal H}^{1}_{X})}.$
\end{lemma}
\begin{proof} Let $\alpha \in H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})=H^0(X,a^X_*\sF_{{\mathbb Z}})$. Then by (\ref{eqsur}),
there exists a Zariski open cover
$\{U_i\}$ of $X$ such that $\alpha|_{(U_i)_{an}} = Im(\beta_{i})$ where
$\beta_{i} \in H^{1}(Y_{an},{\mathbb Z}).$ Therefore $\alpha|_{(U_{i})_{an}} \to 0
\in H^{2}((U_i)_{an},{\mathbb Z})$ as shown in the commutative diagram below (where
$U$ stands for any of the $U_i$)
$$\diagram H^{1}(Y_{an},{\mathbb Z}) \dto \rto & H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}}) \rto
\dto & H^{2}(X_{an},{\mathbb Z}) \dto \\
H^{1}((\pi^{an})^{-1}(U_{an}),{\mathbb Z}) \rto & H^{0}(U_{an},{\mathcal F}_{{\mathbb Z}})
\rto & H^{2}(U_{an}{\mathbb Z}) \enddiagram$$
This finishes the proof of the lemma.
\end{proof}
If $\{X_i\}_{i\in I}$ is an admissible family of subsets
for the constructible sheaf $R^1\pi^{an}_*{\mathbb Z}$ on $X_{an}$, then
(lemma~\ref{const1}) it is also an admissible family for $\sF_{{\mathbb Z}}$ and
for $a^X_*\sF_{{\mathbb Z}}=\sG_{{\mathbb Z}}$. We fix such an admissible family once and
for all, and fix base points $x_i\in U_i$ with corresponding reduced
fibers $F_i=\pi^{-1}(x_i)_{red}$. Let
$F=\cup_iF_i=\pi^{-1}(\{x_i\mid i\in I\})$.
By the proper base change theorem, the stalk
$(R^1\pi^{an}_*{\mathbb Z})_{x_i}$ is naturally identified with
$H^1((F_i)_{an},{\mathbb Z})$; thus $R^1\pi^{an}_*{\mathbb Z}\mid_{(U_i)_{an}}$ is a local
system with fiber $H^1((F_i)_{an},{\mathbb Z})$. Note that the stalk
$(\sF_{{\mathbb Z}})_{x_i}$ has the resulting description
\begin{equation}\label{stalk}
(\sF_{{\mathbb Z}})_{x_i}={\rm image}\,\left(H^1(Y_{an},{\mathbb Z})\to
H^1((F_i)_{an},{\mathbb Z})\right).
\end{equation}
By mixed Hodge theory [D], we deduce that $(\sF_{{\mathbb Z}})_{x_i}$ naturally
supports a pure Hodge structure of weight 1, which is a quotient Hodge
structure of $H^1(Y_{an},{\mathbb Z})$ (depending only on $i\in I$, and not on the
chosen base point $x_i\in U_i$), as well as a Hodge sub-structure of
$H^1((F_i)_{an},{\mathbb Z})$. Finally note also that $\sF_{{\mathbb Z}}\mid_{(U_i)_{an}}$
is a constant sheaf whose fiber supports this pure Hodge structure ({\it i.e.\/},\ is
the underlying lattice).
\begin{lemma} $H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})$ carries a pure Hodge
structure of weight one, such that $H^{1}(Y_{an},{\mathbb Z}) \to
H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})$ is a morphism of Hodge structures.
\end{lemma}
\begin{proof}
>From lemma~\ref{const1}(iii) there exists an
exact sequence of abelian groups
$$0 \to H^{0}(X_{an}, {\mathcal F}_{{\mathbb Z}}) \to
\bigoplus_{i\in I}(\sF_{{\mathbb Z}})_{x_i} \to
\bigoplus_{i_0,i_1\in I,\;i_0\leq
i_1}(\sF_{{\mathbb Z}})_{x_{i_1}}$$
The natural surjective maps $(\sF_{{\mathbb Z}})_{x_i}\to (\sF_{{\mathbb Z}})_{x_j}$
(for $i\leq j$) are maps of pure Hodge structures of weight one, which are
quotients of $H^{1}(Y_{an},{\mathbb Z})$. Hence $H^{0}(X_{an}, {\mathcal F}_{{\mathbb Z}})$ is
identified with the kernel of a morphism of pure Hodge structures of
weight~1, and hence itself supports a pure Hodge structure of weight one.
Also it is clear from the construction that the composition
\[H^1(Y_{an},{\mathbb Z})\to H^0(X_{an},\sF_{{\mathbb Z}})\hookrightarrow
\bigoplus_{i\in I}(\sF_{{\mathbb Z}})_{x_i}\]
is a direct sum of the natural quotient maps $H^1(Y_{an},{\mathbb Z})\to
(\sF_{{\mathbb Z}})_{x_i}$, and hence is a morphism of Hodge structures. Hence
$H^1(Y_{an},{\mathbb Z})\to H^0(X_{an},\sF_{{\mathbb Z}})$ is one as well.
\end{proof}
\begin{propose} \label{MHS} $H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}}) \to
H^{2}(X_{an},{\mathbb Z})$ is morphism of Hodge structures, i.e., the Hodge
structures on $H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})$ and $H^{2}(X_{an}, {\mathbb Z})$
are compatible.
\end{propose}
\begin{proof} Let $F_i=\pi^{-1}(x_i)$ as above, and let $F = \cup_{i\in I}
F_i$. We note that the natural map $H^{0}(X_{an}, R^{1}\pi_{*}^{an}{\mathbb Z}) \to
H^{1}(F_{an},{\mathbb Z})$ is an injection (any section in the kernel must vanish
in all
stalks). This implies that the map $H^{2}(X_{an},{\mathbb Z}) \to
H^{2}(Y_{an},F_{an},{\mathbb Z})$ (which is a morphism of mixed Hodge structures)
is injective in the following commutative diagram (here
$G_i=(\sF_{{\mathbb Z}})_{x_i}$).
$$\diagram \displaystyle{\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})}{Im(H^{1}(Y_{an},{\mathbb Z})}}
\rto|<<\ahook \dto|<<\ahook &
\displaystyle{\frac{H^{0}(X_{an},
R^{1}\pi^{an}_{*}{\mathbb Z})}{Im(H^{1}(Y_{an},{\mathbb Z})}}
\ddlto|<<\ahook \rto|<<\ahook & H^{2}(X_{an},{\mathbb Z}) \ddto|<<\ahook \\
\displaystyle{\frac{\oplus_{i}G_{i}}{Im(H^{1}(Y_{an},{\mathbb Z})}}
\dto|<<\ahook \\
\displaystyle{\frac{\oplus_{i}H^{1}((F_{i})_{an},
{\mathbb Z})}{Im(H^{1}(Y_{an},{\mathbb Z}))}} \rrto|<<\ahook & &
H^{2}(Y_{an},F_{an},{\mathbb Z}) \enddiagram$$
We are done, because all the arrows in the above diagram are injections,
and the vertical arrows (on the left and right borders), as well as the
lower horizontal arrow, are morphisms of mixed Hodge structures.
\end{proof}
Let $A \subset B$ be an inclusion of abelian groups. Let $A \subset A^s
\subset B$ denote the saturation of $A$ in $B,$ i.e., $A^{s}$ is the
smallest subgroup of $B$ containing $A$ such that
$\displaystyle{\frac{B}{A^{s}}}$ is torsion free. Let $H^{0}(X_{an}, {\mathcal
F}_{{\mathbb Z}})^{s}$ be the saturation of $H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})$ in
$H^{0}(X_{an}, R^{1}\pi^{an}_{*}{\mathbb Z}).$
\begin{lemma} $\displaystyle{Ker(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z})) = \frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))}}.$
\end{lemma}
\begin{proof} It is easy to see, from lemma~\ref{incl}, that
\[Ker(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z})) \supset \frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))}.\]
We will prove, using lemma~\ref{[PS]}, that given any element
$$\alpha \in ker(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z})),$$
and any preimage $\beta_{1} \in H^{0}(X_{an},R^{1}\pi^{an}_{*}{\mathbb Z}),$
some non-zero (integral) multiple of $\beta_{1}$ lies in
$H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}}) \subset H^{0}(X_{an},R^{1}\pi^{an}_{*}{\mathbb Z}).$
This will prove the assertion of the lemma.
Since $\alpha \in H^{1}(X,{\mathcal H}^{1}_{X}),$ there exists a finite
Zariski open cover $\{U_{i}\}$ of $X$ such that $\alpha \mapsto 0$ in
$H^{2}((U_{i})_{an},{\mathbb Z})$ for all $i.$ Let $U$ denote any one of these
$U_{i}$'s and consider again the above commutative diagram with exact
rows and columns.
$$\hspace{-1cm}\diagram & & & H^{1}(Y_{an},{\mathbb Z}) \dto \\
& & H^{1}(U_{an},{\mathbb Z}) \rto \dto & H^{1}(\pi^{-1}(U_{an},{\mathbb Z}) \dto \\
& & H^{2}(X_{an},U_{an},{\mathbb Z}) \rto \dto &
H^{2}(Y_{an},\pi^{-1}(U_{an}),{\mathbb Z}) \dto \\
H^{1}(Y_{an},{\mathbb Z}) \rto \dto &
H^{0}(X_{an},R^{1}\pi^{an}_{*}{\mathbb Z}) \rto \dto & H^{2}(X_{an},{\mathbb Z}) \rto \dto
& H^{2}(Y_{an},{\mathbb Z}) \dto \\
H^{1}(\pi^{-1}(U_{an}),{\mathbb Z}) \rto &
\Gamma(U_{an},R^{1}\pi_{*}^{an}{\mathbb Z}) \rto & H^{2}(U_{an},{\mathbb Z}) \rto &
H^{2}(\pi^{-1}(U_{an},{\mathbb Z}) \enddiagram$$
We wish to apply lemma~\ref{[PS]} to this diagram; for this, we need to
know that this diagram arises by applying the cohomology functor to a
suitable $9$-diagram in the category of complexes of abelian groups. It is
clear that the above diagram arises by applying the cohomology functor to
the following $9$-diagram, where all the objects are in the (bounded
below) derived category of sheaves of abelian groups on $X$, and the rows
and columns are exact triangles; here $K_i$ are suitable cones.
$$\diagram K_{1} \rto \dto & K_{2} \rto \dto & K_{3} \dto \rto & \\
{\mathbb Z}_{X} \rto \dto & R\pi_{*}{\mathbb Z}_{Y} \rto \dto & C_{1} \rto \dto & \\
Rj_{*}{\mathbb Z}_{U} \rto \dto & Rj_{*}R\pi_{*}{\mathbb Z}_{\pi^{-1}(U)} \rto \dto &
Rj_{*}C_{2} \rto \dto & \\
& & & \enddiagram$$
Applying the functor $R\Gamma(X,-)$ yields a $9$-diagram in the derived
category of abelian groups. Using Cartan-Eilenberg resolutions, this
$9$-diagram in the derived category is seen to be the image of a
$9$-diagram where all the objects are complexes of abelian groups and the
rows and columns are short exact sequences of complexes. Since the
arguments are standard, we omit the details.
Returning to our cohomology diagram, note that the relative cohomology
sequences
$$\to H^{1}(U_{an},{\mathbb Z}) \to H^{2}(X_{an},U_{an},{\mathbb Z}) \to H^{2}(X_{an},{\mathbb Z})
\to H^{2}(U_{an},{\mathbb Z}) \to$$
and
$$\hspace{-1cm}\to H^{1}((\pi^{an})^{-1}(U_{an}),{\mathbb Z}) \to
H^{2}(Y_{an},(\pi^{an})^{-1}(U_{an}),{\mathbb Z}) \to
H^{2}(Y_{an},{\mathbb Z}) \to H^{2}((\pi^{an})^{-1}(U_{an},{\mathbb Z})$$
are sequences in the category of mixed Hodge structures by
[D]~(8.3.9).
Since $\alpha \to 0 \in H^{2}(Y_{an},{\mathbb Z})$ therefore $\alpha_{{\mathbb Q}} \in
W_{1}H^{2}(X_{an},{\mathbb Q})$ by [D],~Proposition~8.2.5. This implies, by
[D],~Theorem~2.3.5 ({\it i.e.\/},\ strictness of morphisms of mixed Hodge structures
with respect to $W$) that we can choose $\beta_{2} \in H^{2}(X_{an},
U_{an},{\mathbb Z})$ such that
\[(\beta_{2})_{{\mathbb Q}} \in W_{1}H^{2}(X_{an},U_{an},{\mathbb Q}),\;\;\beta_{2} \mapsto
n \alpha,\;\;n\in{\mathbb Z}_{>0}.\]
Let
\[\beta_2\mapsto\gamma_{2}\in
H^{2}(Y_{an},(\pi^{an})^{-1}(U_{an}),{\mathbb Z}) \cong {\mathbb Z}(-1)^k,\]
for some $k\geq 0$, where the last isomorphism is because $Y$ is
non-singular; then
\[(\gamma_{2})_{{\mathbb Q}} \in W_{1}H^{2}(Y_{an},(\pi^{an})^{-1}(U_{an}),{\mathbb Q})=0,\]
{\it i.e.\/},\ $\gamma_2=0$. So we can choose a preimage $\delta_{2} \in
H^{1}((\pi^{an})^{-1}(U_{an}),{\mathbb Z})$ of $\gamma_{2}$ to be zero.
On the other hand, chasing the diagram the other way, we get $n\beta_{1}
\in H^{0}(X_{an},R^{1}\pi^{an}_{*}{\mathbb Z})$ which lifts $n
\alpha$, and $n\beta\mapsto n\gamma_{1} \in
H^{0}(U_{an},R^{1}\pi^{an}_{*}{\mathbb Z})$; now take a lift $n\delta_{1} \in
H^{1}((\pi^{an})^{-1}(U_{an},{\mathbb Z}).$ of $n\gamma_1$.
By lemma~\ref{[PS]}, we know that $n\delta_{1} \equiv \delta_{2}=0$
modulo the images of $H^{1}(Y_{an},{\mathbb Z})$ and $H^{1}(U_{an},{\mathbb Z}).$ Therefore
\[n\gamma_{1} \in Im(H^{1}(Y_{an},{\mathbb Z})\to
H^{0}(U_{an},R^{1}\pi^{an}_{*}{\mathbb Z})).\]
This proves that
$n\beta_{1}|_{U_{an}}$ comes from $H^{1}(Y_{an},{\mathbb Z}).$
Since $X$ has a finite cover by such open sets $U,$ we see that $\beta_{1}
\in H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}.$
\end{proof}
\begin{cor}\label{corexseq}
There exists a short exact sequence of mixed Hodge structures
$$0 \to
\frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))} \to
H^{1}(X,{\mathcal H}^{1}_{X}) \to
Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z})) \to 0.$$
\end{cor}
Let $Im(H^{1}(Y_{an},{\mathbb Z}))^{s}$ denote the saturation of $H^{1}(Y_{an},{\mathbb Z})$
in $H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}.$ Then
\[\frac{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}{Im(H^{1}(Y_{an},{\mathbb Z})} =
\left(\frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}}
{Im(H^{1}(Y_{an},{\mathbb Z})}\right)_{\rm torsion}.\] Since
$NS(X) \subset F^{1}H^{2}(X_{an},{\mathbb Z})$ it follows that it has finite
intersection with $\displaystyle{\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))}}$ which is a pure Hodge structure of
weight one. On the other hand, $H^{2}(X_{an},{\mathbb Z})_{\rm torsion} \subset
NS(X)$ from the exponential sequence. Thus,
$$\left(
\frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z})}
\right)_{\rm torsion} = NS(X) \cap
\frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z})}.$$
Hence we get the exact sequence
$$0 \to
\displaystyle{\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}} \to
\displaystyle{\frac{H^{1}(X,{\mathcal H}^{1}_{X})}{NS(X)}}
\stackrel{f}{\to} \displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}} \to 0,\hspace{2mm}(+)$$
It is clear that the third term is pure of type $(1,1)$ as it lies
inside $\displaystyle{\frac{NS(Y)}{Im(NS(X))}}.$
Let $A = \displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}}$ and let
$$M = f^{-1}(A_{\rm torsion}).$$
We then have a short exact sequence of mixed Hodge structures
$$0 \to M \to \frac{H^{1}(X,{\mathcal H}^{1}_{X})}{NS(X)} \to
\frac{A}{A_{\rm torsion}} \cong {\mathbb Z}(-1)^{r} \to 0,\hspace{2mm}(++)$$
where the third term is free of rank $r$ and pure of type $(1,1),$ and $M$
is a pure Hodge structure of weight $1.$ Further, all of the underlying
abelian groups are free.
We recall some facts about extensions of mixed Hodge structures (see [C],
for example). Let $H$ be a finitely generated abelian
group which supports a pure Hodge structure of weight one, and $G$ a
finitely generated abelian group, regarded as a pure Hodge structure of
type $(0,0)$. Then there is a natural identification of the abelian group
${\rm Ext}\,^{1}_{\bf MHS}(G(-1),H)$ of extensions in the category ${\bf MHS}$ of
mixed Hodge structures with the abelian group ${\rm Hom}(G,J(H))$, where
$$J(H) = J^{1}(H) = \frac{H_{{\mathbb C}}}{F^{1}H_{{\mathbb C}}{}+{}Im(H)};$$
here $H_{{\mathbb C}} = H \otimes_{{\mathbb Z}}{\mathbb C}$ and $F$ gives the Hodge filtration.
In particular we have
\[{\rm Ext}\,^1_{\bf MHS}({\mathbb Z}(-1),H)=J(H).\]
If
\[0\to H\to E\to G(-1)\to 0\]
is an extension of mixed Hodge structures, let $\psi_E:G\to J(H)$ be the
corresponding homomorphism (which we call the {\em extension class map} of
$E$). This may be described as follows: there is an identification
\[\alpha:\frac{H_{{\mathbb C}}}{F^1H_{{\mathbb C}}}\by{\cong}\frac{E_{{\mathbb C}}}{F^1E_{{\mathbb C}}},\]
giving
\[\beta:J(H)=\frac{H_{{\mathbb C}}}{F^1H_{{\mathbb C}}+H}\by{\cong}\frac{E_{{\mathbb C}}}{F^1E_{{\mathbb C}}+H},\]
and $\psi_E$ is the composition
\[G\cong \frac{E}{H}\to \frac{E_{{\mathbb C}}}{F^1E_{{\mathbb C}}+H}
\by{\beta^{-1}}J(H).\]
In case $G={\mathbb Z}^{\oplus r}$ is free abelian, we have that ${\rm Hom}_{\bf
MHS}({\mathbb Z}(-1),G)$ is naturally identified with $\ker \psi_E$. Also, if $G$
is free abelian and $H$ is polarizable, then $J(H)$ is an abelian
variety, and for an extension $E$, the homomorphism $\psi_E:G\to J(H)$ is
the 1-motive over ${\mathbb C}$ associated to the mixed Hodge structure $E$ by the
procedure in [D], (10.1.3).
In particular, the sequence of mixed Hodge structures $(+)$ is an
extension of a pure Hodge structure of type $(1,1)$ by a pure weight one
Hodge structure and hence gives rise to an extension class homomorphism,
$$\psi: \frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{NS(X)}
\to J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{(Im(H^{1}(Y_{an},{\mathbb Z})))^{s}}\right).$$
Similarly, the sequence of mixed Hodge structures $(++)$ gives rise to a
related homomorphism
$$\psi_{1}: {\mathbb Z}^{\oplus r} \cong \frac{A}{A_{tors}} \to J(M),$$
which is in fact a 1-motive. Note that $J(M)$ is isogenous to
$\displaystyle{J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{(Im(H^{1}(Y_{an},{\mathbb Z})))^{s}}\right)}$ which in turn is
isogenous to $\displaystyle{J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))}\right)}.$
By the above remarks, our main result Theorem~\ref{(1,1)} is
equivalent to proving {\em $\psi_{1}$ is injective}.
\section{Proof of the Main Theorem}
\subsection{Construction of a 1-motive}
The aim of this section is to directly construct a certain 1-motive over
${\mathbb C}$. The proof of the Main Theorem will be by showing that it is
isogenous to that associated to $(H^1(X,\sH^1_X)/NS(X))\otimes{\mathbb Z}(1)$.
Let $\pi : Y \to X$ be a desingularization of $X$ as before and let $U
\subset X$ be a Zariski open subset. We have an exact sequence of groups
$$Pic^{0}(Y) \to Pic(\pi^{-1}(U)) \to H^{1}(\pi^{-1}(U),{\mathcal H}^{1}_{Y})
\to 0.$$
We sheafify this on $X=X_{Zar}$ to get an exact sequence of
sheaves
$$Pic^{0}(Y)_{X} \to R^{1}\pi_{*}{\mathcal O}^{*}_{Y} \to R^{1}\pi_{*}{\mathcal
H}^{1}_{Y} \to 0$$
where $Pic^{0}(Y)_{X}$ is the constant sheaf on $X$ associated to the
group $Pic^{0}(Y).$ Define ${\mathcal F}$ to be the Zariski sheaf
$${\mathcal F}:=Im(Pic^{0}(Y)_{X} \to R^{1}\pi_{*}{\mathcal O}^{*}_{Y})$$ on $X.$
Hence we have short exact sequence
of sheaves
\begin{equation}\label{eq}
0 \to {\mathcal F} \to R^{1}\pi_{*}{\mathcal O}^{*}_{Y} \to
R^{1}\pi_{*}{\mathcal H}^{1}_{Y} \to 0
\end{equation}
\begin{lemma} \label{compare}(1) There is an injective map
\[\mu:J(H^0(X_{an},\sF_{{\mathbb Z}})^s)\to H^0(X,\sF),\]
whose image $H^{0}(X,{\mathcal F})^{0}$ is a subgroup of finite index, such
that the natural map $Pic^{0}(Y) \to H^{0}(X,{\mathcal F})$ factors
through $\mu$. The induced map $Pic^{0}(Y) \to H^{0}(X,{\mathcal F})^{0}$ is
that determined by the map on Hodge structures $H^1(Y_{an},{\mathbb Z})\to
H^0(X_{an},\sF_{{\mathbb Z}})^s$. Thus $H^{0}(X,{\mathcal F})^{0}$ is the group of
${\mathbb C}$-points of an abelian variety, such that $Pic^{0}(Y) \to
H^{0}(X,{\mathcal F})^{0}$ is a homomorphism of abelian varieties.
(2)\quad $\displaystyle{J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z})}\right)}$ is isogenous to
$\displaystyle{\frac{H^{0}(X,{\mathcal F})^{0}}{Im(Pic^{0}(Y))}},$ and hence
the latter is isogenous to $J(M).$
\end{lemma}
\begin{proof}
Let $x \in X$ be any point and $F_{x} = \pi^{-1}(x)_{red}.$ Let $Y_x =
Spec({\mathcal O}_{X,x}) \times_{X} Y$ and $\displaystyle{F_{x}^{n} =
Spec\left(\frac{{\mathcal O}_{X,x}}{{\mathcal M}_{x}^{n}}\right) \times_{X}Y}$,
where ${\mathcal M}\subset {\mathcal O}_{X,x}$ is the maximal ideal.
For each $n$ we have the restriction maps $h_{n}: Pic(F^{n}_{x}) \to
Pic(F_{x}).$
We claim that the kernel of $h_{n},$ for each $n,$ is a ${\mathbb C}$-vector
space. To see this consider the short exact sequence of Zariski sheaves
$$\diagram 0 \rto & {\mathcal I}_{F^{n}_{x}/F_{x}} \rto^{exp} & {\mathcal
O}^{*}_{F^{n}_{x}} \rto^{h_n} & {\mathcal O}^{*}_{F_{x}} \rto & 0
\enddiagram$$
where $exp$ denotes the exponential map, which makes sense as ${\mathcal
I}_{F^{n}_{x}/F_{x}}$ is nilpotent.
Considering the associated cohomology sequence we get
$$0 \to H^{1}(F_{x}, {\mathcal I}_{F^{n}_{x}|F_{x}}) \to
Pic(F^{n}_{x}) \longby{h_n} Pic(F_{x})$$
which proves the kernel of $h_{n}$, for each $n$, is a ${\mathbb C}$-vector space.
For each $n$ and for each $x \in X,$ we have a commutative diagram
$$\diagram Pic^{0}(Y) \rrtou ^{f_{n}} \rto \drto^{g} & Pic(Y_{x}) \dto
\rto & Pic(F^{n}_{x}) \dlto_{h_{n}} \\
& Pic(F_{x}) \enddiagram$$
Let $f_{n}$ be the composition $Pic^{0}(Y) \to Pic(Y_{x}) \to
Pic(F^{n}_{x}).$ Then, $Ker(f_{n})$ and $Ker(g)$ are both closed
subgroups of $Pic^{0}(Y)$ hence are compact (topological) groups.
Since $Ker(h_{n})$ is a ${\mathbb C}$-vector space it follows that
$Ker(f_{n}) = Ker(g),$ as any continuous homomorphism from a
compact group to a ${\mathbb C}$-vector space is zero (note that $Pic^{0}(Y),$
$Pic^{0}(F_{x}^{n}),$ and $Pic^{0}(F_{x})$ are isomorphic to the
corresponding analytic groups, by GAGA, and hence from the exponential
sequence carry natural topologies, such that the restriction homomorphisms
are continuous).
Passing to the inverse limit we have a commutative diagram
$$\diagram Pic^{0}(Y) \rto^{f} \drto^{g} & Pic(Y_{x}) \dto \rto|<<\ahook &
Pic(\hat{Y_{x}}) \dlto_{h} \\
& Pic(F_{x}) \enddiagram$$
where $\hat{Y_{x}}$ stands for the completion of $Y_{x}$ along $F_x$.
By Grothendieck's Formal Function Theorem [H, Ch.III, Th.11.1] and the
fact that $Pic(\hat{Y_{x}}) \to \liminv{n}(Pic(F^{n}_{x}))$ is an
isomorphism [H, Ch.II, Ex.9.6], we have that $Pic(Y_{x}) \to
Pic(\hat{Y_{x}})$ is an injection. Thus it follows that $Ker(f) =
Ker(g).$ We have from the definition of ${\mathcal F}$ that the stalk of
${\mathcal F}$ at $x$, ${\mathcal F}_{x} = Im(Pic^{0}(Y) \to Pic(Y_{x})).$ By
our analysis so far we have proved that the natural map ${\mathcal F}_{x} \to
Pic(F_{x})$ is an inclusion, and it clearly factors through the the
subgroup $Pic^{0}(F_{x}).$
By the results of Du Bois [DB] there exists a commutative triangle
$$\diagram H^{1}(F_{x},{\mathbb C}) \rto \drto & H^{1}(F_{x},{\mathcal O}_{F_{x}})
\dto^{\alpha} \\
& \displaystyle{\frac{H^{1}(F_{x},{\mathbb C})}{F^{1}H^{1}(F_{x},{\mathbb C})}}
\enddiagram$$
Note that $Ker(\alpha)$ is a ${\mathbb C}$-vector space.
Now $\alpha$ induces a map
$$\beta : Pic^{0}(F_{x}) \cong \frac{H^{1}(F_{x},{\mathcal
O})}{H^{1}(F_{x},{\mathbb Z})} \to J(H^{1}(F_{x},{\mathbb Z})).$$
Thus we have a diagram
$$\diagram H^{1}(F_{x},{\mathcal O}_{F_{x}}) \rto^{\alpha} \dto &
\displaystyle{\frac{H^{1}(F_{x},{\mathbb C})}{F^{1}H^{1}(F_{x},{\mathbb C})}} \dto \\
Pic^{0}(F_{x}) \rto^{\beta} & J(H^{1}(F_{x},{\mathbb Z})) \enddiagram$$
Since $H^{1}(F_{x},{\mathbb Z})$ is a mixed Hodge structure
with weights $0$ and $1$ (by [D2], as $F_{x}$ is a projective variety),
$H^{1}(F_{x},{\mathbb Z})$ injects into
$\displaystyle{\frac{H^{1}(F_{x},{\mathbb C})}{F^{1}H^{1}(F_{x},{\mathbb C})}}$.
Thus it is clear from the above diagram $Ker(\beta) = Ker(\alpha)$ and
so $Ker(\beta)$ is also a ${\mathbb C}$-vector
space. Hence the composite ${\mathcal F}_{x} \to Pic^{0}(F_{x}) \to
J(H^{1}(F_{x},{\mathbb Z}))$ is injective, as ${\mathcal F}_{x}$ is a compact group,
from its definition.
Let ${\mathcal F}_{{\mathbb Z},x} = Im(H^{1}(Y,{\mathbb Z}) \to H^{1}(F_{x},{\mathbb Z}))$ be the stalk
of ${\mathcal F}_{{\mathbb Z}}$ at $x.$ Let ${\mathcal F}_{{\mathbb Z},x}^{s}$ be the saturation
of ${\mathcal F}_{{\mathbb Z},x}$ in $H^{1}(F_{x},{\mathbb Z}).$ The inclusion ${\mathcal
F}_{{\mathbb Z},x} \to H^{1}(F_{x},{\mathbb Z})$ induces a natural map with finite kernel
$J({\mathcal F}_{{\mathbb Z},x}) \to J(H^{1}(F_{x},{\mathbb Z})).$ In fact this map factors
as
$$\diagram J({\mathcal F}_{{\mathbb Z},x}) \rto|>>\tip & J({\mathcal F}^{s}_{{\mathbb Z},x})
\rto & J(H^{1}(F_{x},{\mathbb Z})). \enddiagram$$
The second map is an inclusion and $J({\mathcal F}^{s}_{{\mathbb Z},x})$ is the
image of $J({\mathcal F}_{{\mathbb Z},x})$ in $J(H^{1}(F_{x},{\mathbb Z})).$
We thus have a commutative diagram with surjective and injective maps
as follows (where we identify $Pic^{0}(Y)$ with $J(H^{1}(Y_{an},{\mathbb Z})).$
$$\diagram Pic^{0}(Y) \rto|>>\tip \dto|>>\tip & J({\mathcal F}_{{\mathbb Z},x})
\dto|>>\tip \rto & J({\mathcal F}_{{\mathbb Z},x}^{s}) \dlto|<<\ahook \\
{\mathcal F}_{x} \rto|<<\ahook & J(H^{1}(F_{x},{\mathbb Z})) \enddiagram$$
Since it is clear from the diagram that $J({\mathcal F}^{s}_{{\mathbb Z},x})$ and
${\mathcal F}_{x}$ are both the image of $Pic^{0}(Y)$ in
$J(H^{1}(F_{x},{\mathbb Z}))$ it follows that $J({\mathcal F}^{s}_{{\mathbb Z},x}) \cong {\mathcal
F}_{x}.$ Therefore there exists a map $J({\mathcal F}_{{\mathbb Z},x}) \to {\mathcal
F}_{x}$ which is an isogeny.
We had proved that the sheaf ${\mathcal F}_{{\mathbb Z}}$ was constructible, i.e.,
constant with groups $G_{i}$ over locally closed sets $(U_{i})_{an},$ and
this data gives rise to a flasque resolution of $a_*\sF_{{\mathbb Z}}$ in the
Zariski site (by lemma~\ref{const1} and lemma~\ref{resol}). It is then
clear that analogous results hold also for the sheaf ${\mathcal F}_{{\mathbb Z}}^{s}$
where ${\mathcal F}_{{\mathbb Z}}^{s}$ denotes the saturation of the sheaf ${\mathcal
F}_{{\mathbb Z}}$ in $R^{1}\pi^{an}_{*}{\mathbb Z}$ (which is a torsion-free sheaf). Thus
the abelian varieties $J({\mathcal F}^{s}_{{\mathbb Z},x})$ are constant quotients of
$Pic^{0}(Y)$ over the strata $U_{i},$ hence so are ${\mathcal F}_{x}.$ This
proves that the sheaf ${\mathcal F}$ is a constructible sheaf on $X$ for the
Zariski topology, with admissible family $\{X_{i}\},$ and further (by
lemma~\ref{resol}) ${\mathcal F}$ has a flasque resolution similar to
$a_*{\mathcal F}_{{\mathbb Z}}.$
Taking global sections of the flasque resolution of $a_*{\mathcal
F}^{s}_{{\mathbb Z}}$, we get an exact sequence
$$0 \to H^{0}(X_{an},{\mathcal F}^{s}_{{\mathbb Z}}) \to \bigoplus_{i}{\mathcal
F}^{s}_{{\mathbb Z},x_{i}} \to \bigoplus_{i<j}{\mathcal F}^{s}_{{\mathbb Z},x_{j}}.$$
Also it is clear from the definitions that
$H^{0}(X_{an},{\mathcal F}^{s}_{{\mathbb Z}})=H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^s$.
Applying $J$ on all the terms, we obtain a complex
$$0 \to J(H^{0}(X_{an},{\mathcal F}^{s}_{{\mathbb Z}})) \to \oplus_{i}J({\mathcal
F}^{s}_{{\mathbb Z},x_{i}}) \to \oplus_{i<j}J({\mathcal F}^{s}_{{\mathbb Z},x_{j}})$$
This complex is exact on the left and has
finite homology in the middle, since $H^{0}(\oplus_{i<j}{\mathcal
F}^{s}_{{\mathbb Z},x_{j}})$ is torsion-free.
Similarly taking global sections of the flasque resolution of ${\mathcal
F}$ we get an exact sequence
$$0 \to H^{0}(X,{\mathcal F}) \to \oplus_{i}{\mathcal F}_{x_{i}} \to
\oplus_{i<j}{\mathcal F}_{x_{j}}.$$
There exists a commutative diagram
$$\diagram & J(H^{0}(X,{\mathcal F}_{{\mathbb Z}})^{s}) \ddotted_{\mu} \rto &
\oplus_{i}J({\mathcal
F}_{{\mathbb Z},x_{i}}^{s}) \rto \ddouble & \oplus_{i<j}J({\mathcal
F}_{{\mathbb Z},x_{j}}^{s}) \ddouble \\
0 \rto & H^{0}(X,{\mathcal F}) \rto & \oplus_{i}{\mathcal F}_{x_{i}} \rto &
\oplus_{i<j}{\mathcal F}_{x_j} \enddiagram$$
where the two vertical arrows are isomorphisms. Hence the dotted arrow
$\mu$ exists, and is an inclusion with finite cokernel.
Define
$$H^{0}(X,{\mathcal F})^{0} = Im(J(H^{0}(X,{\mathcal F}_{{\mathbb Z}})^{s}).$$
Clearly this is an abelian variety, and there is an isogeny
$J(H^{0}(X,{\mathcal F}_{{\mathbb Z}})) \to H^{0}(X,{\mathcal F})^{0}.$ Also, by
construction, the natural
map ${\rm Pic}^0(Y)\to H^0(X,\sF)$ clearly factors through the map
\[{\rm Pic}^0(Y)=J(H^1(Y_{an},{\mathbb Z}))\to J(H^{0}(X,{\mathcal F}_{{\mathbb Z}})).\]
Thus we have
an isogeny $\displaystyle{\frac{J(H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}}))}{Im(Pic^{0}(Y))} \to \frac{H^{0}(X, {\mathcal
F})^{0}}{Im(Pic^{0}(Y))}}.$
We finally note that there exists an isogeny
$$\frac{J(H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}}))}{Im(Pic^{0}(Y)} \to
J\left(\frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})}{Im(H^{1}(Y_{an},{\mathbb Z})}\right)$$
since $J(H^{1}(Y_{an},{\mathbb Z})) \cong Pic^{0}(Y).$
This finishes the proof that $\displaystyle{\frac{H^{0}(X,{\mathcal
F})^{0}}{Im(Pic^{0}(Y))}}$ and $\displaystyle{J\left
(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})}{Im(H^{1}(Y_{an},{\mathbb Z})}\right)}$ are isogenous.
\end{proof}
We can now construct a 1-motive, as follows. Since $X$ is normal, we have
that $\pi_{*}{\mathcal O}_{Y}= {\mathcal O}_{X}$, and so we have an exact sequence
$$0 \to Pic(X) \to Pic(Y) \to H^{0}(X,R^{1}\pi_{*}{\mathcal O}^{*}_{Y}).$$
This induces another exact sequence
\[NS(X) \to NS(Y) \to \frac{H^{0}(X,R^{1}\pi_{*}{\mathcal
O}^{*}_{Y})}{Im(Pic^{0}(Y))}.\]
We thus have an {\em injective} map
\begin{equation}\label{eqq}
\frac{NS(Y)}{Im(NS(X))} \to \frac{H^{0}(X,R^{1}\pi_{*}{\mathcal
O}^{*}_{Y})}{Im(Pic^{0}(Y))}\end{equation}
\begin{lemma}\label{lemphi}
The map (\ref{eqq}) induces an (injective) map
\[\phi:\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}
\to \frac{\Gamma(X,{\mathcal F})}{Im(Pic^{0}(Y))}.\]
\end{lemma}
\begin{proof}
Using the short exact sequence of sheaves (\ref{eq}), we get the following
commutative diagram, whose right column is exact,
\[\diagram \displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}} \dto|<<\ahook \rdotted^>>>{\phi} &
\displaystyle{\frac{\Gamma(X,{\mathcal F})}{Im(Pic^{0}(Y))}} \dto \\
\displaystyle{\frac{NS(Y)}{Im(NS(X))}} \rto|<<\ahook &
\displaystyle{\frac{\Gamma(X,R^{1}\pi_{*}{\mathcal O}^{*})}{Im(Pic^{0}(Y))}}
\dto \\ & \Gamma(X,R^{1}\pi_{*}{\mathcal H}^{1}_{Y}) \enddiagram\]
Here, we claim the dotted arrow $\phi$ exists (and is also injective)
because the composition
\[\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z}))}{NS(X)} \to
H^{0}(X,R^{1}\pi_{*}{\mathcal H}^{1}_{Y})\]
is zero. This is obvious as this map can be described in the following
way: given the image in $H^1(Y,\sH^1_Y)=NS(Y)$ of a Zariski locally
trivial cohomology class $\eta\in H^1(X,\sH^1_X)$, consider a line bundle
$L_{\eta}$ on $Y$ which represents it, then consider the line bundle
restricted to open sets $\pi^{-1}(U) \subset Y,$
$L_{\eta}|_{\pi^{-1}(U)},$ (where $U \subset X$ open) and take the Chern
classes of these restrictions. These give a global section of
$R^{1}\pi_{*}{\mathcal H}^{1}_{Y}$ which is zero as the line bundle came from
a locally trivial cohomology class on $X.$
\end{proof}
Since $\displaystyle{\frac{\Gamma(X,{\mathcal F})}{Im(Pic^{0}(Y))}}$ has a
subgroup of finite index which is an abelian variety, $\phi$ determines a
1-motive in an obvious way,
\[B\to \frac{\Gamma(X,{\mathcal F})^0}{Im(Pic^{0}(Y))},\]
where $B$ is the inverse image under $\phi$ of the abelian variety.
\begin{rmk} We do not know if $\Gamma(X,\sF)$ is itself an abelian
variety, {\it i.e.\/},\ if $\Gamma(X,\sF)^0=\Gamma(X,\sF)$.
\end{rmk}
\subsection{Comparison of the two 1-motives}
We now finish the proof of the theorem, by comparing the 1-motive
constructed above using $\phi$ with that constructed earlier, using the
extension class map $\psi$ for the mixed Hodge structure on
$H^1(X,\sH^1_X)$.
Recall that $\{X_i\}_{i\in I}$ is the chosen admissible family of subsets
for $R^1\pi^{an}_*{\mathbb Z}$, and hence for $\sF_{{\mathbb Z}}$ and $\sF$ as well; recall
also the corresponding (irreducible, non-singular) locally closed strata
$\{U_i\}_{i\in I}$. Also recall the choice of points $x_i\in U_i$, and
$F_i=\pi^{-1}(x_i)$, $F=\cup_{i\in I}F_i$.
Suitably blow up $Y$ to get $f:\tilde{Y} \to Y$, with $\tilde{Y}$
non-singular projective, so that the reduced strict transform of $F$ is a
{\em smooth} possibly disconnected subvariety $\tilde{F}.$ Note that there
exists the following diagram.
$$\diagram \tilde{F} \rto|<<\ahook \dto & \tilde{Y} \dto^{f} \\
F \rto|<<\ahook & Y \enddiagram$$
Now consider the following commutative diagram which is a diagram in the
category of mixed Hodge structures
$$\hspace{-2cm}\diagram 0 \rto &
\displaystyle{\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))}}\rto \dto & H^{1}(X,{\mathcal
H}^{1}_{X}) \rto \dto & Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z})) \rto \dto & 0 \\ 0 \rto &
\displaystyle{\frac{H^{1}(F_{an},{\mathbb Z})}{Im(H^{1}(Y_{an},{\mathbb Z}))}} \rto \dto &
H^{2}(Y_{an},F_{an},{\mathbb Z}) \rto \dto & Im(H^{2}(Y_{an},F_{an},{\mathbb Z}) \to
H^{2}(Y_{an},{\mathbb Z})) \dto \rto & 0 \\
0 \rto &
\displaystyle{\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}}
,{\mathbb Z}))}} \rto & H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z}) \rto &
Im(H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z}) \to H^{2}(\tilde{Y_{an}},{\mathbb Z}))
\rto & 0 \enddiagram$$
Let $Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^{s}$ be the saturation of
$Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))$ in $H^{1}(\tilde{F_{an}},{\mathbb Z}).$ Then,
arguing as before,
$$Im(NS(X)) \cap
\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))} =
\frac{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^{s}}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}.$$
So we have a short exact sequence of mixed Hodge structures
$$\diagram 0 \rto &
\displaystyle{\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}}
,{\mathbb Z}))^{s}}} \rto &
\displaystyle{\frac{H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z})}{Im(NS(X))}}
\rto &
\displaystyle{\frac{Im(H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z})
\to H^{2}(\tilde{Y_{an}},{\mathbb Z}))}{Im(NS(X))}} \rto & 0 \enddiagram$$
and a commutative diagram
$$\diagram 0 \rto &
\displaystyle{\frac{H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}}
{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}} \rto \dto &
\displaystyle{\frac{H^{1}(X,{\mathcal H}_{X}^{1})}{NS(X)}} \rto \dto &
\displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X})
\to H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}} \rto \dto & 0 \\
0 \rto &
\displaystyle{\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}}
,{\mathbb Z}))^{s}}} \rto &
\displaystyle{\frac{H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z})}{Im(NS(X))}}
\rto & \displaystyle{\frac{Im(H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z})
\to H^{2}(\tilde{Y_{an}},{\mathbb Z}))}{Im(NS(X))}} \rto & 0 \enddiagram$$
The following diagram commutes by functoriality of the extension class
maps.
$$\diagram Im(H^{1}(X,{\mathcal H}_{X}^{1}) \to H^{2}(Y_{an},{\mathbb Z}))
\rto^{\;\;\;\psi} \dto &
\displaystyle{J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))}\right)} \dto \\
Im(H^{2}(\tilde{Y_{an}},\tilde{F_{an}},{\mathbb Z}) \to H^{2}(\tilde{Y_{an}},{\mathbb Z}))
\rto & \displaystyle{J\left(\frac{H^{1}
(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^{s}}\right)}
\enddiagram$$
where
$$J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}
\right) \to
J\left(\frac{H^{1}
(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^{s}}\right)$$
is induced from the map on the underlying Hodge structures.
Let $$\psi^{'}:Im(H^{1}(X,{\mathcal H}_{X}^{1}) \to H^{2}(Y_{an},{\mathbb Z})) \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}
\right)$$
be the composite in the diagram above.
We also have a natural \lq\lq sheaf theoretic\rq\rq map
$$\phi^{'}: \frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{NS(X)} \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}
\right)$$
which is defined as follows. Let $\eta \in Im(H^{1}(X,{\mathcal H}^{1}_{X})
\to H^{2}(Y_{an},{\mathbb Z})).$ Consider a line bundle $L$ on $Y$ such that
$c_{1}(L)= \eta.$ Then, $L|_{F}$ (where $F = \cup_{i} F_{i}$) gives an
element of $Pic^{0}(F)$ and hence an element of $J(H^{1}(F,{\mathbb Z}))$ via the
mapping $Pic^{0}(F) \to J(H^{1}(F,{\mathbb Z})).$ Under this mapping $NS(X)$
goes to zero, hence we get a well defined mapping
$$\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z}))}{NS(X)}
\to J\left(\frac{H^{1}(F_{an},{\mathbb Z})}{Im(H^{1}(Y_{an},{\mathbb Z}))}\right).$$
Now compose with the map (induced by the morphism of the underlying
Hodge structures)
$$J\left(\frac{H^{1}(F_{an},{\mathbb Z})}{Im(H^{1}(Y_{an},{\mathbb Z}))}\right) \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},
{\mathbb Z}))}\right)$$
to get
$$\phi^{'}: \frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{NS(X)} \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}
\right).$$
It is easy to see the following diagram commutes
$$\diagram \displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^1_X) \to
H^2(Y_{an},{\mathbb Z}))}{Im(NS(X))}} \rto^{\quad\quad\phi} \drto^{\phi^{'}} &
\displaystyle{\frac{H^0(X,{\mathcal F})}{Im(Pic^0(Y))}} \dto \\
& \displaystyle{J\left(
\frac{H^1(\tilde{F_{an}},{\mathbb Z})}{Im(H^1(\tilde{Y_{an}},{\mathbb Z}))}\right)}
\enddiagram$$
where the map
$\displaystyle{\frac{H^{0}(X,{\mathcal F})}{Im(Pic^{0}(Y))}} \to
\displaystyle{J\left(
\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}\right)}$
is the composition
$$\frac{H^{0}(X,{\mathcal F})}{Im(Pic^{0}(Y))} \to
\frac{Pic^{0}(F)}{Im(Pic^{0}(Y))} \to
\frac{Pic^{0}(\tilde{F})}{Im(Pic^{0}(\tilde{Y}))} \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}}
,{\mathbb Z}))}\right).$$
We now note that the map $H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s} \to
H^{1}(F_{an},{\mathbb Z})$ is an injective map. Since $W_{0}H^{1}(F_{an},{\mathbb Q}) =
Ker(H^{1}(F_{an},{\mathbb Q}) \to H^{1}(\tilde{F_{an}},{\mathbb Q}))$ and
$H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s}$ is pure of weight one, it follows that
$$(Ker(H^{1}(F_{an},{\mathbb Q}) \to H^{1}(\tilde{F_{an}},{\mathbb Q})))
\cap Im(H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}}) \to H^{1}(F_{an},{\mathbb Q})) = 0.$$
Hence the composite $$H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}}) \to
H^{1}(F_{an},{\mathbb Z}) \to
H^{1}(\tilde{F_{an}},{\mathbb Z})$$ has finite kernel, and is hence injective (as
${\mathcal F}_{{\mathbb Z}} \subset R^{1}\pi^{an}_{*}{\mathbb Z}$ is torsion-free).
It follows that
$$J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}\right) \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^{s}}
\right), \hspace{2mm}(+++)$$
has finite kernel.
We have the following diagram which shows all the maps we have
constructed so far (the outer border is not yet known to commute).
$$\diagram
\displaystyle{J\left(\frac{H^0(X_{an},{\mathcal F}_{{\mathbb Z}})^s}
{Im(H^{1}(Y_{an},{\mathbb Z}))^s} \right)} \ddto & &
\displaystyle{
\frac{Im(H^1(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}}
\llto_{\psi} \dto^{\phi}|<<\ahook \\
& \displaystyle{\frac{J(H^{0}(X_{an},{\mathcal F}_{{\mathbb Z}})^{s})}
{Im(Pic^{0}(Y))}} \rto|<<\ahook \dto &
\displaystyle{\frac{H^{0}(X,{\mathcal F})}{Im(Pic^{0}(Y))}}
\dto \\
\displaystyle{J\left(\frac{H^1(\tilde{F_{an}},{\mathbb Z})}
{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^s}\right)} &
\displaystyle{\frac{J(H^1(\tilde{F_{an}},{\mathbb Z}))}{Im(Pic^0(\tilde{Y}))}}
\lto &
\displaystyle{
\frac{Pic^0(\tilde{F})}
{Im(Pic^0(\tilde{Y}))}} \lto \enddiagram$$
We will prove that the following subdiagram commutes
$$\diagram
\displaystyle{J\left(\frac{H^0(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}\right)} \dto &
\displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}} \lto_{\psi\quad\quad} \dto^\phi|<<\ahook \\
\displaystyle{J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}
{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^s}\right)} &
\displaystyle{\frac{H^0(X,{\mathcal F})}{Im(Pic^{0}(Y))}} \lto
\enddiagram(*)$$
Note that the composite map
$$\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z})}{Im(NS(Y))}
\to \frac{H^{0}(X,{\mathcal F})}{Im(Pic^{0}(Y))} \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}\right)
$$ is the previously defined
map $\phi^{'},$ and the map
$$\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z})}{Im(NS(Y))}
\to J\left(\frac{H^{0}(X_{an},{\mathcal
F}_{{\mathbb Z}})^{s}}{Im(H^{1}(Y_{an},{\mathbb Z}))^{s}}\right) \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}
{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))}\right)$$
is our previously defined map $\psi^{'}.$
Thus the commutativity of the above diagram is equivalent to proving
$$\psi^{'} = \phi^{'}.$$
Assuming this diagram commutes we finish the proof of Theorem~\ref{(1,1)}
as follows. We claim that the composite map
$$\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))} \to J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}
{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^{s}}\right)$$
has finite kernel, as $\phi$ is injective and the map
$$\frac{H^{0}(X,{\mathcal F})}{Im(Pic^{0}(Y))} \to
J\left(\frac{H^{1}(\tilde{F_{an}},{\mathbb Z})}
{Im(H^{1}(\tilde{Y_{an}},{\mathbb Z}))^s}\right)$$
has finite kernel (combining lemma~\ref{compare} with the fact that
the map in $(+++)$ above has finite kernel).
Thus, $\psi$ has finite kernel.
Now recall the map $$\psi_{1}:
{\mathbb Z}^{r} \cong \frac{A}{A_{tors}} \to J(M),$$
where $A = \displaystyle{\frac{Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to
H^{2}(Y_{an},{\mathbb Z}))}{Im(NS(X))}}.$ It follows immediately that $\psi_{1}$ has
finite kernel. Since $\displaystyle{\frac{A}{A_{tors}}}$ is a free abelian group, it
follows that $\psi_{1}$ is injective. This is equivalent to proving our
main result, Theorem~\ref{(1,1)}, as has been remarked before.
We now finish the final part of the proof by showing the commutativity
of the diagram$~(*).$ Let $Z$ be a smooth projective variety over ${\mathbb C}$
and $W \subset Z$ be a smooth subvariety. Let $\eta \in H^{2}(Z_{an},{\mathbb Z})$
be an algebraic class (i.e., let $\eta \in NS(Z)$), such that $\eta
\mapsto 0 \in H^{2}(W_{an},{\mathbb Z}).$ Then, $\eta$ gives rise to the following
pullback diagram
$$\hspace{-2cm} \diagram 0 \rto &
\displaystyle{\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}}
\rto & H^{2}(Z_{an},W_{an},{\mathbb Z}) \rto & Ker(H^{2}(Z_{an},{\mathbb Z}) \to
H^{2}(W_{an},{\mathbb Z})) \rto & 0 \\
0 \rto & \displaystyle{\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}} \rto
\udouble & B \rto \uto & {\mathbb Z}[\eta] \rto \uto & 0 \enddiagram$$
Thus we have an extension class map
$$\diagram {\mathbb Z} \cong {\mathbb Z}[\eta] \rto^{\psi^{''}\quad\quad} &
\displaystyle{J\left(\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}\right)}.
\enddiagram$$
Again given $\eta$ as above consider $L_{\eta},$ a line bundle on $Z$
which has Chern class $\eta.$ Restrict this line bundle on $W$ to get
$L_{\eta}|_{W} \in Pic^{0}(W_{an}) \cong J(H^{1}(W_{an},{\mathbb Z})).$ This gives
us a well-defined mapping
$$\diagram {\mathbb Z} \cong {\mathbb Z}[\eta] \rto^{\phi^{''}\quad\quad} &
\displaystyle{J\left(\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}\right)}.
\enddiagram$$
We now have the following lemma.
\begin{lemma} \label{abstract} With the above notation $\psi^{''} =
\phi{''},$ i.e., the extension class map and the restriction map
corresponding to the class
$\eta$ are the same.
\end{lemma}
\begin{proof} Consider the following diagram with exact rows and columns
$$\hspace{-1cm}\diagram
\rto & H^{0}(W_{an},{\mathbb Z}) \rto \dto & H^{1}(Z_{an},W_{an},{\mathbb Z}) \rto \dto &
H^{1}(Z_{an},{\mathbb Z}) \rto \dto & H^{1}(W_{an},{\mathbb Z}) \rto \dto & \\
\rto & \displaystyle{\frac{H^{0}(W_{an},{\mathbb C})}{F^{1}H^{0}(W_{an},{\mathbb C})}}
\rto \dto & \displaystyle{\frac{H^{1}(Z_{an},W_{an},{\mathbb C})}{F^{1}
H^{1}(Z_{an},W_{an},{\mathbb C})}} \rto \dto &
\displaystyle{\frac{H^{1}(Z_{an},{\mathbb C})}{F^{1}H^{1}(Z_{an},{\mathbb C})}} \rto
\dto &
\displaystyle{\frac{H^{1}(W_{an},{\mathbb C})}{F^{1}H^{1}(W_{an},{\mathbb C})}} \rto
\dto & \\
\rto & H^{0}(W_{an},{\mathcal O}_{W_{an}}^{*}) \rto \dto & Pic(Z_{an},W_{an})
\rto \dto & Pic(Z_{an}) \rto \dto & Pic(W_{an}) \rto \dto & \\
\rto & H^{1}(W_{an},{\mathbb Z}) \rto \dto & H^{2}(Z_{an},W_{an},{\mathbb Z}) \rto \dto &
H^{2}(Z_{an},{\mathbb Z}) \rto \dto & H^{2}(W_{an},{\mathbb Z}) \rto \dto & \\
\rto & \displaystyle{\frac{H^{1}(W_{an},{\mathbb C})}{F^{1}H^{1}(W_{an},{\mathbb C})}}
\rto \dto & \displaystyle{\frac{H^{2}(Z_{an},W_{an},{\mathbb C})}{F^{1}
H^{2}(Z_{an},W_{an},{\mathbb C})}} \rto \dto &
\displaystyle{\frac{H^{2}(Z_{an},{\mathbb C})}{F^{1}H^{2}(Z_{an},{\mathbb C})}} \rto
\dto &
\displaystyle{\frac{H^{2}(W_{an},{\mathbb C})}{F^{1}H^{2}(W_{an},{\mathbb C})}} \rto
\dto & \\
& & & & \enddiagram$$
The above diagram comes from the following $9$-diagram in the category
of sheaves (where the bottom row defines ${\mathcal O}_{Z_{an}}(-W_{an})^{*},$
and $j:Z_{an}-W_{an} \hookrightarrow Z_{an}$ is the inclusion).
$$\diagram & 0 \dto & 0 \dto & 0 \dto \\
0 \rto & j_{!}{\mathbb Z}_{Z_{an}-W_{an}} \rto \dto & {\mathbb Z}_{Z_{an}} \rto \dto &
{\mathbb Z}_{W_{an}} \rto \dto & 0 \\
0 \rto & {\mathcal O}_{Z_{an}}(-W_{an}) \rto \dto & {\mathcal O}_{Z_{an}} \rto
\dto & {\mathcal O}_{W_{an}} \rto \dto & 0 \\
0 \rto & {\mathcal O}_{Z_{an}}(-W_{an})^{*} \rto \dto & {\mathcal O}_{Z_{an}}^{*}
\rto \dto & {\mathcal O}_{W_{an}}^{*} \rto \dto & 0 \\
& 0 & 0 & 0 \enddiagram$$
Let $\eta \in H^{2}(Z_{an},{\mathbb Z})$ such that $\eta \mapsto 0$ both in
$H^{2}(W_{an},{\mathbb Z})$ and
$\displaystyle{\frac{H^{2}(Z_{an},{\mathbb C})}{F^{1}H^{2}(Z_{an},{\mathbb C})}}$ (i.e.,
$\eta \in NS(Z)$). By a diagram chase as before we get elements
$\delta_{1}$ and $\delta_{2}$ in the group
$\displaystyle{\frac{H^{1}(W_{an},{\mathbb C})}{F^{1}H^{1}(W_{an},{\mathbb C})}}.$ Both
$\delta_{1}$ and $\delta_{2}$ are well-defined in the quotient group
$\displaystyle{\frac{H^{1}(W_{an},{\mathbb C})}{F^{1}
H^{1}(W_{an},{\mathbb C})}/\left(Im(H^{1}(W_{an},{\mathbb Z})
+ Im\left(\frac{H^{1}(Z_{an},{\mathbb C})}{F^{1}
H^{1}(Z_{an},{\mathbb C})}\right)\right)}.$
Now note that
$$\frac{H^{1}(W_{an},{\mathbb C})}{F^{1}
H^{1}(W_{an},{\mathbb C})}/\left(Im(H^{1}(W_{an},{\mathbb Z}) +
Im\left(\frac{H^{1}(Z_{an},{\mathbb C})}{F^{1}H^{1}(Z_{an},{\mathbb C})}\right)\right)
\cong J\left(\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}\right).$$ Hence
we get 2 maps $\eta \mapsto \bar{\delta_{1}}$ and $\eta \mapsto
\bar{\delta_{2}}$ from
$${\mathbb Z} \cong {\mathbb Z}[\eta] \to
J\left(\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}\right)$$
where $\bar{\delta}$ denotes the class of $\delta$ in the quotient
$\displaystyle{J\left(
\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}\right)}.$
We claim that these two maps are nothing but our previously defined
maps ${\phi}^{''}$ and ${\psi}^{''}$ respectively. It is clear that the
map $\eta \mapsto \bar{\delta_{2}}$ is equal to $\phi^{''}(\eta).$ This is
because we got $\bar{\delta_{2}}$ by first taking a lift of $\eta,$ say
$\beta_{2}$ in $Pic(Z_{an}),$ then restricting to $Pic(W_{an})$ to get
$\gamma_{2}$ and finally taking the class $\bar{\delta_{2}} \in
\displaystyle{J\left(\frac{H^{1}(W_{an},{\mathbb Z})}{H^{1}(Z_{an},{\mathbb Z})}\right)}.$
This is exactly how $\phi^{''}(\eta)$ was defined, so $\bar{\delta_{2}}
= \phi^{''}(\eta).$ It is also clear that $\bar{\delta_{1}} =
\psi^{''}(\eta)$ as the extension class map is defined exactly the same
way as the map $\eta \mapsto \bar{\delta_{1}}.$ Now by
lemma~\ref{[PS]}, we have that $\bar{\delta_{1}} = \bar{\delta_{2}}.$
This implies that $\phi^{''} = \psi^{''}.$
\end{proof}
\begin{rmk} In a similar vein, using lemma~\ref{[PS]}, one can show that
the cycle class map with values in Deligne-Beilinson cohomology restricts
to the Abel-Jacobi mapping, on cycles which are homologically trivial.
This is essentially the argument given in [EV], though the need to appeal
to lemma~\ref{[PS]} is not brought out explicitly there.
\end{rmk}
Let $Z= \tilde{Y}$ and $W= \tilde{F}$ in lemma~\ref{abstract}.
Let $\eta \in Im(H^{1}(X,{\mathcal H}^{1}_{X}) \to H^{2}(Y_{an},{\mathbb Z})),$
and let $\eta \mapsto \tilde{\eta} \in H^{2}(\tilde{Y_{an}},{\mathbb Z}).$
Then, $\tilde{\eta}$ satifies the conditions of lemma~\ref{abstract},
i.e., $\tilde{\eta} \in NS(\tilde{Y})$ and $\tilde{\eta} \mapsto 0 \in
H^{2}(\tilde{F_{an}},{\mathbb Z}).$ Clearly,
$$\psi^{'}(\eta) = \psi^{''}(\tilde{\eta}),$$
and
$$\phi^{'}(\eta) = \phi^{''}(\tilde{\eta}).$$
Hence,
$$\psi^{'} = \phi^{'}$$
which proves the commutativity of diagram$~(*).$
This finishes the proof of our main result, Theorem~\ref{(1,1)}.
\section{An example}
In this section we give an example of an integral projective variety $X$
over ${\mathbb C}$ which is not normal, for which we have a strict inclusion
\[NS(X)\propsubset \{\alpha\in H^2(X,{\mathbb Z})\mid\mbox{ $\alpha$ is Zariski
locally trivial and $\alpha_{{\mathbb C}}\in F^1H^2(X,{\mathbb Z})$}\}.\]
Our variety will have the property that its normalization $Y$ is
non-singular, and the normalization map $\pi:Y\to X$ is bijective. Then
$H^2(X,{\mathbb Z})\cong H^2(Y,{\mathbb Z})$ as mixed Hodge structures, and the subspaces of
Zariski locally trivial classes correspond. Hence the desired property of
$X$ is equivalent to the strictness of the first inclusion
\[NS(X)\propsubset NS(Y)\subset H^2(Y,{\mathbb Z}).\]
We will make use of a variant of a construction in [Ha], III, Ex. 5.9 (see
also [Ha], II, Ex. 5.16b). If $V$ is a non-singular variety over ${\mathbb C}$,
then following [Ha], an {\it infinitesimal extension} of $V$ by an
invertible $\sO_V$-module $\sL$ is a scheme $W$ with $W_{red}=V$, such
that the nilradical $\sI$ of $\sO_W$ has square zero (so that it is an
$\sO_V$-module), and there is an $\sO_V$-isomorphism $\sI\cong \sL$. In
other words, there is an exact sequence of sheaves of $\sO_W$-modules
\[0\to \sL\to \sO_W\to \sO_V\to 0.\]
There is a corresponding exact sequence of sheaves
\[0\to \sL\to \sO_W^*\to\sO_V^*\to 0,\]
where $\sL$ is identified with the (multiplicative) subsheaf of units on
$W$ which restrict to $1$ on $V$ (the identification is given on sections
by $s\mapsto 1+s$). The latter exact sheaf sequence gives rise to an
exact sequence on cohomology
\[ H^1(V,\sL)\to {\rm Pic} W\to {\rm Pic} V \by{\delta} H^2(V,\sL).\]
The following is an elaboration of [Ha], III, Ex. 5.9 (the proof is left
as an exercise!).
\begin{lemma}\label{infinitesimal}
\begin{points}
\item There is a natural bijection between isomorphism classes of
infinitesimal extensions of $V$ by $\sL$ and elements
\[\alpha\in H^1(V,\shom_V(\Omega^1_{V/{\mathbb C}},\sL)).\]
\item If $W$ is the infinitesimal extension corresponding to $\alpha$, the
boundary map $\delta=\delta_{\alpha}:{\rm Pic} V\to H^2(V,\sL)$ is expressible
as the composition
\[{\rm Pic} V=H^1(V,\sO_V^*)\by{\dlog} H^1(V,\Omega^1_{V/{\mathbb C}})\by{\cup \alpha}
H^2(V,\sL).\]
\item Let $f:\sL\to\sM$ be a morphism of $\sO_X$-modules, and
$\alpha\mapsto f_*(\alpha)$ under the natural map
\[f_*:H^1(V,\shom_V(\Omega^1_{V/{\mathbb C}},\sL))\to
H^1(V,\shom_V(\Omega^1_{V/{\mathbb C}},\sM)).\]
Let $Z$ be the infinitesimal extension of $V$ by $\sM$ determined by
$f_*(\alpha)$. Then there is a unique morphism of ${\mathbb C}$-schemes
$\tilde{f}:Z\to W$, such that the corresponding morphism of reduced
schemes is the identity on $V$, and such that there are commutative
diagrams with exact rows
\[\begin{array}{ccc}
0\to \sL\to & \sO_W & \to\sO_V\to 0\\
f\downarrow\quad & \tilde{f}^*\downarrow\quad & \mbox{\large $\parallel$}\quad\\
0\to\sM\to & \sO_Z & \to\sO_V\to 0
\end{array}\]
and
\[\begin{array}{cccr}
H^1(V,\sL) & \to {\rm Pic} W \to & {\rm Pic} V & \longby{\delta_{\alpha}}
H^2(V,\sL)\\
\quad f_*\downarrow & \tilde{f}^*\downarrow\quad & \mbox{\large $\parallel$} & \downarrow f_*\quad\\
H^1(V,\sM) & \to {\rm Pic} Z \to & {\rm Pic} V & \longby{\delta_{f_*(\alpha)}}
H^2(V,\sM)
\end{array}\]
\end{points}
\end{lemma}
\begin{ex} In the above lemma, take $V=\P^1_{{\mathbb C}}\times\P^1_{{\mathbb C}}$,
and $\sL=\omega_{V}$, $\sM=\sO_{V}$, $f:\sL\hookrightarrow\sM$ any non-zero
map (since $\omega_{V}\cong\sO_{\P^1}(-2)\boxtimes\sO_{\P^1}(-2)$, such
maps $f$ exist). Infinitesimal extensions of $V$ by $\sL$ are classified
by elements of
\[H^1(V,\shom_V(\Omega^1_{V/{\mathbb C}},\omega_V))\cong H^1(V,\Omega^1_{V/{\mathbb C}}),\]
where we have identified $\shom_V(\Omega^1_{V/{\mathbb C}},\omega_V)$ with
$\Omega^1_{V/{\mathbb C}}$ using the non-degenerate bilinear form
\[\Omega^1_{V/{\mathbb C}}\otimes_{\sO_V}\Omega^1_{V/{\mathbb C}}\to\omega_V\]
given by the exterior product of 1-forms. Thus if $\alpha\in
H^1(V,\Omega^1_{V/{\mathbb C}})$, the corresponding cup-product map
\[H^1(V,\Omega^1_{V/{\mathbb C}})\longby{\cup\alpha} H^2(V,\omega_V)\]
is just the product with $\alpha$ in the commutative graded ring
\[\oplus_{n\geq 0} H^n(V,\Omega^n_{V/{\mathbb C}}).\]
Thus if $\alpha$ is the cohomology class of a divisor $D$ on $V$, then for
any divisor $E$ on $V$, we see that by (ii) of the lemma,
\[\delta_{\alpha}(E)=(D\cdot E)\in {\mathbb C}= H^2(V,\omega_V)\]
is the intersection product of $D$ and $E$ on the non-singular projective
surface $V$.
We will choose $\alpha$ to be the cohomology class of $D=L_1-L_2$, where
$L_1=\P^1_{{\mathbb C}}\times\{0\}$ and $L_2=\{0\}\times\P^1_{{\mathbb C}}$ are elements
of the two rulings on $V=\P^1_{{\mathbb C}}\times\P^1_{{\mathbb C}}$. Since $D^2=-2$,
$\alpha\neq 0$ and $W=(V,\sO(\alpha))$ is a non-trivial infinitesimal
extension of $V$ by $\omega_V$. Note that $H^1(V,\omega_V)=0$, so that
there is an exact sequence
\[0\to {\rm Pic} W\to {\rm Pic} V \longby{(D\cdot \;)} {\mathbb Z}\to 0\]
($(D\cdot L_2)=1$, so the map to ${\mathbb Z}$ is surjective). Here ${\rm Pic} V={\rm Pic}
(\P^1_{{\mathbb C}}\times\P^1_{{\mathbb C}})={\mathbb Z}[L_1]\oplus {\mathbb Z}[L_2]$ is free abelian of rank
2, and as usual, we denote a representative of the class of
$a[L_1]+b[L_2]$ by $\sO_V(a,b)$.
Next, note that
\[f_*(\alpha)\in H^1(V,\shom_V(\Omega^1_{V/{\mathbb C}},\sO_V))=0,\]
since $\Omega^1_{V/{\mathbb C}}\cong \sO_V(-2,0)\oplus \sO_V(0,-2)$. Hence the
infinitesimal extension $Z$ of $V$ by $\sO_V$ determined by $f_*(\alpha)$
is the trivial extension $(V,\sO_V[\epsilon])$, where $\sO_V[\epsilon]$ is
the sheaf of dual numbers over $\sO_V$.
There is an obvious way in which we may regard $Z=(V,\sO_V[\epsilon])$ as
a closed subscheme of
$Y=\P^1_{{\mathbb C}}\times\P^1_{{\mathbb C}}\times\P^1_{{\mathbb C}}=V\times\P^1_{{\mathbb C}}$, whose
underlying reduced scheme is $V\times\{0\}$.
Finally, we define $X$ to be the ${\mathbb C}$-scheme which is the pushout of $Y$ and
$W$ along the morphisms $\tilde{f}:Z\to W$ and the above inclusion $i:Z\hookrightarrow
Y$, so that there is a commutative pushout diagram
\[ \begin{array}{ccc}
Z & \longby{i} & Y=V\times\P^1_{{\mathbb C}}\\
\tilde{f} \downarrow &&\quad \downarrow \pi\\
W & \longby{j} & X
\end{array}\]
Since $\tilde{f}$ is a finite and bijective morphism, we see that for each
affine open $U={\rm Spec \,} A$ in $Y$, $U\cap Z={\rm Spec \,} A/I$ is affine, and finite
over the affine open subscheme $\tilde{f}(U\cap Z)={\rm Spec \,} B\subset W$. The
image of $U$ in $X$ is then defined as the affine scheme ${\rm Spec \,} C$, where
$C$ is the inverse image of $B$ in $A$ under the surjection $A\mbox{$\,\>>>\hspace{-.5cm}\to\hspace{.15cm}$} A/I$.
One shows easily that $C$ is in fact a finitely generated ${\mathbb C}$-subalgebra of
$A$, and $A$ is a finite $C$-module with conductor ideal $I$. Further, the
construction of $C$ localizes well. Hence the local schemes ${\rm Spec \,} C$ can
be glued together to yield the ${\mathbb C}$-scheme $X$.
We claim that this scheme $X$ has the desired properties, {\it i.e.\/},\ $X$ is an
integral projective scheme over ${\mathbb C}$ with normalization $\pi:Y\to X$, such
that $Y$ is non-singular and bijective with $X$, while $NS(X)\to NS(Y)$
is a strict inclusion.
That $X$ is integral with $Y$ as its normalization is clear, from the
description of its affine open sets above. Next, since $\alpha=[D]$, and
$(D\cdot (L_1+L_2))=0$, there is a unique $\sH\in{\rm Pic} W$ such that
$\sH\otimes\sO_V=\sO_V(1,1)$. Also ${\rm Pic} Z\to{\rm Pic} V$ is an isomorphism.
We have ${\rm Pic} Y={\mathbb Z}^{\oplus 3}$, where we may regard the restriction map
${\rm Pic} Y\to{\rm Pic} V\times\{0\}={\rm Pic} V={\mathbb Z}^{\oplus 2}$ as projection on the
first 2 factors. Hence we see that the very ample invertible sheaf
$\sO_Y(1,1,1)$ has the property that there is an isomorphism
$\tilde{f}^*\sH\cong \sO_Y(1,1,1)\otimes\sO_Z$. From the Mayer-Vietoris
sequence of sheaves of rings
\[0\to \sO_X\to \pi_*\sO_Y\oplus j_*\sO_W \to (i\circ \pi)_*\sO_Z\to 0\]
we have a corresponding sequence of sheaves of unit groups
\[0\to \sO_X^*\to \pi_*\sO_Y^*\oplus j_*\sO_W^* \to (i\circ
\pi)_*\sO_Z^*\to 0\]
leading to an exact sequence
\[H^0(Y,\sO_Y^*)\oplus H^0(W,\sO_W^*)\to H^0(Z,\sO_Z^*)\to {\rm Pic} X \to {\rm Pic}
Y\oplus {\rm Pic} W \to {\rm Pic} Z.\]
Hence there exists an invertible sheaf $\sA$ on $X$ with
$\pi^*\sA=\sO_Y(1,1,1)$ (and $j^*\sA=\sH$). Since $\pi$ is finite, and
$\pi^*\sA$ is ample on $Y$, we have that $\sA$ is ample on $X$. Hence $X$
is projective.
>From the exact sequence
\[0\to\omega_V\to\sO_W\to \sO_V\to 0,\]
we see that $H^0(W,\sO_W)\to H^0(V,\sO_V)={\mathbb C}$ is an isomorphism. On the
other hand, we see at once that $H^0(Z,\sO_Z)={\mathbb C}[\epsilon]$ is the ring of
dual numbers. Hence we get analogous formulas for the unit groups. Thus
there is an exact sequence
\[0\to {\mathbb C}\to {\rm Pic} X\to {\rm Pic} Y\oplus {\rm Pic} W \to {\rm Pic} Z \to 0\]
(note that ${\rm Pic} Z={\rm Pic} V$ is a quotient of ${\rm Pic} Y$).
Since ${\rm Pic} W\hookrightarrow {\rm Pic} Z$ is an inclusion ${\mathbb Z}\hookrightarrow {\mathbb Z}^{\oplus 2}$ as a
direct summand, while ${\rm Pic} Y\to{\rm Pic} Z$ is the projection ${\mathbb Z}^{\oplus
3}\mbox{$\,\>>>\hspace{-.5cm}\to\hspace{.15cm}$} {\mathbb Z}^{\oplus 2}$, we see that ${\rm image}\,({\rm Pic} X\to{\rm Pic} Y=NS(Y))$
is a direct summand ${\mathbb Z}^{\oplus 2}\hookrightarrow {\mathbb Z}^{\oplus 3}$. Thus
$NS(X)={\mathbb Z}^{\oplus 2}$ is strictly contained in $NS(Y)={\mathbb Z}^{\oplus 3}$.
$\square$
\end{ex}
|
\section{Introduction}
By definition cross product bialgebras are isomorphic to
factorizations $B=B_1\otimes B_2$, in a braided category say, such that the
given isomorphisms can be characterized universally in terms of certain
projections and injections from the bialgebra into the particular
tensor factors. In this context tensor product bialgebras, bi- or bismash
product bialgebras \cite{Rad1:85,Tak1:81}, doublecross product bialgebras,
bicross product bialgebras and double biproduct bialgebras
\cite{Ma1:90,Ma1:95} are cross product bialgebras. But also the
bicrossed or cocycle bicross product bialgebras
\cite{MS:94,Ma6:94} are cross product
bialgebras in our terminology; their
universal characterization is given in terms of cleft extensions
and coextensions (see \cite{BCM:86,DT:86,Mon1:92,Sch1:94} for a
comprehensive study of cleft or normal basis Hopf-Galois extensions
on the algebra level). The notion of cross product bialgebras
therefore as well includes constructions involving cocycles and dual
cocycles (or simply ``cycles'').
Well known examples of cross product bialgebras are
Drinfel'd's quantum double, the quantum Poincar\'e group, Radford's
4-parameter Hopf algebra, Lusztig's construction of the quantum
enveloping algebra, the quantum Weyl group, the affine
quantum groups $U_q(\hat g)$, the Connes-Moscovici Hopf algebra in
transverse differential geometry \cite{CM:98}, etc.
The universal constructions of cross product bialgebras in
\cite{Rad1:85,Ma1:90,Ma6:94,MS:94,Tak1:81} additionally admit an equivalent
characterization in terms of mutual weak (co\n-)modular, co-cyclic
relations of the tensor factors $B_1$ and $B_2$.
This means that there exist certain weak (co\n-)multiplications,
(co\n-)cycles and (co\n-)actions which are morphisms with tensor
rank $(2,1)$ or $(1,2)$ defined on $B_1$ and $B_2$ and their
two-fold tensor products, such that the multiplication $\m_B$ and the
comultiplication $\Delta_B$ of the bialgebra $B$ can be composed monoidally
by these structure morphisms. For example we encounter a weak left coaction
$\nu_l:B_1\to B_2\otimes B_1$ which has tensor rank $(1,2)$,
a weak cocycle $\sigma:B_2\otimes B_2\to B_1$ with rank $(2,1)$,
etc. All these structure morphisms are subject to certain
relations with rank $(2,2)$, $(3,1)$ and $(1,3)$ which
on the one hand follow from the bialgebra structure of $B=B_1\otimes B_2$
and on the other hand determine the bialgebra structure of $B_1\otimes B_2$
uniquely.
Despite those common properties of the cross product bialgebra constructions
in \cite{Rad1:85,Ma1:90,Ma6:94,MS:94,Tak1:81} no theory
exists which characterizes all of them
as special versions of a single universal construction which in turn
is uniquely determined by its (co\n-)modular co-cyclic structure. A first step
towards this objective has been achieved in
\cite{BD1:98} where we discovered a method to describe cross product bialgebras
without co-cycles, generalizing and uniting \cite{Ma1:90,Ma1:95,Rad1:85}.
The present article considerably extends the results of \cite{BD1:98}.
The main outcome is a universal and (co\n-)modular, co-cyclic
theory of cross product bialgebras with cocycles and cycles which unites
all known constructions within a single setting
and provides several new families of cross product
bialgebras. This is the most comprising
framework for cross product bialgebras so far, which equivalently takes into
account both universal and (co\n-)modular co-cyclic aspects.
Furthermore our construction is designed to work in arbitrary braided
categories\footnote{Therefore we have been able to apply the results
of \cite{BD1:98} to categories of Hopf bimodules and
Yetter-Drinfel'd modules \cite{BD3:98}; we characterized two-sided bi-
or bismash product bialgebras as certain cross product Hopf bimodule
bialgebras. In particular we showed that the double biproduct is a twisted
Hopf bimodule tensor product bialgebra.}.
To derive these results we will proceed in two steps.
We begin with the general definition of cross product bialgebras
and find a universal characterization for them.
We show that so-called cocycle cross product bialgebras
possess a certain (co\n-)modular co\n-cyclic structure.
Then we restrict our consideration to strong
cross product bialgebras. They also admit a universal
characterization and will turn out to be the central objects in our
theory of cross product bialgebras.
In a second step we present a (co\n-)modular co-cyclic
construction method by so-called Hopf data\footnote{Observe that
Hopf data without co-cycles have been introduced in \cite{BD1:98} already.
There should be no confusion with the more general notation of the same
name which will be defined in the present article. In a very similar
context the term ``Hopf datum" has been used in \cite{And1:96} for
special cases of bicrossed product bialgebras \cite{Ma6:94}.}.
A Hopf datum
consists of a pair of objects $(B_1,B_2)$ and weak (co\n-)\-ac\-tions,
(co\n-)\-mul\-ti\-pli\-cations and co-cycles defined on
$B_1$ and $B_2$ which obey certain
interrelated identities with rank $(2,2)$, $(3,1)$ and $(1,3)$.
Then we introduce strong Hopf data for which additional
``strong'' relations hold. We show that strong Hopf data induce the
structure of a strong cross product bialgebra on the tensor
product $B_1\otimes B_2$. Strong Hopf data and strong cross
product bialgebras are different depictions of the same object.
Then we combine the universal characterization and the (co\n-)modular
co-cyclic description in terms of strong Hopf data to
obtain the theory of strong cross product bialgebras.
This is the central result of the article. Eventually we apply our
construction and investigate strong cross product bialgebras according
to their (co\n-)modular co-cyclic structure. In particular we recover all known
constructions \cite{Rad1:85,Ma1:90,MS:94,Ma6:94,BD1:98} and
find various new types of cross product bialgebras. All of them are
special versions of the most general strong cross
product bialgebra construction.
\abs
After introducing preliminary notations and definitions we study
cross product algebras and cross product coalgebras from a
universal point of view in Section \ref{sec-ccpa}. These results will be used
in Section \ref{sec-ccpb} to define cross product bialgebras. From
this general definition we extract cocycle cross product bialgebras
and strong cross product bialgebras. The structure of
cocycle cross product bialgebras
gives useful hints for the design of Hopf data. This will be done in
Section \ref{hopf-data} where we
investigate the connection of Hopf data and cocycle cross product bialgebras.
A Hopf datum is canonically assigned to every cocycle cross
product bialgebra. The converse is not true in general.
But Theorem \ref{hp-cp} states that strong Hopf data yield strong
cross product bialgebras. More precisely strong Hopf data
and strong cross product bialgebras are
equivalent constructions which are in one-to-one
correspondence. The universal (co\n-)modular co-cyclic theory of
strong cross product bialgebras culminates in the
central Theorems \ref{hp-cp} and \ref{central-theory}.
Finally we present a table of special strong
cross product bialgebras which contains all known and several
new types of cross product bialgebras embedded in our most general framework.
The rather intricate proof of Theorem \ref{hp-cp} has been postponed to
Section \ref{proof-hp-cp}.
\abs
In the recent articles \cite{CIMZ:98,CZ:98,Sbg1:98}
cross product bialgebras have been studied as well.
In \cite{CIMZ:98} the universal properties of co-cycle free cross
product bialgebras have been considered. The article
\cite{CZ:98} investigates certain cross product
bialgebras without co-cycles in universal and (co\n-)modular co-cyclic terms.
In \cite{Sbg1:98} a special type of our cocycle cross product
bialgebras with one cocycle has been studied in the special case where
all objects are vector spaces over a field.
\subsection{Preliminaries}
We are working throughout in (strict) braided monoidal categories
\cite{JS:93}. In our article we denote
categories by calligraphic letters $\C$, $\D$, etc. For a braided monoidal
category $\C$ the tensor product is denoted by $\otimes_\C$, the unit
object by $\E_\C$, and the braiding by ${}^\C\Psi$. If it is clear from
the context we omit the index `$\C$' at the various symbols.
We confine ourselves to braided categories which admit split idempotents
\cite{BD:95,Lyu1:95}; for each idempotent $\Pi=\Pi^2:M\to M$ of any
object $M$ in $\C$ there exists an object $M_\Pi$ and a pair of morphisms
$(\inj_\Pi,\proj_\Pi)$ such that $\proj_\Pi\circ\inj_\Pi=\id_{M_\Pi}$
and $\inj_\Pi\circ\proj_\Pi=\Pi$. This is not a severe restriction
of the categories since every braided category
can be canonically embedded into a braided category which admits split
idempotents \cite{BD:95,Lyu1:95}.
We use generalized algebraic
structures like algebra, bialgebra, module, comodule, etc.~in such
categories. We assume the reader is familiar with these
generalizations. A thorough introduction to braided algebraic
structures and the graphical calculus coming along with it
can be found in \cite{FY:89,JS:93,Kas1:95,Lyu1:95,Ma4:93,ReT:90,Tur1:94}.
We denote by $\m:A\otimes A\to A$ the multiplication and by
$\eta:\E\to A$ the unit of an algebra $A$ in $\C$.
$\Delta: C\to C\otimes C$ is meant for the comultiplication and
$\varepsilon:C\to \E$ for the counit of a coalgebra $C$ in $\C$,
$\mu_l:A\otimes M\to M$ is the left action of an algebra $A$ on a
module $M$, and $\nu_l:N\to C\otimes N$ denotes the left coaction of a
coalgebra $C$ on a comodule $N$. Right actions are denoted by $\mu_r$ and
right coactions by $\nu_r$. In the course of our work we often
encounter weak versions of (co\n-)multiplication and (co\n-)action
which do not necessarily obey the relations of `true' (co\n-)algebras
and (co\n-)modules. We will nevertheless use the same symbols as
for proper (co\n-)multiplications and (co\n-)actions -- it should
become clear from the context whether these structure morphisms are weak or
not.
In our article we make use of graphical calculus for (strict) braided monoidal
categories which simplifies intricate categorical equations of
morphisms and helps to uncover their intrinsic structure.
Morphisms will be composed from up to down, i.~e.~the domains of the morphisms
are at the top and the codomaines are at the bottom of the graphics.
Tensor products are represented by horizontal concatenation
in the corresponding order.
We present our own conventions \cite{BD:95,BD:97,BD1:98} in Figure
\ref{fig-conv}. If there is no fear of confusion
we omit the assignment of a specific object to the ends
of the particular strings in the graphics.
\begin{figure}
\begin{equation*}
\begin{array}{c}
\m=\divide\unitlens by 2
\begin{tangle}\cu\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\eta=\divide\unitlens by 2
\ \begin{tangle}\unit\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\Delta=\divide\unitlens by 2
\begin{tangle}\cd\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\varepsilon=\divide\unitlens by 2
\ \begin{tangle}\counit\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\\[8pt]
\mu_l=\divide\unitlens by 2
\begin{tangle}\lu\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\mu_r=\divide\unitlens by 2
\begin{tangle}\ru\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\nu_l=\divide\unitlens by 2
\begin{tangle}\ld\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\nu_r=\divide\unitlens by 2
\begin{tangle}\rd\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\\[8pt]
\Psi=\divide\unitlens by 2
\begin{tangle}\x\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\Psi^{-1}=\divide\unitlens by 2
\begin{tangle}\xx\end{tangle}
\multiply\unitlens by 2
\quad .\quad
\end{array}
\end{equation*}
\caption{Graphical presentation of (weak) multiplication $\m$, unit
$\eta$, (weak) comultiplication $\Delta$, counit $\varepsilon$,
(weak) left action $\mu_l$, (weak) right action $\mu_r$,
(weak) left coaction $\nu_l$, (weak) right coaction $\nu_r$,
braiding $\Psi$, and inverse braiding
$\Psi^{-1}$.}\Label{fig-conv}
\end{figure}
Below we elucidate the graphical calculus to readers unfamiliar
with this technique. The first example is the associativity of the
action of an algebra $A$ on a right $A$-module $M$. The second is the
naturality of the braiding in a braided category.
\begin{equation*}
\begin{array}{ccc}
\mu_r\circ(\mu_r\otimes\id_A)=\mu_r\circ(\id_M\otimes\m_A)
&\quad\text{corresponds to}\quad
&\divide\unitlens by 2
{\begin{tangle}
\hh\hru\hstep\id\\
\ru
\end{tangle}}
=
{\begin{tangle}
\hh\id\hstep\hcu\\
\ru
\end{tangle}}\,\,\,,
\\[10pt]
\Psi\circ (f\otimes g) = (g\otimes f)\circ\Psi
&\quad\text{corresponds to}\quad
&\divide\unitlens by 2
{\begin{tangle}
\O{\sstyle f}\step[2]\O{\sstyle g}\\
\x
\end{tangle}}
\,=\,
{\begin{tangle}
\x\\
\O{\sstyle g}\step[2]\O{\sstyle f}
\end{tangle}}\,\,\,\,\,.
\end{array}
\end{equation*}
Note that throughout the article there is (almost) no need to require
invertibility of the braiding. Therefore most of the results can be
derived if we assume that the underlying category $\C$ is pre-braided.
We do not discuss these generalizations further and confine to braided
categories in what follows.
\section{Cross Product Algebras and Coalgebras}\Label{sec-ccpa}
In the first part of Section \ref{sec-ccpa} we study cross
product algebras. They have been considered also in \cite{Brz1:96}.
It turns out that cross product algebras are universal
constructions. They generalize crossed product algebras
\cite{DT:86,Mon1:92,Sch1:94}.
Cross product coalgebras will be studied
in the second part of Section \ref{sec-ccpa}. Since the results for
cross product coalgebras can be obtained easily by certain
categorical dualization, we omit all the proofs in this case and
refer to the analogous proofs for cross product algebras.
Both structures, cross product algebras and cross product
coalgebras, will be needed later in the definition of cross product
bialgebras.
\subsection{Cross Product Algebras}
\begin{definition}\Label{caat}
Let $(B_1,\m_1,\eta_1)$ be an algebra and $B_2$ be an object in $\C$.
Suppose there are morphisms $\eta_2:\E\to B_2$,
$\varphi_{2,1}:B_2\otimes B_1\to B_1\otimes B_2$, and
$\hat\sigma:B_2\otimes B_2\to B_1\otimes B_2$ such that
the relation
\begin{equation}\mathlabel{cond1-caat}
\varphi_{2,1}=(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\hat\sigma)\circ
(\varphi_{2,1}\otimes\eta_2)
\end{equation}
holds. Suppose further that $B=B_1\otimes B_2$ is an algebra through
\begin{equation}\mathlabel{cond2-caat}
\m_B = (\m_1\otimes\id_{B_2})\circ(\m_1\otimes\hat\sigma)\circ
(\id_{B_1}\otimes\varphi_{2,1}\otimes\id_{B_2})\,,\quad
\eta_B =\eta_1\otimes\eta_2
\end{equation}
then $B$ is called cross product algebra and will be denoted by
$B_1\cpcyalg B_2$.
\end{definition}
\abs
In the subsequent proposition we find equivalent constructive
conditions for cross product algebras. Similar results
have been obtained in \cite{Brz1:96}.
\begin{proposition}\Label{cross-equiv}
The following statements are equivalent.
\begin{enumerate}
\item
$B_1\cpcyalg B_2$ is a cross product algebra.
\item
$(B_1,\m_1,\eta_1)$ is an algebra and $B_2$ is an object in $\C$. There
exist morphisms $\eta_2:\E\to B_2$, $\varphi_{2,1}:B_2\otimes B_1\to
B_1\otimes B_2$ and $\hat\sigma:B_2\otimes B_2\to B_1\otimes B_2$ for
which the subsequent identities hold.
\begin{equation}\mathlabel{cond3-caat}
\begin{gathered}
{\begin{split}
\varphi_{2,1}\circ(\eta_2\otimes\id_{B_1})&=\id_{B_1}\otimes\eta_2\,,\\
\varphi_{2,1}\circ(\id_{B_2}\otimes\eta_1)&=\eta_1\otimes\id_{B_2}\,,
\end{split}
\quad
\begin{split}
\hat\sigma\circ(\eta_2\otimes\id_{B_2})&=\eta_1\otimes\id_{B_2}\,,\\
\hat\sigma\circ(\id_{B_2}\otimes\eta_2)&=\eta_1\otimes\id_{B_2}\,,
\end{split}}\\[5pt]
{\begin{gathered}
\varphi_{2,1}\circ(\id_{B_2}\otimes\m_1)=
(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\varphi_{2,1})\circ
(\varphi_{2,1}\otimes\id_{B_1})\,,\\[5pt]
(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\hat\sigma)\circ
(\varphi_{2,1}\otimes\id_{B_2})\circ(\id_{B_2}\otimes\hat\sigma)\\
=(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\hat\sigma)\circ
(\hat\sigma\otimes\id_{B_2})\,,\\[5pt]
(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\hat\sigma)\circ
(\varphi_{2,1}\otimes\id_{B_2})\circ(\id_{B_2}\otimes\varphi_{2,1})\\
=(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\varphi_{2,1})\circ
(\hat\sigma\otimes\id_{B_1})\,.
\end{gathered}}
\end{gathered}
\end{equation}
\end{enumerate}
\end{proposition}
\begin{proof}
Suppose that $B_1\cpcyalg B_2$ is a cross product algebra.
Since by assumption $(B_1,\m_1,\eta_1)$ is itself an algebra one
obtains the first and second identity of \eqref{cond3-caat}
by unitality of $\m_B$ and $\m_1$. Using these relations
yields the fourth identity of \eqref{cond3-caat}
by application of $\id_{B_2}\otimes\eta_1$ to
\eqref{cond1-caat}. Again using the left unitality of $\m_B$ one concludes
that $(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes
\hat\sigma\circ(\eta_2\otimes\id_{B_2}))=\id_{B_1}\otimes\id_{B_2}$
from which one immediately derives the third identity of
\eqref{cond3-caat}. Now the associativity
$\m_B\circ(\m_B\otimes\id_B)= \m_B\circ(\id_B\otimes\m_B)$ yields
the fifth identity of \eqref{cond3-caat} by application of
$\eta_1\otimes\id\otimes\id\otimes\eta_1\otimes\id\otimes\eta_2$.
The sixth relation is obtained by applying
$\eta_1\otimes\id\otimes\eta_1\otimes\id\otimes\eta_1\otimes\id$.
To derive the seventh identity one has to apply
$\eta_1\otimes\id\otimes\eta_1\otimes\id\otimes\id\otimes\eta_2$.
Conversely suppose the conditions of the second item of the proposition
are fulfilled. Define $\m_B$ according to \eqref{cond2-caat}. Then
${\m_B\circ(\m_B\otimes\id_B)} = (\m_1^{(5)}\otimes\id)\circ
(\id^{(4)}\otimes\hat\sigma)\circ(\id^{(3)}\otimes\varphi_{2,1}\otimes\id)
\circ(\id^{(2)}\otimes\hat\sigma\otimes\id^{(2)})\circ
(\id\otimes\varphi_{2,1}\otimes\id^{(3)})$
where $\m_1^{(n)}:B_1^{\otimes_n}
\to B_1$ is the canonical $n$-fold multiplication, and
$\id^{(n)}$ is the abbreviation of the identity of an $n$-fold tensor
product of (combined) $B_1$'s and $B_2$'s.
On the other hand
\begin{equation*}
\begin{split}
\m_B\circ(\id_B\otimes\m_B) &= (\m_1^{(4)}\otimes\id)\circ
(\id^{(4)}\otimes\hat\sigma)\circ(\id^{(3)}\otimes\varphi_{2,1}\otimes\id)
\circ\\
&\quad\circ(\id^{(2)}\otimes\varphi_{2,1}\otimes\hat\sigma)\circ
(\id\otimes\varphi_{2,1}\otimes\varphi_{2,1}\otimes\id)\\
&=(\m_1^{(5)}\otimes\id)\circ(\id^{(4)}\otimes\hat\sigma)\circ
(\id^{(3)}\otimes\hat\sigma\otimes\id)\circ\\
&\quad\circ(\id^{(2)}\otimes\varphi_{2,1}\otimes\id^{(2)})\circ
(\id\otimes\varphi_{2,1}\otimes\varphi_{2,1}\otimes\id)\\
&= (\m_1^{(5)}\otimes\id)\circ
(\id^{(4)}\otimes\hat\sigma)\circ(\id^{(3)}\otimes\varphi_{2,1}\otimes\id)
\circ\\
&\quad\circ(\id^{(2)}\otimes\hat\sigma\otimes\id^{(2)})\circ
(\id\otimes\varphi_{2,1}\otimes\id^{(3)})
\end{split}
\end{equation*}
where the fourth condition of \eqref{cond3-caat} has been used twice
to obtain the first equation, the sixth relation of \eqref{cond3-caat}
has been applied in the second equation, and with the help of the seventh
relation of \eqref{cond3-caat} we obtained the third equation.
Hence associativity of $\m_B$ has been proven. Using the first
four identities of \eqref{cond3-caat} one easily proves
\eqref{cond1-caat} and unitality
$\m_B\circ(\eta_B\otimes\id_B)=\id_B=\m_B\circ(\id_B\otimes\eta_B)$.
\end{proof}
\abs
\begin{remark} {\normalfont Condition \eqref{cond1-caat}
in Definition \ref{caat} can be replaced equivalently by the identity
$(\id_{B_1}\otimes\hat\sigma)
=(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\hat\sigma)\circ
(\varphi_{2,1}\circ(\id_{B_2}\otimes\eta_1)\otimes\id_{B_2})$.}
\end{remark}
\begin{remark} {\normalfont There is another equivalent definition of
cross product algebras which has been kindly reported to us by
Gigel Militaru. Observe that the morphism $\Gamma$ below is our morphism
$\m_{0,20}$ in \eqref{aux-m-delta}. Then the following statements are
equivalent.
\begin{enumerate}
\item
$B_1\cpcyalg B_2$ is a cross product algebra.
\item
$(B_1,\m_1,\eta_1)$ is an algebra and $B_2$ is an object in $\C$.
There are morphisms $\eta_2:\E\to B_2$ and
$\Gamma:B_2\otimes B_1\otimes B_2\to B_1\otimes B_2$ such that
$B=B_1\otimes B_2$ is an algebra through
\begin{equation}\mathlabel{cond4-caat}
\m_B = (\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\Gamma)\quad\text{and}
\quad\eta_B =\eta_1\otimes\eta_2\,.
\end{equation}
\item
$(B_1,\m_1,\eta_1)$ is an algebra and $B_2$ is an object in $\C$. There
exist morphisms $\eta_2:\E\to B_2$ and
$\Gamma:B_2\otimes B_1\otimes B_2\to B_1\otimes B_2$ such that
\begin{equation}\mathlabel{cond5-caat}
\begin{gathered}
\Gamma\circ(\eta_2\otimes\id_{B_1\otimes B_2})=\id_{B_1\otimes B_2}\,,
\quad
\Gamma\circ(\id_{B_2}\otimes\eta_1\otimes\eta_2)=
\eta_1\otimes\id_{B_2}\,,\\[5pt]
(\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\Gamma)\circ
(\Gamma\otimes\id_{B_1\otimes B_2})\\
=\Gamma\circ(\id_{B_2}\otimes(\m_1\otimes\id_{B_2})\circ
(\id_{B_1}\otimes\Gamma))\,.
\end{gathered}
\end{equation}
\end{enumerate}
The one-to-one correspondence is given by
$\Gamma= (\m_1\otimes\id_{B_2})\circ(\id_{B_1}\otimes\hat\sigma)\circ
(\varphi_{2,1}\otimes\id_{B_2})$ and its inverse
$\varphi_{2,1}=\Gamma\circ(\id_{B_2\otimes B_1}\otimes\eta_2)$ and
$\hat\sigma= \Gamma\circ(\id_{B_2}\otimes\eta_1\otimes\id_{B_2})$.}
\end{remark}
\abs
We will show that cross product algebras are universal
constructions. The first proposition describes equivalent projection and
injection conditions for algebras isomorphic to cross product
algebras (see also \cite{BD1:98} for the cocycle-free case).
The second proposition is closely related to the first one and
characterizes the universal construction explicitely.
\begin{proposition}\Label{univ-constr}
Let $A$ be an algebra in $\C$. Then it holds equivalently
\begin{enumerate}
\item
$A$ is algebra isomorphic to a cross product algebra
$B_1\cpcyalg B_2$.
\item
There is an algebra $(B_1,\m_1,\eta_1)$, an object $B_2$,
morphisms $B_1\overset{\inj_1}\to A\overset{\inj_2}\leftarrow B_2$ and
$\eta_2:\E\to B_2$ where $\inj_1$ is an algebra morphism and
$\inj_2\circ\eta_2=\eta_A$ such that
$\m_A\circ(\inj_1\otimes\inj_2): B_1\otimes B_2\to A$
is an isomorphism in $\C$.
\end{enumerate}
\end{proposition}
\begin{proof}
If $\Lambda:B_1\cpcyalg B_2\to A$ is an algebra isomorphism then
define $\inj_1:=\Lambda\circ(\id_{B_1}\otimes\eta_2)$
and $\inj_2:=\Lambda\circ(\eta_1\otimes\id_{B_2})$.
Using the particular definition of $\m_{B_1\cpcyalg B_2}$ in
\eqref{cond2-caat} and the identities \eqref{cond1-caat},
it is verified immediately that $\inj_1$ is an algebra morphism
since $\Lambda$ is an algebra morphism. Similarly one proves that
$\inj_2\circ\eta_2=\eta_A$ and $\m_A\circ(\inj_1\otimes\inj_2)=\Lambda$
which is therefore an isomorphism.
If on the other hand the conditions of the second statement of
Proposition \ref{univ-constr} is fulfilled, then
$\Lambda:=\m_A\otimes(\inj_1\otimes\inj_2$) is an isomorphism by
assumption. Therefore $B=B_1\otimes B_2$ is canonically an algebra
through $\m_B:=\Lambda^{-1}\circ\m_A\circ(\Lambda\otimes\Lambda)$ and
$\eta_B:=\Lambda^{-1}\circ\eta_A$. We will show that this defines a
cross product algebra structure on $B$ by
$\varphi_{2,1}:=\Lambda^{-1}\circ\m_A\circ(\inj_2\otimes\inj_1)$
and $\hat\sigma:=\Lambda^{-1}\circ\m_A\circ(\inj_2\otimes\inj_2)$.
Using the explicit expression for $\Lambda$,
inserting several times $\Lambda\circ\Lambda^{-1}$ and using the above
definitions, as well as the fact that $\inj_1$ is
algebra morphism, one obtains the following identities.
\begin{equation*}
\begin{split}
\m_B &=\Lambda^{-1}\circ\m_A^{(3)}\circ
(\id\otimes\Lambda\circ\Lambda^{-1}\otimes\id)\circ
(\inj_1\otimes\m_A\circ(\inj_2\otimes\inj_1)\otimes\inj_2)\\
&=\Lambda^{-1}\circ\m_A^{(4)}\circ
(\inj_1\otimes(\inj_1\otimes\inj_2)\circ\varphi_{2,1}\otimes\inj_2)\\
&=\Lambda^{-1}\circ\m_A^{(3)}\circ(\inj_1\otimes\inj_1\otimes\inj_2)\circ
(\m_1\otimes\hat\sigma)\circ(\id\otimes\varphi_{2,1}\otimes\id)\\
&=(\m_1\otimes\id)\circ(\m_1\otimes\hat\sigma)\circ
(\id\otimes\varphi_{2,1}\otimes\id)\,.
\end{split}
\end{equation*}
Similarly one derives $\Lambda\circ(\eta_1\otimes\eta_2)=\eta_A$, hence
$\eta_B=\eta_1\otimes\eta_2$. Therefore
$(B,\m_B,\eta_B)=B_1\cpcyalg B_2$.\end{proof}
\abs
\begin{proposition}\Label{univ-constr1}
Let $B_1\cpcyalg B_2$ be a cross product algebra, and $A$ be
an algebra. Suppose there are morphisms $\alpha:B_1\to A$ and
$\beta:B_2\to A$ such that
\begin{enumerate}
\item
$\alpha$ is an algebra morphism.
\item
$\beta\circ\eta_2=\eta_A\,.$
\item
$\m_A\circ(\alpha\otimes\beta)\circ\varphi_{2,1}
= \m_A\circ(\beta\otimes\alpha)\,.$
\item
$\m_A\circ(\alpha\otimes\beta)\circ\hat\sigma
= \m_A\circ(\beta\otimes\beta)\,.$
\end{enumerate}
Then there exists a unique algebra morphism $\gamma: B_1\cpcyalg B_2\to A$
obeying the identities $\gamma\circ(\id_{B_1}\otimes\eta_2)=\alpha$ and
$\gamma\circ(\eta_1\otimes\id_{B_2})=\beta$.
\end{proposition}
\begin{proof}
Define $\gamma:=\m_A\circ(\alpha\otimes\beta)$. Then by assumption 1
and 2 of Proposition \ref{univ-constr1} it follows
$\gamma\circ(\id_{B_1}\otimes\eta_2)=\alpha$,
$\gamma\circ(\eta_1\otimes\id_{B_2})=\beta$, and
$\gamma\circ(\eta_1\otimes\eta_2)=\eta_A$.
Using consecutively that $\alpha$ is an algebra morphism,
assumption 4, and assumption 3 of Proposition \ref{univ-constr1} then
yields $\gamma\circ\m_{B_1\cpcyalg B_2} = \m_A^{(4)}\circ
(\alpha\otimes\alpha\otimes(\alpha\otimes\beta)\circ\hat\sigma)\circ
(\id\otimes\varphi_{2,1}\otimes\id)=\m_A^{(4)}\circ
(\alpha\otimes(\alpha\otimes\beta)\circ\varphi_{2,1}\otimes\beta)
=\m_A\circ(\m_A\circ(\alpha\otimes\beta)\otimes\m_A\circ(\alpha\otimes\beta))
=\m_A\circ(\gamma\otimes\gamma)$.
Therefore $\gamma$ is an algebra morphism obeying the conditions
of Proposition \ref{univ-constr1}. Suppose that there is another
$\gamma'$ meeting the same stipulations as $\gamma$. Then
$\gamma=\m_A\circ(\alpha\otimes\beta)=\m_A\circ
(\gamma'\circ(\id\otimes\eta_2)\otimes\gamma'\circ(\eta_1\otimes\id))
=\gamma'\circ\m_{B_1\cpcyalg B_2}\circ(\id\otimes\eta_2\otimes
\eta_1\otimes\id)=\gamma'$.
In the last equation unital properties of cross product algebra
have been used. This proves uniqueness of $\gamma$.\end{proof}
\abs
\begin{remark}{\normalfont The conditions of Proposition \ref{univ-constr1}
can be realized on $A=B_1\cpcyalg B_2$ with $\alpha=\id_{B_1}\otimes\eta_2$
and $\beta=\eta_1\otimes\id_{B_2}$. Then of course $\gamma=\id$.}
\end{remark}
\abs
In the following we consider generalized smash product algebras \cite{BD1:98}
and demonstrate that they are special cases of cross product
algebras\footnote{We adopt the naming "smash product algebras" from
\cite{CIMZ:98}.}. A (generalized) {\it smash product algebra}
$B_1\times_{\phi_{2,1}}B_2$ consists of
algebras $B_1$ and $B_2$, and a morphism
$\phi_{2,1}:B_2\otimes B_1\to B_1\otimes B_2$,
such that $B=B_1\otimes B_2$ is an algebra through
$\m_B=(\m_1\otimes\m_2)\circ(\id_{B_1}\otimes\phi_{2,1}\otimes\id_{B_2})$
and $\eta_B=\eta_1\otimes\eta_2$.
The next proposition shows that smash product algebras are indeed special
cases of cross product algebras under certain natural conditions.
\abs
\begin{proposition}\Label{smash-ccpa}
Suppose there is a morphism $\varepsilon_1:B_1\to \E$ with
$\varepsilon_1\circ\eta_1=\id_{\E}$. Then
\begin{entry}
\item[\ ]
$B_1\cpcyalg B_2$ is a cross product algebra
with $\hat\sigma = \eta_1\otimes\m_2$ for some morphism
$\m_2:B_2\otimes B_2\to B_2$.
\item[$\Leftrightarrow$]
$B_1\times_{\varphi_{2,1}}B_2$ is a smash product algebra.
\end{entry}
\end{proposition}
\begin{proof}
The proposition can be proven easily by
\eqref{cond3-caat} and triviality of $\hat\sigma$.\end{proof}
\abs
\begin{remark}{\normalfont Crossed product algebras
$A\#_\sigma H$ \cite{DT:86,Mon1:92}
are special examples of cross product algebras through
$\varphi_{2,1}:=(\mu_l\otimes\id_H)\circ(\Delta_H\otimes\id_A)$ and
$\hat\sigma :=(\sigma\otimes\m_H)\circ(\id_H\otimes\Psi_{H,H}\otimes\id_H)
\circ(\Delta_H\otimes\Delta_H)$
where $A$ is an algebra, $\mu_l:H\otimes A\to A$ is a left $H$-measure
on $A$, and $\sigma:H\otimes H\to A$ is a (convolution invertible)
cocycle.}
\end{remark}
\abs
The next proposition yields criteria under which a morphism
$f:B_1\cpcyalg B_1\to A$ is multiplicative.
\begin{proposition}\Label{alg-morph-equiv}
Let $A$ be an algebra and $B=B_1\cpcyalg B_2$ be a cross product
algebra. Then the subsequent statements are equivalent.
\begin{enumerate}
\item\Label{alg-morph-equiv1}
The morphism $f:B\to A$ is multiplicative,
i.~e.~$f\circ\m_B=\m_A\circ(f\otimes f)$.
\item\Label{alg-morph-equiv2}
The identities
\begin{equation*}
\begin{split}
\m_A\circ(f\otimes f)\circ
(\id_{B_1}\otimes\eta_2\otimes\id_{B_1}\otimes\id_{B_2}) &=
f\circ\m_B\circ(\id_{B_1}\otimes\eta_2\otimes\id_{B_1}\otimes\id_{B_2})\,,\\
\m_A\circ(f\otimes f)\circ
(\eta_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\id_{B_2}) &=
f\circ\m_B\circ(\eta_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\id_{B_2})\\[-12pt]
\end{split}
\end{equation*}
are satisfied.
\item\Label{alg-morph-equiv3}
The identities
\begin{equation*}
\begin{split}
\m_A\circ(f\otimes f)\circ
(\id_{B_1}\otimes\id_{B_2}\otimes\id_{B_1}\otimes\eta_2) &=
f\circ\m_B\circ(\id_{B_1}\otimes\id_{B_2}\otimes\id_{B_1}\otimes\eta_2)\,,\\
\m_A\circ(f\otimes f)\circ
(\id_{B_1}\otimes\id_{B_2}\otimes\eta_1\otimes\id_{B_2}) &=
f\circ\m_B\circ(\id_{B_1}\otimes\id_{B_2}\otimes\eta_1\otimes\id_{B_2})\\[-12pt]
\end{split}
\end{equation*}
are satisfied.
\item\Label{alg-morph-equiv4}
The identities
\begin{equation*}
\begin{split}
\m_A\circ(f\otimes f)\circ
(\id_{B_1}\otimes\eta_2\otimes\id_{B_1}\otimes\id_{B_2}) &=
f\circ\m_B\circ(\id_{B_1}\otimes\eta_2\otimes\id_{B_1}\otimes\id_{B_2})\,,\\
\m_A\circ(f\otimes f)\circ
(\id_{B_1}\otimes\id_{B_2}\otimes\eta_1\otimes\id_{B_2}) &=
f\circ\m_B\circ(\id_{B_1}\otimes\id_{B_2}\otimes\eta_1\otimes\id_{B_2})\,,\\
\m_A\circ(f\otimes f)\circ
(\eta_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\eta_2) &=
f\circ\m_B\circ(\eta_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\eta_2)\\[-12pt]
\end{split}
\end{equation*}
are satisfied.
\end{enumerate}
\end{proposition}
\begin{proof}
The non-trivial part of the proposition can be verified easily with the help
of the identity
$\m_B\circ(\id\otimes\eta_2\otimes\eta_1\otimes\id)=\id\otimes\id$,
the associativity of $A$ and $B$, and the assumptions
\ref{alg-morph-equiv}.\ref{alg-morph-equiv2},
\ref{alg-morph-equiv}.\ref{alg-morph-equiv3}
or \ref{alg-morph-equiv}.\ref{alg-morph-equiv4} respectively.
\end{proof}
\subsection{Cross Product Coalgebras}
Cross product coalgebras are somehow dual constructions to
cross product algebras. We will omit proofs since they can be
obtained from the corresponding proofs for cross product
algebras by an obvious kind of dualization.
\abs
\begin{definition}\Label{ccat}
Let $(C_2,\Delta_2,\varepsilon_2)$ be a coalgebra and $C_1$ be an
object in $\C$. Suppose there exist morphisms $\varepsilon_1:B_1\to\E$,
$\varphi_{1,2}:C_1\otimes C_2\to C_2\otimes C_1$,
$\hat\rho:C_1\otimes C_2\to C_1\otimes C_1$ such that the relation
$\varphi_{1,2}=(\varepsilon_1\otimes\varphi_{1,2})\circ
(\hat\rho\otimes\id_{C_2})\circ(\id_{C_1}\otimes\Delta_2)$
holds. Then $C=C_1\otimes C_2$ is called cross product coalgebra if
it is a coalgebra through $\Delta_C= (\id_{B_1}\otimes
\varphi_{1,2}\otimes\id_{B_2})\circ(\hat\rho\otimes\Delta_2)\circ
(\id_{B_1}\otimes\Delta_2)$ and $\varepsilon_C=
\varepsilon_1\otimes\varepsilon_2$. We denote $C$ by $C_1\cpcyco C_2$.
\end{definition}
\abs
\begin{proposition}
The following statements are equivalent.
\begin{enumerate}
\item
$C_1\cpcyco C_2$ is a cross product coalgebra.
\item
$(C_2,\Delta_2,\varepsilon_2)$ is a coalgebra and $C_1$ is an object in
$\C$. There are morphisms $\varepsilon_1:B_1\to\E$ and
$\varphi_{1,2} : C_1\otimes C_2\to C_2\otimes C_1$ such that
\begin{equation*}
\begin{gathered}
{\begin{split}
(\id_{C_2}\otimes\varepsilon_1)\circ\varphi_{1,2}&=
\varepsilon_1\otimes\id_{C_2}\,,\\
(\varepsilon_1\otimes\id_{C_2})\circ\varphi_{1,2}&=\id_{C_2}\otimes
\varepsilon_1\,,
\end{split}
\quad
\begin{split}
(\id_{C_1}\otimes\varepsilon_1)\circ\hat\rho &=\id_{C_1}\otimes
\varepsilon_2\,,\\
(\varepsilon_1\otimes\id_{C_1})\circ\hat\rho &=
\id_{C_1}\otimes\varepsilon_2\,,
\end{split}}\\[5pt]
{\begin{gathered}
(\Delta_2\otimes\id_{C_1})\circ\varphi_{1,2}=
(\id_{C_2}\otimes\varphi_{1,2})\circ(\varphi_{1,2}\otimes\id_{C_2})\circ
(\id_{C_1}\otimes\Delta_2)\,,\\[5pt]
(\hat\rho\otimes\id_{C_1})\circ(\id_{C_1}\otimes\varphi_{1,2})\circ
(\hat\rho\otimes\id_{C_2})\circ(\id_{C_1}\otimes\Delta_2)\\
= (\id_{C_1}\otimes\hat\rho)\circ(\hat\rho\otimes\id_{C_2})\circ
(\id_{C_1}\otimes\Delta_2)\,,\\[5pt]
(\varphi_{1,2}\otimes\id_{C_1})\circ(\id_{C_1}\otimes\varphi_{1,2})\circ
(\hat\rho\otimes\id_{C_2})\circ(\id_{C_1}\otimes\Delta_2)\\
=(\id_{C_2}\otimes\hat\rho)\circ(\varphi_{1,2}\otimes\id_{C_2})\circ
(\id_{C_1}\otimes\Delta_2)\,.
\end{gathered}}
\end{gathered}
\end{equation*}
\end{enumerate}
~\endproof
\end{proposition}
\abs
Like cross product algebras the cross product coalgebras are
universal constructions, too.
\begin{proposition}\Label{couniv-constr}
Let $C$ be a coalgebra. Then
\begin{entry}
\item[\ ]
$C$ is coalgebra isomorphic to a cross product coalgebra
$C_1\cpcyco C_2$.
\item[$\Leftrightarrow$]
There is a coalgebra $(C_2,\Delta_2,\varepsilon_2)$ and
an object $C_1$ in $\C$, morphisms $C_1\overset{\proj_1}\leftarrow C
\overset{\proj_2}\to C_2$ and $\varepsilon_1:C_1\to \E$ where
$\proj_2$ is a coalgebra morphism and
$\varepsilon_1\circ\proj_1=\varepsilon_C$ such that
$(\proj_1\otimes\proj_2)\circ\Delta_C:C\to C_1\otimes C_2$
is an isomorphism in $\C$.\endproof
\end{entry}
\end{proposition}
\abs
\begin{proposition}
Let $C_1\cpcyco C_2$ be a cross product coalgebra and
$C$ be a coalgebra. Suppose that there exist morphisms $a$ and $b$
such that
\begin{enumerate}
\item
$a:C\to C_1$ and $\varepsilon_1\circ a=\varepsilon_C\,.$
\item
$b:C\to C_2$ is a coalgebra morphism.
\item
$\varphi_{1,2}\circ(a\otimes b)\circ\Delta_C=(b\otimes a)\circ\Delta_C\,.$
\item
$\hat\rho\circ(a\otimes b)\circ\Delta_C=(a\otimes a)\circ\Delta_C\,.$
\end{enumerate}
Then there exists a unique coalgebra morphism $c:C\to C_1\cpcyco C_2$
obeying the identities $a=(\id_{C_1}\otimes\varepsilon_2)\circ c$ and
$b=(\varepsilon_1\otimes\id_{C_2})\circ c$.\endproof
\end{proposition}
\abs
Co-smash product coalgebras are special cases of cross
product coalgebras. More precisely, a {\it co-smash product coalgebra}
$C=C_1\times^{\phi_{1,2}}C_2$ is given by coalgebras $C_1$ and $C_2$,
and a morphism $\phi_{1,2}:C_1\otimes C_2\to C_2\otimes C_1$,
such that $C=C_1\otimes C_2$ is a coalgebra through
$\Delta_C=(\id_{C_1}\otimes\phi_{1,2}\otimes\id_{C_2})\circ
(\Delta_1\otimes\Delta_2)$ and
$\varepsilon_C=\varepsilon_{C_1}\otimes\varepsilon_{C_2}$.
Then the corresponding result of Proposition \ref{smash-ccpa} holds
for co-smash product coalgebras.
\begin{proposition}
Suppose $C_2$ is a coalgebra and there is a morphism $\eta_2:\E\to C_2$
with $\varepsilon_2\circ\eta_2=\id_{\E}$. Then
\begin{entry}
\item[\ ]
$(C_1,C_2,\varphi_{1,2},\hat\rho)$ is a cross product coalgebra
with $\hat\rho=\Delta_1\otimes\varepsilon_2$ for some morphism
$\Delta_2:C_2\to C_2\otimes C_2$.
\item[$\Leftrightarrow$]
$C_1\times^{\varphi_{1,2}}C_2$ is a co-smash product coalgebra.\endproof
\end{entry}
\end{proposition}
\abs
\begin{remark}[$\pi$-Symmetry]\Label{pi-sym}{\normalfont We would like
to point out the following important observation to the reader. Definition
\ref{caat} is not dual to Definition \ref{ccat}.
Rather, both definitions can be obtained from each other by
a combination of duality (followed by the usual exchanges of
multiplication and comultiplication, unit and counit, cocycle and cycle,
etc.) and use of the opposite tensor product (followed by exchange of indices
"1 $\leftrightarrow$ 2", and later also by exchange of "left/right
(coaction) $\leftrightarrow$ right/left (action)").
This is not an accidential fact but has to do
with the subsequent definition of cross product bialgebras where we
use precisely these algebra and coalgebra structures. Only then are we
able to get all the known cross product bialgebras
\cite{Rad1:85,Ma1:90,MS:94,Ma6:94,BD1:98}.
This kind of symmetry between cross product algebra and
cross product coalgebra will be called
\textbf{$\mathbf{\pi}$-symmetry} henceforth. In terms of graphical calculus
$\pi$-symmetry can be interpreted easily as rotation of the graphic
by the angle $\pi$ along an axis normal to the planar graphic
followed by the above mentioned exchanges of
indices and morphism types. We will often apply this principle to obtain
$\pi$-symmetric results simply by $\pi$-rotation.
Hence cross product bialgebras - as defined below - are therefore
$\pi$-symmetric invariant rather than dual symmetric invariant.
Dualization of our definition would lead to another kind of cross
product bialgebras. The corresponding results for these dual
versions follow from our results straightforwardly by dualization.}
\end{remark}
\section{Cross Product Bialgebras}\Label{sec-ccpb}
In Section \ref{sec-ccpb} we investigate cross product bialgebras
from a universal point of view.
They are simultaneously cross product algebras and
cross product coalgebras with compatible bialgebra structure.
We find a universal description for the isomorphism classes of
cross product bialgebras. Then we will consider
special cases called cocycle and strong cross product bialgebras
which are universal constructions as well. But in addition the given
isomorphism class of a cocycle/strong cross product bialgebra
can be characterized by certain interrelated
(co\n-)modular co-cyclic structures on the tensor factors.
Strong cross product bialgebras are the pivotal objects
for the studies in Section \ref{hopf-data}.
\subsection{Cross Product Bialgebras -- General Definition}
\begin{definition}\Label{cbat}
A bialgebra $B$ is called cross product bialgebra
if its underlying algebra is a cross product algebra
$B_1\cpcyalg B_2$, and its underlying coalgebra is a
cross product coalgebra $B_1\cpcyco B_2$ on the same objects.
The cross product bialgebra $B$ will be denoted by $B_1\cpcybi B_2$.
A cross product bialgebra is called normalized if
$\varepsilon_1\circ\eta_1=\id_{\E}$ (and then equivalently
$\varepsilon_2\circ\eta_2=\id_{\E}$).
\end{definition}
\abs
Cross product bialgebras are universal in the following sense.
\begin{theorem}\Label{proj-cycle}
Let $B$ be a bialgebra in $\C$.
Then the subsequent equivalent conditions hold.
\begin{enumerate}
\item\Label{proj-cycle3}
$B$ is bialgebra isomorphic to a normalized cross product
bialgebra $B_1\cpcybi B_2$.
\item\Label{proj-cycle1}
There are idempotents $\Pi_1,\Pi_2\in\End(B)$ such that
\begin{equation}\mathlabel{pi-rel1}
\begin{gathered}
\m_B\circ(\Pi_1\otimes\Pi_1)=\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)\,,\qquad
\Pi_1\circ\eta_B=\eta_B\,,\\
(\Pi_2\otimes\Pi_2)\circ\Delta_B
=(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2\,,\qquad
\varepsilon_B\circ\Pi_2=\varepsilon_B\,,
\end{gathered}
\end{equation}
and the sequence
$B\otimes B\xrightarrow{\m_B\circ(\Pi_1\otimes\Pi_2)}B
\xrightarrow{(\Pi_1\otimes\Pi_2)\circ\Delta_B} B\otimes B$
is a splitting of the idempotent $\Pi_1\otimes\Pi_2$ of $B\otimes B$.
\item\Label{proj-cycle2}
There are objects $B_1$ and $B_2$ and
morphisms $B_1\overset{\inj_1}\to B\overset{\proj_1}\to B_1$
and $B_2\overset{\inj_2}\to B\overset{\proj_2}\to B_2$ where
$\inj_1$ is algebra morphism, $\proj_2$ is coalgebra morphism and
$\proj_j\circ\inj_j=\id_{B_j}$ for $j\in\{1,2\}$ such that
$\m_B\circ(\inj_1\otimes\inj_2):B_1\otimes B_2\to B$ and
$(\proj_1\otimes\proj_2)\circ\Delta_B:B\to B_1\otimes B_2$
are mutually inverse isomorphisms.
\end{enumerate}
\end{theorem}
\begin{proof}
``\ref{proj-cycle1}.$\Rightarrow$\ref{proj-cycle2}.'':
For $j\in\{1,2\}$ we define $\inj_j$, $\proj_j$ to be the morphisms
splitting the idempotent $\Pi_j$ as $\Pi_j=\inj_j\circ\proj_j$ and
$\proj_j\circ\inj_j=\id_{B_j}$ for some objects $B_j$. Then with the
help of Theorem \ref{proj-cycle}.\ref{proj-cycle1} it follows
$\m_B\circ(\inj_1\otimes\inj_2)\circ(\proj_1\otimes\proj_2)\circ\Delta_B
=\m_B\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B=\id_B$ and
$(\proj_1\otimes\proj_2)\circ\Delta_B\circ\m_B\circ(\inj_1\otimes\inj_2)
= (\proj_1\otimes\proj_2)\circ(\Pi_1\otimes\Pi_2)\circ
\Delta_B\circ\m_B\circ(\Pi_1\otimes\Pi_2)\circ(\inj_1\otimes\inj_2)
=(\proj_1\circ\inj_1\otimes\proj_2\circ\inj_2)=\id_{B_1\otimes B_2}$. Hence
$(\m_B\circ(\inj_1\otimes\inj_2))^{-1}=(\proj_1\otimes\proj_2)\circ\Delta_B$.
Now we define $\m_1:=\proj_1\circ\m_B\circ(\inj_1\otimes\inj_1)$,
$\eta_1:=\proj_1\circ\eta_B$,
$\Delta_2:=(\proj_2\otimes\proj_2)\circ\Delta_B\circ\inj_2$, and
$\varepsilon_2:=\varepsilon_B\circ\inj_2$. Then using
\eqref{pi-rel1} and the (co\n-)algebra property of $B$ one verifies
easily that $(B_1,\m_1,\eta_1)$ is an algebra and
$(B_2,\Delta_2,\varepsilon_2)$ is a coalgebra.
Furthermore $\inj_1$ is an algebra morphism because
$\inj_1\circ\m_1 = \Pi_1\circ\m_B\circ(\inj_1\otimes\inj_1)
= \m_B\circ(\inj_1\otimes\inj_1)$, and
$\inj_1\circ\eta_1=\Pi_1\circ\eta_B = \eta_B$. In a $\pi$-symmetric manner
one proves that $\proj_2$ is a coalgebra morphism.
\nl
``\ref{proj-cycle2}.$\Rightarrow$\ref{proj-cycle1}.'':
Conversely we define $\Pi_j:=\inj_j\circ\proj_j$ for $j\in\{1,2\}$.
Since by assumption $\inj_1$ is an algebra morphism it follows
$\Pi_1\circ\eta_B=\Pi_1\circ\inj_1\circ\eta_1=\inj_1\circ\eta_1=\eta_B$
and using $\proj_1\circ\inj_1=\id_{B_1}$ it holds
$\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1) =
\inj_1\circ\proj_1\circ\inj_1\circ\m_1\circ(\proj_1\otimes\proj_1)
= \m_B\circ(\Pi_1\otimes\Pi_1)$.
In $\pi$-symmetrical analogy it will be proven that
$(\Pi_2\otimes\Pi_2)\circ\Delta_B=(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2$
and $\varepsilon_B\circ\Pi_2=\varepsilon_B$.
From Theorem \ref{proj-cycle}.\ref{proj-cycle2} we conclude
$(\Pi_1\otimes\Pi_2)\circ\Delta_B\circ\m_B\circ(\Pi_1\otimes\Pi_2)
=(\inj_1\otimes\inj_2)\circ\id_{B_1\otimes B_2}\circ
(\proj_1\otimes\proj_2)=\Pi_1\otimes\Pi_2$ and
$\id_B=\m_B\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B=
\m_B\circ(\Pi_1\otimes\Pi_2)\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B$.
Thus $\m_B\circ(\Pi_1\otimes\Pi_2)$ and $(\Pi_1\otimes\Pi_2)\circ\Delta_B$
split the idempotent $(\Pi_1\otimes\Pi_2)$.
\nl
``\ref{proj-cycle2}.$\Rightarrow$\ref{proj-cycle3}.'':
By assumption $(B_1,\m_1,\eta_1)$ is an algebra and $\inj_1$ is
an algebra morphism, $\inj_2$ is a morphism in $\C$, and
$\m_B\circ(\inj_1\otimes\inj_2)$ is an isomorphism.
Define $\eta_2:=\proj_2\circ\eta_B:\E\to B_2$. Then
$\inj_2\circ\eta_2 =\Pi_2\circ\eta_B =\m_B\circ(\Pi_1\circ\eta_B\otimes
\Pi_2\circ\eta_B) =\m_B\circ(\inj_1\otimes\inj_2)\circ(\proj_1\otimes\proj_2)
\circ\Delta_B\circ\eta_B =\eta_B$. Therefore all conditions of
Proposition \ref{univ-constr}.2 are fulfilled implying that
$B$ is algebra isomorphic to a cross product algebra
$B_1\cpcyalg B_2$. From the proof of Proposition \ref{univ-constr} one
reads off that the isomorphism is given by
$\Lambda=\m_B\circ(\inj_1\otimes\inj_2)$. In a $\pi$-symmetric way
one shows that $B$ is coalgebra isomorphic to a cross product
coalgebra $B_1\cpcyco B_2$ on the same tensor product with
isomorphism $\tilde\Lambda=(\proj_1\otimes\proj_2)\circ\Delta_B$. Thus
by assumption $\tilde\Lambda=\Lambda^{-1}$. Since $B$ is a bialgebra
it follows then from Definition \ref{cbat} that $B_1\otimes B_2$ is
cross product bialgebra $B_1\cpcybi B_2$ and
$\Lambda:B_1\cpcybi B_2\to B$ is bialgebra isomorphism. Furthermore
the identities $\id_\E=\varepsilon_B\circ\eta_B=\varepsilon_B\circ
\inj_2\circ\eta_2=\varepsilon_2\circ\proj_2\circ\inj_2\circ\eta_2=
\varepsilon_2\circ\eta_2$ show that $B_1\cpcybi B_2$ is normalized.
\nl
``\ref{proj-cycle3}.$\Rightarrow$\ref{proj-cycle2}.'':
Given a bialgebra isomorphism $\Lambda:B_1\cpcybi B_2\to B$.
Then in particular $\Lambda:B_1\cpcyalg B_2\to B$ is an algebra
isomorphism and $\Lambda:B_1\cpcyco B_2\to B$ is a coalgebra isomorphism.
We define $\inj_1 :=\Lambda\circ(\id_{B_1}\otimes\eta_2)$,
$\inj_2 :=\Lambda\circ(\eta_1\otimes\id_{B_2})$,
$\proj_1 :=(\id_{B_1}\otimes\varepsilon_2)\circ\Lambda^{-1}$, and
$\proj_2 :=(\varepsilon_1\otimes\id_{B_2})\circ\Lambda^{-1}$.
From Propositions \ref{univ-constr} and \ref{couniv-constr} one
derives that $\inj_1$ is an algebra morphism, $\proj_2$ is a coalgebra
morphism, and $\Lambda=\m_B\circ(\inj_1\otimes\inj_2)=
\big((\proj_1\otimes\proj_2)\circ\Delta_B\big)^{-1}$.
By assumption $B_1\cpcybi B_2$ is normalized from which follows
$\proj_1\circ\inj_1=(\id_{B_1}\otimes\varepsilon_2)\circ
\Lambda^{-1}\circ\Lambda\circ(\id_{B_1}\otimes\eta_2)=\id_{B_1}$ and
similarly $\proj_2\circ\inj_2=\id_{B_2}$.\end{proof}
\abs
The next corollary of Theorem \ref{proj-cycle} will be used
frequently in the following.
\begin{corollary}\Label{proj-cycle-eqs}
Under the conditions of Theorem \ref{proj-cycle} it holds
\begin{equation}\mathlabel{cons-proj-cycle1}
\begin{gathered}
\begin{split}
\Pi_2\circ\eta_B&=\eta_B\,,\\
\varepsilon_B\circ\Pi_1&=\varepsilon_B\,,
\end{split}
\quad
\begin{split}
\Pi_1\circ\Pi_2&=\eta_B\circ \varepsilon_B\,,\\
\Pi_2\circ\Pi_1&=\eta_B\circ \varepsilon_B\,,
\end{split}
\quad
\begin{split}
\Pi_2\circ\m_B &=\Pi_2\circ\m_B\circ(\Pi_2\otimes\id_B)\,,\\
\Delta_B\circ \Pi_1&=(\id_B\otimes\Pi_1)\circ\Delta_B\circ\Pi_1\,,
\end{split}\\[5pt]
{\begin{split}
\Pi_1\circ\m_B\circ(\Pi_1\otimes\id_B)&=
\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)\,,\\
(\id_B\otimes\Pi_2)\circ\Delta_B\circ \Pi_2&=
(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ \Pi_2\,.
\end{split}}
\end{gathered}
\end{equation}
The structure morphisms $\varphi_{1,2}$, $\varphi_{2,1}$, $\hat\rho$,
and $\hat\sigma$ of the cross product bialgebra
$B_1\cpcybi B_2$ can be expressed through the projections
and injections on $B$ as
\begin{equation}\mathlabel{struc-morph}
\begin{gathered}
\inj_1\circ\eta_1=\eta_B\,,\quad
\varepsilon_1\circ\proj_1=\varepsilon_B\,,\quad
\inj_2\circ\eta_2=\eta_B\,,\quad
\varepsilon_2\circ\proj_2= \varepsilon_B\,,\\[5pt]
{\begin{split}
\m_1&=\proj_1\circ\m_B\circ(\inj_1\otimes\inj_1)\,,\\
\varphi_{1,2} &=(\proj_2\otimes\proj_1)\circ\Delta_B\circ\m_B\circ
(\inj_1\otimes\inj_2)\,,\\
\varphi_{2,1} &=(\proj_1\otimes\proj_2)\circ\Delta_B\circ\m_B\circ
(\inj_2\otimes\inj_1)\,,
\end{split}}
\quad
{\begin{split}
\Delta_2&=(\proj_2\otimes\proj_2)\circ\Delta_B\circ\inj_2\,,\\
\hat\rho &=(\proj_1\otimes\proj_1)\circ\Delta_B\circ\m_B\circ
(\inj_1\otimes\inj_2)\,,\\
\hat\sigma &=(\proj_1\otimes\proj_2)\circ\Delta_B\circ\m_B\circ
(\inj_2\otimes\inj_2)\,.
\end{split}}
\end{gathered}
\end{equation}
\end{corollary}
\begin{proof}
The identities $\inj_1\circ\eta_1=\eta_B$ and
$\varepsilon_2\circ\proj_2=\varepsilon_B$ hold since $\inj_1$ is an
algebra morphism and $\proj_2$ is a coalgebra morphism, whereas
the identities $\inj_2\circ\eta_2=\eta_B$ and
$\varepsilon_1\circ\proj_1=\varepsilon_B$, as well as the equations
$\Pi_2\circ\eta_B=\eta_B$ and $\varepsilon_B\circ\Pi_1=\varepsilon_B$
have been shown in the proof of Theorem \ref{proj-cycle}.
From the proofs of Propositions \ref{univ-constr} and \ref{couniv-constr}
one can directly read off the structure of the morphisms
$\varphi_{1,2}$, $\varphi_{2,1}$, $\hat\rho$, and $\hat\sigma$
given in \eqref{struc-morph} using
$\Lambda=\m_B\circ(\inj_1\otimes\inj_2)$ and
$\Lambda^{-1}=(\proj_1\otimes\proj_2)\circ\Delta_B$.
Since $\inj_1$ is algebra morphism, $\proj_2$ is coalgebra
morphism and $\proj_j\circ\inj_j=\id_{B_j}$ for $j\in\{1,2\}$ it
follows $\m_1=\proj_1\circ\m_B\circ(\inj_1\otimes\inj_1)$ and
$\Delta_2=(\proj_2\otimes\proj_2)\circ\Delta_B\circ\inj_2$.
Because of \eqref{pi-rel1} and Theorem \ref{proj-cycle}.\ref{proj-cycle1}
it follows
$\Pi_1\circ\Pi_2=(\id_B\otimes\varepsilon_B)\circ(\Pi_1\otimes\Pi_2)
\circ\Delta_B\circ\m_B\circ{(\Pi_1\otimes\Pi_2)}\circ(\eta_B\otimes\id_B)=
(\Pi_1\circ\eta_B\otimes\varepsilon_B\circ\Pi_2)=\eta_B\circ\varepsilon_B$.
Using $\Pi_2\circ\eta_B=\eta_B$ and $\varepsilon_B\circ\Pi_1=\varepsilon_B$
from above we obtain in a similar way
$\Pi_2\circ\Pi_1=\eta_B\circ\varepsilon_B$.
Using consecutively Theorem \ref{proj-cycle}.\ref{proj-cycle1} two times,
\eqref{pi-rel1}, the identity
$\Pi_2=(\varepsilon_B\circ\Pi_1\otimes\Pi_2)\circ\Delta_B$,
again Theorem \ref{proj-cycle}.\ref{proj-cycle1}, and the identity
$\varepsilon_B\circ\Pi_1=\varepsilon_B$
one obtains the following series of equations.
\begin{equation*}
\begin{split}
\Pi_2\circ\m_B&=\Pi_2\circ\m_B\circ(\m_B\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B
\otimes\id_B)\\
&=\Pi_2\circ\m_B^{(3)}\circ(\Pi_1\otimes\Pi_1\otimes\Pi_2)\circ
(\id_B\otimes\Delta_B\circ\m_B)\circ\\
&\quad\circ(\Pi_1\otimes\Pi_2\otimes\id_B)\circ(\Delta_B\otimes\id_B)\\
&=\Pi_2\circ\m_B\circ(\Pi_1\otimes\Pi_2)\circ
(\m_B\circ(\Pi_1\otimes\Pi_1)\otimes\id_B)\circ\\
&\quad\circ(\id_B\otimes\Delta_B\circ\m_B)\circ(\id_B\otimes\Pi_2\otimes\id_B)
\circ(\Delta_B\otimes\id_B)\\
&=\Pi_2\circ\m_B\circ(\Pi_2\otimes\id_B)
\end{split}
\end{equation*}
and $\pi$-symmetrically
$\Delta_B\circ\Pi_1= (\id_B\otimes\Pi_1)\circ\Delta_B\circ\Pi_1$.
Finally one obtains in a similar manner
\begin{equation*}
\begin{split}
\Pi_1\circ\m_B\circ(\Pi_1\otimes\id_B) &= \Pi_1\circ\m_B\circ
(\Pi_1\otimes\m_B\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B)\\
&=\Pi_1\circ\m_B^{(3)}\circ(\Pi_1\otimes\Pi_1\otimes\Pi_2)\circ
(\id_B\otimes\Delta_B)\\
&=\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_2)\circ
(\m_B\circ(\Pi_1\otimes\Pi_1)\otimes\Pi_2)\circ(\id_B\otimes\Delta_B)\\
&=(\Pi_1\otimes\varepsilon_B\circ\Pi_2)\circ\Delta_B\circ\m_B
\circ(\Pi_1\otimes\Pi_2)\circ\\
&\quad\circ(\m_B\circ(\Pi_1\otimes\Pi_1)\otimes\Pi_2)
\circ(\id_B\otimes\Delta_B)\\
&=\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)
\end{split}
\end{equation*}
and $(\id_B\otimes\Pi_2)\circ\Delta_B\circ\Pi_2=
(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2$ by $\pi$-symmetry.\end{proof}
\begin{remark}\Label{alpha-ip}
{\normalfont The statement of Theorem \ref{proj-cycle} can be
refined in the following sense. For a given bialgebra $B$
the tuples $(B,\Lambda,B_1\cpcybi B_2)$
and $(B,\inj_1,\inj_2,\proj_1,\proj_2)$ obeying the conditions of
Theorem \ref{proj-cycle}.1 and \ref{proj-cycle}.3
respectively, are in one-to-one correspondence. This correspondence
has been constructed in the proof of Theorem \ref{proj-cycle}.}
\end{remark}
\abs
Theorem \ref{proj-cycle} and the previous remark imply a useful corollary.
Given a normalized cross product bialgebra $B_1\cpcybi B_2$.
We set $\inj_{\alpha\,1}:= \id_{B_1}\otimes\eta_2\,,$ $\inj_{\alpha\,2}:=
\eta_1\otimes\,\id_{B_2}\,,$ $\proj_{\alpha\,1}:=\id_{B_1}\otimes\,\varepsilon_2$ and
$\proj_{\alpha\,2}:=\varepsilon_1\otimes\id_{B_2}$. Then we define
\begin{equation}\mathlabel{m-delta-ijk}
\begin{split}
\m_{\bowtie\,i,jk}&:=\proj_{\alpha\,i}\circ\m_{B_1\cpcybi B_2}\circ
(\inj_{\alpha\,j}\otimes\inj_{\alpha\,k})\,,\\
\Delta_{\bowtie\, ij,k}&:=(\proj_{\alpha\,i}\otimes\proj_{\alpha\,j})\circ
\Delta_{B_1\cpcybi B_2}\circ\inj_{\alpha\,k}
\end{split}
\end{equation}
for $i,j,k\in\{1,2\}$. On the other hand suppose that $B$ is a bialgebra which
obeys the conditions of Theorem \ref{proj-cycle}.3. Then we define
\begin{equation}\mathlabel{m-delta-ijk1}
\m_{B\,i,jk}:=\proj_i\circ\m_B\circ(\inj_j\otimes\inj_k)\,,\quad
\Delta_{B\,ij,k}:=(\proj_i\otimes\proj_j)\circ\Delta_B\circ\inj_k
\end{equation}
for $i,j,k\in\{1,2\}$.
\begin{corollary}\Label{co-act-inv}
Let $B$ be bialgebra and suppose that the tuples
$(B,\Lambda,B_1\cpcybi B_2)$ and
$(B,\inj_1,\inj_2,\proj_1,\proj_2)$ are related according to
Theorem \ref{proj-cycle}. Then $\m_{\bowtie\,i,jk}=\m_{B\,i,jk}$ and
$\Delta_{\bowtie\,ij,k}= \Delta_{B\,ij,k}$.
\end{corollary}
\begin{proof} If the conditions of Theorem \ref{bialg-cocycle2} hold,
then by construction (see the proof of the theorem) the injections and
projections of $B$ and the morphisms $\inj_{\alpha\,1}$, $\inj_{\alpha\,2}$,
$\proj_{\alpha\,1}$ and $\proj_{\alpha\,2}$ are related by
$\inj_1 :=\Lambda\circ\inj_{\alpha\,1}$,
$\inj_2 :=\Lambda\circ\inj_{\alpha\,2}$,
$\proj_1 :=\proj_{\alpha\,1}\circ\Lambda^{-1}$, and
$\proj_2 :=\proj_{\alpha\,2}\circ\Lambda^{-1}$, where
$\Lambda:B_1\otimes B_2\to B$ is the given bialgebra isomorphism.
Thus $\m_{\bowtie\,i,jk}=\m_{B\,i,jk}$ and $\Delta_{\bowtie\,ij,k}=
\Delta_{B\,ij,k}$ follow directly.\end{proof}
\abs
Hence we will use the notation $\m_{i,jk}$ and $\Delta_{ij,k}$
for the corresponding morphisms henceforth.
\subsection{Cocycle Cross Product Bialgebras}
Up to now a (co\n-)modular co-cyclic structure of
cross product bialgebras and their tensor factors $B_1$ and $B_2$ did not
emerge. In Definition \ref{type-alpha} below we will restrict our
considerations to cocycle cross product bialgebras for which (co\n-)modular
co-cyclic structures appear in a very natural way. The universal property
of cocycle cross product bialgebras will be discussed
subsequently. We define strong cross product
bialgebras in Definition \ref{strong-type-alpha}. They are the basic
objects which eventually constitute the universal,
(co\n-)modular co-cyclic theory of cross product bialgebras.
In Definition \ref{cbat} the morphisms $\m_1$,
$\eta_1$, $\varepsilon_1$, $\Delta_2$, $\eta_2$, $\varepsilon_2$, etc.~occur.
Definition \ref{type-alpha} below requires additional morphisms
$\Delta_1$, $\m_2$, $\mu_l$, $\mu_r$, $\nu_l$, $\nu_r$, $\sigma$, and $\rho$.
This is the stage where we find it convenient to start working with
graphical calculus. All $\m$, $\Delta$, $\eta$, $\varepsilon$,
$\mu$, and $\nu$ will be presented by the graphics displayed
in Figure \ref{fig-conv} respectively\footnote{Note that some of these
morphisms might be weak. That is, weak (co\n-)\-mul\-ti\-plications, weak
(co\n-)\-act\-ions, etc.}. The cocycle and cycle
morphisms $\sigma$ and $\rho$ will be presented henceforth graphically
by $\sigma=\begin{tangle}\hh \cu* \end{tangle}:B_2\otimes B_2\to B_1$ and
$\rho=\begin{tangle}\hh \cd* \end{tangle}:B_2\to B_1\otimes B_1$.
\begin{definition}\Label{type-alpha}
A normalized cross product bialgebra $B_1\cpcybi B_2$ is called
cocycle cross product bialgebra if there exist additional morphisms
\begin{equation*}
\begin{gathered}
\Delta_1:B_1\to B_1\otimes B_1\,,\quad\m_2:B_2\otimes B_2\to B_2\,,\\
\mu_l:B_2\otimes B_1\to B_1\,,\quad\mu_r:B_2\otimes B_1\to B_2\,,\\
\nu_l:B_1\to B_2\otimes B_1\,,\quad\nu_r:B_2\to B_2\otimes B_1\,,\\
\sigma:B_2\otimes B_2\to B_1\,,\quad\rho:B_2\to B_1\otimes B_1
\end{gathered}
\end{equation*}
such that $\varphi_{1,2}$, $\varphi_{2,1}$, $\hat\sigma$ and $\hat\rho$
are of the form
\begin{equation*}
\begin{split}
\varphi_{1,2}&=(\m_2\otimes\m_1)\circ(\id_{B_1}\otimes\Psi_{B_1,B_2}\otimes
\id_{B_2})\circ(\nu_l\otimes\nu_r)\,,\\
\varphi_{2,1}&=(\mu_l\otimes\mu_r)\circ(\id_{B_2}\otimes\Psi_{B_2,B_1}
\otimes\id_{B_1})\circ(\Delta_2\otimes\Delta_1)\,,\\
\hat\sigma&=(\sigma\otimes\m_2\circ(\mu_r\otimes\id_{B_2}))\circ
(\id_{B_2}\otimes\Psi_{B_2,B_2}\otimes\id_{B_1\otimes B_2})\circ
(\Delta_2\otimes(\nu_r\otimes\id_{B_2})\circ\Delta_2)\,,\\
\hat\rho&=(\m_1\circ(\id_{B_1}\otimes\mu_l)\otimes\m_1)\circ
(\id_{B_1\otimes B_2}\otimes\Psi_{B_1,B_1}\otimes\id_{B_1})\circ
((\id_{B_1}\otimes\nu_l)\circ\Delta_1\otimes\rho)\,.
\end{split}
\end{equation*}
Furthermore we require the following (co\n-)unital identities
\begin{equation}\mathlabel{type-alpha1}
\begin{gathered}
\Delta_1\circ\eta_1=\eta_1\otimes\eta_2\,,\quad
(\id_{B_1}\otimes\varepsilon_1)\circ\Delta_1=
(\varepsilon_1\otimes\id_{B_1})\circ\Delta_1=\id_{B_1}\,,\\
\varepsilon_2\circ\m_2=\varepsilon_2\otimes\varepsilon_2\,,\quad
\m_2\circ(\id_{B_2}\otimes\eta_2)=\m_2\circ(\eta_2\otimes\id_{B_2})=
\id_{B_2}\,,\\[5pt]
\mu_l\circ(\eta_2\otimes\id_{B_1})=\id_{B_1}\,,\quad
\mu_l\circ(\id_{B_2}\otimes\eta_1)=\eta_1\circ\varepsilon_2\,,\quad
\varepsilon_1\circ\mu_l=\varepsilon_2\otimes\varepsilon_1\,,\\
\mu_r\circ(\id_{B_2}\otimes\eta_1)=\id_{B_2}\,,\quad
\mu_r\circ(\eta_2\otimes\id_{B_1})=\eta_2\circ\varepsilon_1\,,\quad
\varepsilon_2\circ\mu_r=\varepsilon_2\otimes\varepsilon_1\,,\\
(\varepsilon_2\otimes\id_{B_1})\circ\nu_l=\id_{B_1}\,,\quad
(\id_{B_2}\otimes\varepsilon_1)\circ\nu_l=\eta_2\circ\varepsilon_1\,,
\quad\nu_l\circ\eta_1=\eta_2\otimes\eta_1\,,\\
(\id_{B_2}\otimes\varepsilon_1)\circ\nu_r=\id_{B_2}\,,\quad
(\varepsilon_2\otimes\id_{B_1})\circ\nu_r=\eta_1\circ\varepsilon_2\,,
\quad\nu_r\circ\eta_2=\eta_2\otimes\eta_1
\end{gathered}
\end{equation}
and the ``projection" relations
\begin{equation}\mathlabel{type-alpha2}
\begin{split}
(\m_2\otimes\id_{B_1})\circ(\id_{B_2}\otimes\nu_r)&=
(\m_2\otimes\id_{B_1})\circ(\mu_r\otimes\varphi_{1,2})\circ
(\id_{B_2}\otimes(\rho\otimes\id_{B_2})\circ\Delta_2)\,,\\
(\mu_l\otimes\id_{B_1})\circ(\id_{B_2}\otimes\Delta_1)&=
(\m_1\circ(\id_{B_1}\otimes\sigma)\otimes\id_{B_1})\circ
(\varphi_{2,1}\otimes\nu_l)\circ(\id_{B_2}\otimes\Delta_1)\,.
\end{split}
\end{equation}
Cocycle cross product bialgebras will be denoted subsequently
by $B_1\cpcybi[\mathfrak{c}] B_2$.
\end{definition}
\begin{remark}\Label{unit-prop}
{\normalfont For a cocycle cross product bialgebra
the following (co\n-)unital identities can be derived easily.
\begin{equation*}
\begin{gathered}
\Delta_2\circ\eta_2=\eta_2\otimes\eta_2\,,\quad\varepsilon_1\circ\m_1=
\varepsilon_1\otimes\varepsilon_1\,,\\
\varepsilon_1\circ\sigma=\varepsilon_2\otimes\varepsilon_2\,,\quad
\rho\circ\eta_2=\eta_1\otimes\eta_1\,,\\
\sigma\circ(\eta_2\otimes\id_{B_2})=\sigma\circ(\id_{B_2}\otimes\eta_2)
= \eta_1\circ\varepsilon_2\,,\\
(\id_{B_1}\otimes\varepsilon_1)\circ\rho=
(\varepsilon_1\otimes\id_{B_1})\circ\rho=
\eta_1\circ\varepsilon_2\,.
\end{gathered}
\end{equation*}}
\end{remark}
\begin{remark}\Label{alpha-struc}{\normalfont
By definition, cocycle cross product bialgebras always come with
certain structure morphisms $\m_1$, $\m_2$, $\Delta_1$, $\Delta_2$, etc.
There might be two different cocycle cross product bialgebras
with the same underlying bialgebra structure. It is understood
henceforth, that the notation $B_1\cpcybi[\mathfrak{c}] B_2$ always refers to
the complete structure of cocycle cross product bialgebras.}
\end{remark}
\abs
The graphics of the morphisms $\varphi_{1,2}$, $\varphi_{2,1}$,
$\hat\rho$ and $\hat\sigma$ of a cocycle cross product
bialgebra $B_1\cpcybi[\mathfrak{c}] B_1$ are given by
\begin{equation}\mathlabel{phi-sigma-rho}
\begin{array}{c}
\hstretch 125
\vstretch 85
\varphi_{1,2}
:=
\divide\unitlens by 3
\begin{tangle}
\hh\step\id\step\id\\
\hh\ld\step\rd\step\\
\id\step\hx\step\id\\
\hh\cu\step\cu\\
\hh\hstep\id\Step\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\varphi_{2,1}
:=
\divide\unitlens by 3
\begin{tangle}
\hh\hstep\id\Step\id\\
\hh\cd\step\cd\\
\id\step\hx\step\id\\
\hh\lu\step\ru\\
\hh\step\id\step\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\hat\rho
:=
\hstretch 150
\vstretch 85
\divide\unitlens by 3
\begin{tangle}
\hh\hstep\id\Step\id\\
\hh\cd\step\cd*\\
\hh \id\hstep\hld\step\id\step\id\\
\id\hstep\id\hstep\hx\step\id\\
\hh \id\hstep\hlu\step\id\step\id\\
\hh \cu\step\cu\\
\hh\hstep\id\Step\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\hat\sigma
:=
\divide\unitlens by 3
\begin{tangle}
\hh\hstep\id\Step\id\\
\hh\cd\step\cd \\
\hh\id\step\id\step\hrd\hstep\id \\
\id\step\hx\hstep\id\hstep\id \\
\hh\id\step\id\step\hru\hstep\id \\
\hh\cu*\step\cu\\
\hh\hstep\id\Step\id
\end{tangle}
\multiply\unitlens by 3
\end{array}
\end{equation}
\abs
Then the graphical shapes of the multiplication and the
comultiplication look like
\begin{equation}\mathlabel{type-alpha3}
\begin{array}{c}
\m_{B_1\cpcybi[\mathfrak{c}] B_1}
=
\hstretch 150
\divide\unitlens by 3
\begin{tangle}
\hh \id\step\cd\step\cd\step\id \\
\id\step\id\step\hx\step\id\step\id \\
\hh \id\step\lu\step\ru\step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\hh \id\Step\id\hstep\id\step\id\step\hrd\hstep\id \\
\cu\hstep\id\step\hx\hstep\id\hstep\id \\
\hh \step\id\step\hstep\id\step\id\step\hru\hstep\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id
\end{tangle}
\multiply\unitlens by 3
\qquad\text{and}\qquad
\Delta_{B_1\cpcybi[\mathfrak{c}] B_1}
=
\divide\unitlens by 3
\begin{tangle}
\hstep\id\Step\cd \\
\hh \cd\step\cd*\step[1.5]\id \\
\hh \id\hstep\hld\step\id\step\id\hstep\step\id \\
\id\hstep\id\hstep\hx\step\id\hstep\cd \\
\hh \id\hstep\hlu\step\id\step\id\hstep\id\Step\id \\
\hh \cu\step\cu\hstep\id\Step\id \\
\hh \hstep\id\step\ld\step\rd\step\id \\
\hstep\id\step\id\step\hx\step\id\step\id \\
\hh \hstep\id\step\cu\step\cu\step\id \end{tangle}
\multiply\unitlens by 3
\end{array}
\end{equation}
\abs
It is an easy exercise to prove that Definition \ref{type-alpha} is compatible
with trivial (co\n-)actions $\mu_l$, $\mu_r$, $\nu_l$, $\nu_r$ and trivial
(co\n-)cycles $\sigma$ and $\rho$ respectively, in the sense that no
additional identities involving (co\n-)units have to be required for the
remaining morphisms which define this special cocycle
cross product bialgebra.
The following theorem is the central basic statement in Section \ref{sec-ccpb}.
It describes universality of cocycle cross product bialgebras.
\begin{theorem}\Label{bialg-cocycle2}
Let $B$ be a bialgebra in $\C$. Then the following statements are
equivalent.
\begin{enumerate}[1.]
\item
$B$ is bialgebra isomorphic to a cocycle cross product
bialgebra $B_1\cpcybi[\mathfrak{c}] B_2$.
\item
There are idempotents $\Pi_1,\Pi_2\in \End(B)$ such that the
conditions of Theorem \ref{proj-cycle}.\ref{proj-cycle1} and the
following ``projection" relations hold.
\begin{align}
\notag
(\mathbf{\gamma_1})&:&(\id\otimes\Pi_1)\circ\Delta_B
&=\delta_{(1,2,1)}\circ\gamma_1\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B\,,\\
\mathlabel{techn2-cond}
(\mathbf{\gamma_2})&:&\m_B\circ(\Pi_2\otimes\id)
&= \m_B\circ(\Pi_1\otimes\Pi_2)\circ\gamma_2\circ\delta_{(2,1,2)}\,,\\
\notag
(\mathbf{\delta_j})&:
&(\Pi_j\otimes\id)\circ\delta_{(0,0,0)}\circ(\id\otimes\Pi_j)
&=(\Pi_j\otimes\id)\circ\delta_{(0,j,0)}\circ(\id\otimes\Pi_j)
\end{align}
for $j\in\{1,2\}$. In \eqref{techn2-cond} we used the abbreviations
\begin{equation*}
\begin{split}
\gamma_j &:= (\m_B\otimes\m_B)\circ(\id\otimes\Pi_j\otimes\id\otimes\id)
\circ(\id\otimes\Psi_{B,B}\otimes\id)\circ(\Delta_B\otimes\Delta_B)\,,\\
\delta_{(i,j,k)} &:=(\m_B\otimes\id)\circ(\Pi_i\otimes\Pi_j\otimes\Pi_k)
\circ(\id\otimes\Delta_B)\,.
\end{split}
\end{equation*}
for $i,j,k\in\{0,1,2\}$, and we set $\Pi_0:=\id_A$.
\item
There are objects $B_1$ and $B_2$ and
morphisms $B_1\overset{\inj_1}\to B\overset{\proj_1}\to B_1$
and $B_2\overset{\inj_2}\to B\overset{\proj_2}\to B_2$ for which
the conditions of Theorem \ref{proj-cycle}.\ref{proj-cycle2} and the
``projection" relations \eqref{techn2-cond} hold, with
$\Pi_j:=\inj_j\circ\proj_j$, $j\in\{1,2\}$.
\end{enumerate}
\end{theorem}
\begin{proof}
Obviously statement $2.$ and $3.$ are equivalent due to Theorem
\ref{proj-cycle}. Hence suppose statements
$2.$ and $3.$ hold. Then by Theorem \ref{proj-cycle} there
exists a normalized cross product bialgebra $B_1\cpcybi B_2$ which is
isomorphic to $B$ via the isomorphism $\Lambda=\m_B\circ
(\inj_1\otimes\inj_2)$. We define
\begin{equation}\mathlabel{proj-struc}
{\begin{split}
\m_2 &:=\proj_2\circ\m_B\circ(\inj_2\otimes\inj_2)\,,\\
\sigma &:= \proj_1\circ\m_B\circ(\inj_2\otimes\inj_2)\,,\\
\mu_l &:= \proj_1\circ\m_B\circ(\inj_2\otimes\inj_1)\,,\\
\mu_r &:= \proj_2\circ\m_B\circ(\inj_2\otimes\inj_1)\,,
\end{split}}
\quad
{\begin{split}
\Delta_1 &:= (\proj_1\otimes\proj_1)\circ\Delta_B\circ\inj_1\,,\\
\rho &:= (\proj_1\otimes\proj_1)\circ\Delta_B\circ\inj_2\,,\\
\nu_l &:= (\proj_2\otimes\proj_1)\circ\Delta_B\circ\inj_1\,,\\
\nu_r &:= (\proj_2\otimes\proj_1)\circ\Delta_B\circ\inj_2\,.\\
\end{split}}
\end{equation}
The morphisms $\m_1$, $\Delta_2$, $\eta_1$, $\eta_2$, $\varepsilon_1$, and
$\varepsilon_2$ are given in terms of the projections and injections
by the corresponding relations in \eqref{struc-morph} in Corollary
\ref{proj-cycle-eqs}. With the help of these data we define
structure morphisms $\tilde{\varphi}_{1,2}$, $\tilde{\varphi}_{2,1}$,
$\hat{\tilde\sigma}$, and $\hat{\tilde\rho}$ analogous to
\eqref{phi-sigma-rho}. We will show that the morphisms
$\m_{B_1\otimes B_2}$ and $\Delta_{B_1\otimes B_2}$ defined
with these structure morphisms according to \eqref{type-alpha3}
precisely coincide with $\m_{B_1\cpcybi B_2}$ and
$\Delta_{B_1\cpcybi B_2}$ respectively, if the assumptions of
statement $2.$ (or $3.$) are satisfied. Before we prove this
we will provide several auxiliary identities.
From \eqref{pi-rel1} and Corollary \ref{proj-cycle-eqs} we obtain
\begin{gather}\mathlabel{aux1}
{\begin{split}
(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\m_B\circ(\Pi_2\otimes\id_B) &=
(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2\circ\m_B\circ(\Pi_2\otimes\id_B)\\
&= (\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\m_B\,,
\end{split}}
\\
\mathlabel{aux2}
{\begin{split}
\Pi_1\circ\m_B^{(3)}\circ(\Pi_1\otimes\Pi_1\otimes\Pi_1)
&=\Pi_1\circ\m_B\circ(\id_B\otimes\Pi_1\circ\m_B)\circ
(\Pi_1\otimes\Pi_1\otimes\Pi_1)\\
&=\Pi_1\circ\m_B\circ(\id_B\otimes\Pi_1\circ\m_B)\circ
(\Pi_1\otimes\Pi_1\otimes\id_B)\\
&=\Pi_1\circ\m_B^{(3)}\circ(\Pi_1\otimes\Pi_1\otimes\id_B)\,.
\end{split}}
\end{gather}
Furthermore it holds
\begin{equation}\mathlabel{aux3}
(\Pi_1\otimes\id_B)\circ\delta_{(1,0,2)}
= (\Pi_1\otimes\Pi_2)\circ\Delta_B\circ\m_B\circ(\Pi_1\otimes\id_B)
\end{equation}
since the following equations are satisfied.
\begin{equation*}
\begin{split}
\m_B\circ(\Pi_1\otimes\Pi_2)\circ(\m_B\otimes\id_B)\circ(\Pi_1\otimes
\Delta_B)
&= \m_B\circ(\Pi_1\otimes\id_B)\circ(\m_B\circ(\Pi_1\otimes\Pi_1)\otimes
\Pi_2)\circ\\
&\quad\circ(\id_B\otimes\Delta_B)\\
&= \m_B\circ(\Pi_1\otimes\m_B\circ(\Pi_1\otimes\Pi_2)\circ\Delta_B)\\
&= \m_B\circ(\Pi_1\otimes\id_B)\,.
\end{split}
\end{equation*}
Then \eqref{aux3} follows from Theorem \ref{proj-cycle}.\ref{proj-cycle2}.
Observe that Corollary \ref{proj-cycle-eqs} implies
the additional conditions
\eqref{techn2-cond} have not been used to derive \eqref{aux3}.
Subsequently we will prove $\m_{B_1\otimes B_2}=\m_{B_1\cpcybi B_2}$
graphically. Henceforth we use the notation
$\Pi_j=$\hbox to 0.5truecm{\hfil$\divide\unitlens by 2
\begin{tangle}\hh\O{\sstyle j}\end{tangle}$\hfil}
for $j\in\{1,2\}$.
\unitlens 17pt %
\begin{equation}\mathlabel{aux7}
\begin{split}
\m_{B_1\otimes B_2}&=
\quad
\divide\unitlens by 2
\vstretch 60\hstretch 75
\begin{tangle}
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\step[4]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[5]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\\
\id\step\cd\step[2]\cd\step[3]\cd\\
\id\step\O{\sstyle 2}\step[2]\O{\sstyle 2}\step[2]\O{\sstyle 1}%
\step[2]\O{\sstyle 1}\step[3]\O{\sstyle 2}\step[2]\O{\sstyle 2}\\
\id\step\id\step[2]\x\step[2]\id\step[3]\id\step[2]\id\\
\id\step\cu\step[2]\cu\step[3]\id\step[2]\id\\
\id\step[2]\O{\sstyle 1}\step[4]\O{\sstyle 2}\step[4]\id\step[2]\id\\
\cu\step[3]\cd\step[2]\cd\step\id\\
\step\O{\sstyle 1}\step[4]\O{\sstyle 2}\step[2]\O{\sstyle 2}\step[2]%
\O{\sstyle 2}\step[2]\O{\sstyle 1}\step\id\\
\step\id\step[4]\id\step[2]\x\step[2]\id\step\id\\
\step\id\step[4]\cu\step[2]\cu\step\id\\
\step\nw1\step[4]\O{\sstyle 1}\step[4]\O{\sstyle 2}\step[2]\id\\
\step[2]\Cu\step[4]\cu\\
\step[4]{\hstr{90}\vstr{80}\O{\sstyle\proj_1}}\step[7]%
{\hstr{90}\vstr{80}\O{\sstyle\proj_2}}
\end{tangle}
\quad=\quad
\begin{tangle}
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\step[4]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[4]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\\
\id\step\cd\step[2]\cd\step[2]\cd\\
\id\step\id\step[2]\O{\sstyle 2}\step[2]\id%
\step[2]\id\step[2]\O{\sstyle 2}\step[2]\id\\
\id\step\id\step[2]\x\step[2]\id\step[2]\id\step[2]\id\\
\id\step\cu\step[2]\cu\step[2]\id\step\cd\\
\id\step[2]\O{\sstyle 1}\step[4]\id\step[3]\id\step\id\step[2]\id\\
\cu\step[3]\cd\step[2]\id\step\O{\sstyle 1}\step[2]\O{\sstyle 2}\\
\step\id\step[4]\O{\sstyle 2}\step[2]\O{\sstyle 2}\step[2]\id\step\cu\\
\step\id\step[4]\id\step[2]\x\step[2]\id\\
\step\id\step[4]\cu\step[2]\cu\\
\step\nw1\step[4]\O{\sstyle 1}\step[4]\id\\
\step[2]\Cu\step[4]\id\\
\step[4]{\hstr{90}\vstr{80}\O{\sstyle\proj_1}}\step[6]%
{\hstr{90}\vstr{80}\O{\sstyle\proj_2}}
\end{tangle}
\quad =\quad
\begin{tangle}
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\step[4]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[4]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\\
\id\step\cd\step[2]\cd\step[2]\cd\\
\id\step\id\step[2]\id\step[2]\id%
\step[2]\id\step[2]\O{\sstyle 2}\step[2]\id\\
\id\step\id\step[2]\x\step[2]\id\step[2]\id\step[2]\id\\
\id\step\cu\step[2]\cu\step[2]\id\step[2]\id\\
\id\step[2]\O{\sstyle 1}\step[4]\id\step[3]\id\step[2]\id\\
\cu\step[3]\cd\step[2]\id\step[2]\id\\
\step\id\step[4]\O{\sstyle 2}\step[2]\id\step[2]%
\id\step[2]\id\\
\step\id\step[4]\id\step[2]\x\step[2]\id\\
\step\id\step[4]\cu\step[2]\cu\\
\step\nw1\step[4]\O{\sstyle 1}\step[4]\id\\
\step[2]\Cu\step[4]\id\\
\step[4]{\hstr{90}\vstr{80}\O{\sstyle\proj_1}}\step[6]%
{\hstr{90}\vstr{80}\O{\sstyle\proj_2}}
\end{tangle}\\
&= \quad
\divide\unitlens by 2
\vstretch 60\hstretch 75
\begin{tangle}
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[3]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\\
\id\step[2]\cu\step[3]\id\\
\id\step[2]\cd\step[2]\cd\\
\id\step[2]\id\step[2]\id\step[2]\O{\sstyle2}\step[2]\id\\
\id\step[2]\id\step[2]\x\step[2]\id\\
\nw1\step\cu\step[2]\cu\\
\step\cu\step[4]\id\\
\step[2]{\hstr{90}\vstr{80}\O{\sstyle\proj_1}}\step[5]%
{\hstr{90}\vstr{80}\O{\sstyle\proj_2}}
\end{tangle}
\quad = \quad
\begin{tangle}
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_1}}\step[2]%
{\hstr{90}\vstr{80}\O{\sstyle\inj_2}}\\
\id\step[2]\cu\step\dd\\
\id\step[3]\cu\\
\nw1\step[2]\cd\\
\step\cu\step[2]\d\\
\step[2]{\hstr{90}\vstr{80}\O{\sstyle\proj_1}}\step[4]%
{\hstr{90}\vstr{80}\O{\sstyle\proj_2}}
\end{tangle}
\quad = (\proj_1\otimes\proj_2)\circ\Delta_B\circ\m_B^{(4)}\circ
(\inj_1\otimes\inj_2\otimes\inj_1\otimes\inj_2)\\
&=\Lambda^{-1}\circ\m_B\circ(\Lambda\otimes\Lambda)=
\m_{B_1\cpcybi B_2}\,.
\end{split}
\end{equation}
\unitlens 15pt %
To derive the first identity of \eqref{aux7} we used the specific form of
the structure morphisms $\tilde{\varphi}_{1,2}$, $\tilde{\varphi}_{2,1}$,
$\hat{\tilde\sigma}$, and $\hat{\tilde\rho}$.
Then \eqref{pi-rel1}, \eqref{cons-proj-cycle1}, and the third
identity of \eqref{techn2-cond} for $j=1$ yield the second equality
of \eqref{aux7}. With Theorem \ref{proj-cycle}.\ref{proj-cycle1},
\eqref{cons-proj-cycle1} and
\eqref{aux1} we derive the third equation, whereas \eqref{aux2} and
again use of Theorem \ref{proj-cycle}.\ref{proj-cycle1} yield the fourth
identity of \eqref{aux7}. The fifth equality comes from application of the
second ``projection" relation of \eqref{techn2-cond}, and for the derivation
of the sixth identity we used \eqref{aux3}. Finally the definition of
$\Lambda$, given in the proof of Theorem \ref{proj-cycle}, yields the
result. In a $\pi$-symmetric way the identity
$\Delta_{B_1\otimes B_2}=\Delta_{B_1\cpcybi B_2}$ will be shown.
The (co\n-)unital identities \eqref{type-alpha1} can be verified
straightforwardly from the definitions, the assumptions and
Corollary \ref{proj-cycle-eqs}. It remains to prove the ``projection''
relations \eqref{type-alpha2}. Observe that
\begin{equation}\mathlabel{aux4}
\Pi_1=\Lambda\circ(\id_{B_1}\otimes\eta_2\circ\varepsilon_2)
\circ\Lambda^{-1}\,,\quad
\Pi_2=\Lambda\circ(\eta_1\circ\varepsilon_1\otimes\id_{B_2})\circ
\Lambda^{-1}\,.
\end{equation}
Taking into account the relation
$\Delta_B=(\Lambda\otimes\Lambda)\circ\Delta_{B_1\cpcybi B_2}\circ
\Lambda^{-1}$ and the relation $\m_B=\Lambda\circ\m_{B_1\cpcybi B_2}\circ
(\Lambda^{-1}\otimes\Lambda^{-1})$ one obtains with the help of
\eqref{aux4} and \eqref{type-alpha1}
\begin{equation}\mathlabel{aux5}
\begin{split}
\Delta_B\circ\Pi_1 &= (\Lambda\otimes\Lambda\circ(\id_{B_1}\otimes\eta_2))
\circ(\id_{B_1}\otimes\nu_l)\circ\Delta_1\circ(\id_{B_1}\otimes
\varepsilon_2)\circ\Lambda\,,\\
\Pi_1\circ\m_B &=\Lambda\circ(\id\otimes\eta_2)\circ\m_1\circ
(\m_1\otimes\sigma)\circ(\id\otimes\varphi_{2,1}\otimes\id)\circ
(\Lambda^{-1}\otimes\Lambda^{-1})\,.
\end{split}
\end{equation}
Gluing the identities of \eqref{aux5} and $\Pi_1$ according to the
left and right hand side of the third equation of \eqref{techn2-cond}
(for $j=1$), using \eqref{aux4} again, and eventually multiplying both
resulting sides with
$\vert\circ(\Lambda\otimes\Lambda)\circ(\eta_1\otimes\id\otimes\id\otimes
\eta_2)$ and $(\id\otimes\varepsilon_2\otimes\id\otimes\varepsilon_2)\circ
(\Lambda^{-1}\otimes\Lambda^{-1})\circ\vert$ yields
$(\mu_l\otimes\id_{B_1})\circ(\id_{B_2}\otimes\Delta_1)=
(\m_1\circ(\id_{B_1}\otimes\sigma)\otimes\id_{B_1})\circ
(\varphi_{2,1}\otimes\nu_l)\circ(\id_{B_2}\otimes\Delta_1)$
which is the second identity of \eqref{type-alpha2}. Analogously
the first equation of \eqref{type-alpha2} can be derived.
This proves that $B_1\cpcybi B_2$ is cocycle cross product
bialgebra.
Conversely suppose that $\Lambda:B_1\cpcybi[\mathfrak{c}] B_2\to B$ is an isomorphism
of bialgebras. To prove statement $2.$ it suffices to verify relations
\eqref{techn2-cond}. Like in the proof
``\ref{proj-cycle3}.$\Rightarrow$\ref{proj-cycle2}.'' of Theorem
\ref{proj-cycle} we define
$\inj_1 :=\Lambda\circ(\id_{B_1}\otimes\eta_2)$,
$\inj_2 :=\Lambda\circ(\eta_1\otimes\id_{B_2})$,
$\proj_1 :=(\id_{B_1}\otimes\varepsilon_2)\circ\Lambda^{-1}$, and
$\proj_2 :=(\varepsilon_1\otimes\id_{B_2})\circ\Lambda^{-1}$.
Then the identities \eqref{aux5} will be proven for
$\Pi_j:=\inj_j\circ\proj_j$ in the way described above. And the
previously performed gluing of the identities \eqref{aux5} yields
\begin{equation*}
\begin{split}
(\Pi_1\otimes\id)\circ\delta_{(0,0,0)}\circ(\id\otimes\Pi_1) &=
(\Lambda\circ(\id\otimes\eta_2)\otimes\Lambda\circ(\id\otimes\eta_2))\circ
(\m_1\circ(\m_1\otimes\sigma)\otimes\id)\circ\\
&\quad\circ(\id\otimes\varphi_{2,1}\otimes\nu_l)\circ
(\Lambda^{-1}\otimes\Delta_1\circ(\id\otimes\varepsilon_2)\circ\Lambda^{-1})\,,
\\
(\Pi_1\otimes\id)\circ\delta_{(0,1,0)}\circ(\id\otimes\Pi_1) &=
(\Lambda\circ(\id\otimes\eta_2)\otimes\Lambda\circ(\id\otimes\eta_2))\circ
(\m_1\circ(\id\otimes\mu_l)\otimes\id)\circ\\
&\quad\circ(\Lambda^{-1}\otimes\Delta_1\circ
(\id\otimes\varepsilon_2)\circ\Lambda^{-1})\,.
\end{split}
\end{equation*}
Therefore by assumption \eqref{type-alpha2} the third identity
of \eqref{techn2-cond} for $j=1$ follows. Applying $\pi$-symmetry
the third equation of \eqref{techn2-cond} for $j=2$ will be derived.
Then all conditions are satisfied which have been needed to
derive the first four identities in \eqref{aux7}. Hence
\begin{equation*}
\begin{split}
\m_{B_1\cpcybi[\mathfrak{c}] B_2} &=
(\proj_1\circ\m_B^{(3)}\otimes\proj_1\circ\m_B)\circ
(\id^{(2)}\otimes\Psi_{B,B}\circ(\id\otimes\Pi_2)\otimes\id)\circ\\
&\quad\circ(\inj_1\otimes\Delta_B\circ\m_B\circ(\inj_2\otimes\inj_1)\otimes
\Delta_B\circ\inj_2)
\end{split}
\end{equation*}
On the other hand it holds by assumption
$\m_{B_1\cpcybi[\mathfrak{c}] B_2}
=\Lambda^{-1}\circ\m_B\circ(\Lambda\otimes\Lambda)
= (\proj_1\otimes\proj_2)\circ\Delta_B\circ\m_B^{(4)}\circ
(\inj_1\otimes\inj_2\otimes\inj_1\otimes\inj_2)
= (\proj_1\otimes\proj_2)\circ\delta_{(0,0,0)}\circ
(\inj_1\otimes\m_B^{(3)}\circ(\inj_2\otimes\inj_1\otimes\inj_2))$,
where we used \eqref{aux3} which has been derived under the more
general assumption of Theorem \ref{proj-cycle}. Multiplying both
expressions of $\m_{B_1\cpcybi[\mathfrak{c}] B_2}$ with
$\vert\circ(\eta_1\otimes\proj_2\otimes\proj_1\otimes\proj_2)$ and
$(\inj_1\otimes\inj_2)\circ\vert$, and using that $\inj_1$ is algebra
morphism yields
\begin{equation*}
\begin{split}
&(\Pi_1\circ\m_B\otimes\Pi_1\circ\m_B)\circ
(\id\otimes\Psi_{B,B}\circ(\id\otimes\Pi_2)\otimes\id)\circ
(\Delta_B\circ\m_B\circ(\Pi_2\otimes\Pi_1)\otimes\Delta_B\circ\Pi_2)\\
&=(\Pi_1\otimes\Pi_1)\circ\Delta_B\circ\m_B^{(3)}\circ
(\Pi_2\otimes\Pi_1\otimes\Pi_2)
\end{split}
\end{equation*}
from which the second identity of \eqref{techn2-cond} can be derived
easily with the help of Theorem \ref{proj-cycle}.\ref{proj-cycle1}.
Similarly by $\pi$-symmetry the first equation of \eqref{techn2-cond}
will be proven.\end{proof}
\abs
\begin{remark}\Label{yu-cond}
{\normalfont The ``projection" relations $(\gamma_1)$ and
$(\gamma_2)$ in \eqref{techn2-cond} can be derived from the
identities
\begin{equation}\mathlabel{alt-gamma}
(\mathbf{\beta_1}):\ \beta_1=\beta_1\circ(\id\otimes\Pi_1)
\quad\text{and}\quad
(\mathbf{\beta_2}):\ \beta_2=(\Pi_2\otimes\id)\circ\beta_2
\end{equation}
where $\beta_1:=(\Pi_1\circ\m_B\otimes\id)\circ
(\id\otimes\Psi_{B,B})\circ(\Delta_B\circ\Pi_1\otimes\id)$ and
$\beta_2:=(\id\otimes\Pi_2\circ\m_B)\circ
(\Psi_{B,B}\otimes\id)\circ(\id\otimes\Delta_B\circ\Pi_2)$.
However, in general the conditions $(\mathbf{\beta_1})$ and
$(\mathbf{\beta_2})$ are not equivalent to the conditions
$(\mathbf{\gamma_1})$ and $(\mathbf{\gamma_2})$ in
Theorem \ref{bialg-cocycle2}.}
\end{remark}
\begin{proposition}\Label{alg-coalg-triv}
Under the equivalent conditions of Theorem \ref{bialg-cocycle2}
the following statements hold.
\begin{enumerate}
\item
$\mu_l=\varepsilon_2\otimes\id_{B_1}$ is trivial $\Leftrightarrow$
$\Pi_1\circ\m_B\circ(\id_B\otimes\Pi_1)=
\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)$.
\item
$\mu_r=\id_{B_2}\otimes\varepsilon_1$ is trivial
$\Leftrightarrow$ $\Pi_2\circ\m_B=\Pi_2\circ\m_B\circ(\Pi_2\otimes\Pi_2)$
$\Leftrightarrow$ $\proj_2$ is algebra morphism.
\item
$\nu_l=\eta_2\otimes\id_{B_1}$ is trivial $\Leftrightarrow$
$\Delta_B\circ\Pi_1=(\Pi_1\otimes\Pi_1)\circ\Delta_B\circ\Pi_1$
$\Leftrightarrow$ $\inj_1$ is coalgebra morphism.
\item
$\nu_r=\id_{B_2}\otimes\eta_1$ is trivial $\Leftrightarrow$
$(\Pi_2\otimes\id_B)\circ\Delta_B\circ\Pi_2=
(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2$.
\item
$\sigma=\eta_1\circ(\varepsilon_2\otimes\varepsilon_2)$ is trivial
$\Leftrightarrow$ $\hat\sigma=\eta_1\otimes\m_2$ $\Leftrightarrow$
$\m_B\circ(\Pi_2\otimes\Pi_2)=\Pi_2\circ\m_B\circ(\Pi_2\otimes\Pi_2)$
$\Leftrightarrow$ $\inj_2$ is algebra morphism.
\item
$\rho= (\eta_1\otimes\eta_1)\circ\varepsilon_2$ is trivial
$\Leftrightarrow$ $\hat\rho=\Delta_1\otimes\varepsilon_2$ $\Leftrightarrow$
$(\Pi_1\otimes\Pi_1)\circ\Delta_B=(\Pi_1\otimes\Pi_1)\circ\Delta_B\circ\Pi_1$
$\Leftrightarrow$ $\proj_1$ is coalgebra morphism.
\item
$\mu_l$ and $\sigma$ are trivial
$\Leftrightarrow$ $\Pi_1\circ\m_B=\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)$
$\Leftrightarrow$ $\proj_1$ is algebra morphism.
\item
$\nu_r$ and $\rho$ are trivial $\Leftrightarrow$
$\Delta_B\circ\Pi_2=(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2$
$\Leftrightarrow$ $\inj_2$ is coalgebra morphism.
\end{enumerate}
\end{proposition}
\begin{proof}
We prove Proposition \ref{alg-coalg-triv}.1 and \ref{alg-coalg-triv}.5.
The remaining statements can be derived in a similar
manner or follow directly by $\pi$-symmetric reasoning.
Without loss of generality we may assume
$\Pi_1=\id_{B_1}\otimes\eta_2\circ\varepsilon_2$ and
$\Pi_2=\eta_1\circ\varepsilon_1\otimes\id_{B_2}$.
Ad 1.: Suppose that $\mu_l$ is trivial. This means that
$(\id_{B_1}\otimes\varepsilon_2)\circ\varphi_{2,1}=
\varepsilon_2\otimes\id_{B_1}$. Using \eqref{cond2-caat} and the unital
identities of \eqref{cond3-caat} then yields
$\Pi_1\circ\m_B\circ(\id_B\otimes\Pi_1)=
(\m_{B_1}\otimes\eta_2\circ\varepsilon_2)\circ
(\id_{B_1}\otimes\varphi_{2,1}\otimes\varepsilon_2)=
(\m_{B_1}\otimes\eta_2)\circ(\id_{B_1}\otimes\mu_l\otimes\varepsilon_2)=
(\m_{B_1}\otimes\eta_2)\circ(\id_{B_1}\otimes\varepsilon_2\otimes
\id_{B_1}\otimes\varepsilon_2)$. On the other hand from the unital identities of
\eqref{cond3-caat} immediately follows
$\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)=(\m_{B_1}\otimes\eta_2)\circ
(\id_{B_1}\otimes\varepsilon_2\otimes\id_{B_1}\otimes\varepsilon_2)$.
Conversely if $\Pi_1\circ\m_B\circ(\id_B\otimes\Pi_1)=
\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)$ holds then analogous calculations as
before yield $(\m_{B_1}\otimes\eta_2\circ\varepsilon_2)\circ
(\id_{B_1}\otimes\varphi_{2,1}\otimes\varepsilon_2)=(\m_{B_1}\otimes\eta_2)\circ
(\id_{B_1}\otimes\varepsilon_2\otimes\id_{B_1}\otimes\varepsilon_2)$. And then
the triviality of $\mu_l=(\id_{B_1}\otimes\varepsilon_2)\circ\varphi_{2,1})$
follows straightforwardly.
Ad 5.: Suppose that $\m_B\circ(\Pi_2\otimes\Pi_2)=\Pi_2\circ
\m_B\circ(\Pi_2\otimes\Pi_2)$. Then $\m_2=\proj_2\circ\m_B\circ
(\inj_2\otimes\inj_2)$ is associative for as $\m_2\circ(\id\otimes\m_2)=
\proj_2\circ\m_B\circ(\inj_2\otimes\Pi_2\circ\m_B\circ(\inj_2\otimes\inj_2))=
\proj_2\circ\m_B\circ(\id_B\otimes\m_B)\circ(\inj_2\otimes\inj_2\otimes\inj_2)
=\proj_2\circ\m_B\circ(\m_B\otimes\id_B)\circ(\inj_2\otimes\inj_2\otimes\inj_2)
=\m_2\circ(\m_2\otimes\id_{B_2})$. Hence $B_2$ is an algebra. And
$\inj_2\circ\m_2=\Pi_2\circ\m_B\circ(\inj_2\otimes\inj_2)=
\m_B\circ(\inj_2\otimes\inj_2)$. Therefore $\inj_2$ is algebra morphism.
Conversely, if $\inj_2$ is algebra morphism then
$\m_B\circ(\Pi_2\otimes\Pi_2)=\inj_2\circ\m_2\circ(\proj_2\otimes\proj_2)
=\inj_2\circ\proj_2\circ\inj_2\circ\m_2\circ(\proj_2\otimes\proj_2)
=\Pi_2\circ\m_B\circ(\Pi_2\otimes\Pi_2)$.
Let $\inj_2$ be algebra morphism. Then from the last identity of
\eqref{struc-morph} and the second identity of \eqref{cons-proj-cycle1}
we derive $\hat\sigma=(\proj_1\otimes\proj_2)\circ\Delta_B\circ\m_B\circ
(\inj_2\otimes\inj_2)=(\proj_1\otimes\proj_2)\circ\Delta_B\circ\inj_2\circ\m_2
=\eta_1\otimes\m_2$. If on the other hand $\hat\sigma=\eta_1\otimes\m_2$ then
we use \eqref{cond2-caat} and \eqref{cond3-caat} to obtain
$\m_B\circ(\inj_2\otimes\inj_2)=\m_B\circ
(\eta_1\otimes\id_{B_2}\otimes\eta_1\otimes\id_{B_2})=
(\eta_1\otimes\id_{B_2})\circ\m_2=\inj_2\circ\m_2$ which shows that $\inj_2$ is
algebra morphism.
It is an easy exercise to prove that triviality of
$\hat\sigma$ and triviality of $\sigma$ are equivalent.\end{proof}
\begin{remark}{\normalfont
Under the equivalent conditions of Theorem \ref{proj-cycle}
similar results like in Proposition \ref{alg-coalg-triv} can be shown
for general cross product bialgebras if $\m_2$, $\Delta_1$, $\mu_l$,
$\mu_r$, $\nu_l$ and $\nu_r$ will be defined formally as in
\eqref{proj-struc}.}
\end{remark}
\subsection{Strong Cross Product Bialgebras}
In the following we define strong cross product
bialgebras. They will be studied in more detail in Theorems \ref{hp-cp}
and \ref{central-theory} in connection with
so-called strong Hopf data. It turns out that strong
cross product bialgebras are the central objects in our universal,
(co\n-)modular co-cyclic theory of cross product bialgebras
\footnote{strong cross product bialgebras and
strong Hopf data provide a unifying universal and (co\n-)modular
co-cyclic theory of cross product bialgebras. But
they are probably not the most general setting to meet
the same demands. Therefore our notation ``strong'' should be
specified further. However, in order to avoid terminological
blow-up we will henceforth use our notation, always having
in mind that there might be a weaker definition of ``strong''.
This is certainly an interesting direction for further study.}.
\begin{definition}\Label{strong-type-alpha}
A cocycle cross product bialgebra $B_1\cpcybi[\mathfrak{c}] B_2$
is called strong cross product bialgebra if in addition the identities
\begin{gather}\mathlabel{strong-bi1}
\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\hld\step\id\\
\hh\id\hstep\x\\
\hh\hru\step\id\\
\end{tangle}
=
\begin{tangle}
\ld\step\counit
\end{tangle}
\quad ,
\quad
\begin{tangle}
\hh\id\step\hld\\
\hh\x\hstep\id\\
\hh\id\step\hru\\
\end{tangle}
=
\begin{tangle}\unit\step\ru
\end{tangle}
\\
\mathlabel{strong-bi2}
\divide\unitlens by 2
\vstretch 60
\begin{tangle}
\Ld\step\id\\
\rd\step\hx\\
\id\step\hx\step\id\\
\hcu*\step\id\step\id
\end{tangle}
=
\begin{tangle}
\unit\step\unit\step\vstr{40}\id\step\id\\
\vstr{40}\id\step\id\step\id\step\counit
\end{tangle}
\quad ,
\quad
\begin{tangle}
\id\step\id\step\hcd*\\
\id\step\hx\step\id\\
\hx\step\lu\\
\id\step\Ru
\end{tangle}
=
\begin{tangle}
\unit\vstr{40}\step\id\step\id\step\id\\
\vstr{40}\id\step\id\step\counit\step\counit
\end{tangle}
\divide\unitlens by 3
\multiply\unitlens by 4
\vstretch 100
\quad ,
\quad
\vstretch 120
\begin{tangle}
\hh\id\step\hld\\
\hh\cu*\hstep\id
\end{tangle}
=
\vstretch 90
\begin{tangle}
\hh\step\id\\
\hh\counit\step\id\\
\hh\unit\step\id
\end{tangle}
\quad ,
\quad
\vstretch 120
\begin{tangle}
\hh\id\hstep\cd*\\
\hh\hru\step\id
\end{tangle}
=
\vstretch 90
\begin{tangle}
\hh\id\\
\hh\id\step\counit\\
\hh\id\step\unit
\end{tangle}
\\[3pt]
\mathlabel{strong-bi3}
\divide\unitlens by 3
\def5{6}
\vstretch 50
\begin{tangle}
\vstr{70}\id\Step\id\Step\cd*\\
\id\Step\id\step\cd\step\id\\
\id\Step\hx\Step\id\step\id\\
\id\step\dd\ld\step\dd\step\id\\
\hx\step\id\step\hx\Step\id\\
\id\step\d\lu\step\id\Step\id\\
\id\Step\ru\step\id\Step\id
\end{tangle}
=
\vstretch 100
\begin{tangle}
\vstr{70}\id\step\id\step[1.5]\cd*\\
\hh\id\step\id\step\cd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\x\step\x\step[1.5]\id\\
\hh\id\step{\hstr{200}\hru}\step\id\step[1.5]\id
\end{tangle}
\quad,\quad
\vstretch 50
\begin{tangle}
\id\Step\id\step\ld\Step\id\\
\id\Step\id\step\rd\d\step\id\\
\id\Step\hx\step\id\step\hx\\
\id\step\dd\step\ru\dd\step\id\\
\id\step\id\Step\hx\Step\id\\
\id\step\cu\step\id\Step\id\\
\vstr{70}\cu*\Step\id\Step\id
\end{tangle}
=
\vstretch 100
\begin{tangle}
\hh\id\step[1.5]\id\step{\hstr{200}\hld}\step\id\\
\hh\id\step[1.5]\x\step\x\\
\hh\id\step[1.5]\id\step\x\step\id\\
\hh\id\step[1.5]\cu\step\id\step\id\\
\vstr{70}\cu*\step[1.5]\id\step\id
\end{tangle}
\end{gather}
are fullfilled. We denote strong cross product bialgebras
by$\,$\footnote{The deeper meaning of this notation will become clear
in Section \ref{hopf-data}.} $B_1\overset{\mathfrak{h}}\bowtie B_2$.
\end{definition}
\begin{remark}[!]\Label{non-except-bialg}
{\normalfont A tedious calculation shows that the ``projection'' relations
$(\mathbf{\gamma_1})$, $(\mathbf{\gamma_2})$, $(\mathbf{\delta_1})$,
and $(\mathbf{\delta_2})$ in \eqref{techn2-cond} as well as
\eqref{type-alpha2} are redundant for strong cross product
bialgebras. Therefore, if we use the morphisms
$\m_{i,jk}$ and $\Delta_{ij,k}$ from
Corollary \ref{co-act-inv} an alternative
definition of strong cross product bialgebras can be given as follows.
\abs
\emph{A strong cross product bialgebra is a cross product bialgebra
for which the identities \eqref{strong-bi1}, \eqref{strong-bi2} and
\eqref{strong-bi3} formally hold for the corresponding morphisms
$\m_{i,jk}$ and $\Delta_{ij,k}$.}
\abs
We call a strong cross product bialgebra \textit{regular} if all
defining identities have rank
$(2,2)$, $(1,3)$ or $(3,1)$.
We call it \textit{pure} if \eqref{strong-bi1}, \eqref{strong-bi2}
and \eqref{strong-bi3} are redundant.}
\end{remark}
\abs
Any bialgebra $B$ which is isomorphic to a certain cocycle
cross product
bialgebra has injections and projections $\inj_1$, $\inj_2$, $\proj_1$,
and $\proj_2$ which are uniquely determined by the cocycle
cross product bialgebra and the given isomorphism
(see Theorem \ref{bialg-cocycle2} and Remark \ref{alpha-ip}).
In the special case
of strong cross product bialgebras the structure
morphisms obey the additional identities
\eqref{strong-bi1} -- \eqref{strong-bi3}.
Using \eqref{struc-morph} and \eqref{proj-struc} these relations can be
translated easily into equations of the idempotents $\Pi_1$ and $\Pi_2$
of the bialgebra $B$ as follows.
\begin{equation}\mathlabel{pi-strong}
\begin{gathered}
(\Pi_2\otimes\id_B)\circ\Phi_{\id_B}\circ(\Pi_1\otimes\Pi_1)=
(\Pi_2\otimes\id_B)\circ\Delta_B\circ\Pi_1\otimes\varepsilon_B\,,\\[5pt]
((\Pi_1\otimes\Pi_1)\circ\Phi_{\Pi_2}\otimes\id_B)\circ
(\Pi_2\otimes\Psi_{B,B})\circ(\Delta_B\circ\Pi_1\otimes\Pi_2)
=\eta_B\otimes\eta_B\otimes\Pi_1\otimes\varepsilon_B\,,\\[5pt]
(\Pi_1\otimes\id_B)\circ\delta_{(2,2,0)}\circ(\id_B\otimes\Pi_1)=
\eta_B\circ\varepsilon_B\otimes\Pi_1\,,\\[5pt]
\begin{split}
&(\Pi_1\otimes\Pi_2\circ\m_B\circ(\id_B\otimes\Pi_2)\otimes\id_B\otimes\Pi_1)
\circ(\Psi_{B,B}\otimes\Phi_{\Pi_2}\otimes\id_B)\circ\\
&\circ(\id_B\otimes(\Psi_{B,B}\otimes\id_B\otimes\id_B)\circ
(\Pi_1\otimes(\Delta_B\circ\Pi_1\otimes\id_B)\circ\Delta_B\circ\Pi_2))\\
=\,\,&((\Pi_1\otimes\Pi_2\circ\m_B)\circ(\Psi_{B,B}\otimes\id_B)\circ
(\id_B\otimes\Delta_B)\otimes\Pi_1\otimes\Pi_1)\circ\\
&\circ(\id_B\otimes(\Psi_{B,B}\otimes\id_B)\circ(\id_B\otimes(\Pi_1\otimes\id_B)
\circ\Delta_B\circ\Pi_2))
\end{split}\\[5pt]
\text{and the corresponding $\pi$-symmetric counterparts.}
\end{gathered}
\end{equation}
We used $\Phi_f:=(\m_B\otimes\id_B)\circ(f\otimes\Psi_{B,B})\circ
(\Delta_B\otimes\id_B)$ for $f:B\to B$, and $\delta_{(i,j,k)}$ of
Theorem \ref{bialg-cocycle2}.
\abs
Recall Remark \ref{pi-sym} and observe that our construction of
(strong/cocycle) cross product bialgebras is invariant under
$\pi$-symmetry. However, cross product bialgebras in general fail to be
invariant if duality or rotational symmetry along a vertical axis
(in the plane of the paper) will be transformed separately. Without
problems the dual versions of cross product bialgebras can be defined using
our original definition. The corresponding results follow immediately.
If the category $\C$ has (right) duality then the following proposition shows
how dual cross product bialgebras can be constructed explicitely.
\begin{proposition}
Suppose that $\C$ is a braided category with duality functor
$(\_)^\vee:\C\to\C^{\mathrm{op}}_{\mathrm{op}}$ where
$\C^{\mathrm{op}}_{\mathrm{op}}$ is the opposite category with opposite
tensor product. If $(B;B_1,B_2,\inj_1,\inj_2,\proj_1,\proj_2)$ is a
(strong/cocycle) cross product bialgebra in $\C$ then
the dual tuple $(B^\vee;B_1^\vee,B_2^\vee,\inj_1^\vee,\inj_2^\vee,
\proj_1^\vee,\proj_2^\vee)$ is a dual (strong/cocycle)
cross product bialgebra in $\C^{\mathrm{op}}_{\mathrm{op}}$.
\phantom{xxxxx}\endproof
\end{proposition}
\section{Hopf Data}\Label{hopf-data}
In Section \ref{sec-ccpb} we studied cross product bialgebras
from a universal point of view. We answered the question under which
conditions a bialgebra is isomorphic to a cross product bialgebra.
Now we present an explicit (co\n-)-modular co-cyclic
construction method in terms of Hopf data.
A Hopf datum consists of two objects with
certain interrelated (co\n-)modular co-cyclic identities. In Theorem
\ref{hp-cp} we will show that so-called strong Hopf data and
strong cross product bialgebras are different descriptions of
the same objects. Universality of our ``strong'' construction will be
demonstrated in Theorem \ref{central-theory}.
We postpone the lengthy proof of Theorem
\ref{hp-cp} to Section \ref{proof-hp-cp}.
In the subsequent definition of Hopf data occur two objects
$B_1$ and $B_2$, and morphisms
\begin{equation}\mathlabel{hp-morph}
\begin{split}
\m_1 &:B_1\otimes B_1\to B_1\,,\\
\Delta_1 &:B_1\to B_1\otimes B_1\,,\\
\eta_1 &:\E\to B_1\,,\\
\varepsilon_1 &:B_1\to\E\,,\\
\mu_l &:B_2\otimes B_1\to B_1\,,\\
\mu_r &:B_2\otimes B_1\to B_2\,,\\
\sigma &:B_2\otimes B_2\to B_1\,,
\end{split}
\qquad
\begin{split}
\m_2 &:B_2\otimes B_2\to B_2\,,\\
\Delta_2 &:B_2\to B_2\otimes B_2\,,\\
\eta_2 &:\E\to B_2\,,\\
\varepsilon_2 &:B_2\to\E\,,\\
\nu_l &:B_1\to B_2\otimes B_1\,,\\
\nu_r &:B_2\to B_2\otimes B_1\,,\\
\rho &: B_2\to B_1\otimes B_1\,.
\end{split}
\end{equation}
Again we use the graphical presentation of Figure \ref{fig-conv} for these
morphisms, and we represent $\sigma$ and $\rho$
by $\sigma=\begin{tangle}\hh \cu* \end{tangle}:B_2\otimes B_2\to B_1$ and
$\rho=\begin{tangle}\hh \cd* \end{tangle}:B_2\to B_1\otimes B_1$.
Similarly as in \eqref{phi-sigma-rho} we define morphisms
$\varphi_{1,2}$, $\varphi_{2,1}$, $\hat\sigma$, and $\hat\rho$ by
\begin{equation*}
\begin{split}
\varphi_{1,2}&:=(\m_2\otimes\m_1)\circ(\id_{B_1}\otimes\Psi_{B_1,B_2}\otimes
\id_{B_2})\circ(\nu_l\otimes\nu_r)\,,\\
\varphi_{2,1}&:=(\mu_l\otimes\mu_r)\circ(\id_{B_2}\otimes\Psi_{B_2,B_1}
\otimes\id_{B_1})\circ(\Delta_2\otimes\Delta_1)\,,\\
\hat\sigma&:=(\sigma\otimes\m_2\circ(\mu_r\otimes\id_{B_2}))\circ
(\id_{B_2}\otimes\Psi_{B_2,B_2}\otimes\id_{B_1\otimes B_2})\circ
(\Delta_2\otimes(\nu_r\otimes\id_{B_2})\circ\Delta_2)\,,\\
\hat\rho&:=(\m_1\circ(\id_{B_1}\otimes\mu_l)\otimes\m_1)\circ
(\id_{B_1\otimes B_2}\otimes\Psi_{B_1,B_1}\otimes\id_{B_1})\circ
((\id_{B_1}\otimes\nu_l)\circ\Delta_1\otimes\rho)
\end{split}
\end{equation*}
or graphically
\begin{equation}\mathlabel{phi-sigma-rho2}
\begin{array}{c}
\hstretch 125
\vstretch 85
\varphi_{1,2}
:=
\divide\unitlens by 3
\begin{tangle}
\hh\step\id\step\id\\
\hh\ld\step\rd\step\\
\id\step\hx\step\id\\
\hh\cu\step\cu\\
\hh\hstep\id\Step\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\varphi_{2,1}
:=
\divide\unitlens by 3
\begin{tangle}
\hh\hstep\id\Step\id\\
\hh\cd\step\cd\\
\id\step\hx\step\id\\
\hh\lu\step\ru\\
\hh\step\id\step\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\hat\rho
:=
\hstretch 150
\vstretch 85
\divide\unitlens by 3
\begin{tangle}
\hh\hstep\id\Step\id\\
\hh\cd\step\cd*\\
\hh \id\hstep\hld\step\id\step\id\\
\id\hstep\id\hstep\hx\step\id\\
\hh \id\hstep\hlu\step\id\step\id\\
\hh \cu\step\cu\\
\hh\hstep\id\Step\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\hat\sigma
:=
\divide\unitlens by 3
\begin{tangle}
\hh\hstep\id\Step\id\\
\hh\cd\step\cd \\
\hh\id\step\id\step\hrd\hstep\id \\
\id\step\hx\hstep\id\hstep\id \\
\hh\id\step\id\step\hru\hstep\id \\
\hh\cu*\step\cu\\
\hh\hstep\id\Step\id
\end{tangle}
\multiply\unitlens by 3
\end{array}
\end{equation}
Occasionally we also use the graphical abbreviations
\begin{equation}\mathlabel{phi-sigma-rho3}
\begin{array}{c}
\varphi_{1,2}=
\divide\unitlens by 2
\begin{tangle}\ox{12}\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\varphi_{2,1}=
\divide\unitlens by 2
\begin{tangle}\ox{21}\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\hat\rho
=
\divide\unitlens by 2
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox 2{\hat\rho}\\
\hh\hstep\id\step\id
\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\hat\sigma=
\divide\unitlens by 2
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox 2{\hat\sigma}\\
\hh\hstep\id\step\id
\end{tangle}
\multiply\unitlens by 2
\end{array}
\end{equation}
\begin{definition}\Label{hp}
The tuple $\mathfrak{h}=
\big((B_1,\m_1,\eta_1,\Delta_1,\varepsilon_1),(B_2,\m_2,\eta_2,\Delta_2,
\varepsilon_2);\mu_l,\mu_r,\nu_l,\nu_r,\rho,\sigma\big)$ is called Hopf datum if
\begin{enumerate}
\item
$(B_1,\m_1,\eta_1)$ is an algebra and $\varepsilon_1:B_1\to \E$
is an algebra morphism.
\item
$(B_2,\Delta_2,\varepsilon_2)$ is a coalgebra and $\eta_2:\E\to B_2$
is a coalgebra morphism.
\item
$(B_1,\nu_l)$ is left $B_2$-comodule.
\item
$(B_2,\mu_r)$ is right $B_1$-module.
\item The identities
\begin{equation}\mathlabel{hp1}
\begin{gathered}
\Delta_1\circ\eta_1=\eta_1\otimes\eta_2\,,\quad
(\id_{B_1}\otimes\varepsilon_1)\circ\Delta_1=
(\varepsilon_1\otimes\id_{B_1})\circ\Delta_1=\id_{B_1}\,,\\
\varepsilon_2\circ\m_2=\varepsilon_2\otimes\varepsilon_2\,,\quad
\m_2\circ(\id_{B_2}\otimes\eta_2)=\m_2\circ(\eta_2\otimes\id_{B_2})=
\id_{B_2}\,,\\[5pt]
\mu_r\circ(\eta_2\otimes\id_{B_1})=\eta_2\circ\varepsilon_1\,,\quad
\varepsilon_2\circ\mu_r=\varepsilon_2\otimes\varepsilon_1\,,\\
(\id_{B_2}\otimes\varepsilon_1)\circ\nu_l=\eta_2\circ\varepsilon_1\,,
\quad\nu_l\circ\eta_1=\eta_2\otimes\eta_1\,,\\[5pt]
\mu_l\circ(\eta_2\otimes\id_{B_1})=\id_{B_1}\,,\quad
\mu_l\circ(\id_{B_2}\otimes\eta_1)=\eta_1\circ\varepsilon_2\,,\quad
\varepsilon_1\circ\mu_l=\varepsilon_2\otimes\varepsilon_1\,,\\
(\id_{B_2}\otimes\varepsilon_1)\circ\nu_r=\id_{B_2}\,,\quad
(\varepsilon_2\otimes\id_{B_1})\circ\nu_r=\eta_1\circ\varepsilon_2\,,
\quad\nu_r\circ\eta_2=\eta_2\otimes\eta_1\,,\\[5pt]
\sigma\circ(\eta_2\otimes\id_{B_2})=\sigma\circ(\id_{B_2}\otimes\eta_2)=
\eta_1\circ\varepsilon_2\,,\quad
\varepsilon_1\circ\sigma=\varepsilon_2\otimes\varepsilon_2\,,\\
(\id_{B_1}\otimes\varepsilon_1)\circ\rho=
(\varepsilon_1\otimes\id_{B_1})\circ\rho=\eta_1\circ\varepsilon_2\,,\quad
\rho\circ\eta_2=\eta_1\otimes\eta_1
\end{gathered}
\end{equation}
are satisfied.
\item The subsequent compatibility relations hold.
\end{enumerate}
\def5{5}
\begin{equation*}
{\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\step[2]\object{\ }\\
\hh\hcu\hstep\id\\
\hh\hstep\hcu\\
\step\object{B_2}
\end{tangle}
=
\begin{tangle}
\step[2]\object{\ }\\
\hh\id\step\id\step\id\\
\hh\id\hstep\ffbox 2{\hat\sigma}\\
\hh\ru\step\id\\
\cu\\
\step\object{B_2}
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Weak associativity of $\m_2$,}
\end{array}}
\qquad
{\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\hstep\object{B_1}\\
\hh\hcd\\
\hh\id\hstep\hcd\\
\step[2]\object{\ }
\end{tangle}
=
\begin{tangle}
\step\object{B_1}\\
\hstep\cd\\
\hh\hstep\id\step\ld\\
\hh\ffbox 2{\hat\rho}\hstep\id\\
\hh\hstep\id\step\id\step\id\\
\step[2]\object{\ }\\
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Weak coassociativity of $\Delta_1$,}
\end{array}}
\end{equation*}
\begin{equation*}
{\begin{array}{c}
\divide\unitlens by 2
\vstretch 110
\begin{tangle}
\hh\hstep\id\step\id\step\id\\
\hh\ffbox2{\hat\sigma}\hstep\id\\
\hh\hstep\id\step\lu\\
\hstep\cu
\end{tangle}
\vstretch 100
\multiply\unitlens by 2
=
\divide\unitlens by 2
\begin{tangle}
\id\Step\ox{21}\\
\ox{21}\Step\id\\
\id\Step\cu*\\
\hstr{150}\cu
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Weak associativity of $\mu_l$,}
\end{array}}
\qquad
{\begin{array}{c}
\divide\unitlens by 2
\vstretch 110
\begin{tangle}
\cd\\
\hh\rd\step\id\\
\hh\id\hstep\ffbox2{\hat\rho}\\
\hh\id\step\id\step\id
\end{tangle}
\vstretch 100
\multiply\unitlens by 2
=
\divide\unitlens by 2
\begin{tangle}
\step\hstr{150}\cd\\
\cd*\Step\id\\
\id\Step\ox{12}\\
\ox{12}\Step\id
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Weak coassociativity of $\nu_r$,}
\end{array}}
\end{equation*}
\begin{equation*}
{\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\hh\id\hstep\hcu\\
\lu
\end{tangle}
=
\begin{tangle}
\ox{21}\step\id\\
\hh\id\Step\lu\\
\hstr{150}\cu
\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\divide\unitlens by 2
\begin{tangle}
\hh\hcu\hstep\id\\
\hstep\ru
\end{tangle}
=
\begin{tangle}
\id\step\ox{21}\\
\hh\ru\Step\id\\
\hstr{150}\cu
\end{tangle}
\\
\\
\text{Module-algebra compatibility,}
\end{array}}
\qquad
{\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\ld\\
\hh\id\hstep\hcd
\end{tangle}
=
\begin{tangle}
\hstr{150}\cd\\
\hh\id\Step\ld\\
\ox{12}\step\id
\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\divide\unitlens by 2
\begin{tangle}
\hstep\rd\\
\hh\hcd\hstep\id
\end{tangle}
=
\begin{tangle}
\hstr{150}\cd\\
\hh\rd\Step\id\\
\id\step\ox{12}
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Comodule-coalgebra compatibility,}
\end{array}}
\end{equation*}
\begin{equation*}
\begin{array}{c}
\divide\unitlens by 2
\vstretch 110
\begin{tangle}
\hh\hstep\id\step\id\step\id\\
\hh\ffbox2{\hat\sigma}\hstep\id\\
\hh\hstep\id\step\cu*\\
\hh\hstep\hstr{150}\cu
\end{tangle}
\vstretch 100
=
\begin{tangle}
\hh\id\Step\id\step\id\\
\hh\id\step[1.5]\ffbox2{\hat\sigma}\\
\ox{21}\step\id\\
\hh\id\Step\cu*\\
\hstr{125}\cu
\end{tangle}
\quad ,\quad
\vstretch 110
\begin{tangle}
\hh\hstep\hstr{150}\cd\\
\hh\cd*\step\id\\
\hh\id\hstep\ffbox2{\hat\rho}\\
\hh\id\step\id\step\id
\end{tangle}
\vstretch 100
=
\begin{tangle}
\step\hstr{125}\cd\\
\hh\hstep\cd*\Step\id\\
\hstep\id\step\ox{12}\\
\hh\ffbox2{\hat\rho}\step[1.5]\id\\
\hh\hstep\id\step\id\Step\id
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Cocycle and cycle compatibilities,}
\end{array}
\end{equation*}
\begin{equation*}
\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\object{B_1}\step[2]\object{B_1}\\
\cu\\
\cd\\
\object{\ }\step[2]\object{\ }
\end{tangle}
=
\begin{tangle}
\hh \cd\step\cd \\
\hh \id\hstep\hld\step\id\step\id \\
\id\hstep\id\hstep\hx\step\id \\
\hh \id\hstep\hlu\step\id\step\id \\
\hh \cu\step\cu
\end{tangle}
\multiply\unitlens by 2
\quad ,\quad
\divide\unitlens by 2
\begin{tangle}
\object{B_2}\step[2]\object{B_2}\\
\cu\\
\cd\\
\object{\ }\step[2]\object{\ }
\end{tangle}
=
\begin{tangle}
\hh \cd\step\cd \\
\hh \id\step\id\step\hrd\hstep\id \\
\id\step\hx\hstep\id\hstep\id \\
\hh \id\step\id\step\hru\hstep\id \\
\hh \cu\step\cu
\end{tangle}
\multiply\unitlens by 2
\\
\\
\text{Algebra-coalgebra compatibility,}
\end{array}
\end{equation*}
\def5{6}
\begin{equation*}
{\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\ox{21}\\
\hh\dh\step\ddh\\
\hh\ffbox2{\hat\rho}\\
\hh\hstep\id\step\id
\end{tangle}
=
\multiply\unitlens by 2
\divide\unitlens by 3
\begin{tangle}
\step\Cd\step[2]\Cd\\
\cd*\step[2]\cd\step\id\step[3]\ld\\
\vstr{110}\id\step[2]\ox{\sstyle 12}\step[2]\hx\step[2]\dd\step\id\\
\id\step[2]\id\step[2]\x\step[1]\x\step[2]\id\\
\vstr{110}\id\step[2]\ox{\sstyle 21}\step[2]\hx\step[2]\d\step\id\\
\cu\step[2]\cu*\step\id\step[3]\lu\\
\step\Cu\step[2]\Cu
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\divide\unitlens by 2
\begin{tangle}
\hh\hstep\ru\\
\hcd
\end{tangle}
\multiply\unitlens by 2
=
\divide\unitlens by 3
\begin{tangle}
\hcd\step\cd \\
\id\step\hx\step\ld \\
\ru\step\hx\step\id \\
\cu\step\ru
\end{tangle}
\multiply\unitlens by 3
\\
\\
\text{Module-coalgebra compatibility,}
\end{array}}
\quad\qquad
{\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\hstep\hcu\\
\hh\ld
\end{tangle}
\multiply\unitlens by 2
=
\divide\unitlens by 3
\begin{tangle}
\ld\step\cd \\
\id\step\hx\step\ld \\
\ru\step\hx\step\id \\
\cu\step\hcu
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\divide\unitlens by 2
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox2{\hat\sigma}\\
\hh\hdd\step\hd\\
\ox{12}
\end{tangle}
\multiply\unitlens by 2
=
\divide\unitlens by 3
\begin{tangle}
\Cd\step[2]\Cd\\
\rd\step[3]\id\step\cd*\step[2]\cd\\
\vstr{110}\id\step\d\step[2]\hx\step[2]\ox{\sstyle 12}\step[2]\id\\
\id\step[2]\x\step\x\step[2]\id\step[2]\id\\
\vstr{110}\id\step\dd\step[2]\hx\step[2]\ox{\sstyle 21}\step[2]\id\\
\ru\step[3]\id\step\cu\step[2]\cu*\\
\Cu\step[2]\Cu
\end{tangle}
\multiply\unitlens by 3
\\
\\
\text{Comodule-algebra compatibility,}
\end{array}}
\end{equation*}
\begin{equation*}
\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\ox{21}\\
\hh\id\step[2]\id\\
\ox{12}
\end{tangle}
\multiply\unitlens by 2
=
\divide\unitlens by 3
\begin{tangle}
\cd\step\cd \\
\rd\step\hx\step\ld \\
\id\step\hx\step\hx\step\id \\
\ru\step\hx\step\lu \\
\cu\step\cu
\end{tangle}
\multiply\unitlens by 3
\\
\\
\text{Module-comodule compatibility,}
\end{array}
\end{equation*}
\begin{equation*}
\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox2{\hat\sigma}\\
\hh\hstep\id\step\id\\
\hh\ffbox2{\hat\rho}\\
\hh\hstep\id\step\id
\end{tangle}
\multiply\unitlens by 2
=
\divide\unitlens by 3
\begin{tangle}
\step\Cd\step[3]\Cd\\
\cd*\step[2]\cd\step\cd*\step[2]\cd\\
\vstr{110}\id\step[2]\ox{\sstyle 12}\step[2]\hx\step[2]\ox{\sstyle 12}%
\step[2]\id\\
\id\step[2]\id\step[2]\x\step\x\step[2]\id\step[2]\id\\
\vstr{110}\id\step[2]\ox{\sstyle 21}\step[2]\hx\step[2]\ox{\sstyle 21}%
\step[2]\id\\
\cu\step[2]\cu*\step\cu\step[2]\cu*\\
\step\Cu\step[3]\Cu
\end{tangle}
\multiply\unitlens by 3
\\
\\
\text{Cycle-cocycle compatibility.}
\\
\text{(End of Definition \ref{hp})}
\end{array}
\end{equation*}
\end{definition}
\def5{4}
\abs
Observe that Definition \ref{hp} is $\pi$-symmetric invariant.
It will be verified in the next proposition that every cocycle cross
product bialgebra canonically induces a Hopf datum.
\begin{proposition}\Label{cp-hp}
Let $B_1\cpcybi[\mathfrak{c}]B_2$ be a cocycle cross product
bialgebra with corresponding structure morphisms
$\m_1$, $\eta_1$, $\Delta_1$ ,$\varepsilon_1$, $\m_2$, $\eta_2$, $\Delta_2$,
$\varepsilon_2$, $\mu_l$, $\mu_r$, $\nu_l$, $\nu_r$, $\rho$, and $\sigma$.
Then $\big((B_1,\m_1,\eta_1,\Delta_1,\varepsilon_1),
(B_2,\m_2,\eta_2,\Delta_2,\varepsilon_2);
\mu_l,\mu_r,\nu_l,\nu_r,\rho,\sigma\big)$ is a Hopf datum.
\end{proposition}
\begin{proof}
By definition cocycle cross product bialgebras are
cocycle cross product algebras and cycle cross product coalgebras
in particular. Hence $(B_1,\m_1\eta_1)$ is an algebra and
$(B_2,\Delta_2,\varepsilon_2)$
is a coalgebra. Since cocycle cross product bialgebras are
normalized one deduces with \eqref{type-alpha1} that
$\eta_2:\E\to B_2$ is an algebra morphism and
$\varepsilon_1:B_1\to \E$ is a coalgebra morphism. Then the conditions
of Definition \ref{hp}.1 and \ref{hp}.2 hold. The first sixteen
(co\n-)\-unital identities of \eqref{hp1} hold by assumption
(see \eqref{type-alpha1}). Then from \eqref{phi-sigma-rho} one easily derives
\begin{equation}\mathlabel{cp-hp1}
{\begin{split}
\varphi_{1,2}\circ(\eta_1\otimes\id_{B_2})&=\nu_r\,,\\
\varphi_{1,2}\circ(\id_{B_1}\otimes\eta_2)&=\nu_l\,,
\end{split}}
\quad
{\begin{split}
(\varepsilon_1\otimes\id_{B_2})\circ\varphi_{2,1}&=\mu_r\,,\\
(\id_{B_1}\otimes\varepsilon_2)\circ\varphi_{2,1}&=\mu_l\,,
\end{split}}
\quad
{\begin{split}
\hat\rho\circ(\eta_1\otimes\id_{B_1})&=\rho\,\\
(\id_{B_2}\otimes\varepsilon_2)\circ\hat\sigma&=\sigma\,.
\end{split}}
\end{equation}
Since $B_1\cpcybi[\mathfrak{c}] B_2$ is especially cocycle cross product
algebra, the identities \eqref{cond3-caat} are satisfied. Composing
the fifth equation of \eqref{cond3-caat} with
$(\varepsilon_1\otimes\id_{B_2})\circ\vert$ and using
\eqref{cp-hp1} proves that $(B_2,\mu_r)$ is right $B_1$-module.
Similarly the weak associativity of $\mu_l$ is shown by composing
with $(\id_{B_1}\otimes\varepsilon_2)\circ\vert$. From \eqref{cp-hp1}
and \eqref{cond3-caat} it will be concluded straightforwardly that
$\sigma\circ(\eta_2\otimes\id_{B_2}) =(\id_{B_1}\otimes\varepsilon_2)
\circ\hat\sigma\circ(\eta_2\otimes\id_{B_2})
=(\id_{B_1}\otimes\varepsilon_2)\circ(\eta_1\otimes\id_{B_2})
=\eta_1\circ\varepsilon_2$
and similarly $\sigma\circ(\eta_2\otimes\id_{B_2})=\eta_1\circ\varepsilon_2$.
Since $B_1\cpcybi[\mathfrak{c}] B_2$ is cross product bialgebra,
$\varepsilon_1\otimes\varepsilon_2$ is algebra morphism, and
therefore one easily shows that $\varepsilon_1\circ\sigma=
\varepsilon_2\otimes\varepsilon_2$. Eventually all other (co\n-)unital
identities of \eqref{hp1} follow then by $\pi$-symmetry.
The weak associativity for $\m_2$ follows now from the sixth equation
of \eqref{cond3-caat} by adjoining $(\varepsilon_1\otimes\id_{B_2})\circ\vert$
on both sides. The module-algebra compatibility will be derived
from the fifth equation of \eqref{cond3-caat} through
composition with $(\id_{B_1}\otimes\varepsilon_2)\circ\vert$.
Making use of the bialgebra identity $\Delta_B\circ\m_B=
(\m_B\otimes\m_B)\circ(\id_B\otimes\Psi_{B,B}\otimes\id_B)\circ
(\Delta_B\otimes\Delta_B)$ for $B=B_1\cpcybi[\mathfrak{c}] B_2$ we prove
the algebra-coalgebra compatibility by application of
$\vert\circ(\id_{B_1}\otimes\eta_2\otimes\id_{B_1}\otimes\eta_2)$
and $(\id_{B_1}\otimes\id\varepsilon_2\otimes\id_{B_1}\otimes
\varepsilon_2)\circ\vert$ on both sides of the bialgebra identity.
The first equation of the module-coalgebra compatibility will be shown
similarly by composing with $\vert\circ(\eta_1\otimes\id_{B_2}\otimes
\id_{B_1}\otimes\eta_2)$ and ${(\varepsilon_1\otimes\id_{B_2}\otimes
\varepsilon_1\otimes\id_{B_2})\circ\vert}$. Application of
$\vert\circ(\eta_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\eta_2)$
and $(\id_{B_1}\otimes\varepsilon_2\otimes\id_{B_1}\otimes
\varepsilon_2)\circ\vert$ yields the second equation of
the module-coalgebra compatibility. The module-comodule compatibility
is derived from the bialgebra identity by composing with
$\vert\circ(\eta_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\eta_2)$
and $(\varepsilon_1\otimes\id_{B_2}\otimes\id_{B_1}\otimes\varepsilon_2)
\circ\vert$, whereas the cycle-cocycle compatibility
comes from composition with
$\vert\circ(\eta_1\otimes\id_{B_2}\otimes\eta_1\otimes\id_{B_2})$
and $(\id_{B_1}\otimes\varepsilon_2\otimes\id_{B_1}\otimes
\varepsilon_2)\circ\vert$. Application of
$(\id_{B_1}\otimes\varepsilon_2)\circ\vert$ to the sixth equation of
\eqref{cond3-caat} and use of the sixth identity of \eqref{cp-hp1}
yield the cocycle compatibility.
All remaining
compatibility relations of Definition \ref{hp} will be proven by
application of $\pi$-symmetry to the former results.\end{proof}
\subsection{Strong Hopf Data}
The converse of Proposition \ref{cp-hp} is not true in general.
However, in the following we show that so-called strong Hopf
data yield cross product bialgebras. The definition of strong Hopf data
is closely related to the definition of strong cross
product bialgebras.
\begin{definition}\Label{strong-hp}
A Hopf datum $\mathfrak{h}$ is called strong
if in addition the following identities hold.
\begin{gather}\mathlabel{act-coact-triv}
\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\hld\step\id\\
\hh\id\hstep\x\\
\hh\hru\step\id\\
\end{tangle}
=
\begin{tangle}
\ld\step\counit
\end{tangle}
\quad ,
\quad
\begin{tangle}
\hh\id\step\hld\\
\hh\x\hstep\id\\
\hh\id\step\hru\\
\end{tangle}
=
\begin{tangle}\unit\step\ru
\end{tangle}
\\
\mathlabel{cocycle-triv1}
\divide\unitlens by 2
\vstretch 60
\begin{tangle}
\Ld\step\id\\
\rd\step\hx\\
\id\step\hx\step\id\\
\hcu*\step\id\step\id
\end{tangle}
=
\begin{tangle}
\unit\step\unit\step\vstr{40}\id\step\id\\
\vstr{40}\id\step\id\step\id\step\counit
\end{tangle}
\quad ,
\quad
\begin{tangle}
\id\step\id\step\hcd*\\
\id\step\hx\step\id\\
\hx\step\lu\\
\id\step\Ru
\end{tangle}
=
\begin{tangle}
\unit\vstr{40}\step\id\step\id\step\id\\
\vstr{40}\id\step\id\step\counit\step\counit
\end{tangle}
\vstretch 100
\divide\unitlens by 3
\multiply\unitlens by 4
\quad ,
\quad
\vstretch 120
\begin{tangle}
\hh\id\step\hld\\
\hh\cu*\hstep\id
\end{tangle}
=
\vstretch 90
\begin{tangle}
\hh\step\id\\
\hh\counit\step\id\\
\hh\unit\step\id
\end{tangle}
\quad ,
\quad
\vstretch 120
\begin{tangle}
\hh\id\hstep\cd*\\
\hh\hru\step\id
\end{tangle}
=
\vstretch 90
\begin{tangle}
\hh\id\\
\hh\id\step\counit\\
\hh\id\step\unit
\end{tangle}
\\[3pt]
\mathlabel{strong-hopf-combo}
\divide\unitlens by 3
\def5{6}
\vstretch 50
\begin{tangle}
\vstr{70}\id\Step\id\Step\cd*\\
\id\Step\id\step\cd\step\id\\
\id\Step\hx\Step\id\step\id\\
\id\step\dd\ld\step\dd\step\id\\
\hx\step\id\step\hx\Step\id\\
\id\step\d\lu\step\id\Step\id\\
\id\Step\ru\step\id\Step\id
\end{tangle}
=
\vstretch 100
\begin{tangle}
\vstr{70}\id\step\id\step[1.5]\cd*\\
\hh\id\step\id\step\cd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\x\step\x\step[1.5]\id\\
\hh\id\step{\hstr{200}\hru}\step\id\step[1.5]\id
\end{tangle}
\quad ,
\quad
\vstretch 50
\begin{tangle}
\id\Step\id\step\ld\Step\id\\
\id\Step\id\step\rd\d\step\id\\
\id\Step\hx\step\id\step\hx\\
\id\step\dd\step\ru\dd\step\id\\
\id\step\id\Step\hx\Step\id\\
\id\step\cu\step\id\Step\id\\
\vstr{70}\cu*\Step\id\Step\id
\end{tangle}
=
\vstretch 100
\begin{tangle}
\hh\id\step[1.5]\id\step{\hstr{200}\hld}\step\id\\
\hh\id\step[1.5]\x\step\x\\
\hh\id\step[1.5]\id\step\x\step\id\\
\hh\id\step[1.5]\cu\step\id\step\id\\
\vstr{70}\cu*\step[1.5]\id\step\id
\end{tangle}
\end{gather}
\end{definition}
\abs
Strong Hopf data are invariant under $\pi$-symmetry.
This fact will be used extensively in Section \ref{proof-hp-cp} where we prove
Theorem \ref{hp-cp}.
\begin{remark}\Label{non-except-hopf}
{\normalfont Observe that except of \eqref{strong-hopf-combo}
and the first and the second identity of \eqref{cocycle-triv1} the
defining identities of a strong Hopf datum are either of rank
$(2,2)$, $(1,3)$ or $(3,1)$. Therefore
in the same way as in Remark \ref{non-except-bialg}
we call a strong Hopf datum \textit{regular} if the
defining identities are of either rank $(2,2)$, $(1,3)$ or $(3,1)$.
We call the strong Hopf datum \textit{pure} if \eqref{act-coact-triv},
\eqref{cocycle-triv1} and \eqref{strong-hopf-combo} are redundant.}
\end{remark}
\subsection{Main Results}
This is the central part of the article. In the subsequent
Theorems \ref{hp-cp} and \ref{central-theory} we present
the universal (co\n-)modular co-cyclic theory of (strong) cross
product bialgebras. In Theorem \ref{hp-cp} we describe the (co\n-)modular
co-cyclic construction of strong cross product bialgebras.
Theorem \ref{central-theory} exhibits its universality.
\begin{theorem}\Label{hp-cp}
Let $\mathfrak{h}=
\big((B_1,\m_1,\eta_1,\Delta_1,\varepsilon_1),(B_2,\m_2,\eta_2,\Delta_2,
\varepsilon_2);\mu_l,\mu_r,\nu_l,\nu_r,\rho,\sigma\big)$ be a strong
Hopf datum. Then $B_1\otimes B_2$ with $\varphi_{1,2}$,
$\varphi_{2,1}$, $\hat\sigma$, and $\hat\rho$ defined according to
\eqref{phi-sigma-rho2} is a strong cross product bialgebra
$B_1\overset{\mathfrak{h}}\bowtie B_2$. Strong Hopf data and strong
cross product bialgebras are in one-to-one correspondence.
\endproof
\end{theorem}
\abs
Every strong Hopf datum $\mathfrak{h}$ therefore induces a strong
cross product bialgebra $B_1\overset{\mathfrak{h}}\bowtie B_2$. For any
bialgebra $B$ which is isomorphic to $B_1\overset{\mathfrak{h}}\bowtie B_2$ the
additional relations \eqref{act-coact-triv} -- \eqref{strong-hopf-combo} (or by
one-to-one correspondence the equations \eqref{strong-bi1} --
\eqref{strong-bi3}) imply the additional relations \eqref{pi-strong} for the
idempotents $\Pi_1$ and $\Pi_2$ of $B$.
\begin{theorem}\Label{central-theory}
Let $B$ be a bialgebra in $\C$. Then the following statements are
equivalent.
\begin{entry}
\item[1.]
There is a strong Hopf datum $\mathfrak{h}$ such that the
corresponding strong cross product
bialgebra $B_1\overset{\mathfrak{h}}\bowtie B_2$
is bialgebra isomorphic to $B$.
\item[2.]
There are idempotents $\Pi_1,\Pi_2\in \End(B)$ such that
the conditions of Theorem \ref{proj-cycle}.2 and the
``strong projection'' relations \eqref{pi-strong} hold.
\item[3.]
There are objects $B_1$ and $B_2$ and
morphisms $B_1\overset{\inj_1}\to B\overset{\proj_1}\to B_1$
and $B_2\overset{\inj_2}\to B\overset{\proj_2}\to B_2$ such that
the conditions of Theorem \ref{proj-cycle}.3 and
the ``strong projection'' relations \eqref{pi-strong} hold
with $\Pi_j:= \inj_j\circ\proj_j$.
\end{entry}
\end{theorem}
\begin{proof}
Theorem \ref{central-theory} can be derived
straightforwardly from Theorem \ref{hp-cp}, Theorem \ref{proj-cycle},
(Theorem \ref{bialg-cocycle2}), Remark \ref{alpha-ip},
Remark \ref{non-except-bialg} and the relations \eqref{pi-strong}.\end{proof}
\abs
\abs
In the sequel we will consider special versions of strong cross
product bialgebras. They admit equivalent universal and (co\n-)modular
co-cyclic descriptions and can be derived from the most general
construction given in Theorem \ref{hp-cp}.
In particular all known constructions of cocycle cross product bialgebras
will be recovered, and additionally we find several new types of cocycle
cross product bialgebras. In Proposition \ref{bialg-cocycle6}
we describe comprehensively the universal and (co\n-)modular
co-cyclic properties of one of these special types.
Theorem \ref{hp-cp} and Proposition \ref{alg-coalg-triv}
provide the necessary tools to describe the various subclasses.
We distinguish the different special versions of strong
cross product bialgebras with the help of the boxes
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\mu_l}\ffbox1{\normalfont\mu_r}\\
\hh\ffbox1{\normalfont\sigma}\ffbox1{\normalfont\rho}\\
\hh\ffbox1{\normalfont\nu_l}\ffbox1{\normalfont\nu_r}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3$
where the particular entries will be left blank ``$\ $" or take the values
$\bullet$ for the (weak) (co\n-)actions, $\blacktriangledown$ for $\sigma$ and
$\blacktriangle$ for $\rho$ dependent on the respective morphisms
are trivial or not. For example a strong cross product bialgebra
with trivial cocycle $\sigma=\eta_1\circ(\varepsilon_2\otimes\varepsilon_2)$
and trivial right coaction $\nu_r=\id_{B_2}\otimes\eta_1$ will be
represented by the classification box
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\bullet}\\
\hh\ffbox1{\normalfont\ }\ffbox1{\normalfont\blacktriangle}\\
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\ }
\end{tangle}\,$.
\begin{figure}
\begin{equation*}
\unitlens 8pt
\providecommand{\0}{\ffbox1{}}
\providecommand{\9}{\ffbox1{\bullet}}
\providecommand{\7}{\ffbox1{\blacktriangledown}}
\providecommand{\8}{\ffbox1{\blacktriangle}}
\providecommand{\4}[1]{\Put(0,5)[lt]{\begin{tangle}\vstr{250}\hstr{1#100}%
\nw1\end{tangle}}}
\providecommand{\5}{\Put(0,0)[cc]{\vdots}}
\providecommand{\morelines}[3]{\Put(#2,#3)[cc]{\text{#1 more lines}}}
\begin{tangles}{ccccc}
\HH\9\9 &\step[3]& &\step[3]&\\
\HH\7\8 && &&\\
\HH\9\9 && &&\\
\vstr{200}\step[5]\hstr{200}\dd\d\step[-1]{\hstr{700}\nw3} && &&\\
\HH\9\9\Step\9\9 &&\9\9 &&\\
\HH\7\8\Step\7\8 &&\7\0 &&\\
\HH\9\0\Step\0\9 &&\9\9 &&\\
\vstr{200}\hstr{400}\dd\id\X\id\d &\morelines{3}{5}{20}& %
\vstr{200}\hstr{200}\step[2]\sw3\step\dd\d\step[-1]\nw3%
\step[-1]{\hstr{600}\nw3}\Step&&\\
\HH\0\9\Step\9\9\Step\9\0\Step\9\0 && %
\9\0\Step\9\9\Step\9\9\Step\0\9 &&\9\9\\
\HH\7\8\Step\7\8\Step\7\8\Step\7\8 &&%
\7\0\Step\7\0\Step\7\0\Step\7\0 && \0\0\\
\HH\9\0\Step\0\0\Step\9\0\Step\0\9 &&%
\9\9\Step\0\9\Step\9\0\Step\9\9 && \9\9\\
\vstr{200}\hstr{400}\d\id\X\id\dd%
&\morelines{4}{0}{20}& \vstr{200}\hstr{400}\sw1\step[-1]\sw2\id\sw2%
\vstr{200}\step[-1]\nw2\id\sw1\step[-1]\nw2\sw1\step[-1]\nw2\id\nw1
&\morelines{4}{10}{20}&\vstr{200}\id\\
\HH\0\9\Step\9\0 &&\9\0\Step\9\0\Step\9\9\Step\0\0\Step\0\9\Step\0\9 &&\0\9\\
\HH\7\8\Step\7\8 &&\7\0\Step\7\0\Step\7\0\Step\7\0\Step\7\0\Step\7\0 &&\0\0\\
\HH\0\0\Step\0\0 &&\0\9\Step\9\0\Step\0\0\Step\9\9\Step\0\9\Step\9\0 &&\9\9\\
\vstr{200}\hstr{200}\d\dd &\morelines{4}{-20}{20}&\vstr{200}\hstr{400}\nw1\step[-1]\nw2%
\id\sw1\step[-1]\nw2\XX\step[-2]\sw1\id\sw2\id\sw1 &\morelines{6}{-4}{20}& %
\vstr{200}\sw3\Step\id\nw3\Step\\
\HH\0\0 && \9\0\Step\0\0\Step\0\0\Step\0\9 && \0\0\step\0\9\step\0\9\\
\HH\7\7 && \hh\7\0\Step\7\0\Step\7\0\Step\7\0 && \hh\0\0\step\0\0\step\0\0\\
\HH\0\0 && \0\0\Step\0\9\Step\9\0\Step\0\0 && \9\9\step\9\0\step\0\9\\%
\vstr{200}\step[6]\hstr{720}\nw3&&\vstr{200}\hstr{200}\nw3\step\d\dd%
\step[-1]\sw3\Step &\morelines{4}{-30}{20}&\vstr{200}\nw3\Step\id\sw3\Step\\
\HH && \0\0 && \0\0\\
\HH && \hh\7\0 && \hh\0\0\\
\HH && \0\0 && \0\9\\
&& && \step[-5.5]{\hstr{640}\se3}\step[-1]\id\\
\HH && && \0\0\\
\HH && && \0\0\\
\HH && && \0\0
\end{tangles}
\end{equation*}
\caption{Graph representing the various special versions of strong
cross product bialgebras.}\Label{fig-class}
\end{figure}
\unitlens 12pt
Thereby $2^6=64$ types of special strong cross products can be
obtained; $2^4=16$ of them are co-cycle free. In a similar (dual) way
the dual strong cross product bialgebras will be denoted
by a classification box
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\mu_l}\ffbox1{\normalfont\mu_r}\\
\hh\ffbox1{\normalfont\sigma}\ffbox1{\normalfont\rho}\\
\hh\ffbox1{\normalfont\nu_l}\ffbox1{\normalfont\nu_r}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3$
with entries $\bullet$ for the (co\-)actions, $\blacktriangle$ for the
cocycle $\sigma$, and $\blacktriangledown$ for the cycle $\rho$.
Then analogously $64$ special types of dual strong cross
product bialgebras can be obtained from which $2^4=16$
(co\n-cycle free) types coincide with the corresponding 16 types
of strong cross product bialgebras.
Therefore the total number of different
types of strong and dual strong cross product
bialgebras which is classified by our scheme is $2^6+(2^6-2^4)=7\cdot 2^4=112$.
However, henceforth we do not tell between a certain type and its dual or
$\pi$-symmetric counterparts - it is an easy exercise to obtain one from the
other\footnote{Observe that a $\pi$-rotation of the classification box in the
plane of the graphic yields the corresponding $\pi$-symmetric counterpart
of a certain type of cross product bialgebra. And analogously a reflection of
the classification box along a horizontal line yields the dual counterpart.}.
Up to this $\mathbb{Z}_2\times\mathbb{Z}_2$-symmetry we obtain $33$ different
classification boxes.
In Figure \ref{fig-class} we present a graph where the boxes
(modulo $\mathbb{Z}_2\times\mathbb{Z}_2$-symmetry) are the vertices, and
each box is linked with its descending ``next neighbour" special versions.
We stratify the whole graph into three layers. The layers consist of boxes
with 2, 1 or 0 co-cycles respectively.
The most general type of strong cross product bialgebra
(Theorem \ref{hp-cp}) is at the top of the graph in the first row.
In the second row all descendants with one trivial morphism
$\mu_l$, $\mu_r$, $\sigma$, $\rho$, $\nu_l$, or $\nu_r$ are listed,
in the third row all special types with two trivial morphisms are listed, etc.
The cross product bialgebra constructions from
\cite{Rad1:85,Ma1:90,MS:94,Ma6:94,BD1:98} are descendants of the special cases
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\bullet}\\
\hh\ffbox1{\normalfont\ }\ffbox1{\normalfont\ }\\
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\bullet}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3\,$
and
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\ }\\
\hh\ffbox1{\normalfont\blacktriangledown}%
\ffbox1{\normalfont\blacktriangle}\\
\hh\ffbox1{\normalfont\ }\ffbox1{\normalfont\bullet}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3\,$
which are the co-cycle free cross product bialgebras of \cite{BD1:98}
and the braided versions of bicrossed product bialgebras
\cite{Ma6:94,MS:94} respectively\footnote{A special symmetric version of
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\bullet}\\
\hh\ffbox1{\normalfont\blacktriangledown}\ffbox1{\normalfont\ }\\
\hh\ffbox1{\normalfont\ }\ffbox1{\normalfont\bullet}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3$ has been studied recently in \cite{Sbg1:98}.}.
In the following tables we describe the different types of cocycle cross
product bialgebras in more detail; we omit the description of the
various cross product bialgebras studied in
\cite{Rad1:85,Ma1:90,MS:94,Ma6:94,BD1:98}.
In the second column of the tables the
structure of the multiplication $\m_B$ and the comultiplication $\Delta_B$
is given. In the third column we recall the corresponding equivalent
properties of Proposition \ref{alg-coalg-triv}, and in the last column
the status of the strong cross product bialgebra is listed.
\abs
\unitlens 15pt
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline Type&Structure morphisms $\m_B$, $\Delta_B$
&Proposition \ref{alg-coalg-triv}&Status\\
\hline &&&\\[-12pt]
\hline $\divide\unitlens by 2\begin{tangle}\classbox 111111\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{\nu_l}{\nu_r}\sigma\rho
{\hh\id\step\cd\step\cd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\lu\step\ru\step\id\\
\hh\id\Step\id\hstep\cd\step\cd\\
\hh\id\Step\id\hstep\id\step\id\step\hrd\hstep\id\\
\cu\hstep\id\step\hx\hstep\id\hstep\id\\
\hh\step\id\step\hstep\id\step\id\step\hru\hstep\id\\
\hh\step\id\step[1.5]\cu*\step\cu\\
\step\cu\Step\id}
{\hstep\id\Step\cd\\
\hh\cd\step\cd*\step[1.5]\id\\
\hh \id\hstep\hld\step\id\step\id\hstep\step\id\\
\id\hstep\id\hstep\hx\step\id\hstep\cd\\
\hh\id\hstep\hlu\step\id\step\id\hstep\id\Step\id\\
\hh\cu\step\cu\hstep\id\Step\id\\
\hh\hstep\id\step\ld\step\rd\step\id\\
\hstep\id\step\id\step\hx\step\id\step\id\\
\hh\hstep\id\step\cu\step\cu\step\id}$
&
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111110\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{\nu_l}{}\sigma\rho
{\hh \id\step\cd\step\cd\step\id \\
\id\step\id\step\hx\step\id\step\id \\
\hh \id\step\lu\step\ru\step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\cu\hstep\id\step\hx\step\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\cd\step[1.5]\hstr{225}\hcd\\
\id\step\ld\step\hcd*\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\id\step\lu\step\hcu\step\id\\
\hh{\hstr{300}\cu}\hstep\ld\step\hcd\\
\hh\step\id\step[1.5]\id\step\x\step\id\\
\hh\step\id\step[1.5]\hcu\step\id\step\id}$
&\ref{alg-coalg-triv}.4
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111101\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{}{\nu_r}\sigma\rho
{\hh \id\step\cd\step\cd\step\id \\
\id\step\id\step\hx\step\id\step\id \\
\hh \id\step\lu\step\ru\step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\hh \id\Step\id\hstep\id\step\id\step\hrd\hstep\id \\
\cu\hstep\id\step\hx\hstep\id\hstep\id \\
\hh \step\id\step\hstep\id\step\id\step\hru\hstep\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\hstep\id\Step\cd \\
\hh \cd\step\cd*\step[1.5]\id \\
\id\step\hx\step\id\hstep\cd \\
\hh \cu\step\cu\hstep\id\Step\id \\
\hh \hstep\id\Step\id\step\rd\step\id \\
\hstep\id\Step\hx\step\id\step\id \\
\hh \hstep\id\Step\id\step\cu\step\id}$
&\ref{alg-coalg-triv}.3
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111011\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{\nu_l}{\nu_r}{}\rho
{\hh\id\step\cd\step\cd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\lu\step\ru\step\id\\
\hh\id\Step\id\hstep\cd\step\cd\\
\hh\id\Step\id\hstep\id\step\id\step\hrd\hstep\id\\
\cu\hstep\id\step\hx\hstep\id\hstep\id\\
\hh\step\id\step[1.5]\id\step\id\step\hru\hstep\id\\
\hh\step\id\step[1.5]\cu*\step\cu\\
\step\cu\Step\id}
{\cd\step\cd\\
\hh\id\step\ld\step\rd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\cu\step\cu\step\id}$
&\ref{alg-coalg-triv}.6
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 011110\end{tangle}$
&$\mDelta{}{\mu_r}{\nu_l}{}\sigma\rho
{\hh\id\step\id\step\hcd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\ru\step\id\\
\hh\hstep\id\step\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\d\hstep\cu*\step\cu\\
\hh\step\cu\step[2]\id}
{\hh\hstep\id\step[2]\cd\\
\hh\cd\step\cd*\hstep\d\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\cu\step\id\\
\hh\hstep\id\step\ld\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\id\step\hcu\step\id\step\id}$
&\ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.4
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111100\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{}{}\sigma\rho
{\hh \id\step\cd\step\cd\step\id \\
\id\step\id\step\hx\step\id\step\id \\
\hh \id\step\lu\step\ru\step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\cu\hstep\id\step\hx\step\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\hh\hstep\id\step[2]\cd\\
\hh\cd\step\cd*\hstep\d\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\cu\hstep\cd\\
\hh\hstep\id\step[2]\x\step\id}$
&\ref{alg-coalg-triv}.3, \ref{alg-coalg-triv}.4
&
\\
\hline
\end{tabular}
\end{center}
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline Type&Structure morphisms $\m_B$, $\Delta_B$
&Proposition \ref{alg-coalg-triv}&Status\\
\hline &&&\\[-12pt]
\hline$\divide\unitlens by 2\begin{tangle}\classbox 101110\end{tangle}$
&$\mDelta{\mu_l}{}{\nu_l}{}\sigma\rho
{\hh \id\step\cd\step\id\Step\id \\
\id\step\id\step\hx\Step\id \\
\hh \id\step\lu\step\id\Step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\cu\hstep\id\step\hx\step\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\cd\step[1.5]\hstr{225}\hcd\\
\id\step\ld\step\hcd*\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\id\step\lu\step\hcu\step\id\\
\hh{\hstr{300}\cu}\hstep\ld\step\hcd\\
\hh\step\id\step[1.5]\id\step\x\step\id\\
\hh\step\id\step[1.5]\hcu\step\id\step\id}$
&\ref{alg-coalg-triv}.2, \ref{alg-coalg-triv}.4
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 101101\end{tangle}$
&$\mDelta{\mu_l}{}{}{\nu_r}\sigma\rho
{\hh \id\step\cd\step\id\Step\id \\
\id\step\id\step\hx\Step\id \\
\hh \id\step\lu\step\id\Step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\cu\hstep\id\step\hx\step\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\hstep\id\Step\cd \\
\hh \cd\step\cd*\step[1.5]\id \\
\id\step\hx\step\id\hstep\cd \\
\hh \cu\step\cu\hstep\id\Step\id \\
\hh \hstep\id\Step\id\step\rd\step\id \\
\hstep\id\Step\hx\step\id\step\id \\
\hh \hstep\id\Step\id\step\cu\step\id}$
&\ref{alg-coalg-triv}.2, \ref{alg-coalg-triv}.3
& pure
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 101011\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{}{\nu_r}{}\rho
{\hh\id\step\cd\step\id\Step\id\\
\id\step\id\step\hx\Step\id\\
\hh\id\step\lu\step\id\Step\id\\
\hh\id\Step\id\hstep\cd\step\cd\\
\cu\hstep\id\step\hx\step\id\\
\hh\step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\cd\step\cd\\
\hh\id\step\ld\step\rd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\cu\step\cu\step\id}$
&\ref{alg-coalg-triv}.2, \ref{alg-coalg-triv}.6
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111001\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{}{\nu_r}\sigma{}
{\hh \id\step\cd\step\cd\step\id \\
\id\step\id\step\hx\step\id\step\id \\
\hh \id\step\lu\step\ru\step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\hh \id\Step\id\hstep\id\step\id\step\hrd\hstep\id \\
\cu\hstep\id\step\hx\hstep\id\hstep\id \\
\hh \step\id\step\hstep\id\step\id\step\hru\hstep\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id}
{\cd\step\cd\\
\hh\id\Step\id\step\rd\step\id \\
\id\Step\hx\step\id\step\id \\
\hh\id\Step\id\step\cu\step\id}$
&\ref{alg-coalg-triv}.3, \ref{alg-coalg-triv}.6
& pure
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111010\end{tangle}$
&$\mDelta{}{\mu_r}{\nu_l}{\nu_r}{}\rho
{\hh\id\step\cd\step\cd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\lu\step\ru\step\id\\
\hh\id\Step\id\hstep\cd\step\cd\\
\cu\hstep\id\step\hx\step\id\\
\hh\step\id\step[1.5]\cu*\step\cu\\
\step\cu\step[2]\id}
{\cd\step\cd\\
\hh\id\step\ld\step\id\Step\id\\
\id\step\id\step\hx\Step\id\\
\hh\id\step\cu\step\id\Step\id}$
&\ref{alg-coalg-triv}.8
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 011011\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{\nu_l}{}{}\rho
{\hh\id\step\id\step\hcd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\hcu\step\ru\hstep{\hstr{300}\cd}\\
\hstep\id\step\hcd\step\rd\step\id\\
\hstep\id\step\id\step\hx\step\id\step\id\\
\hstep\id\step\hcu*\step\ru\step\id\\
\hstep{\hstr{225}\hcu}\step[1.5]\cu}
{\cd\step\cd\\
\hh\id\step\ld\step\rd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\cu\step\cu\step\id}$
&\ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.5
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 110011\end{tangle}$
&
$\mDelta{\mu_l}{\mu_r}{\nu_l}{\nu_r}{}{}{
\hh\id\step\cd\step\cd\step\id \\
\hh\id\step\id\step\x\step\id\step\id \\
\hh\id\step\lu\step\ru\step\id \\
\cu\step\cu}
{\cd\step\cd\\
\hh\id\step\ld\step\rd\step\id \\
\hh\id\step\id\step\x\step\id\step\id \\
\hh\id\step\cu\step\cu\step\id}$
&\ref{alg-coalg-triv}.5 , \ref{alg-coalg-triv}.6
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 010011\end{tangle}$
&$\mDelta{}{\mu_r}{\nu_l}{\nu_r}{}{}
{\hh\id\step[2]\id\step\hcd\step\id\\
\hh\id\step[2]\x\step\id\step\id\\
\hh\id\step[2]\id\step\ru\step\id\\
\cu\step\cu}
{\cd\step\cd\\
\hh\id\step\ld\step\rd\step\id\\
\hh\id\step\id\step\x\step\id\step\id\\
\hh\id\step\hcu\step\hcu\step\id}$
&\ref{alg-coalg-triv}.6, \ref{alg-coalg-triv}.7
& pure
\\
\hline
\end{tabular}
\end{center}
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline Type&Structure morphisms $\m_B$, $\Delta_B$
&Proposition \ref{alg-coalg-triv}&Status\\
\hline &&&\\[-12pt]
\hline$\divide\unitlens by 2\begin{tangle}\classbox 011100\end{tangle}$
&$\mDelta{}{\mu_r}{}{}\sigma\rho
{\hh\id\step\id\step\hcd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\ru\step\id\\
\hh\hstep\id\step\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\d\hstep\cu*\step\cu\\
\hh\step\cu\step[2]\id}
{\hh\hstep\id\step[2]\cd\\
\hh\cd\step\cd*\hstep\d\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\cu\hstep\cd\\
\hh\hstep\id\step[2]\x\step\id}$
& \ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.3, \ref{alg-coalg-triv}.4
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 101010\end{tangle}$
&$\mDelta{}{\mu_r}{}{\nu_r}{}\rho
{\hh\id\step\cd\step\id\Step\id\\
\id\step\id\step\hx\Step\id\\
\hh\id\step\lu\step\id\Step\id\\
\hh\id\Step\id\hstep\cd\step\cd\\
\cu\hstep\id\step\hx\step\id\\
\hh\step\id\step[1.5]\cu*\step\cu\\
\step\cu\Step\id}
{\cd\step\cd\\
\hh\id\step\ld\step\id\Step\id\\
\id\step\id\step\hx\Step\id\\
\hh\id\step\cu\step\id\Step\id}$
& \ref{alg-coalg-triv}.2, \ref{alg-coalg-triv}.8
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 111000\end{tangle}$
&$\mDelta{}{}{\nu_l}{\nu_r}{}\rho
{\hh\id\step\cd\step\cd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\lu\step\ru\step\id\\
\hh\id\Step\id\hstep\cd\step\cd\\
\cu\hstep\id\step\hx\step\id\\
\hh\step\id\step[1.5]\cu*\step\cu\\
\step\cu\Step\id}
{\hcd\step\hcd\\
\id\step\hx\step\id}$
& \ref{alg-coalg-triv}.3, \ref{alg-coalg-triv}.8
& pure
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 001011\end{tangle}$
&$\mDelta{\mu_l}{\mu_r}{}{}{}{\rho}
{\hh\id\step\id\step\hcd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\ru\step\id\\
\hh\hstep\id\step\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\d\hstep\cu*\step\cu\\
\hh\step\cu\step[2]\id}
{\cd\step\cd\\
\hh\id\step\ld\step\rd\step\id\\
\id\step\id\step\hx\step\id\step\id\\
\hh\id\step\cu\step\cu\step\id}$
& \ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.2, \ref{alg-coalg-triv}.6
&
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 011001\end{tangle}$
&$\mDelta{\mu_l}{}{\nu_l}{}{}{\rho}
{\hh\id\step\id\step\hcd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\hcu\step\ru\hstep{\hstr{300}\cd}\\
\hstep\id\step\hcd\step\rd\step\id\\
\hstep\id\step\id\step\hx\step\id\step\id\\
\hstep\id\step\hcu*\step\ru\step\id\\
\hstep{\hstr{225}\hcu}\step[1.5]\cu}
{\cd\step\cd \\
\hh\id\Step\id\step\rd\step\id \\
\id\Step\hx\step\id\step\id \\
\hh\id\Step\id\step\cu\step\id}$
& \ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.3, \ref{alg-coalg-triv}.6
& pure
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 011010\end{tangle}$
&$\mDelta{}{\mu_r}{\nu_l}{}{}\rho
{\hh\id\step\id\step\hcd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\ru\step\id\\
\hh\hstep\id\step\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\d\hstep\cu*\step\cu\\
\hh\step\cu\step[2]\id}
{\cd\step\cd\\
\hh\id\step\ld\step\id\Step\id\\
\id\step\id\step\hx\Step\id\\
\hh\id\step\cu\step\id\Step\id}$
& \ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.8
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 001010\end{tangle}$
&$\mDelta{}{\mu_r}{}{}{}\rho
{\hh\id\step\x\step[2]\id\\
\hh\cu\hstep\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\d\hstep\cu*\step\cu\\
\hh\step\cu\step[2]\id}
{\cd\step\cd\\
\hh\id\step\ld\step\id\Step\id\\
\id\step\id\step\hx\Step\id\\
\hh\id\step\cu\step\id\Step\id}$
& \ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.2, \ref{alg-coalg-triv}.8
& regular
\\
\hline$\divide\unitlens by 2\begin{tangle}\classbox 011000\end{tangle}$
&$\mDelta{}{}{\nu_l}{}{}\rho
{\hh\id\step\id\step\hcd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\cu\step\ru\step\id\\
\hh\hstep\id\step\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\hstep\d\hstep\cu*\step\cu\\
\hh\step\cu\step[2]\id}
{\hcd\step\hcd\\
\id\step\hx\step\id}$
& \ref{alg-coalg-triv}.1, \ref{alg-coalg-triv}.3, \ref{alg-coalg-triv}.8
& pure
\\
\hline
\end{tabular}
\end{center}
\unitlens 12pt
\abs
There are various redundant and special relations among the defining
identities of the particular cases listed above. They can be derived
from Theorem \ref{hp-cp}.
Exemplarily we will describe the explicit structure of
the cross product bialgebras of type
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\bullet}\\
\hh\ffbox1{\normalfont\blacktriangledown}\ffbox1{\normalfont\ }\\
\hh\ffbox1{\normalfont\ }\ffbox1{\normalfont\bullet}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3\,$. They consist of two objects $B_1$ and $B_2$
such that
\begin{enumerate}
\item $(B_1,\m_1,\eta_1,\Delta_1,\varepsilon_1)$ is a bialgebra.
\item $(B_2,\Delta_2,\varepsilon_2,\mu_r,\nu_r)$ is
$B_1$-module coalgebra and $B_1$-comodule coalgebra and
$\eta_2:\E\to B_2$ is coalgebra morphism.
\item
\begin{equation}\mathlabel{hp1-epsilon}
\begin{gathered}
\varepsilon_2\circ\m_2=\varepsilon_2\otimes\varepsilon_2\,,\quad
\m_2\circ(\id_{B_2}\otimes\eta_2)=\m_2\circ(\eta_2\otimes\id_{B_2})=
\id_{B_2}\,,\\%[5pt]
\mu_r\circ(\eta_2\otimes\id_{B_1})=
(\id_{B_2}\otimes\varepsilon_1)\circ\nu_l=\eta_2\circ\varepsilon_1\,,
\quad\nu_l\circ\eta_1=\eta_2\otimes\eta_1\,,\\%[5pt]
\mu_l\circ(\eta_2\otimes\id_{B_1})=\id_{B_1}\,,\quad
\mu_l\circ(\id_{B_2}\otimes\eta_1)=\eta_1\circ\varepsilon_2\,,\\%[5pt]
\varepsilon_1\circ\mu_l=\varepsilon_2\otimes\varepsilon_1\,,
\quad\nu_r\circ\eta_2=\eta_2\otimes\eta_1\,,\\%[5pt]
\sigma\circ(\eta_2\otimes\id_{B_2})=\sigma\circ(\id_{B_2}\otimes\eta_2)=
\eta_1\circ\varepsilon_2\,,\quad
\varepsilon_1\circ\sigma=\varepsilon_2\otimes\varepsilon_2\,,\\%[5pt]
(\id_{B_1}\otimes\varepsilon_1)\circ\rho=
(\varepsilon_1\otimes\id_{B_1})\circ\rho=\eta_1\circ\varepsilon_2\,,\quad
\rho\circ\eta_2=\eta_1\otimes\eta_1\,.
\end{gathered}
\end{equation}
\item The weak associativity of $\m_2$ and of $\mu_l$ in Definition
\ref{hp} hold.
\item The module-algebra compatibility of Definition \ref{hp} is satisfied.
\item The cocycle compatibility of $\hat\sigma$ holds.
\item The algebra-coalgebra compatibility of $B_2$ holds.
\item The module-coalgebra compatibility of $\mu_l$, the
comodule-algebra compatibility of $\nu_r$, the
module-comodule compatibility and the cycle-cocycle compatibility
are respectively given by
\begin{equation*}
\begin{split}
&\Delta_1\circ\mu_l=
(\mu_l\otimes\m_1\circ(\id_{B_1}\otimes\mu_l))\circ
(\id_{B_2}\otimes\Psi_{B_1\otimes B_2,B_1}\otimes\id_{B_1})\circ\\
&\quad\circ((\nu_r\otimes\id_{B_2})\circ\Delta_2\otimes\Delta_1)\,,\\[5pt]
&(\id_{B_2}\otimes\m_1)\circ(\Psi_{B_1,B_2}\otimes\id_{B_1})\circ
(\id_{B_1}\otimes\nu_r)\circ\hat\sigma\\
&=(\m_2\otimes\m_1\circ(\m_1\otimes\sigma)\circ
(\id_{B_1}\otimes\varphi_{2,1}\otimes\id_{B_2}))\circ
(\id_{B_2}\otimes\Psi_{B_1\otimes B_2,B_2}\otimes\id_{B_1\otimes B_2})\circ\\
&\quad\circ
((\nu_r\otimes\id_{B_2})\circ\Delta_2\otimes(\nu_r\otimes\id_{B_2})\circ
\Delta_2)\,,\\[5pt]
&(\id_{B_2}\otimes\m_1)\circ(\Psi_{B_1,B_2}\otimes\id_{B_1})\circ
(\id_{B_1}\otimes\nu_r)\circ\varphi_{2,1}\\
&=(\mu_r\otimes\m_1\circ(\id_{B_1}\otimes\mu_l))\circ
(\id_{B_2}\otimes\Psi_{B_1\otimes B_2,B_1}\otimes\id_{B_1})\circ
((\nu_r\otimes\id_{B_2})\circ\Delta_2\otimes\Delta_1)\,,\\[5pt]
&\Delta_1\circ\sigma=
(\sigma\otimes\m_1\circ(\m_1\otimes\sigma)\circ
(\id_{B_1}\otimes\varphi_{2,1}\otimes\id_{B_2}))\circ
(\id_{B_2}\otimes\Psi_{B_1\otimes B_2,B_2}\otimes\id_{B_1\otimes B_2})\circ\\
&\quad\circ
((\nu_r\otimes\id_{B_2})\circ\Delta_2\otimes(\nu_r\otimes\id_{B_2})\circ
\Delta_2)\,.
\end{split}
\end{equation*}
\end{enumerate}
The set of defining identities of the idempotents $\Pi_1$ and
$\Pi_2$ and of the projections and injections $\proj_1$, $\proj_2$,
$\inj_1$, $\inj_2$ can be determined similarly. Eventually the
universal properties of the cross product bialgebras of type
$\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\ffbox1{\normalfont\bullet}\ffbox1{\normalfont\bullet}\\
\hh\ffbox1{\normalfont\blacktriangledown}\ffbox1{\normalfont\ }\\
\hh\ffbox1{\normalfont\ }\ffbox1{\normalfont\bullet}
\end{tangle}
\divide\unitlens by 2
\multiply\unitlens by 3$
will be described by the following proposition. We will use the
notations of Corollary \ref{co-act-inv} and Remark \ref{non-except-bialg}.
\begin{proposition}\Label{bialg-cocycle6}
Let $B$ be a bialgebra in $\C$. Then the following equivalent conditions
are satisfied.
\begin{enumerate}
\item
$B$ is isomorphic to a cross product
bialgebra $B_1\bowtie B_2$ where
$\Delta_{21,1}=\eta_2\otimes\id_{B_1}$
and $\Delta_{11,2}= (\eta_1\otimes\eta_1)\circ\varepsilon_2$ are trivial.
\item
There are idempotents $\Pi_1,\Pi_2\in\End(B)$ such that
\begin{enumerate}
\item
$\m_B\circ(\Pi_1\otimes\Pi_1)=\Pi_1\circ\m_B\circ(\Pi_1\otimes\Pi_1)$,
\item
$\Pi_1$ is coalgebra morphism,
\item
$(\Pi_2\otimes\Pi_2)\circ\Delta_B=
(\Pi_2\otimes\Pi_2)\circ\Delta_B\circ\Pi_2$,
\item
$\Pi_1\circ\eta_B=\eta_B$ and $\varepsilon_B\circ\Pi_2=\varepsilon_B$,
\item
$\m_B\circ(\Pi_1\otimes\Pi_2)$ and $(\Pi_1\otimes\Pi_2)\circ\Delta_B$
split the idempotent $\Pi_1\otimes\Pi_2$.
\end{enumerate}
\item
There are objects $B_1$ and $B_2$ and
morphisms $B_1\overset{\inj_1}\longrightarrow A\overset{\proj_1}
\longrightarrow B_1$ and $B_2\overset{\inj_2}\longrightarrow
A\overset{\proj_2}\longrightarrow B_2$ such that
\begin{enumerate}
\item
$\inj_1$ is algebra and coalgebra morphism,
\item
$\proj_1$ is coalgebra morphism,
\item
$\proj_2$ is coalgebra morphism,
\item
$\proj_j\circ\inj_j=\id_{B_j}$ for $j\in\{1,2\}$.
\item
$\m_A\circ(\inj_1\otimes\inj_2):B_1\otimes B_2\to A$ is isomorphism
with inverse
$(\proj_1\otimes\proj_2)\circ\Delta_A$.\endproof
\end{enumerate}
\end{enumerate}
\end{proposition}
\section*{Concluding Remarks}
We found a universal theory of strong cross product
bialgebras with an equivalent (co\n-)modular co-cyclic
characterization in terms of strong Hopf data.
The theory unites all known cross product bialgebras
\cite{Rad1:85,Ma1:90,MS:94,Ma6:94,Ma1:95,BD1:98} in a
single construction. Furthermore various new types of
cross product bialgebras arise out of the most general
construction. The (co\n-)modular co-cyclic structure of strong
cross product bialgebras corresponds canonically to a strong Hopf
datum which in turn completely determines the bialgebra structure.
Thus Hopf data basically provide a pattern for the
realization of cross product bialgebras in terms of
explicit examples - a task to be done in future
investigations.
There is no conceptual explanation yet
for the understanding of the strong conditions in Definition
\ref{strong-hp}. A canonical origin of these conditions may be found
in higher dimensional categorical constructions of cross product bialgebras
where the two tensor factors of the strong cross product bialgebras
will be considered as object of two different monoidal categories.
Since we are working throughout
in braided categories, the results of the article may now be used to
investigate cross product bialgebras in various types of braided categories
(see \cite{BD1:98,BD3:98} for applications in Hopf bimodule categories).
Our directions of study of cross product bialgebras are particularly
concerned with these questions as well as with extension theory
and cohomological considerations (see \cite{Hof1:90,Sin1:72,Swe1:68}).
\section{Proof of Theorem \ref{hp-cp}}\Label{proof-hp-cp}
This section is exclusively devoted to the proof of
Theorem \ref{hp-cp}. In what follows we denote by
$\mathfrak{h}=\big((B_1,\m_1,\eta_1,\Delta_1,\varepsilon_1),
(B_2,\m_2,\eta_2,\Delta_2,\varepsilon_2);\mu_l,\mu_r,\nu_l,\nu_r,
\rho,\sigma\big)$ a strong Hopf datum. Although many of the subsequent results
hold for more general Hopf data, we do not explicitely point out this fact
in the particular lemmas and propositions. The reader will easily
verify which of the following results hold under more general assumptions.
Before we start proving the theorem we would like to explain some
useful notations and results on Hopf data which will be used subsequently.
\subsection{Basic Properties of Strong Hopf Data}
For a finite set $I=\{i_1,\dots,i_r\}$ of indices
$i_k\in\{1,2\}$, $k\in\{1,\dots, r\}$ we denote
$B_I:=B_{i_1}\otimes\cdots\otimes B_{i_r}$.
Suppose now $r=\left\vert I\right\vert$ is the lenght of the sequence $I$.
Given morphisms $f:B_I\otimes B_2\otimes B_J\to B_K$ and
$g:B_K\to B_I\otimes B_2\otimes B_J$
we define their \emph{relativization} $f^{[r+1]}$ and $g_{[r+1]}$
in the $(r+1)$st domaine index and codomaine index respectively by
\begin{equation}\mathlabel{rel-morph1}
f^{[r+1]}:=\,
\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hstep\obj{B_I}\step[2]\obj{B_1}\step[1.5]\obj{B_J}\\
\hh\hstep\id\step{\hstr{200}\hld}\step\id\\
\hh\hstep\id\step\id\step\x\\
\hh\ffbox3{f}\hstep\id\\
\hh\step[1.5]\id\Step\id\\
\step[1.5]\obj{B_K}\Step\obj{B_1}
\end{tangle}
\quad\text{and}\quad
g_{[r+1]}:=\,
\begin{tangle}
\obj{B_2}\Step\obj{B_K}\\
\hh\id\Step\id\\
\hh\id\hstep\ffbox3{g}\\
\hh\x\step\id\step\id\\
\hh\id\step{\hstr{200}\hru}\step\id\\
\obj{B_I}\step[1.5]\obj{B_2}\step[1.5]\obj{B_J}
\end{tangle}
\end{equation}
We say that two morphisms $f$ and $f'$
coincide relatively if $f^{[r+1]}=f'{}^{[r+1]}$. Similarly
$g_{[r+1]}=g'_{[r+1]}$ means that $g$ and $g'$ coincide relatively.
Instead of \eqref{rel-morph1} we will often use the obvious shorthand notation
\begin{equation}\mathlabel{rel-morph2}
\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\hstep\id\step\hld\hstep\id\\
\hh\hstep\id\step\id\step\id\\
\hh\ffbox3{f}\\
\hh\step[1.5]\id
\end{tangle}
\,=\,
\begin{tangle}
\hh\hstep\id\step\hld\hstep\id\\
\hh\hstep\id\step\id\step\id\\
\hh\ffbox3{f^\prime}\\
\hh\step[1.5]\id
\end{tangle}
\quad\text{and}\quad
\begin{tangle}
\hh\step[1.5]\id\\
\hh\ffbox3{g}\\
\hh\hstep\id\step\id\step\id\\
\hh\hstep\id\hstep\hru\step\id
\end{tangle}
\,=\,
\begin{tangle}
\hh\step[1.5]\id\\
\hh\ffbox3{g^\prime}\\
\hh\hstep\id\step\id\step\id\\
\hh\hstep\id\hstep\hru\step\id
\end{tangle}
\end{equation}
for $f^{[r+1]}=f'{}^{[r+1]}$ and $g_{[r+1]}=g'_{[r+1]}$ respectively.
This means that the identities \eqref{rel-morph1} result from \eqref{rel-morph2}
by braiding the threads with endpoints in the middle of the graphics with
all neighbouring strings on the right, respectively on the left and
then completing vertically these threads to the bottom, respectively to the
top of the graphics.
Besides \eqref{phi-sigma-rho2} and \eqref{phi-sigma-rho3} we will use the definitions
\unitlens 15pt %
\begin{equation}\mathlabel{phi-sigma-rho4}
\begin{array}{c}
\varphi_{1,1}
:=
\divide\unitlens by 2
\begin{tangle}\ox{11}\end{tangle}
\multiply\unitlens by 2
:=
\divide\unitlens by 3
\begin{tangle}
\ld\step[2]\id\\\id\step\x\\\lu\step[2]\id
\end{tangle}
\multiply\unitlens by 3
\quad ,\quad
\varphi_{2,2}
:=
\divide\unitlens by 2
\begin{tangle}\ox{22}\end{tangle}
\multiply\unitlens by 2
:=
\divide\unitlens by 3
\begin{tangle}
\id\step[2]\rd\\
\x\step\id\\
\id\step[2]\ru
\end{tangle}
\end{array}
\end{equation}
All subsequent lemmas can be proven straighforwardly with the help of the
definition of (strong) Hopf data. We will therefore only sketch the main
steps of the derivation of the proofs.
\begin{lemma}\Label{weak-strong}
Let $\mathfrak{h}$ be a strong Hopf datum. Then the identities
\begin{equation}\mathlabel{cocycle-triv2}
\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\hh\hstep\hld\hstep\id\\
\hh\hstep\id\step\id\\
\hh\ffbox 2{\hat\sigma}\\
\hh\hstep\id\step\id
\end{tangle}
=
\begin{tangle}
\hh\step\hld\hstep\id\\
\hh\unit\step\cu
\end{tangle}
\quad , \quad
\begin{tangle}
\hstep\id\step\ld\\
\hh\ffbox 2{\hat\sigma}\hstep\id\\
\hh\hstep\id\step\id\step\id
\end{tangle}
=
\begin{tangle}
\unit\step\id\step\ld\\
\id\step\hcu\step\id
\end{tangle}
\quad , \quad
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox 2{\hat\rho}\\
\hh\hstep\id\step\id\\
\hh\hstep\id\hstep\hru
\end{tangle}
=
\begin{tangle}
\hh\cd\\
\hh\id\hstep\hru\step\counit
\end{tangle}
\quad ,
\quad
\begin{tangle}
\hh\id\step\id\step\id\hstep\\
\hh\id\hstep\ffbox 2{\hat\rho}\\
\ru\step\id
\end{tangle}
=
\begin{tangle}
\id\step\hcd\step\id\\
\ru\step\id\step\counit
\end{tangle}
\end{equation}
are satisifed.
\end{lemma}
\begin{proof}
The first identity in \eqref{cocycle-triv2} has been obtained
from \eqref{cocycle-triv1} and \eqref{act-coact-triv}.
With the help of \eqref{cocycle-triv1} and \eqref{strong-hopf-combo}
the second identity will be derived. Application of $\pi$-symmetry
completes the proof.\end{proof}
\abs
The relative associativity of $\m_2$ and $\mu_l$, and by $\pi$-symmetry
the relative coassociativity of $\Delta_2$ and $\nu_r$ will be shown
in the following lemma.
\begin{lemma}\Label{strong-hd3}
For a strong Hopf datum $\mathfrak{h}$ the identities
\begin{equation*}
\begin{split}
\vstretch 60
\divide\unitlens by 2
\begin{tangle}
\ld\step\id\step\id\\
\id\step[2]\hcu\\
\hstr{125}\cu
\end{tangle}
=
\begin{tangle}
\ld\step\id\step\id\\
\cu\step\id\\
\step\cu
\end{tangle}
\quad ,\quad
\begin{tangle}
\id\step\ld\step\id\\
\id\step\cu\\
\cu
\end{tangle}
&=
\vstretch 60
\divide\unitlens by 2
\begin{tangle}
\id\step\ld\step\id\\
\hcu\step[2]\id\\
\hstep\hstr{125}\cu
\end{tangle}
\quad ,\quad
\begin{tangle}
\id\step\id\step\ld\\
\id\step\hcu\\
\hstr{150}\hcu
\end{tangle}
=
\begin{tangle}
\id\step\id\step\ld\\
\hcu\step\id\\
\hstep\hstr{150}\hcu
\end{tangle}
\\
\vstretch 60
\divide\unitlens by 2
\begin{tangle}
\ld\step\id\step\id\\
\cu\step\id\\
\step\lu[2]
\end{tangle}
=
\begin{tangle}
\ld\step\id\step\id\\
\nw1\step\lu\\
\step\lu[2]
\end{tangle}
\quad &,\quad
\vstretch 60
\divide\unitlens by 2
\begin{tangle}
\id\step\ld\step\id\\
\hcu\step[2]\id\\
\hstep\hstr{125}\lu[2]
\end{tangle}
=
\begin{tangle}
\id\step\hld\step\id\\
\id\step\hstr{150}\lu\\
\hstr{125}\lu[2]
\end{tangle}
\end{split}
\end{equation*}
and the corresponding $\pi$-symmetric versions of the relative coassociativity
of $\Delta_1$ and $\nu_r$ hold.
\end{lemma}
\begin{proof}
The relative associativity of $\m_2$ and $\mu_l$ will be derived from the
respective weak associativity of Definition \ref{hp} taking into account
\eqref{cocycle-triv1} and \eqref{cocycle-triv2} of Lemma
\ref{weak-strong}.\end{proof}
\begin{remark}
{\normalfont For a Hopf data satisfying
$\divide\unitlens by 2\begin{tangle}
\vstr{70}\hh\id\hstep\cu*\\
\vstr{35}\ru
\end{tangle}
=
\begin{tangle}
\vstr{70}\id\hstep\counit\hstep\counit
\end{tangle}$\
and
$\divide\unitlens by 2\begin{tangle}
\vstr{35}\hstep\ld\\
\vstr{70}\hh\cd*\hstep\id
\end{tangle}
=
\begin{tangle}
\vstr{70}\unit\hstep\unit\hstep\id
\end{tangle}$
the multiplication $\m_2$ and the comultiplication $\Delta_1$
are (co\n-)associative.}
\end{remark}
\begin{lemma}\Label{strong-hd2}
The relative versions of the morphisms $\varphi_{1,2}$ and $\varphi_{2,1}$
are given by
\begin{equation}\mathlabel{rel-phi}
\begin{array}{c}
\divide\unitlens by 2
\begin{tangle}
\ox{12}\\
\vstr{50}\id\step[2]\id\\
\vstr{60}\id\step\ru
\end{tangle}
=
\vstretch 60
\begin{tangle}
\id\step[2]\rd\\
\x\step\id\\
\id\step[2]\hcu\\
\id\step[1.5]\ru
\end{tangle}
\quad\text{and}\quad
\vstretch 100
\begin{tangle}
\vstr{60}\ld\step\id\\
\vstr{50}\id\step[2]\id\\
\ox{21}
\end{tangle}
=
\vstretch 60
\begin{tangle}
\hstep\ld\step[1.5]\id\\
\hcd\step[2]\id\\
\id\step\x\\
\lu\step[2]\id
\end{tangle}
\end{array}
\end{equation}
\end{lemma}
\begin{proof}
Using that $(B_2,\mu_r)$ is a right module and applying Lemma \ref{weak-strong}
yields the first identity of \eqref{rel-phi}.
\end{proof}
\abs
Then the next lemma follows from the module-algebra and comodule-coalgebra
compatibilities of Definition \ref{hp}.
\begin{lemma}\Label{strong-hd4}
The identities
\begin{gather}\mathlabel{mod-alg-rel}
\divide\unitlens by 2
\begin{tangle}
\hh\hld\step[1.5]\id\step\id\\
\hh{\hstr{200}\cu}\step\id\\
\hh\step\hstr{400}\hru
\end{tangle}
=
\begin{tangle}
\hh\hld\step[1.5]\hru\\
\cu
\end{tangle}
\quad ,
\quad
\begin{tangle}
\hh\hstr{400}\hld\\
\hh\id\step\hstr{200}\cd\\
\hh\id\step\id\step[1.5]\hru
\end{tangle}
=
\begin{tangle}
\hstep\cd\\
\hh\hld\step[1.5]\hru
\end{tangle}
\\
\mathlabel{mod-alg-rel2}
\divide\unitlens by 2
\begin{tangle}
\hh\id\step\hld\step\id\\
\hh\cu\step[1.5]\id\\
\hstep\Ru
\end{tangle}
\;=\;
\begin{tangle}
\hh\id\step[1.5]\hld\step\id\\
\hh\id\step\cd\step\id\\
\hh\id\step\id\step\x\\
\hh\id\step{\hstr{200}\hlu}\step\id\\
\hh\Ru\step\id\\
\hh\hstr{300}\cu
\end{tangle}
\quad ,
\quad
\begin{tangle}
\Ld\\
\hh\id\step[1.5]\cd\\
\hh\id\step\hru\step\id
\end{tangle}
\;=\;
\begin{tangle}
\hh\hstr{300}\cd\\
\hh\id\step\hstr{200}\ld\\
\hh\id\step{\hstr{200}\hrd}\step\id\\
\hh\x\step\id\step\id\\
\hh\id\step\cu\step\id\\
\hh\id\step\hru\step[1.5]\id
\end{tangle}
\end{gather}
are satisfied for a strong Hopf datum $\mathfrak{h}$.
\end{lemma}
\begin{lemma}\Label{strong-hd5} Let $\mathfrak{h}$ be strong Hopf datum. Then
$(B_1,\m_1,\eta_1,\nu_l)$ is a left $B_2$-comodule algebra and
$(B_2,\Delta_2,\varepsilon_2,\mu_r)$ is a right $B_1$-module coalgebra.
\end{lemma}
\begin{proof}
We use Lemma \ref{weak-strong} and the second identity of the comodule-algebra
compatibility of Definition \ref{hp} to show that
$(B_1,m_1,\eta_1,\nu_l)$ is a left $B_2$-comodule algebra. In $\pi$-symmetric
manner it will be proven that $(B_2,\Delta_2,\varepsilon_2,\mu_r)$ is
a right $B_1$-module coalgebra.\end{proof}
\begin{lemma}\Label{strong-hd8}
For a strong Hopf datum $\mathfrak{h}$
the relativizations of the algebra-coalgebra compatibilities,
the (left) module-coalgebra compatibility, and the (right)
comodule-algebra compatibility are respectively given by
\begin{gather}\mathlabel{alg-coalg-rel}
\divide\unitlens by 2
\begin{tangle}
\cu\\
\hh\hstr{200}\cd\\
\hh\id\step[1.5]\hru
\end{tangle}
=
\begin{tangle}
\hh\cd\step\cd\\
\hh\id\step\x\step\id\\
\hh\cu\step\cu\\
\hh\hstep\id\step[1.5]\hru
\end{tangle}
\quad ,\quad
\begin{tangle}
\hh\hld\step[1.5]\id\\
\hh\hstr{200}\cu\\
\cd
\end{tangle}
=
\begin{tangle}
\hh\hstep\hld\step[1.5]\id\\
\hh\cd\step\cd\\
\hh\id\step\x\step\id\\
\hh\cu\step\cu\\
\end{tangle}
\\
\mathlabel{rel-mod-coalg}
\divide\unitlens by 2
\begin{tangle}
\lu\\
\cd\\
\hh\id\step[1.5]\hru
\end{tangle}
=
\begin{tangle}
\cd\step\hcd\\
\hh\rd\step\x\step\id\\
\hh\id\step\x\step\lu\\
\lu\step\cu\\
\hh\step\id\step[1.5]\hru
\end{tangle}
\quad ,\quad
\begin{tangle}
\hstep\lu\\
\hstep\cd\\
\hh\hru\step[2]\id
\end{tangle}
=
\begin{tangle}
\cd\step\hcd\\
\hh\rd\step\x\step\id\\
\hh\id\step\x\step\lu\\
\lu\step\cu\\
\hh\hstep\hru\step[2]\id
\end{tangle}
\\
\mathlabel{rel-comod-alg}
\divide\unitlens by 2
\begin{tangle}
\hh\hld\\
\cu\\
\step\rd
\end{tangle}
=
\begin{tangle}
\hh\step\hld\\
\cd\step\rd\\
\hh\rd\step\x\step\id\\
\hh\id\step\x\step\lu\\
\hcu\step\cu
\end{tangle}
\quad ,\quad
\begin{tangle}
\hh\step[2]\hld\\
\cu\\
\step\rd
\end{tangle}
=
\begin{tangle}
\hh\step[3]\hld\\
\cd\step\rd\\
\hh\rd\step\x\step\id\\
\hh\id\step\x\step\lu\\
\hcu\step\cu
\end{tangle}
\end{gather}
\end{lemma}
\begin{proof}
Since $(B_1,\m_1,\eta_1,\nu_l)$ is a is a left comodule algebra by Lemma
\ref{strong-hd5} the first identity in \eqref{phi-prod} follows from the
relative asociativity of $\m_2$ according to Lemma \ref{strong-hd3}.
The verification of \eqref{rel-mod-coalg} needs a little more
calculation. We prove the first identity of \eqref{rel-mod-coalg}.
The second one can be derived with similar techniques.
We start with the second module-coalgebra compatibility in Definition \ref{hp}.
We apply the relativization (with $\mu_r$) to the second tensor factor
on both sides of the graphic. The left hand side of this relative
module-coalgebra compatibility yields the left hand side of the
first identity in \eqref{rel-mod-coalg} if we consecutively apply
modularity of $(B_2,\mu_r)$, the second relation of \eqref{cocycle-triv1} and
the second relation of \eqref{act-coact-triv}.
To obtain the right hand side of the first identity of \eqref{rel-mod-coalg}
we transform the right hand side of the relative module-coalgebra compatibility
using successively the third relation of \eqref{cocycle-triv1}, modularity
of $(B_2,\mu_r)$, the first equation of \eqref{rel-phi}, the modularity of $B_2$,
the second identity of \eqref{cocycle-triv1}, and eventually again modularity of
$B_2$.
All other identities can be derived similarly, in particular because of
$\pi$-symmetric reasons.
\end{proof}
\abs
In the subsequent Lemmas \ref{strong-hd6} and \ref{strong-hd7} the
entwining properties of the morphisms $\varphi_{1,1}$, $\varphi_{2,2}$,
$\varphi_{1,2}$ and $\varphi_{2,1}$ will be investigated.
\begin{lemma}\Label{strong-hd6}
The morphism $\varphi_{1,1}$ entwines with the
multiplication $\m_1$, and $\varphi_{2,2}$ entwines with the
comultiplication $\Delta_2$ according to
\begin{equation}\mathlabel{entw1}
\begin{split}
\divide\unitlens by 2
\begin{tangle}
\cu\step\id\\
\step\ox{11}
\end{tangle}
=
\begin{tangle}
\id\step[2]\ox{11}\\
\ox{11}\step[2]\id\\
\id\step[2]\cu
\end{tangle}
\quad &,\quad
\divide\unitlens by 2
\begin{tangle}
\id\step\cu\\
\ox{11}
\end{tangle}
=
\begin{tangle}
\ox{11}\step[2]\id\\
\id\step[2]\ox{11}\\
\cu\step[2]\id
\end{tangle}
\\
\divide\unitlens by 2
\begin{tangle}
\ox{22}\\
\id\step\cd
\end{tangle}
=
\begin{tangle}
\cd\step[2]\id\\
\id\step[2]\ox{22}\\
\ox{22}\step[2]\id
\end{tangle}
\quad &,\quad
\divide\unitlens by 2
\begin{tangle}
\step\ox{22}\\
\cd\step\id
\end{tangle}
=
\begin{tangle}
\id\step[2]\cd\\
\ox{22}\step[2]\id\\
\id\step[2]\ox{22}
\end{tangle}
\\
\divide\unitlens by 2
\begin{tangle}
\cu\step\id\\
\step\ox{22}
\end{tangle}
=
\begin{tangle}
\vstr{50}\id\step[2]\cd\step[2]\id\\
\id\step[2]\id\step[2]\ox{22}\\
\vstr{50}\id\step[2]\id\step[2]\rd\step\id\\
\vstr{50}\id\step[2]\x\step\id\step\id\\
\vstr{50}\x\step[2]\lu\step\id\\
\vstr{50}\id\step[2]{\hstr{300}\ru}\step\id\\
\vstr{50}\id\step[2]\hstr{200}\cu
\end{tangle}
=
\begin{tangle}
\id\step\id\step\Rd\\
\hh\id\step\x\Step\id\\
\Put(0,0)[lb]{\begin{tangle}\hh\x\\
\hh\id\step\id\end{tangle}}\Step\ox{21}\\
\hh\id\step{\hstr{200}\hru}\Step\id\\
\id\step\hstr{150}\cu
\end{tangle}
\quad &,\quad
\divide\unitlens by 2
\begin{tangle}
\ox{11}\\
\id\step\cd
\end{tangle}
=
\begin{tangle}
\vstr{50}{\hstr{200}\cd}\step[2]\id\\
\vstr{50}\id\step{\hstr{300}\ld}\step[2]\id\\
\vstr{50}\id\step\rd\step[2]\x\\
\vstr{50}\id\step\id\step\x\step[2]\id\\
\vstr{50}\id\step\lu\step[2]\id\step[2]\id\\
\ox{11}\step[2]\id\step[2]\id\\
\vstr{50}\id\step[2]\cu\step[2]\id
\end{tangle}
=
\begin{tangle}
{\hstr{150}\cd}\step\id\\
\hh\id\Step{\hstr{200}\hld}\step\id\\
\ox{12}\step\Put(0,0)[lb]{\begin{tangle}\hh\id\step\id\\
\hh\x\end{tangle}}\\
\hh\id\Step\x\step\id\\
\Lu\step\id\step\id
\end{tangle}
\end{split}
\end{equation}
\end{lemma}
\begin{proof}
For the proof of the first identity of Lemma \ref{strong-hd6} we use Lemma
\ref{strong-hd5} and the fourth equation of Lemma \ref{strong-hd3}. To verify
the second identity of \eqref{entw1}, the module-algebra compatibility for
Hopf data, Lemma \ref{strong-hd2} and the comodularity of $(B_1,\nu_l)$
have to be applied successively. Using the module-algebra compatibility
and the $\pi$-symmetric version of the fourth identity of Lemma \ref{strong-hd3}
yields the fifth identity of \eqref{entw1}. Simple calculations yield the sixth
identity of \eqref{entw1}. The remaining relations of the lemma
follow by $\pi$-symmetric reasoning.\end{proof}
\begin{lemma}\Label{strong-hd7}
The (relative) entwining identities for $\varphi_{1,2}$ and $\varphi_{2,1}$
are given by
\begin{gather}
\divide\unitlens by 3
\multiply\unitlens by 2
\hstretch 80 \vstretch 80
\mathlabel{phi-prod}
\begin{tangle}
\ox{21}\\
\hh\id\step[1.5]\cd
\end{tangle}
=
\begin{tangle}
\hh{\hstr{160}\cd}\step\cd\\
\hh\id\Step\x\step\id\\
\ox{21}\step\ru
\end{tangle}
\quad ,
\quad
\begin{tangle}
\hh\cu\step[1.5]\id\\
\hstep\ox{12}
\end{tangle}
=
\begin{tangle}
\ld\step\ox{12}\\
\hh\id\step\x\Step\id\\
\hh\cu\step\hstr{160}\cu
\end{tangle}
\\[5pt]
\mathlabel{phi-prod2}
\divide\unitlens by 3
\multiply\unitlens by 2
\hstretch 80 \vstretch 80
\begin{tangle}
\hh\id\step[1.5]\cu\\
\ox{21}
\end{tangle}
=
\begin{tangle}
\ox{21}\Step\id\\
\id\Step\ox{21}\\
\hh\hstr{160}\cu\step\id
\end{tangle}
\quad ,\quad
\begin{tangle}
\hstep\ox{12}\\
\hh\cd\step[1.5]\id
\end{tangle}
=
\begin{tangle}
\hh\id\Step\hstr{160}\cd\\
\ox{12}\Step\id\\
\id\Step\ox{12}
\end{tangle}
\\[5pt]
\mathlabel{phi-prod-rel}
\divide\unitlens by 3
\multiply\unitlens by 2
\hstretch 80 \vstretch 70
\begin{tangle}
\hh\hld\step[1.5]\id\step\id\\
\hh{\hstr{160}\cu}\step\id\\
\step\ox{21}
\end{tangle}
=
\begin{tangle}
\hh\hld\step[1.5]\id\Step\id\\
\id\step[2]\ox{21}\\
\ox{21}\step[2]\id\\
\id\step[2]\cu
\end{tangle}
\quad ,
\quad
\begin{tangle}
\ox{12}\\
\hh\id\step\hstr{160}\cd\\
\hh\id\step\id\step[1.5]\hru
\end{tangle}
=
\begin{tangle}
\cd\step[2]\id\\
\id\step[2]\ox{12}\\
\ox{12}\step[2]\id\\
\hh\id\Step\id\step[1.5]\hru
\end{tangle}
\\[5pt]
\mathlabel{phi-prod-rel2}
\divide\unitlens by 3
\multiply\unitlens by 2
\hstretch 80 \vstretch 80
\begin{tangle}
\step\ox{21}\\
\cd\step\id\\
\hh\id\step[1.5]\hru\step\id
\end{tangle}
\enspace=\enspace
\begin{tangle}
\hh\hstr{160}\cd\hstep\cd\\
\hh{\hstr{160}\hrd}\step\x\Step\id\\
\id\step\hx\step\ox{21}\\
\hh\lu\step\cu\Step\id\\
\hh\step\id\step\hru\step[2.5]\id
\end{tangle}
\quad ,
\quad
\begin{tangle}
\hh\id\step\hld\step[1.5]\id\\
\id\step\cu\\
\ox{12}
\end{tangle}
\enspace=\enspace
\begin{tangle}
\hh\id\step[2.5]\hld\step\id\\
\hh\id\Step\cd\step\rd\\
\ox{12}\step\hx\step\id\\
\hh\id\Step\x\step{\hstr{160}\hlu}\\
\hh\hstr{160}\cu\hstep\cu
\end{tangle}
\end{gather}
\end{lemma}
\begin{proof}
We use Lemma \ref{strong-hd5} for $B_1$ and the relative associativity of $\m_2$
(Lemma \ref{strong-hd3}) to obtain the first identity of \eqref{phi-prod}.
The first identity of \eqref{phi-prod2}
will be used explicitely in the proof of Proposition \ref{co-assoc}.
Below we will give its detailed derivation.
\unitlens 10pt %
\begin{equation*}
\hstretch 80 \vstretch 80
\varphi_{2,1}\circ(B_2\otimes\m_1)=
\begin{tangle}
\hh\hstep\id\step\cd\step\cd\\
\hh\hstep\id\step\id\step\x\step\id\\
\hh\cd\hstep\cu\step\cu\\
\hh\id\step\x\Step\id\\
\lu\step\Ru\\
\end{tangle}
=
\begin{tangle}
\hh\hstep{\hstr{160}\cd}\step\cd\step\cd\\
\hh\hstep\id\Step\x\step\x\step\id\\
\hh\cd\step\cd\hstep\x\step\id\step\id\\
\hh\id\step\x\step\id\hstep\id\step\ru\step\id\\
\hh\lu\step\ru\hstep\id\step\Ru\\
\hh\step\id\step{\hstr{120}\lu}\step\id\\
\hh\step{\hstr{200}\cu}\step\id
\end{tangle}
=
\begin{tangle}
\hh\cd\step\cd\Step\id\\
\hh\id\step\x\step\nw 1\step\id\\
\hh\lu\hstep\cd\step\cd\hstep\id\\
\hh\step\id\hstep\id\step\x\step\id\hstep\id\\
\hh\step\id\hstep\ru\step\ru\hstep\id\\
\hh\step\id\hstep\id\Step\id\step\cd\\
\hh\step\id\hstep\nw 1\step\x\step\id\\
\hh\step\d\step\lu\step\ru\\
\step[1.5]\cu\step\id
\end{tangle}
=(\m_1\otimes B_2)\circ\varphi_{2,11}
\end{equation*}
\unitlens 15pt %
where the first equation comes from the relative bialgebra property
of $B_1$ according to \eqref{alg-coalg-rel}.
The second equation can be verified with the help of the
module-algebra compatibility of Definition \ref{hp}, and using
that $(B_2,\mu_r)$ is a right module. In the third identity we
use the relative associativity of $\Delta_1$ proven in Lemma
\ref{strong-hd3}. Finally the result follows because
$(B_2,\Delta_2,\varepsilon_2,\mu_r)$ is a module coalgebra by Lemma
\ref{strong-hd5}.
The left hand side of the first identity of \eqref{phi-prod-rel}
will be transformed to the right hand side of the
identity by using successively the algebra-coalgebra compatibility for
$\Delta_2$ and $\m_2$, the comodularity of $(B_1,\nu_l)$, the first
relation of \eqref{act-coact-triv}, the first identity of \eqref{mod-alg-rel},
the relative associativity of $\mu_l$ according to Lemma
\ref{strong-hd3} and finally the second relation of \eqref{rel-phi}.
The first identity of \eqref{phi-prod-rel2} immediately follows from
the relative module coalgebra properties \eqref{rel-mod-coalg}.
All other relations of the lemma can be derived easily by
applying $\pi$-symmetry.\end{proof}
\begin{lemma}
For the strong Hopf datum $\mathfrak{h}$ the following identities are
satisfied.
\unitlens 15pt %
\begin{gather}
\mathlabel{rel-cocycle-assoc1}
\divide\unitlens by 2
\def5{2}
\begin{tangle}
\hh\hld\step[1.5]\id\step\id\\
\hh\hstr{200}\cu\hstep\id\\
\hh\step\hstr{200}\cu*
\end{tangle}
=
\def5{4}
\begin{tangle}
\hh\hld\step\cu*\\
\Lu
\end{tangle}
\quad ,\quad
\def5{2}
\begin{tangle}
\hh\hstr{200}\cd*\\
\hh\id\step\hstr{200}\cd\\
\hh\id\step\id\step[1.5]\hru
\end{tangle}
=
\def5{4}
\begin{tangle}
\hstep\Rd\\
\hh\cd*\step\hru
\end{tangle}
\\
\mathlabel{rel-cocycle-assoc2}
\divide\unitlens by 2
\def5{2}
\begin{tangle}
\hh\id\step\hld\step\id\\
\hh\cu\step[1.5]\id\\
\hh\hstep\hstr{200}\cu*
\end{tangle}
=
\begin{tangle}
\hh\id\step\hld\step[1.5]\id\\
\hh\id\step\hstr{200}\cu\\
\hh\hstr{200}\cu*
\end{tangle}
\quad ,\quad
\begin{tangle}
\hh\hstr{200}\cd*\\
\hh\id\step[1.5]\cd\\
\hh\id\step\hru\step\id
\end{tangle}
=
\begin{tangle}
\hh\step\hstr{200}\cd*\\
\hh{\hstr{200}\cd}\step\id\\
\hh\id\step[1.5]\hru\step\id
\end{tangle}
\\
\mathlabel{rel-cocycle-assoc3}
\divide\unitlens by 2
\def5{4}
\begin{tangle}
\hh\id\hstep\id\step\hld\\
\hh\id\hstep\cu\hstep\id\\
\hh\cu*\step\id
\end{tangle}
=
\begin{tangle}
\cu*\step\id
\end{tangle}
\quad ,\quad
\begin{tangle}
\hh\id\step\cd*\\
\hh\id\hstep\cd\hstep\id\\
\hh\hru\step\id\hstep\id
\end{tangle}
=
\begin{tangle}
\id\step\cd*
\end{tangle}
\end{gather}
\end{lemma}
\begin{proof}
The proof follows straightforwardly from the cocycle and cycle
compatibilities of Definition \ref{hp} and the identities
\eqref{act-coact-triv} and \eqref{cocycle-triv1}.\end{proof}
\begin{lemma}
The subsequent relations involving $\hat\rho$ and $\hat\sigma$ hold
in $\mathfrak{h}$.
\begin{gather}
\mathlabel{hat-prod}
\divide\unitlens by 2
\vstretch 110
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox2{\hat\sigma}\\
\hh\hstep\id\hstep\cd
\end{tangle}
=
\begin{tangle}
\hh\hstep\cd\Step\cd\\
\hstep\id\step\ox{22}\step\id\\
\hh\ffbox2{\hat\sigma}\step[1.5]\cu\\
\hh\hstep\id\step\id\step[2.5]\id
\end{tangle}
\quad ,\quad
\begin{tangle}
\hh\cu\hstep\id\\
\hh\ffbox2{\hat\rho}\\
\hh\hstep\id\step\id
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step[2.5]\id\step\id\\
\hh\cd\step[1.5]\ffbox2{\hat\rho}\\
\id\step\ox{11}\step\id\\
\hh\cu\Step\cu
\end{tangle}
\\
\mathlabel{hat-prod-rel}
\divide\unitlens by 2
\vstretch 110
\begin{tangle}
\hh\hld\hstep\id\hstep\id\\
\hh\cu\hstep\id\\
\hh\ffbox2{\hat\sigma}\\
\hh\hstep\id\step\id
\end{tangle}
=
\begin{tangle}
\hh\hstep\hld\step\id\step\id\\
\hh\cd\hstep\ffbox2{\hat\sigma}\\
\hh\id\step\x\step\id\\
\hh{\hstr{200}\hlu}\step\cu
\end{tangle}
\quad ,\quad
\begin{tangle}
\hh\hstep\id\step\id\\
\hh\ffbox2{\hat\rho}\\
\hh\hstep\id\hstep\cd\\
\hh\hstep\id\hstep\id\hstep\hru
\end{tangle}
=
\begin{tangle}
\hh\hstep\cd\step\hstr{200}\hrd\\
\hh\hstep\id\step\x\step\id\\
\hh\ffbox2{\hat\rho}\hstep\cu\\
\hh\hstep\id\step\id\step\hru
\end{tangle}
\end{gather}
\end{lemma}
\begin{proof} Because of $\pi$-symmetry we will only
demonstrate the first identities of \eqref{hat-prod} and \eqref{hat-prod-rel}.
Applying to the left hand side of \eqref{hat-prod}
the algebra-coalgebra compatibility (for $\Delta_2$ and $\m_2$)
and the entwining property of $\varphi_{2,2}$ (Lemma \ref{strong-hd6})
yields the result.
To obtain the first identity of \eqref{hat-prod-rel} we transform its left hand
side consecutively using the relative bialgebra property \eqref{alg-coalg-rel}
of $\Delta_2$ and $\m_2$, the comodularity of $(B_1,\nu_l)$, the first relation
of \eqref{mod-alg-rel}, the first identity of \eqref{rel-cocycle-assoc1}, the
relative associativity of $\m_2$, and again the fact that $(B_1,\nu_l)$ is left
comodule.\end{proof}
\begin{remark}{\normalfont Note that \eqref{cocycle-triv2} and
\eqref{rel-cocycle-assoc1} are special cases of \eqref{hat-prod-rel}.
Furthermore \eqref{rel-cocycle-assoc3} implies \eqref{cocycle-triv1}.}
\end{remark}
\subsection{Special Properties of Strong Hopf Data}
Before we will prove Theorem \ref{hp-cp} we have to provide several
auxiliary definitions and specific properties of strong Hopf Data.
Similarly as in Remark \ref{alpha-ip} we define
$\proj_1=\id_{B_1}\otimes\varepsilon_2$,
$\proj_2=\varepsilon_1\otimes\id_{B_2}$,
$\proj_0=\id_B$, $\inj_1=\id_{B_1}\otimes\eta_2$,
$\inj_2=\eta_1\otimes\id_{B_2}$, and $\inj_0=\id_B$. In this context we
occasionally use the notation $B_0:=B:=B_0\otimes B_1$.
Given a strong Hopf datum $\mathfrak{h}$ we can build the morphisms
$\m_B:B\otimes B\to B$ and $\Delta_B:B\to B\otimes B$ like in
\eqref{type-alpha3} as
\begin{equation}\mathlabel{strong-mult-comult}
\begin{array}{c}
\m_B :=
\hstretch 150
\divide\unitlens by 3
\begin{tangle}
\hh \id\step\cd\step\cd\step\id \\
\id\step\id\step\hx\step\id\step\id \\
\hh \id\step\lu\step\ru\step\id \\
\hh \id\Step\id\hstep\cd\step\cd \\
\hh \id\Step\id\hstep\id\step\id\step\hrd\hstep\id \\
\cu\hstep\id\step\hx\hstep\id\hstep\id \\
\hh \step\id\step\hstep\id\step\id\step\hru\hstep\id \\
\hh \step\id\step[1.5]\cu*\step\cu \\
\step\cu\Step\id
\end{tangle}
\multiply\unitlens by 3
\qquad\text{and}\qquad
\Delta_B:=
\divide\unitlens by 3
\begin{tangle}
\hstep\id\Step\cd\\
\hh \cd\step\cd*\step[1.5]\id\\
\hh \id\hstep\hld\step\id\step\id\hstep\step\id\\
\id\hstep\id\hstep\hx\step\id\hstep\cd\\
\hh \id\hstep\hlu\step\id\step\id\hstep\id\Step\id\\
\hh \cu\step\cu\hstep\id\Step\id\\
\hh \hstep\id\step\ld\step\rd\step\id\\
\hstep\id\step\id\step\hx\step\id\step\id\\
\hh \hstep\id\step\cu\step\cu\step\id \end{tangle}
\multiply\unitlens by 3
\end{array}
\end{equation}
Similarly as in \eqref{m-delta-ijk} and \eqref{m-delta-ijk1} we define
\unitlens 12pt %
\begin{equation}\mathlabel{aux-m-delta}
\begin{gathered}
\m_{i,jk} :=\proj_i\circ\m_B\circ(\inj_j\otimes\inj_k)=
\vstretch 80
\begin{tangle}
\tu{\sstyle i\vert jk}\\
\end{tangle}
\quad
\text{for all $i,j,k\in\{0,1,2\}$}\,,
\\
\Delta_{ij,k}:=(\proj_i\otimes\proj_j)\circ\Delta_B\circ\inj_k=
\vstretch 80
\begin{tangle}
\td{\sstyle ij\vert k}\\
\end{tangle}
\quad
\text{for all $i,j,k\in\{0,1,2\}$}\,.
\end{gathered}
\end{equation}
and
\begin{equation}\mathlabel{aux-m-delta1}
\begin{gathered}
\m^*_{0,20}:=\hat\sigma\circ(\mu_r\otimes\id_{B_2})\,,\quad
\m^*_{l,2m}:=\proj_l\circ\m^*_{0,20}\circ(\id_{B_2}\otimes\inj_m)
\\
\Delta^*_{01,0}:=(\id_{B_1}\otimes\nu_l)\circ\hat\rho\,,\quad
\Delta^*_{l1,m}:=(\proj_l\otimes\id_{B_1})\circ\Delta^*_{01,0}\circ\inj_m
\end{gathered}
\end{equation}
for $l,m\in\{0,1,2\}$. In particular \eqref{aux-m-delta} implies
\begin{equation*}
\begin{split}
\m_1&=\m_{1,11}\,,\\
\m_2&=\m_{2,22}\,,\\
\m_B&=\m_{0,00}\,,
\end{split}
\quad
\begin{split}
\Delta_1&=\Delta_{11,1}\,,\\
\Delta_2&=\Delta_{22,2}\,,\\
\Delta_B&=\Delta_{00,0}\,,
\end{split}
\quad
\begin{split}
\mu_l&=\m_{1,21}\,,\\
\mu_r&=\m_{2,21}\,,\\
\sigma&=\m_{1,22}\,,
\end{split}
\quad
\begin{split}
\nu_l&=\Delta_{21,1}\,,\\
\nu_r&=\Delta_{21,2}\,,\\
\rho&=\Delta_{11,2}\,.
\end{split}
\end{equation*}
\abs
The following morphisms
$\ixi rsklijmn,\iy rskli,\yi rsklj, \iy rskl2^*,
\yi rskl1^* :B_k\otimes B_l\to B_r\otimes B_s$
turn out to be useful in the sequel. Let $i,j,m,n,r,s,k,l\in\{0,1,2\}$, then
\begin{equation}\mathlabel{x-def}
\begin{split}
\ixi rsklijmn &:= (\m_{r,im}\otimes\m_{s,jn})\circ
(\id_{B_i}\otimes\Psi_{B_j,B_m}\otimes\id_{B_n})\circ
(\Delta_{ij,k}\otimes\Delta_{mn,l})\,,\\[3pt]
\iy rskli &:=(\m_{r,il}\otimes\id_{B_s})\circ
(\id_{B_i}\otimes\Psi_{B_s,B_l})\circ
(\Delta_{is,k}\otimes\id_{B_l})\,,\\
\yi rsklj &:=(\id_{B_r}\otimes\m_{s,kj})\circ
(\Psi_{B_k,B_r}\otimes\id_{B_j})\circ
(\id_{B_k}\otimes\Delta_{rj,l})\,,\\[3pt]
\iy rskl2^* &:=(\m^*_{r,2l}\otimes\id_{B_s})\circ
(\id_{B_2}\otimes\Psi_{B_s,B_l})\circ
(\Delta_{2s,k}\otimes\id_{B_l})\,,\\
\yi rskl1^* &:=(\id_{B_r}\otimes\m_{s,k1})\circ
(\Psi_{B_k,B_r}\otimes\id_{B_1})\circ
(\id_{B_k}\otimes\Delta^*_{r1,l})\,.
\end{split}
\end{equation}
For a strong Hopf datum $\mathfrak{h}$ we obtain the following bra-ket
decomposition of morphisms \eqref{x-def}.
\begin{lemma}\Label{bra-ket}
\begin{equation}\mathlabel{bra-ket-decomp}
\begin{gathered}
\ixi rskli0mn = (\id_{B_r}\otimes\m_{s,1s})\circ
(\iy r1kmi\otimes\id_{B_s}) \circ
(\id_{B_k}\otimes\yi ms2ln )\circ
(\Delta_{k2,k}\otimes\id_{B_l})
\end{gathered}
\end{equation}
for any $i,k,l,m,n,r,s\in\{0,1,2\}$.
\end{lemma}
\begin{proof}
The lemma follows straightforwardly from the
(co\n-)associativity of $\m_B$ and $\Delta_B$ and the identities
$\m_{0,12}=\id_{B_0}=\Delta_{12,0}$.\end{proof}
\abs
Below we define another set of auxiliary morphisms which will be used
in the course of the proof of Theorem \ref{hp-cp}.
\begin{gather}\mathlabel{aux-tau-t}
\divide\unitlens by 2
\tau^0_t:=
\begin{tangle}
\hh\cd\step\cd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\id\\
\id\step\id\step\ox{\sstyle 22}
\end{tangle}
\enspace ,
\enspace
\tau^1_t:=
\begin{tangle}
\hh\cd\step\cd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\cd\\
\id\step\id\step\ox{\sstyle 22}\step\id\\
\hh\id\step\id\step\id\Step\cu\\
\end{tangle}
\enspace ,
\enspace
\tau^2_t:=
\def5{5}
\begin{tangle}
\hh\cd\step\cd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\cd\\
\id\step\id\step\ox{\sstyle 22}\step\id\\
\hh\id\step\id\step\id\Step\cu*\\
\end{tangle}
\enspace ,
\enspace
\tau^{3}_t:=
\begin{tangle}
\hh\cd\step\cd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\cd\\
\id\step\id\step\ox{\sstyle 22}\step\rd\\
\hh\id\step\id\step\id\Step\x\step\id\\
\hh\id\step\id\step\id\step[1.5]\cd\hstep\hstr{200}\hlu\\
\end{tangle}
\\
\mathlabel{aux-tau-b}
\divide\unitlens by 2
\tau^0_b:=
\begin{tangle}
\ox{\sstyle 11}\step\id\step\id\\
\hh\id\step{\hstr{200}\hld}\step\id\step\id\\
\hh\id\step\id\step\x\step\id\\
\hh\id\step\cu\step\cu\\
\end{tangle}
\enspace ,
\enspace
\tau^1_b:=
\begin{tangle}
\hh\cd\Step\id\step\id\step\id\\
\id\step\ox{\sstyle 11}\step\id\step\id\\
\hh\cu\step{\hstr{200}\hld}\step\id\step\id\\
\hh\hstep\id\step[1.5]\id\step\x\step\id\\
\hh\hstep\id\step[1.5]\cu\step\cu\\
\end{tangle}
\enspace ,
\enspace
\tau^2_b:=
\def5{5}
\begin{tangle}
\hh\cd*\Step\id\step\id\step\id\\
\id\step\ox{\sstyle 11}\step\id\step\id\\
\hh\cu\step{\hstr{200}\hld}\step\id\step\id\\
\hh\hstep\id\step[1.5]\id\step\x\step\id\\
\hh\hstep\id\step[1.5]\cu\step\cu\\
\end{tangle}
\enspace ,
\enspace
\tau^{3}_b:=
\begin{tangle}
\hh{\hstr{200}\hrd}\hstep\cu\step[1.5]\id\step\id\step\id\\
\hh\id\step\x\Step\id\step\id\step\id\\
\lu\step\ox{\sstyle 11}\step\id\step\id\\
\hh\step\cu\step{\hstr{200}\hld}\step\id\step\id\\
\hh\step[1.5]\id\step[1.5]\id\step\x\step\id\\
\hh\step[1.5]\id\step[1.5]\cu\step\cu\\
\end{tangle}
\end{gather}
Besides we use the following definitions.
\begin{equation}\mathlabel{aux-tr-red}
\begin{gathered}
\sigma^{\mathrm{tr}}:=
\divide\unitlens by 2
\def5{5}
\begin{tangle}
\hh\cd\Step\cd\\
\id\step\ox{\sstyle 22}\step\id\\
\hh\id\step\id\Step\cu*
\end{tangle}
\enspace ,
\enspace
\rho^{\mathrm{tr}}:=
\begin{tangle}
\hh\cd*\Step\id\step\id\\
\id\step\ox{\sstyle 11}\step\id\\
\hh\cu\Step\cu
\end{tangle}
\enspace ,
\enspace
(\tau^3_t)^{\mathrm{red}_1}:=
\begin{tangle}
\hh\cd\step\cd\step[1.5]\id\\
\hh\id\step\x\step\id\step[1.5]\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\cd\\
\id\step\id\step\ox{\sstyle 22}\step\rd\\
\hh\id\step\id\step\id\Step\x\step\id\\
\hh\id\step\id\step\id\Step\id\step\hstr{200}\hlu\\
\end{tangle}
\enspace ,
\enspace
(\tau^3_b)^{\mathrm{red}_1}:=
\begin{tangle}
\hh{\hstr{200}\hrd}\step\id\Step\id\step\id\step\id\\
\hh\id\step\x\Step\id\step\id\step\id\\
\lu\step\ox{\sstyle 11}\step\id\step\id\\
\hh\step\cu\step{\hstr{200}\hld}\step\id\step\id\\
\hh\step[1.5]\id\step[1.5]\id\step\x\step\id\\
\hh\step[1.5]\id\step[1.5]\cu\step\cu\\
\end{tangle}
\\
\divide\unitlens by 2
(\tau^3_t)^{\mathrm{red}_2}:=
\begin{tangle}
\hh\cd\Step\cd\\
\id\step\ox{\sstyle 22}\step\rd\\
\hh\id\step\id\Step\x\step\id\\
\hh\id\step\id\Step\id\step\hstr{200}\hlu
\end{tangle}
\enspace ,
\enspace
(\tau^3_b)^{\mathrm{red}_2}:=
\begin{tangle}
\hh{\hstr{200}\hrd}\step\id\Step\id\step\id\\
\hh\id\step\x\Step\id\step\id\\
\lu\step\ox{\sstyle 11}\step\id\\
\hh\step\cu\Step\cu
\end{tangle}
\enspace ,
\enspace
(\tau^3_t)^{\mathrm{red}_3}:=
\begin{tangle}
\hh\cd\step\cd\step\id\\
\hh\id\step\x\step\id\step\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\rd\\
\vstr{50}\id\step\id\step\x\step\id\\
\hh\id\step\id\step\id\Step\hstr{200}\hlu
\end{tangle}
\enspace ,
\enspace
(\tau^3_b)^{\mathrm{red}_3}:=
\begin{tangle}
\hh{\hstr{200}\hrd}\Step\id\step\id\step\id\\
\vstr{50}\id\step\x\step\id\step\id\\
\hh\lu\step{\hstr{200}\hld}\step\id\step\id\\
\hh\step\id\step\id\step\x\step\id\\
\hh\step\id\step\cu\step\cu
\end{tangle}
\end{gathered}
\end{equation}
Observe that the morphisms $(\tau_t)^{\mathrm{red}}$ and
$(\tau_b)^{\mathrm{red}}$ in \eqref{aux-tr-red}
can be expressed with the help $\tau^3_t$ and $\tau^3_b$
respectively. For instance
$(\tau^3_t)^{\mathrm{red}_1}=(\id_{B_2}\otimes\id_{B_1}\otimes
\id_{B_2}\otimes\varepsilon_2\otimes\id_{B_2}\otimes\id_{B_1})\circ\tau^3_t$.
\begin{lemma}\Label{tau-delta}
For a strong Hopf datum $\mathfrak{h}$ the following reduction
identities are satisfied.
\begin{gather}
\mathlabel{tau-R0-aux}
\yi i0201=(\yi i1201\otimes\id_{B_2})\circ\tau^0_t\,,
\quad
\iy 010i1=\tau^0_b\circ(\id_{B_1}\otimes\iy 012i1)\,,
\\[5pt]
\mathlabel{tau-R-aux}
\yi i0200=(\yi i1200\otimes\id_{B_2})\circ\tau^1_t\,,
\quad
\iy 010i0=\tau^1_b\circ(\id_{B_1}\otimes\iy 012i0)\,,
\\[5pt]
\mathlabel{tau-R2-aux}
\yi i1200
=(\id_B\otimes\m_1)\circ(\yi i1201\otimes\id_{B_1})\circ\tau^2_t\,,
\quad
\iy 012i0
=\tau^2_b\circ(\id_{B_2}\otimes\iy 012i2)\circ(\Delta_2\otimes\id_B)\,,
\\[5pt]
\mathlabel{tau-check-aux}
\yi i1220
=(\id_B\otimes\m_1)\circ(\yi i1221\otimes\id_{B_1})\circ\sigma^{\mathrm{tr}}\,,
\enspace
\iy 112i0
=\rho^{\mathrm{tr}}\circ(\id_{B_2}\otimes\iy 112i2)\circ(\Delta_2\otimes\id_B)\,.
\end{gather}
\end{lemma}
\begin{proof}
We only consider the first identities of
\eqref{tau-R0-aux} and \eqref{tau-R-aux}.
The second identities in each row are $\pi$-symmetric analogues.
The identities \eqref{tau-R2-aux} follow from
\eqref{tau-R0-aux} by application of the mapping
$f\mapsto(\id\otimes\m_{1,02})\circ(f\otimes\id_{B_2})
\circ(\id\otimes\Delta_2)$. Relations \eqref{tau-check-aux} are
special cases of \eqref{tau-R2-aux} through composition with $\inj_2$.
In a first step we derive \eqref{tau-R0-aux} and \eqref{tau-R-aux}
for the case i=2. For \eqref{tau-R0-aux} we obtain
\begin{equation}\mathlabel{phi-phi}
\yi 20201 =
\divide\unitlens by 3
\multiply\unitlens by 2
\begin{tangle}
\id\step\ox{\sstyle 12}\\
\hh\x\Step\id\\
\id\step\ox{\sstyle 21}
\end{tangle}
=
\begin{tangle}
\hh\step\id\step{\hstr{200}\cd\step\id}\\
\step\id\step\id\Step\ox{\sstyle 12}\\
\step\id\step\ox{\sstyle 12}\Step\id\\
\hh\step\x\Step\id\Step\id\\
\hh\sw1{\hstr{200}\cd}\step\id\Step\id\\
\hh\id\step\id\Step\x\Step\id\\
\hh\id\step{\hstr{400}\hlu}\step{\hstr{400}\hru}
\end{tangle}
=
\begin{tangle}
\vstretch 110
\vstr{55}\hcd\step[2]\hcd\step\id\\
\vstr{55}\id\step\x\step\id\step\id\\
\vstr{55}\id\step\id\step[2]\ru\step\id\\
\id\step\id\Step\ox{\sstyle 22}\\
\id\step\ox{\sstyle 12}\Step\id\\
\hh\x\Step\id\Step\id\\
\vstr{44}\id\step\Lu\Step\id
\end{tangle}
\end{equation}
where the second equation is an immediate consequence of
the relative entwining property \eqref{phi-prod-rel} of $\varphi_{1,2}$
and the third equation follows from \eqref{rel-phi}
of Lemma~\ref{strong-hd2} and the right module property of
$(B_2,\mu_r)$.
To derive the first identiy of \eqref{tau-R-aux} for i=2
we use \eqref{phi-prod}, the relative entwining property
\eqref{phi-prod-rel} of $\varphi_{1,2}$, \eqref{act-coact-triv}
of Definition \ref{strong-hp}, the coassociativity of $\Delta_2$ and the
entwining property \eqref{entw1} of $\varphi_{2,2}$.
Using \eqref{hat-prod-rel}, coassociativity of $\Delta_2$ and
the entwining property \eqref{entw1} of $\varphi_{2,2}$ we obtain
the relations $(\id_{B_1}\otimes\tau^j_t)\circ f=
(f\otimes\id_{B_2})\circ\tau^j_t\,$ for $j\in\{0,1\}$. We set
$f:=\left((\Psi_{B_2,B_1}\otimes\id_{B_1})\circ(\id_{B_2}\otimes\hat\rho)
\otimes\id_{B_2}\right)\circ(\id_{B_2\otimes B_1}\otimes\Delta_2)$.
These relations allow us to perform easliy the step from i=2 to i=0.
Finally, the case i=1 will be obtained from the corresponding identities
for i=0 by composition with $\varepsilon_2$.
\end{proof}
\begin{lemma}
For a strong Hopf datum $\mathfrak{h}$ the second module-coalgebra
compatibility, the first comodule-algebra compatibility and the
cycle-cocycle compatibility of Definition \ref{hp} can be respectively
converted to
\begin{gather}
\hat\rho\circ\varphi_{2,1}
=\rho^{\mathrm{tr}}\circ\left(\id_{B_2}\otimes\ixi 11212011\right)
\circ(\Delta_2\otimes\id_{B_1})\,,
\notag
\\
\mathlabel{phi-hat-new}
\varphi_{1,2}\circ\hat\sigma
=(\id_{B_2}\otimes\m_1)\circ\left(\ixi 21222021\otimes\id_{B_1}\right)
\circ\sigma^{\mathrm{tr}}\,,
\\
\hat\rho\circ\hat\sigma
=(\id_{B_1}\otimes\m_1)\circ(\rho^{\mathrm{tr}}\otimes\id_{B_1})\circ
\left(\id_{B_2}\otimes\ixi 11222001\otimes\id_{B_1}\right)\circ
(\id_{B_2}\otimes\sigma^{\mathrm{tr}})\circ(\Delta_2\otimes\id_{B_2})\,.
\notag
\end{gather}
\end{lemma}
\begin{proof}
The morphisms on right hand side of the corresponding
compatibility relations can be decomposed with the help of
\eqref{bra-ket-decomp}. Then the reduction formulas
\eqref{tau-check-aux} will be applied to the particular tensor factors.
And finally we use again \eqref{bra-ket-decomp} to obtain the
result.\end{proof}
\abs
The previous lemma leads to
\begin{lemma}
\begin{equation}\mathlabel{yi-red-adv}
\divide\unitlens by 2
\def5{3}
\yi 01221 =
\begin{tangle}
\hh\step\id\step[3]\id\\
\hhstep\ffbox6{(\tau^3_t)^{\mathrm{red}_2}}\\
\vstr{110}\hh\id\step{\hstr{200}\cd*}\step\id\step\id\\
\hh\x\step{\hstr{200}\hld}\step\id\step\id\\
\hh\id\step\x\step\id\step\id\step\id\\
\hh\id\step\id\step{\hstr{200}\hlu}\step\id\step\id\\
\hh\id\step\id\Step\x\step\id\\
\hh\id\step{\hstr{200}\cu}\step\cu
\end{tangle}
\quad,\quad
\iy 11202 =
\begin{tangle}
\hh\cd\step{\hstr{200}\cd}\step\id\\
\hh\id\step\x\Step\id\step\id\\
\hh\id\step\id\step{\hstr{200}\hrd}\step\id\step\id\\
\hh\id\step\id\step\id\step\x\step\id\\
\hh\id\step\id\step{\hstr{200}\hru}\step\x\\
\vstr{110}\hh\id\step\id\step{\hstr{200}\cu*}\step\id\\
\hhstep\ffbox6{(\tau^3_b)^{\mathrm{red}_2}}\\
\hh\step\id\step[3]\id
\end{tangle}
\quad,\quad
\ixi 11222001 =
\begin{tangle}
\hh\step\id\step[4]\id\\
\hhstep\ffbox7{(\tau^3_t)^{\mathrm{red}_2}}\\
\vstr{110}\hh\cd\step{\hstr{200}\cd*}\step\id\step\id\\
\hh\id\step\x\Step\id\step\id\step\id\\
\id\step\id\step\ox{\sstyle 21}\step\id\step\id\\
\id\step\id\step\ox{\sstyle 12}\step\id\step\id\\
\hh\id\step\id\step\id\Step\x\step\id\\
\vstr{110}\hh\id\step\id\step{\hstr{200}\cu*}\step\cu\\
\hhstep\ffbox7{(\tau^3_b)^{\mathrm{red}_2}}\\
\hh\step\id\step[4]\id
\end{tangle}
\end{equation}
\begin{equation}\mathlabel{yi-red-adv2}
\divide\unitlens by 2
\yi 21201 =
\begin{tangle}
\hh\step[1.5]\id\step[1.5]\id\step[1.5]\id\\
\ffbox6{(\tau^3_t)^{\mathrm{red}_3}}\\
\hh\step\id\Step\id\step[1.5]\id\step\id\\
\ffbox4{\yi 21211}\hstep\id\step\id\\
\hh\hstep\id\step[3]\x\step\id\\
\hh\hstep{\hstr{300}\cu}\step\cu
\end{tangle}
\quad,\quad
\iy 10212 =
\begin{tangle}
\hh\cd\step{\hstr{300}\cd}\\
\hh\id\step\x\step[3]\id\\
\id\step\id\hstep\ffbox4{\iy 21212}\\
\hh\id\step\id\step[1.5]\id\Step\id\\
\hhstep\ffbox6{(\tau^3_b)^{\mathrm{red}_3}}\\
\hh\step\id\step[1.5]\id\step[1.5]\id
\end{tangle}
\end{equation}
\begin{equation}\mathlabel{yi-red-adv3}
\divide\unitlens by 2
\yi 01201 =
\begin{tangle}
\hh\step[1.5]\id\step[1.5]\id\step[1.5]\id\\
\ffbox6{(\tau^3_t)^{\mathrm{red}_1}}\\
\hh\hstep\id\step[1.5]\id\step[1.5]\id\step\id\step\id\\
\ffbox4{\yi 01201^*}\hstep\id\step\id\\
\hh\hstep\id\step[1.5]\id\step[1.5]\x\step\id\\
\hh\hstep\id\step[1.5]{\hstr{150}\cu}\step\cu
\end{tangle}
\quad,\quad
\iy 01202 =
\begin{tangle}
\hh\cd\step{\hstr{150}\cd}\step[1.5]\id\\
\hh\id\step\x\step[1.5]\id\step[1.5]\id\\
\id\step\id\hstep\ffbox4{\iy 01202^*}\\
\hh\id\step\id\step\id\step[1.5]\id\step[1.5]\id\\
\hhstep\ffbox6{(\tau^3_b)^{\mathrm{red}_1}}\\
\hh\step\id\step[1.5]\id\step[1.5]\id
\end{tangle}
\end{equation}
\end{lemma}
\begin{proof}
The first identity of \eqref{yi-red-adv} has been derived
by successive application of the left module-algebra
compatibility of Definition \ref{hp} and the relations
\eqref{mod-alg-rel} and \eqref{rel-cocycle-assoc1}.
The second identity is its $\pi$-symmetric counterpart.
To get the third identity we use \eqref{bra-ket-decomp},
the first and the second equation of \eqref{yi-red-adv},
the relations \eqref{rel-cocycle-assoc2} and the module-comodule
compatibility.
The first identity in \eqref{yi-red-adv2} is obtained from the left
module-algebra compatibility and \eqref{mod-alg-rel}.
The identities in \eqref{yi-red-adv3} are derived from
\eqref{yi-red-adv2} with the help of \eqref{hat-prod-rel}.\end{proof}
\abs
The next lemma is a straightforward implication of the previous result.
\begin{lemma}
\begin{equation}\mathlabel{ixi-rel}
\divide\unitlens by 4
\multiply\unitlens by 3
\hstretch 120
\begin{tangle}
\hh\hstep\id\step\id\\
\ffbox2{\ixi 11222001}\\
\hh\hstep\id\step\id\\
\hh\hstep\id\hstep\hru
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step\id\\
\ffbox2{\ixi 11222221}\\
\hh\hstep\id\step\id\\
\hh\hstep\id\hstep\hru
\end{tangle}
\quad,\quad
\begin{tangle}
\hh\hstep\hld\hstep\id\\
\hh\hstep\id\step\id\\
\ffbox2{\ixi 11222001}\\
\hh\hstep\id\step\id
\end{tangle}
=
\begin{tangle}
\hh\hstep\hld\hstep\id\\
\hh\hstep\id\step\id\\
\ffbox2{\ixi 11222111}\\
\hh\hstep\id\step\id
\end{tangle}
\end{equation}
\end{lemma}
\begin{proof}
Both identities follow from the third identity of \eqref{yi-red-adv}
using \eqref{act-coact-triv}, \eqref{rel-cocycle-assoc1} and
\eqref{cocycle-triv1}.\end{proof}
\begin{lemma}\Label{sigma-Delta}
Given a strong Hopf datum $\mathfrak{h}$, then it holds
\begin{gather}
\mathlabel{sigma-Delta-rel}
\divide\unitlens by 2
\def5{6}
\begin{tangle}
\cu*\\
\cd\\
\hh\id\step[1.5]\hru
\end{tangle}
=
\multiply\unitlens by 2
\begin{tangle}
\vstr{60}\hh\hstep\id\step\id\\
\vstr{80}\ffbox2{\ixi 11222020}\\
\vstr{60}\hh\hstep\id\step\id\\
\hstr{50}\vstr{50}\hh\step\id\step[1.5]\hru\\
\end{tangle}
\quad,
\quad
\vstretch 100
\divide\unitlens by 2
\begin{tangle}
\hh\hld\step[1.5]\id\\
\cu\\
\cd*
\end{tangle}
=
\multiply\unitlens by 2
\begin{tangle}
\hstr{50}\vstr{50}\hh\step\hld\step[1.5]\id\\
\vstr{60}\hh\hstep\id\step\id\\
\vstr{80}\ffbox2{\ixi 11220011}\\
\vstr{60}\hh\hstep\id\step\id\\
\end{tangle}
\\
\mathlabel{hat-sigma-Delta}
\hstretch 70
\begin{tangle}
\vstr{60}\hh\hstep\id\step\id\\
\hh\ffbox2{\hat\sigma}\\
\vstr{60}\hh\cd\hstep\id\\
\vstr{60}\hh\id\hstep\hru\hstep\id
\end{tangle}
=
\begin{tangle}
\vstr{60}\hh\step\id\Step\id\\
\vstr{150}\hh\ffbox4{\ixi 10222020}\\
\vstr{60}\hh\hstep\id\Step\id\step\id\\
\vstr{60}\hh\hstep\id\step[1.5]\hru\step\id
\end{tangle}
\quad,
\quad
\begin{tangle}
\vstr{60}\hh\hstep\id\hstep\hld\hstep\id\\
\vstr{60}\hh\hstep\id\hstep\cu\\
\hh\ffbox2{\hat\rho}\\
\vstr{60}\hh\hstep\id\step\id
\end{tangle}
=
\begin{tangle}
\vstr{60}\hh\hstep\id\step\hld\step[1.5]\id\\
\vstr{60}\hh\hstep\id\step\id\Step\id\\
\vstr{150}\hh\ffbox4{\ixi 01220011}\\
\vstr{60}\hh\step\id\Step\id
\end{tangle}
\end{gather}
\end{lemma}
\begin{proof}
We prove the first identity of \eqref{sigma-Delta-rel} graphically.
\begin{equation}\mathlabel{sigma-Delta-expl}
\divide\unitlens by 2
\def5{6}
\begin{tangle}
\cu*\\
\cd\\
\hh\id\step[1.5]\hru
\end{tangle}
=
\multiply\unitlens by 2
\vstretch 60
\begin{tangle}
\hh\hstep\id\step\id\\
\vstr{80}\ffbox2{\ixi 11222000}\\
\hh\hstep\id\step\id\\
\hstr{50}\vstr{50}\hh\step\id\step[1.5]\hru\\
\end{tangle}
=
\begin{tangle}
\hh\step\id\step\id\\
\vstr{40}\ffbox3{\sigma^{\mathrm{tr}}}\\
\hh\hstep\id\step\id\step\id\\
\vstr{80}\ffbox2{\ixi 11222001}\hstep\id\\
\vstr{40}\hh\hstep\id\step\id\step\id\\
\hh\vstr{50}{\hstr{50}\step\id\step[1.5]\hru\hhstep}\hstr{125}\ru\\
\vstr{40}\hh\hstep\id\step[0.75]\id
\end{tangle}
=
\begin{tangle}
\hh\step\id\step\id\\
\vstr{40}\ffbox3{\sigma^{\mathrm{tr}}}\\
\hh\hstep\id\step\id\step\id\\
\vstr{80}\ffbox2{\ixi 11222221}\hstep\id\\
\vstr{40}\hh\hstep\id\step\id\step\id\\
\hh\vstr{50}{\hstr{50}\step\id\step[1.5]\hru\hhstep}\hstr{125}\ru\\
\vstr{40}\hh\hstep\id\step[0.75]\id
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step\id\\
\vstr{80}\ffbox2{\ixi 11222020}\\
\hh\hstep\id\step\id\\
\hstr{50}\vstr{50}\hh\step\id\step[1.5]\hru\\
\end{tangle}
\end{equation}
The first equation in the graphic has been obtained with the help of the
cycle-cocycle compatibility (Definition \ref{hp}),
\eqref{act-coact-triv} and \eqref{cocycle-triv1}. In the second
equation we use \eqref{bra-ket-decomp} and \eqref{tau-check-aux}.
The third equation is \eqref{ixi-rel}. Application of
\eqref{tau-check-aux} to the fourth diagram yields the fifth relation.
The identities \eqref{hat-sigma-Delta} follow from
\eqref{sigma-Delta-rel} with the help of the entwining properties of
$\varphi_{2,2}$ and $\varphi_{1,1}$ respectively.
\end{proof}
\begin{lemma}\Label{phi-sigma-rel}
Let $\mathfrak{h}$ be a strong Hopf datum. Then
\unitlens 15pt %
\begin{equation}\mathlabel{phi-sigma}
\divide\unitlens by 2
\begin{tangle}
\hh\id\step\id\step\id\\
\hh\id\hstep\ffbox 2{\hat\sigma}\\
\ru\step\id\\
\ox{\sstyle 22}
\end{tangle}
=
\hstretch 120
\begin{tangle}
\hh\id\Step\cd\step\id\\
\ox{22}\hstep\ffbox2{\yi 21220}\\
\hh\id\Step\x\step\id\\
\cu\step\ru
\end{tangle}
\end{equation}
and the corresponding $\pi$-symmetric version for $\hat\rho$ holds.
\end{lemma}
\begin{proof}
The identity follows straighforwardly from \eqref{phi-hat-new}.
\end{proof}
\begin{lemma}
\begin{equation}\mathlabel{yi-red-adv4}
\unitlens = 8pt
\yi 00221^*=
\begin{tangle}
\hh\hstep\hstr{200}\cd\step\id\\
\hstep\id\Step\ox{\sstyle 22}\\
\ffbox3{\yi 01221^*}\step[1.5]\id\\
\hh\hstep\id\step\id\step\id\Step\id
\end{tangle}
\quad,\quad
\iy 11002^*=
\begin{tangle}
\hh\id\Step\id\step\id\step\id\\
\id\step[1.5]\ffbox3{\iy 11202^*}\\
\ox{\sstyle 11}\Step\id\\
\hh\hstr{200}\id\step\cu
\end{tangle}
\end{equation}
\begin{equation}\mathlabel{yi-red-adv5}
\unitlens = 8pt
\yi 00221=
\begin{tangle}
\hh\hstep\hstr{200}\cd\step\id\\
\hstep\id\Step\ox{\sstyle 22}\\
\ffbox3{\yi 01221}\step[1.5]\id\\
\hh\hstep\id\step\id\step\id\Step\id
\end{tangle}
\quad,\quad
\iy 11002=
\begin{tangle}
\hh\id\Step\id\step\id\step\id\\
\id\step[1.5]\ffbox3{\iy 11202}\\
\ox{\sstyle 11}\Step\id\\
\hh\hstr{200}\id\step\cu
\end{tangle}
\end{equation}
\end{lemma}
\begin{proof}
The identities \eqref{yi-red-adv4} follow from
\eqref{rel-cocycle-assoc1}, \eqref{mod-alg-rel}, whereas
\eqref{yi-red-adv5} will be derived from \eqref{yi-red-adv}
and the entwining properties \eqref{entw1}, \eqref{phi-prod2} of
$\varphi_{2,2}$, $\varphi_{2,1}$.
\end{proof}
\unitlens 12pt %
\subsection{Proof of Theorem \ref{hp-cp}}
Given a strong Hopf datum $\mathfrak{h}$ we have to prove that the object
$B=B_1\otimes B_2$ provided with the structure given in Theorem \ref{hp-cp},
is a strong cross product bialgebra. The (co\n-)unital identities
\eqref{type-alpha1} can be verified easily. The ``strong conditions"
of Definition \ref{strong-type-alpha} hold by construction.
Using the first and the second identity of Lemma \ref{cocycle-triv2} it follows
straightforwardly that the ``projection relations" \eqref{type-alpha2} are
fulfilled. It remains to show that $B$ is a bialgebra.
\begin{proposition}\Label{co-assoc}
Let $\mathfrak{h}$ be a strong Hopf datum and $\varphi_{1,2}$, $\varphi_{2,1}$,
$\hat\rho$, and $\hat\sigma$ be the morphisms defined in \eqref{phi-sigma-rho2}.
Then $B=B_1\otimes B_2$ is a cocycle cross product algebra
$B_1\cpcyalg B_2$ and a cycle cross product coalgebra $B_1\cpcyco B_2$.
\end{proposition}
\begin{proof}
According to Proposition \ref{cross-equiv} the identities \eqref{cond3-caat}
have to be verified in order to prove that $B$ is a cross product
algebra. Similar $\pi$-symmetric procedures are needed to demonstrate that
$B$ is a cross product coalgebra.
Without difficulties the unital identities of \eqref{cond3-caat} can be
verified. The fifth relation of \eqref{cond3-caat} has been proven
in Lemma \ref{strong-hd7} in the first identity of \eqref{phi-prod2}.
The seventh identity of \eqref{cond3-caat} will be proven subsequently.
\begin{equation*}
\unitlens 10pt %
\begin{split}
&\begin{tangle}
\hh\id\step\id\Step\id\\
\hh\hhstep\ffbox2{\hat\sigma}\step[1.5]\id\\
\id\step\ox{21}\\
\hh\cu\step[2]\id
\end{tangle}
=\enspace
\def5{3}
\begin{tangle}
\hh\cd\Step\cd\step\id\\
\id\step\ox{22}\step\id\step\id\\
\hh\hhstep\ffbox2{\hat\sigma}\step[1.5]\cu\hstep\cd\\
\hh\id\step\id\step[2.5]\x\step\id\\
\hh\id\step{\hstr{250}\lu}\step\ru\\
\hh{\hstr{350}\cu}\step\id
\end{tangle}
=
\hstretch 80 \vstretch 80
\begin{tangle}
\hh{\hstr{160}\cd}\Step\cd\step\hstr{160}\cd\\
\id\Step\ox{22}\step\hx\Step\id\\
\id\Step\id\Step\hx\step\ox{21}\\
\id\Step\ox{21}\step\ru\Step\id\\
\ox{21}\Step\id\step\hstr{120}\cu\\
\hh\id\Step{\hstr{160}\cu*}\step[2.5]\id\\
\hh{\hstr{240}\cu}\step[3.5]\id
\end{tangle}
=
\def5{5}
\begin{tangle}
\hh\cd\Step{\hstr{200}\cd}\step[3.5]\id\\
\id\step\ox{22}\Step\hcd\Step\cd\\
\id\step\id\Step\ox{22}\step\x\Step\id\\
\hh\id\step\d\step[1.5]\id\Step\x\step[1.5]\cd\step\cd\\
\hh\d\step\d\step\d\step\cd\hstep\d\step\id\step\x\step\id\\
\hh\hstep\d\step\d\step\x\step\id\step\id\step
\hstr{160}\hlu\hstep\hru\\
\hh\step\d\step{\hstr{160}\hlu\hstep\hru}\step\Ru\step\id\\
\step[1.5]\ox{21}\step\id\Step\hstr{120}\cu\\
\hh\step[1.5]\id\Step\cu*\step[3.5]\id\\
\hh\step[1.5]{\hstr{200}\cu}\step[4]\id
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step[1.5]{\hstr{240}\cd}\step[3.5]\id\\
\hh\hstep\id\step[1.5]\id\Step{\hstr{160}\cd}\Step\cd\\
\hh\hstep\id\step[1.5]\id\Step\id\Step{\hstr{160}\x}\step\id\\
\hh\hstep\id\step[1.5]\id\Step\id\step\hstr{160}\cd\hstep\hru\\
\hh\hstep\id\step[1.5]\id\Step\x\Step\nw1\nw1\\
\hh\hstep\id\step[1.5]\id\step\sw1{\hstr{160}\cd\hstep\cd}\d\\
\hh\cd\step\hrd\hstep\id\step{\hstr{160}\hrd}\step\x\step{\hstr{160}\hld}
\hstep\id\\
\hh\id\step\x\hstep\id\hstep\id\step\id\step\x\step\x\step\id\hstep\id\\
\hh\id\step\id\step\hru\hstep\id\step{\hstr{160}\hru}\step\x\step
{\hstr{160}\hlu}\hstep\id\\
\hh\id\step\id\step\x\step{\hstr{160}\cu\hstep\cu}\hstep\id\\
\hh\id\step{\hstr{160}\hlu\hstep\x\step\dd}\step\dd\\
\ox{21}\step\id\Step\Ru\sw2\\
\hh\id\Step\cu*\Step\hstr{160}\cu\\
\hh{\hstr{200}\cu}\step[3.5]\id
\end{tangle}
\\
&=
\def5{5}
\vstretch 80 \hstretch 80
\begin{tangle}
\hh\hstep\id\step[2.5]{\hstr{160}\cd}\step[3]\id\\
\hh\hstep\id\step[2.5]\id\step{\hstr{160}\cd}\step[1.5]\cd\\
\hh\hstep\id\step[2.5]\id\step\id\Step{\hstr{120}\x}\step\id\\
\hh\hstep\id\step[2.5]\id\step\id\step{\hstr{160}\cd}\hstep\ru\\
\hh\cd\Step\id\step\x\Step\id\hstep\id\\
\id\step\ox{22}\step\id\step\ox{21}\hstep\id\\
\id\step{\hstr{160}\dh}\step\hx\step\ox{12}\hstep\id\\
\hh\nw1\step{\hstr{160}\hlu}\step\x\Step\id\hstep\id\\
\step\ox{21}\step\id\step\Ru\ddh\\
\hh\step\id\Step\cu*\step{\hstr{160}\cu}\\
\hh\step{\hstr{200}\cu}\step[2.5]\id
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step[2.5]{\hstr{160}\cd}\step[3]\id\\
\hh\hstep\id\step[2.5]\id\step\hstr{160}\cd\hstep\cd\\
\hh\hstep\id\step[2.5]\id\step\id\Step\x\Step\id\\
\hh\hstep\id\step[2.5]\id\step\id\step{\hstr{160}\cd}
\nw1\step\id\\
\hh\cd\Step\id\step\x\Step\id\step\ru\\
\id\step\ox{22}\step\id\step\ox{21}\step\id\\
\hh\id\step\id\Step\x\step\id\Step\id\step\id\\
\hh\nw1{\hstr{320}\hlu}\step{\hstr{160}\hru}\step\sw1\step\id\\
\step\ox{21}\step\ox{22}\Step\id\\
\hh\step\id\Step\cu*\Step\hstr{160}\cu\\
\hh\step{\hstr{200}\cu}\step[3.5]\id
\end{tangle}
=
\begin{tangle}
\hh\step\id\step[2.5]{\hstr{160}\cd}\step\cd\\
\hh\step\id\step[2.5]\id\Step\x\step\id\\
\step\id\step[2.5]\ox{21}\step\ru\\
\hh{\hstr{160}\cd}\step\cd\step[1.5]\id\step\id\\
\hh\id\Step\x\step\id\step[1.5]\id\step\id\\
\ox{21}\step\ru\step\ddh\step\id\\
\id\Step\id\step\ox{22}\step[1.5]\id\\
\hh\id\Step\cu*\Step\hstr{120}\cu\\
\hh{\hstr{200}\cu}\step[3.25]\id
\end{tangle}
=
\begin{tangle}
\id\Step\ox{21}\\
\ox{21}\Step\d\\
\hh\id\step[1.5]\cd\Step\cd\\
\id\step[1.5]\id\step\ox{22}\step\id\\
\hh\id\step[1.5]\cu*\Step\cu\\
\hh{\hstr{160}\cu}\step[3]\id
\end{tangle}
\end{split}
\end{equation*}
\unitlens 12pt %
where the first identity has been obtained from \eqref{hat-prod}.
We use the weak associativity of $\mu_l$ and
the module-algebra compatibility of $\mu_r$ according to Definition \ref{hp}
to get the second equation in the graphic. Then the entwining property
of $\varphi_{2,2}$ (see Lemma \ref{strong-hd6}) yields the third relation.
In the fourth identity we use \eqref{rel-cocycle-assoc3} and the relative
coassociativity of $\Delta_1$ corresponding to Lemma \ref{strong-hd3}.
To derive the fifth identity the module-comodule compatibility of
Definition \ref{hp} has been used. In the sixth equation we apply
\eqref{phi-prod-rel2}, and in the seventh equation we use \eqref{phi-prod}.
Hence the seventh identity of \eqref{cond3-caat} has been verified.
Finally we will prove the sixth identity of \eqref{cond3-caat}. Its left
hand side will be transformed according to
\begin{equation}
\mathlabel{assoc-2l}
\divide\unitlens by 3
\multiply\unitlens by 2
\def5{5}
\begin{split}
&\hstretch 80 \vstretch 80
\begin{tangle}
\hh\id\Step\id\step[3]\id\\
\hh\id\step[1.5]\ffbox4{\hat\sigma}\\
\ox{\sstyle 21}\step[3]\id\\
\hh\id\step[1.5]\cd\Step\cd\\
\id\step[1.5]\id\step\ox{\sstyle 22}\step\id\\
\hh\id\step[1.5]\cu*\Step\cu\\
\hh{\hstr{160}\cu}\step[3]\id
\end{tangle}
=
\begin{tangle}
\hh\step\id\step[2.5]\cd\Step\cd\\
\step\id\step[2.5]\id\step\ox{\sstyle 22}\step\id\\
\hh\step\id\Step\ffbox2{\hat\sigma}\step[1.5]\cu\\
\hh{\hstr{160}\cd}\step\cd\hstep\d\Step\id\\
\hh\id\Step\x\step\id\step\id\Step\id\\
\hh\id\Step\id\step{\hstr{160}\hru}\step\id\Step\id\\
\ox{\sstyle 21}\step\ox{\sstyle 22}\Step\id\\
\hh\id\Step\cu*\Step\hstr{160}\cu\\
\hh{\hstr{200}\cu}\step[3.5]\id
\end{tangle}
=
\begin{tangle}
\hh\step[1.5]\id\step[4]\cd\Step\cd\\
\step[1.5]\id\Step\sw2\Step\ox{\sstyle 22}\step\id\\
\hh\step[1.5]\id\step{\hstr{160}\cd\hstep\cd}\step\cu\\
\hh\step[1.5]\id\step{\hstr{160}\hrd}\step\id\step{\hstr{320}\hrd}\nw1
\hstep\d\\
\hh\step[1.5]\id\step\id\step\id\step\x\Step\id\step\id\step\id\step\\
\step[1.5]\id\step\id\step\hx\step\ox{\sstyle 21}\step\id\step\id\\
\hh\step\cd\hstep\cu*\step\cu\step[1.5]\ffbox2{\hat\sigma}\hstep\id\\
\hh\sw1\step\x\Step\id\step[1.5]\sw1\sw1\step\id\\
\hh\id\Step\id\step{\hstr{320}\hru}\hstep\sw1\sw1\Step\id\\
\hh\id\Step\id\step{\hstr{400}\hru}\sw1\step[3]\id\\
\ox{\sstyle 21}\step{\hstr{100}\ox{\sstyle 22}}\Step\sw2\\
\Put(0,0)[lb]{\begin{tangle}\hh\id\Step\cu*\\ \hh{\hstr{200}\cu}
\end{tangle}}\step[5.5]\cu
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\hstr{240}\step\cd\hstr{160}\step\cd\\
\hstep\id\step[3]\id\step[3]\ox{\sstyle 22}\Step\id\\
\hh\cd\Step\cd\step{\hstr{240}\cd}\hstep\hstr{160}\cu\\
\id\step\ox{\sstyle 22}\step\Put(0,0)[lb]{\begin{tangle}
\hh\id\step{\hstr{320}\hrd}\step\nw1\hstep\nw1\\
\hh\x\Step\id\Step\nw1\hstep\nw1
\end{tangle}}\\
\id\step\id\Step\Put(0,0)[lb]{\begin{tangle}
\hh\x\\ \hh\id\step\d\end{tangle}}
\Step\ox{\sstyle 21}\step[3]\id\step[1.5]\hstr{160}\hd\\
\hh\id\step\nw1\step\id\step[1.5]\hru\step[1.5]\cd\step{\hstr{240}\cd}
\step\id\\
\Put(0,0)[lb]{\begin{tangle}
\hh\d\step[1.5]\cu*\\
\hstep\ox{\sstyle 21}\\
\hh\hstep\id\Step\nw1
\end{tangle}}
\step[4.5]
\Put(0,0)[lb]{\begin{tangle}
\ox{\sstyle 22}\\ \id\Step\hx
\end{tangle}}
\step[3]
\Put(0,0)[lb]{\begin{tangle}
\hh\id\step{\hstr{320}\hrd}\\
\hh\x\Step\id\\
\step\ox{\sstyle 21}
\end{tangle}}
\vstr{160}\step[4]\id\step\id\\
\hh\hstep\nw1\Step\id\step{\hstr{160}\cu\hstep\hru\step}\cu*\step\id\\
\hh\step[1.5]\nw1\step\def5{2}{\hstr{160}\cu*\step}
{\hstr{560}\hru}\step[1.5]\id\\
\hh\step[2.5]{\hstr{160}\cu}\step[3]\hstr{400}\cu
\end{tangle}
\\
&=
\hstretch 80 \vstretch 80
\begin{tangle}
\hh\hstep\id\step[4.5]{\hstr{240}\cd}\Step{\hstr{160}\cd}\\
\hstep\id\step[3]\Put(0,0)[lb]{\begin{tangle}
\hh\hstep\sw1\\ \hh\cd\end{tangle}}
\step[4.5]\ox{\sstyle 22}\Step\id\\
\hh\hstep\id\step[3]\id\hstep\cd\Step{\hstr{160}\cd\hstep\cu}\\
\hcd\Step\ddh\hstep\id\step\ox{\sstyle 22}\Step
\Put(0,0)[lb]{\begin{tangle}\hh\nw1\step\nw1\\
\hh\step\d\step[1.5]\nw1
\end{tangle}}\\
\id\step\ox{\sstyle 22}\step
\Put(0,0)[lb]{\begin{tangle}
\hh\id\step{\hstr{320}\hrd}\nw1\\
\hh\x\Step\id\step\id
\end{tangle}}
\step[6.5]\id\step[2.5]\id\\
\id\step
\Put(0,0)[lb]{\begin{tangle}
\hh\id\Step\x\\
\hh\def5{2}{\hstr{160}\cu*}\step\d
\end{tangle}}
\step[4]\ox{\sstyle 21}\step\id\step[2.5]\id\step[2.5]\id\\
\hh\id\Step\id\step[2.5]\hru\Step\id\step\id\step{\hstr{240}\cd}\step\id\\
\ox{\sstyle 21}\step[2.5]{\hstr{100}\ox{\sstyle 22}}\step
\Put(0,0)[lb]{\begin{tangle}
\hh\id\step{\hstr{320}\hrd}\\
\hh\x\Step\id
\end{tangle}}
\step[4]\id\step\id\\
{\hstr{160}\dh}\step\nw2\step[1.5]\dh\Step\hx\step\ox{\sstyle 21}%
\step\id\step\id\\
\hh\step\nw1\Step\id\step{\hstr{160}\cu\hstep\hru\step}\cu*\step\id\\
\hh\Step\nw1\step\def5{2}{\hstr{160}\cu*\step}
{\hstr{560}\hru}\step[1.5]\id\\
\hh\step[3]{\hstr{160}\cu}\step[3]{\hstr{400}\cu}
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step[6]{\hstr{120}\cd}\Step\cd\\
\hstep\id\step[4]\sw2\step[2.5]\ox{\sstyle 22}\step\id\\
\hh\hstep\id\step[3.5]\cd\step[2.5]\cd\step[1.5]\cu\\
\hcd\Step\hddcd\step{\hstr{100}\ox{\sstyle 22}}\step\nw2\step\nw2\\
\id\step\ox{\sstyle 22}\step\Put(0,0)[lb]{
\begin{tangle}
\hh\id\step{\hstr{160}\hrd}\step[1.5]\nw1\Step\nw1\\
\hh\x\step\nw1\step\cd\Step\cd
\end{tangle}
}\step[8]\hstep\id\\
\id\step\Put(0,0)[lb]{
\begin{tangle}
\hh\id\Step\x\\
\hh\def5{2}{\hstr{160}\cu*}\step\id
\end{tangle}
}\step[4]\ox{\sstyle 21}\step\id\step\ox{\sstyle 22}\step\id\hstep\id\\
\ox{\sstyle 21}\Step\id\step\ox{\sstyle 12}\step\Put(0,0)[lb]{
\begin{tangle}
\hh\id\step{\hstr{160}\hrd}\\
\hh\x\step\id
\end{tangle}
}\step[3]\hcu*\hstep\id\\
\hh\id\Step\nw1\step\id\step\id\Step\x\step{\hstr{160}\hlu}\step[1.5]\id
\step\id\\
\hh\id\step[3]\d\hstep\nw1{\hstr{160}\cu}\step\id\step\sw1\hstep\sw1\sw1\\
\hh\id\step[3.5]\d\step\x\step\sw1\sw1\hstep\sw1\sw1\\
\hh\id\step[4]\cu*\step{\hstr{160}\hru}\step[-1]\Ru\hstep\sw1\sw1\\
{\hstr{180}\cu}\step[1.5]\Put(0,0)[lb]{
\begin{tangle}
\hh{\hstr{400}\hru}\sw1\\
\hh\hstr{200}\cu
\end{tangle}
}
\end{tangle}
\enspace=\enspace
\begin{tangle}
\hh\hstep\id\step[5]{\hstr{240}\cd}\step[2.5]{\hstr{160}\cd}\\
\hstep\id\step[4.5]\hdd\step[3]{\hstr{100}\ox{\sstyle 22}}\Step\nw2\\
\hh\hstep\id\step[3.5]{\hstr{160}\cd}\Step\cd\Step\nw1\step[3]\id\\
\hstep\id\step[3]\hdd\Step\ox{\sstyle 22}\step\nw2\Step\nw2\Step\id\\
\hh\cd\Step\cd\Step{\hstr{160}\hrd}\step\nw1\Step\nw1\Step\id\step\id\\
\id\step\ox{\sstyle 22}\step\x\step\nw2\step\nw2\Step\rd\hstep\ffbox2{\hat\sigma}\\
\hh\id\step\id\Step\x\step{\hstr{160}\cd\hstep\cd}\step\x\step\id
\step\id\step\id\\
\id\step\cu*\step\id\step\Put(0,0)[lb]{
\begin{tangle}
\hh{\hstr{160}\hrd}\step\x\step{\hstr{160}\hld}\\
\hh\id\step\x\step\x\step\id
\end{tangle}
}\step[6]\id\step\lu\step\id\step\id\\
\ox{\sstyle 21}\Step\id\step\Put(0,0)[lb]{
\begin{tangle}
\hh{\hstr{160}\hru}\step\x\step{\hstr{160}\hlu}\\
\hh\hstr{160}\cu\hstep\cu
\end{tangle}
}\step[6]\id\Step\id\step\id\step\id\\
\hh\id\Step\nw1\step\nw1\step\nw1\Step{\hstr{160}\x}\Step\id\step\id
\step\id\\
\hh\nw1\Step\nw1\step\nw1\step{\hstr{160}\cu}\Step\id\step\sw1\step\id
\step\id\\
\hh\step\nw1\Step\nw1\step{\hstr{160}\x}\Step\sw1\sw1\step\sw1\step\id\\
\hh\Step\nw1\Step\cu*\Step{\hstr{320}\hru}\step[-2]{\hstr{240}\ru}\step
\sw1\step\sw1\\
\hh\step[3]\nw1\step\dd\step[2.5]{\hstr{640}\hru}\step\sw1\\
\hh\step[4]\cu\step[3]\hstr{400}\cu
\end{tangle}
\end{split}
\end{equation}
where we used \eqref{phi-prod} and \eqref{hat-prod} in the first equation,
and \eqref{hat-sigma-Delta} and the right modularity of $(B_2,\mu_r)$ in
the second equation. To get the third identity in the graphic one has to
apply \eqref{phi-sigma} and again the fact that $(B_2,\mu_r)$ is right module.
In the fourth relation we use \eqref{phi-prod} and the relative comodule
property of $\nu_r$ according to Lemma \ref{strong-hd3}. The fifth
identity can be
verified with the help of \eqref{rel-phi}. The module-comodule compatibility,
the associativity of $\Delta_2$ and the entwining property \eqref{entw1} of
$\varphi_{2,2}$ yield the final equation in the graphic.
For the right hand side of the sixth identity of \eqref{cond3-caat} we get
\begin{equation}\mathlabel{assoc-2r}
\divide\unitlens by 4
\multiply\unitlens by 3
\begin{split}
&\hstretch 80 \vstretch 80
\begin{tangle}
\hh\id\step\id\step[3]\id\\
\hh\hhstep\ffbox2{\hat\sigma}\step[2.5]\id\\
\hh\id\hstep\cd\Step\cd\\
\id\hstep\id\step\ox{22}\step\id\\
\hh\id\hstep\cu*\Step\cu\\
\hh\cu\step[3]\id
\end{tangle}
=
\begin{tangle}
\hh\cd\Step\cd\step\id\\
\id\step\ox{22}\step\id\hstep\hcd\\
\hh\id\step\id\Step\cu\hstep\hrd\hstep\id\\
\hh\hhstep\ffbox2{\hat\sigma}\Step\x\hstep\id\hstep\id\\
\hh\def5{2}
\id\step{\hstr{200}\cu*}\step\hru\hstep\id\\
\hh{\hstr{180}\cu}\step[2.25]\cu
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step[3]\id\step[1.5]\hstr{200}\cd\\
\hh\cd\Step\cd\step{\hstr{320}\hrd}\hstep\id\\
\id\step\ox{22}\step\hx\Step\id\hstep\id\\
\id\step\id\Step\hx\step\ox{21}\hstep\id\\
\hh\hhstep\ffbox2{\hat\sigma}\hstep\sw1\step{\hstr{160}\hru}
\Step\id\hstep\id\\
\hh\id\step\cu*\Step{\hstr{240}\cu}\hstep\id\\
\hh{\hstr{120}\cu}\step[4]{\hstr{160}\cu}
\end{tangle}
=
\begin{tangle}
\hh\hstep\id\step[3]\id\step[1.5]\hstr{240}\cd\\
\hh\cd\Step\cd\step{\hstr{320}\hrd}\step\id\\
\id\step\ox{22}\step\hx\Step\id\step\id\\
\id\step\d\step\hx\step\ox{21}\step\id\\
\hh\id\step[1.5]\ffbox2{\hat\sigma}\hstep{\hstr{160}\hru}
\step[1.5]\ffbox2{\hat\sigma}\\
\ox{21}\step\id\step{\hstr{120}\Ru}\step\id\\
\hh\id\Step\cu*\step{\hstr{320}\cu}\\
\hh{\hstr{200}\cu}\step[3.5]\id
\end{tangle}
=
\begin{tangle}
\hh\step\id\step[4]\id\step[3]\hstr{200}\cd\\
\hh\step\id\step[4]\id\step[1.5]{\hstr{240}\cd}\step\id\\
\hh{\hstr{160}\cd}\step[2.5]\cd\step{\hstr{160}\hrd}\Step\id\step\id\\
\id\Step{\hstr{100}\ox{22}}\step\hx\step\ox{12}\step\id\\
\hh\id\step[1.5]\cd\Step\x\step\x\Step\id\step\id\\
\id\step[1.5]\id\step\ox{22}\step\hx\step\ox{21}\step\id\\
\hh\id\step[1.5]\cu*\Step\cu\step\d\hstep\id\step[1.5]\ffbox2{\hat\sigma}\\
\ox{21}\step\sw2\step[3]\hd\cu\step\id\\
\hh\id\Step\cu*\step[4.5]{\hstr{160}\hru}\Step\id\\
\hh{\hstr{200}\cu}\step[5]\hstr{240}\cu
\end{tangle}
\\
&=\enspace
\hstretch 80 \vstretch 80
\begin{tangle}
\hh\step[1.25]\id\step[3.75]\id\step[4.5]\hstr{240}\cd\\
\hh\step[1.25]\id\step[3.75]\id\step[2.5]{\hstr{320}\cd}\step\id\\
\hh{\hstr{200}\cd}\Step\cd\Step{\hstr{400}\hrd}\step[1.5]\id\step\id\\
\id\step[2.5]\ox{22}\step\x\step[2.5]\id\step[1.5]\id\step\id\\
\hh\id\Step\dd\Step\x\step{\hstr{160}\cd}\step\cd\step\id\step\id\\
\hh\id\Step\id\step[2.5]\id\step\id\step\id\Step\x\step\id\step\id\step\id\\
\hh\id\step[1.5]\cd\Step\id\step\id\step\id\step\sw1\step{\hstr{160}\hru}
\step\id\step\id\\
\id\step[1.5]\id\step\ox{22}\step\id\step\id\step\id\Step\ox{22}\step\id\\
\id\step[1.5]\id\step\id\Step\id\step\id\step\id\step\ox{12}\Step\id
\step\id\\
\hh\id\step[1.5]\cu*\Step\id\step\id\step\x\Step\id\step[1.5]
\ffbox2{\hat\sigma}\\
\ox{21}\step[2.5]\id\step\hx\step\Lu\Step\id\step\id\\
\hh\id\Step\id\step[2.5]\cu\step\id\step[3]{\hstr{160}\cu}\step\id\\
\hh\def5{2}\id\Step{\hstr{240}\cu*}\step[1.5]{\hstr{640}\hru}
\Step\id\\
\hh{\hstr{280}\cu}\step[3]\hstr{480}\cu
\end{tangle}
=
\begin{tangle}
\hh\step\id\step[4]{\hstr{200}\cd}\Step\cd\\
\step\id\step[4]\id\step[2.5]\ox{22}\step{\hstr{160}\hd}\\
\hh\step\id\step[4]\id\step[1.5]{\hstr{160}\cd}\step\nw1\step\id\\
\cd\step[2.5]\hcd\step\Put(0,0)[lb]{
\begin{tangle}
\hh{\hstr{240}\hrd}\hstep\d\step\ffbox2{\hat\sigma}\\
\hh\id\hstep{\hstr{160}\hld\hstep\hrd}\hstep\id\step\id
\end{tangle}
}\\
\id\Step{\hstr{100}\ox{22}}\step\Put(0,0)[lb]{
\begin{tangle}
\hh\id\step\id\hstep\id\step\x\step\id\hstep\id\step\id\\
\hh\x\hstep\cu\step\d\hstep\id\hstep\id\step\id
\end{tangle}
}\\
\hh\id\step[1.5]\cd\Step\x\step\x\Step\id\hstep\id\hstep\id
\step\id\\
\id\step[1.5]\id\step\ox{22}\step\hx\step\ox{21}\hstep\id\hstep\id
\step\id\\
\hh\id\step[1.5]\cu*\Step\cu\step\id\hstep\dd\Step\hlu\hstep\id\step\id\\
\ox{21}\step[3]\id\step[1.5]\Put(0,0)[lb]{
\begin{tangle}
\hh\id\hstep{\hstr{240}\cu}\hstep\id\step\id\\
\hh{\hstr{320}\hru}\step[-2]{\hstr{160}\Ru}\step\id
\end{tangle}
}\\
\Put(0,0)[lb]{
\begin{tangle}
\hh\def5{2}\id\Step{\hstr{240}\cu*}\\
\hh{\hstr{280}\cu}
\end{tangle}
}\step[6.5]\hstr{200}\cu
\end{tangle}
\end{split}
\end{equation}
In the first equation of \eqref{assoc-2r} we used \eqref{hat-prod}.
Then we applied \eqref{entw1} to get the second equation.
To derive the third identity we make use of the cocycle compatibility
and the weak associativity of $\m_2$ according to Definition \ref{hp}.
In the forth equation we apply the right comodule-coalgebra compatibility,
and the fifth identity follows from \eqref{phi-phi}.
The sixth equation in the graphical calculation \eqref{assoc-2r}
has been obtained from the entwining property of $\varphi_{2,2}$
and the left module-algebra compatibility.
Eventually, the seventh graphic in \eqref{assoc-2l} and in
\eqref{assoc-2r} coincide. This can be verified by applying
Lemma \ref{strong-hd3} for $\m_2$ and $\nu_r$, coassociativity of
$\Delta_2$, the right modularity of $(B_2,\mu_r)$ and
\eqref{mod-alg-rel} of Lemma \ref{strong-hd4}.\end{proof}
\abs
\begin{remark}\Label{dress}
{\normalfont To finish the proof of Theorem \ref{hp-cp} we have to show that
$\Delta_B$ is an algebra morphism. For this purpose we will use a sort of
\emph{dressing transformation} technique. This means we define two sequences
of morphisms $(\beta_\ell^i)_{i=0}^4$ and $(\beta_r^i)_{i=0}^4$
and a sequence of ``dressing transformations" $\left(T^{(i)}\right)_{i=1}^4$
where $\beta_r^0=(\m_B\otimes\m_B)\circ(\id\otimes\Psi_{B,B}\otimes\id)
\circ(\m_B\otimes\m_B)$, $\beta_\ell^0=\Delta_B\circ\m_B$,
$\beta_\ell^4=\beta_r^4$, and $T^{(i)}(\beta_\ell^i)=\beta_\ell^{i-1}$,
$T^{(i)}(\beta_r^i)=\beta_r^{i-1}$ for all $i\in\{1,2,3,4\}$.
This implies $\beta_\ell^0=\beta_r^0$ and therefore the statement.}
\end{remark}
\begin{remark}{\normalfont Once it is proven that $B$ is an algebra,
Proposition \ref{alg-morph-equiv} provides an alternative method to show that
$\Delta_B$ is an algebra morphism. The present proof using dressing
transformations is invariant under $\pi$-symmetry, however.}
\end{remark}
\abs
The ``dressing transformations'' will be defined as
\begin{gather}\mathlabel{dressing}
\divide\unitlens by 3
\multiply\unitlens by 2
\hstretch 80
\vstretch 120
T^{(1)}(f):=
\begin{tangle}
\hh\id\step[1.5]\id\step\id\step\id\\
\hh\vstr{144}\id\hstep\ffbox4{\tau^1_t}\\
\hh\id\step\id\step\id\step\id\step\id\\
\hh\vstr{144}\id\hstep\ffbox3{f}\hstep\id\\
\hh\id\step\id\step\id\step\id\step\id\\
\hh\vstr{144}\hhstep\ffbox4{\tau^1_b}\hstep\id\\
\hh\hstep\id\step\id\step\id\step[1.5]\id
\end{tangle}
\,\,,\,\,
T^{(2)}(f):=
\begin{tangle}
\hh{\hstr{120}\cd}\step\id\step\id\\
\hh\vstr{144}\id\hstep\ffbox4{\tau^2_t}\\
\hh\id\step\id\step\id\step\id\step\id\\
\hh\vstr{144}\id\hstep\ffbox3{f}\hstep\id\\
\hh\id\step\id\step\id\step\id\step\id\\
\hh\vstr{144}\hhstep\ffbox4{\tau^2_b}\hstep\id\\
\hh\hstep\id\step\id\step\hstr{120}\cu
\end{tangle}
\,\, ,\,\,
\hstretch 70
T^{(3)}(g):=
\begin{tangle}
\hh\cd\step[1.5]{\hstr{140}\cd}\step[2]\id\\
\hh\id\step{\hstr{105}\x}\step[2]\id\step[2]\id\\
\hh\vstr{144}\id\step\id\hstep\ffbox6{\tau^{3}_t}\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\id\step\id\\
\hh\vstr{144}\id\step\id\hstep\ffbox4{g}\hstep\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\id\step\id\\
\hh\vstr{144}\hhstep\ffbox6{\tau^{3}_b}\hstep\id\step\id\\
\hh\hstep\id\step[2]\id\step[2]{\hstr{105}\x}\step\id\\
\hh\hstep\id\step[2]{\hstr{140}\cu}\step[1.5]\cu
\end{tangle}
\,\, ,\,\,
T^{(4)}(h):=
\begin{tangle}
\hh\id\step[1.5]{\hstr{210}\hld}\step\id\step\id\\
\hh{\hstr{105}\x}\step[1.5]\id\step\id\step\id\\
\hh\id\step\cd\step\id\step\id\step\hstr{140}\hrd\\
\hh\id\hstep\ffbox5{h}\hstep\id\\
\hh{\hstr{140}\hlu}\step\id\step\id\step\cu\step\id\\
\hh\step\id\step\id\step\id\step[1.5]\hstr{105}\x\\
\hh\step\id\step\id\step{\hstr{210}\hru}\step[1.5]\id\\
\end{tangle}
\end{gather}
for any $f\in\Hom(B_{212},B_{121})$, $g\in\Hom(B_{2122},B_{1121})$
and $h\in\Hom(B_{22122},B_{11211})$.
\abs
The remainder of this section is devoted to the proof of the
following proposition.
\begin{proposition}\Label{b-bialg}
Given a strong Hopf datum $\mathfrak{h}$,
then there exist two sequence of morphisms $(\beta^i_\ell)_{i=0}^4$,
$(\beta^i_r)_{i=0}^4$ with $\beta^0_\ell:=\Delta_B\circ\m_B$ and
$\beta^0_r:=(\m_B\otimes\m_B)\circ(\id_{B}\otimes\Psi_{B,B}\otimes\id_{B})\circ
(\Delta_{B}\otimes\Delta_{B})$ such that
\begin{equation}\mathlabel{beta-lr}
\begin{diagram}[height=15pt]
\beta^0_\ell&\lMapsto{T^{(1)}}&
\beta^1_\ell&\lMapsto{T^{(2)}}&
\beta^2_\ell&\lMapsto{T^{(3)}}&
\beta^3_\ell&\lMapsto{T^{(4)}}&
\beta^4_\ell\\
& & & & & & & &\dEqualsto\\
\beta^0_r&\lMapsto{T^{(1)}}&
\beta^1_r&\lMapsto{T^{(2)}}&
\beta^2_r&\lMapsto{T^{(3)}}&
\beta^3_r&\lMapsto{T^{(4)}}&
\beta^4_r
\end{diagram}
\end{equation}
Therefore $\beta^0_\ell=\beta^0_r$ and $B=B_1\otimes B_2$ is a
bialgebra which proves Theorem \ref{hp-cp}.
\end{proposition}
\begin{proof}
The proof will be splitted into several parts. At first we verify the following
diagram.
\begin{equation}
\begin{diagram}[height=18pt]
\beta^0_r&&\beta^1_r&&\beta^2_r&&\beta^3_r\\
\dCongruent&&\dCongruent&&\dCongruent&&\dCongruent\\
\ixi 00000000&&\ixi 01200000&&\ixi 01202001&&\\
\dEqualsto&&\dEqualsto&&\dEqualsto&&\\
\hstretch 60
\begin{tangle}
\hh\id\step\cd\step\id\step\id\\
\hh\vstr{120}\id\step\id\hstep\ffbox3{\yi 00200}\\
\hh\id\step\id\step[.75]\id\hstep\id\step[.75]\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\iy 01000}\hstep\id\step\id\\
\hh\id\step\id\step\cu\step\id
\end{tangle}
&\lMapsto{T^{(1)}}&
\hstretch 60
\begin{tangle}
\hh\cd\step\id\step\id\\
\hh\vstr{120}\id\hstep\ffbox3{\yi 01200}\\
\hh\id\step\id\step\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\iy 01200}\hstep\id\\
\hh\id\step\id\step\cu
\end{tangle}
&\lMapsto{T^{(2)}}&
\hstretch 60
\begin{tangle}
\hh\cd\step\id\step\id\\
\hh\vstr{120}\id\hstep\ffbox3{\yi 01201}\\
\hh\id\step\id\step\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\iy 01202}\hstep\id\\
\hh\id\step\id\step\cu
\end{tangle}
&\lMapsto{T^{(3)}}&
\hstretch 60 \vstretch 60
\begin{tangle}
\hh{\hstr{150}\cd}\step\cd\Step\cd*\hstep\id\\
{\hstr{90}\rd}\step\hx\step\ox{11}\step\id\hstep\id\\
\hh\id\step[1.5]\x\step\x\Step\cu\hstep\id\\
\hh\id\step\dd\step\x\step\id\step{\hstr{90}\ld}\step\id\\
\hh\id\step\id\step\cd\hstep\id\step\id\hstep\cd\step\x\\
\hh\id\step\x\step\id\hstep\id\step\id\hstep\id\step\x\step\id\\
\hh\x\step\cu\hstep\id\step\id\hstep\cu\step\id\step\id\\
\hh\id\step{\hstr{90}\ru}\step\id\step\x\step\dd\step\id\\
\hh\id\hstep\cd\Step\x\step\x\step[1.5]\id\\
\id\hstep\id\step\ox{22}\step\hx\step{\hstr{90}\lu}\\
\hh\id\hstep\cu*\Step\cu\step\hstr{150}\cu
\end{tangle}
\end{diagram}
\end{equation}
where the identities ($\equiv$) in the first row are definitions and
the equalities ($=$) in the second row are special cases
of \eqref{bra-ket-decomp}. The morphisms in the third row will be obtained by
applying $T^{(1)}$, $T^{(2)}$ and $T^{(3)}$ using the identities
\eqref{tau-R-aux}, \eqref{tau-R2-aux} and \eqref{yi-red-adv3} respectively.
A similar diagram can be set up for the $(\beta^i_\ell)$.
For that we use the following auxilarity definitions
\begin{equation}\mathlabel{aux-gamma0}
\gamma^0_t:=(\id_{B_1}\otimes\Delta_1)\circ\m_{0,20}
\quad,\quad
\gamma^0_b:=\Delta_{01,0}\circ(\m_1\otimes\id_{B_2})
\end{equation}
\begin{equation}
\mathlabel{aux-gamma}
\begin{gathered}
\unitlens = 8pt
\gamma^1_t:=
\begin{tangle}
\hh\step\id\step\cd\step[2.5]\id\\
\hh\step\x\step\id\step[2.5]\id\\
\hh\sw1\step{\hstr{200}\hru}\step[2.5]\id\\
\hh\id\step{\hstr{200}\cd\step}\cd\\
\id\step\id\Step\ox{\sstyle 22}\step\id\\
\id\hstep\ffbox3{\yi 01220}\step[1.5]\hcu\\
\id\step\id\step\id\step\hstr{125}\ox{\sstyle 12}
\end{tangle}
\quad,\quad
\gamma^2_t:=
\begin{tangle}
\hh\step\id\step\cd\step[3]\id\\
\hh\step\x\step\id\step[3]\id\\
\hh\sw1\step\ru\step[3]\id\\
\hh\id\step\hstr{200}\cd\step\cd\\
\id\step\id\Step\ox{\sstyle 22}\Step\id\\
\vstr{115}\id\hstep\ffbox3{\yi 01221}\step\ffbox3{\ixi 21222021}\\
\hh\id\step\id\hstep\id\hstep\hstr{200}\hld\step\id\step\id\\
\hh\id\step\id\hstep\id\hstep\id\step\hstr{200}\x\hstep\dd\\
\hh\id\step\id\hstep\id\hstep\cu\Step\cu
\end{tangle}
\quad,\quad
\unitlens = 9pt
\def5{2}
\gamma^3_t:=
\begin{tangle}
\hh\id\step\cd\hstep\hstr{200}\cd*\hstep\id\\
\hh\x\step\id\hstep\id\step{\hstr{200}\hld}\step\rd\\
\hh\id\step{\hstr{200}\hru}\hstep\id\step\id\step\x\step\id\\
\hh\id\step{\hstr{150}\x}\step\cu\step\id\step\id\\
\hh\id\step\id{\hstr{150}\step\x\step}\id\step\id\\
\id\step\id\step[1.5]\id\step\hstr{50}\ffbox{5}{\ixi 21212001}\step\id\\
\hh\id\step\id\step[1.5]\id\step[1.5]\id\step[1.5]\x\\
\hh\id\step\id\step[1.5]\id\step[1.5]{\hstr{300}\hru}\step\id
\end{tangle}
\\
\unitlens = 8pt
\gamma^1_b:=
\begin{tangle}
\hstep{\hstr{125}\ox{\sstyle 21}}\step\id\step\id\step\id\\
\hcd\step[1.5]\ffbox3{\iy 11200}\hstep\id\\
\id\step\ox{\sstyle 11}\Step\id\step\id\\
\hh\cu\Step{\hstr{200}\cu}\step\id\\
\hh\hstep\id\step[2.5]\hstr{200}\hld\hstep\dd\\
\hh\hstep\id\step[2.5]\id\step\x\\
\hh\hstep\id\step[2.5]\cu\step\id
\end{tangle}
\quad,\quad
\gamma^2_b:=
\begin{tangle}
\hh\step\cd\Step\cd\hstep\id\hstep\id\step\id\\
\hh{\hstr{200}\dd\hstep\x}\step\id\hstep\id\hstep\id\step\id\\
\hh{\hstr{200}\id\step\id\step\hru}\hstep\id\hstep\id\step\id\\
\vstr{115}\hhstep\ffbox3{\ixi 11212011}\step\ffbox3{\iy 11202}\hstep\id\\
\id\Step\ox{\sstyle 11}\Step\id\step\id\\
\hh\hstr{200}\cu\step\cu\hstep\id\\
\hh\step\id\step[3]\ld\step\sw1\\
\hh\step\id\step[3]\id\step\x\\
\hh\step\id\step[3]\cu\step\id
\end{tangle}
\quad,\quad
\unitlens = 9pt
\def5{2}
\gamma^3_b:=
\begin{tangle}
\hh\id\step{\hstr{300}\hld\hstep\id\hstep}\id\step\id\\
\hh\x{\hstr{150}\step\id\step\id\step\id}\step\id\\
\id\hstep{\hstr{50}\ffbox{5}{\ixi 21212001}}\step\id\step[1.5]\id\step\id\\
\hh\id\step\id{\hstr{150}\step\x\step}\id\step\id\\
\hh\id\step\id\step\cd\step{\hstr{150}\x}\step\id\\
\hh\id\step\x\step\id\step\id\hstep{\hstr{200}\hld}\step\id\\
\hh\lu\step{\hstr{200}\hru}\step\id\hstep\id\step\x\\
\hh\step\id\step{\hstr{200}\cu*}\hstep\cu\step\id
\end{tangle}
\end{gathered}
\end{equation}
Then
\begin{equation}
\begin{diagram}[height=18pt]
\beta^0_\ell && \beta^1_\ell && \beta^2_\ell && \beta^3_\ell\\
\dCongruent && \dCongruent && \dCongruent && \dCongruent\\
\Delta_0\circ\m_0 && \Delta_{01,0}\circ\m_{0,20} && &&\\
\dEqualsto&\ldMapsto^{\sstyle T^{(1)}}&\dEqualsto&& &&\\
\hstretch 60
\begin{tangle}
\hh\id\step\cd\step\id\step\id\\
\hh\vstr{120}\id\step\id\hstep\ffbox3{\gamma^0_t}\\
\hh\id\step\id\step[.75]\id\hstep\id\step[.75]\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\gamma^0_b}\hstep\id\step\id\\
\hh\id\step\id\step\cu\step\id
\end{tangle}
&&
\hstretch 60
\begin{tangle}
\hh\cd\step\id\step\id\\
\hh\vstr{120}\id\hstep\ffbox3{\gamma^1_t}\\
\hh\id\step\id\step\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\gamma^1_b}\hstep\id\\
\hh\id\step\id\step\cu
\end{tangle}
&\lMapsto{T^{(2)}}&
\hstretch 60
\begin{tangle}
\hh\cd\step\id\step\id\\
\hh\vstr{120}\id\hstep\ffbox3{\gamma^2_t}\\
\hh\id\step\id\step\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\gamma^2_b}\hstep\id\\
\hh\id\step\id\step\cu
\end{tangle}
&\lMapsto{T^{(3)}}&
\hstretch 60
\begin{tangle}
\hh\cd\step\id\step\id\\
\hh\vstr{120}\id\hstep\ffbox3{\gamma^3_t}\\
\hh\id\step\id\step[0.33]\id\step[0.34]\id\step[0.33]\id\step\id\\
\hh\vstr{120}\hhstep\ffbox3{\gamma^3_b}\hstep\id\\
\hh\id\step\id\step\cu
\end{tangle}
\end{diagram}
\end{equation}
Again the identities ($\equiv$) in the first row are definitions.
The remaining identities will be proven in the subsequent lemmas.
\begin{lemma}
\begin{equation}\mathlabel{tau-L-aux}
\gamma^0_t=(\m_{0,20}\otimes\id_{B_2})\circ\tau^1_t\,,\qquad
\gamma^0_b=\tau^1_b\circ(\id_{B_1}\otimes\Delta_{01,0})
\end{equation}
and hence $\beta^0_\ell=T^{(1)}\left(\beta^1_\ell\right)$.
\end{lemma}
\begin{proof}
The identities are obtained using \eqref{hat-prod} and \eqref{phi-prod}.
\end{proof}
\begin{lemma}
\begin{gather}\mathlabel{beta-1l}
\beta^1_\ell=(\id\otimes\m_1)\circ(\gamma^1_b\otimes\id_{B_1})\circ
(\id_{B_2}\otimes\gamma^1_t)\circ(\Delta_2\otimes\id)
\\
\mathlabel{gamma-12}
\gamma^1_t=(\id\otimes\m_1)\circ(\gamma^2_t\otimes\id_{B_1})\circ\tau^2_t\,,
\quad
\gamma^1_b=\tau^2_b\circ(\id_{B_2}\otimes\gamma^2_b)\circ(\Delta_2\otimes\id)
\end{gather}
and therefore $\beta^1_\ell=T^{(2)}\left(\beta^2_\ell\right)$.
\end{lemma}
\begin{proof}
The proof of the identity \eqref{beta-1l} will be given
in the following graphical calculation.
\begin{equation*}
\beta^1_\ell=
\vstretch 70 \hstretch 70
\begin{tangle}
\ox{21}\step\id\\
\hh\id\step[1.5]\ffbox2{\hat\sigma}\\
\hh\hstr{140}\cu\cd\\
\hh\hstep\ffbox2{\hat\rho}\step[1.5]\id\\
\step\id\step\ox{12}\\
\end{tangle}
=\begin{tangle}
\hstep\ox{21}\step[3]\id\\
\hh\hstep\id\step[1.5]\cd\Step\cd\\
\hstep\id\step[1.5]\id\step\ox{22}\step\id\\
\hh\hstep\id\step\ffbox2{\hat\sigma}\step[1.5]\cu\\
\hh\hstep\id\step[1.5]\nw1\nw1\step[1.5]\id\\
\hh\cd\step[1.5]\ffbox2{\hat\rho}\step\id\\
\id\step\ox{11}\step\id\step[1.5]\id\\
\hh\cu\Step\cu\step[1.5]\id\\
\hstep\id\step[3]\ox{12}
\end{tangle}
=\enspace
\unitlens=7pt
\vstretch 100 \hstretch 100
\begin{tangle}
\hh{\hstr{300}\cd}\step\cd\step[2.5]\id\\
\hh\id\step[3]\x\step\id\step[2.5]\id\\
\hh\id\Step\sw1\step{\hstr{200}\hru}\step[2.5]\id\\
\hh\id\Step\id\step{\hstr{200}\cd\step}\cd\\
\id\Step\id\step\id\Step\ox{\sstyle 22}\step\id\\
\ox{\sstyle 21}\hstep\ffbox3{\yi 01220}\step[1.5]\hcu\\
\hh\hhstep\dd\Step\id\step\id\step\id\step\id\Step\dd\\
\step[-1]\hcd\step[1.5]\ffbox3{\iy 11200}\hstep\ox{\sstyle 12}\\
\step[-1]\id\step\ox{\sstyle 11}\Step\id\step\id\Step\id\\
\hh\step[-1]\cu\Step\hstr{200}\cu\hstep\id\step\id\\
\hh\hhstep\id\step[2.5]\hstr{200}\hld\hstep\dd\step\id\\
\hh\hhstep\id\step[2.5]\id\step\x\step[3]\id\\
\hh\hhstep\id\step[2.5]\cu\step\hstr{300}\cu
\end{tangle}
\end{equation*}
where the first equation holds by definition and the second identity is an
application of \eqref{hat-prod}. In the third equation we use the cycle-cocycle
compatibility, \eqref{bra-ket-decomp}, the (co-)associativity of $\m_1$ and
$\Delta_2$, and \eqref{phi-prod}.
To prove the first identity of \eqref{gamma-12} we transform
in the next calculation the morphism $\gamma^1_t$
with the help of \eqref{tau-check-aux}, the entwining
properties \eqref{entw1} of $\varphi_{2,2}$, and \eqref{phi-prod} of
$\varphi_{1,2}$.
\begin{equation*}
\divide\unitlens by 3
\multiply\unitlens by 2
\gamma^1_t=
\vstretch 80
\begin{tangle}
\hh\step\id\step\cd\step[3]\id\\
\hh\step\x\step\id\step[3]\id\\
\hh\ne1\step{\hstr{200}\hru}\step[3]\id\\
\hh\id\step\hstr{200}\cd\step\cd\\
\id\step\id\step[2]\ox{\sstyle 22}\step[2]\id\\
\id\hstep\ffbox3{\yi 01221}\hstep\ffbox4{\varphi_{1,2}\circ\hat\sigma}\\
\hh\id\step\id\hstep\id\hstep{\hstr{200}\hld\step\id\step\id}\\
\hh\id\step\id\hstep\id\hstep\id\step\hstr{200}\x\step\id\\
\hh\id\step\id\hstep\id\hstep\cu\hstr{200}\step\cu
\end{tangle}
\end{equation*}
From the right hand side of this equation we get the right hand side
of \eqref{gamma-12} by applying \eqref{phi-hat-new} and again the
entwining property \eqref{entw1} of $\varphi_{2,2}$.\end{proof}
\begin{lemma}
\begin{equation}
\unitlens = 8pt
\mathlabel{gamma-34}
\gamma^2_t=
\begin{tangle}
\hh\step\id\step[1.5]\id\step[1.5]\id\\
\hstr{50}\ffbox{11}{\tau^{3}_t}\\
\hh\step\id\hstep\id\hstep\id\hstep\id\step[1.5]\id\step\id\\
{\hstr{50}\ffbox{7}{\gamma^3_t}}\hstep\id\step\id\\
\hh\hstep\id\hstep\id\hstep\id\hstep\id\step\x\step\id\\
\hh\hstep\id\hstep\id\hstep\id\hstep\cu\step\cu
\end{tangle}
\quad,\quad
\gamma^2_b=
\begin{tangle}
\hh\cd\step\cd\hstep\id\hstep\id\hstep\id\\
\hh\id\step\x\step\id\hstep\id\hstep\id\hstep\id\\
\id\step\id\hstep\hstr{50}\ffbox{7}{\gamma^3_b}\\
\hh\id\step\id\step[1.5]\id\hstep\id\hstep\id\hstep\id\\
\hhstep\hstr{50}\ffbox{11}{\tau^{3}_t}\\
\hh\step\id\step[1.5]\id\step[1.5]\id
\end{tangle}
\end{equation}
and therefore $\beta^2_\ell = T^{(3)}(\beta_\ell^3)$.
\end{lemma}
\begin{proof}
The following identities hold.
\begin{equation}\mathlabel{gamma-derive}
\unitlens =7pt
\begin{tangle}
\hh\hstr{200}\cd\step\id\step\id\\
\id\Step\ox{\sstyle 22}\Step\id\\
\hhstep\ffbox3{\yi 01221}\step\ffbox3{\iy 21222}\\
\hh\id\hstep\id\hstep\hstr{200}\hld\hstep\dd\step\id\\
\hh\id\hstep\id\hstep\id\step\x\Step\sw1\\
\hh\id\hstep\id\hstep\cu\step\hstr{200}\cu\\
\object{\sstyle B}\step[1.5]\object{\sstyle B_2}\step[2.5]\object{\sstyle B_1}
\end{tangle}
\;=
\unitlens =12pt
\vstretch 80
\begin{tangle}
\id\step\td{\sstyle 02\vert 2}\step\id\\
\hh\x\Step\id\step\id\\
\id\hstep\ffbox3{\ixi 21212001}\hstep\id\\
\hh\id\step\id\Step\x\\
\hh\id\step\hstr{200}\cu\hstep\id\\
\object{\sstyle B}\step[2]\object{\sstyle B_2}\step[2]\object{\sstyle B_1}
\end{tangle}
\quad ,
\quad
\gamma^2_t=
\unitlens =12pt
\vstretch 80
\begin{tangle}
\vstr{60}\hh\step\id\Step\id\Step\id\\
\vstr{60}\hhstep\ffbox7{(\tau^3_t)^{\mathrm{red}_1}}\\
\hh\id\step\cd\step\id\Step\id\step\id\\
\Put(0,0)[lb]{\begin{tangle}\hh\x\step\id\\
\hh\id\step\hstr{200}\hru\end{tangle}}
\Step\td{\sstyle 02\vert 2}\step\id\step\id\\
\hh\id\step\x\Step\id\step\id\step\id\\
\id\step\id\hstep\ffbox3{\ixi 21212001}\hstep\id\step\id\\
\hh\id\step\id\step\id\Step\x\step\id\\
\hh\id\step\id\step{\hstr{200}\cu}\step\cu\\
\object{\sstyle B_1}\step\object{\sstyle B}\step[2]\object{\sstyle B_2}%
\step[2.5]\object{\sstyle B_1}
\end{tangle}
\end{equation}
The first identity of \eqref{gamma-derive} is a consequence of
\eqref{yi-red-adv5} and the module-comodule compatibility. Using
this result we obtain the second relation in
\eqref{gamma-derive} with the help of \eqref{bra-ket-decomp}
and the entwining property \eqref{entw1} of $\varphi_{2,2}$.
We denote
\begin{equation}
\tau^4_t:=
\unitlens =7pt
\vstretch 100
\begin{tangle}
\hh\cd\step\cd\step\cd\\
\hh\id\step\x\step\id\step\id\step\rd\\
\hh\id\step\id\step{\hstr{200}\hru}\step\id\step\id\hstep\hld\\
\id\step\id\step\ox{\sstyle 22}\step\id\hstep\id\hstep\id\\
\hh\id\step\id\step\id\step[2]{\vstr{50}\hx}\hstep\id\hstep\id\\
\hh\id\step\id\step\id\step[2]\id\step{\hstr{50}\vstr{50}\hx}\hstep\id\\
\hh\id\step\id\step\id\step[2]\id\step\id\hstep\hlu
\end{tangle}
\end{equation}
Then we obtain
\begin{equation}\mathlabel{gamma-derive2}
\begin{tangle}
\hh\id\step\cd\step\id\\
\Put(0,0)[lb]{\begin{tangle}
\hh\x\step\id\\ \hh\id\step\hstr{200}\hru
\end{tangle}}
\Step\td{\sstyle 02\vert 2}\\
\hh\id\step\x\Step\id\\
\id\step\id\hstep\ffbox3{\ixi 21212001}\\
\hh\id\step\id\step\id\Step\id\\
\object{\sstyle B_1}\step\object{\sstyle B}\step\object{\sstyle B_2}%
\step[2]\object{\sstyle B_1}
\end{tangle}
=
\hstretch 80
\begin{tangle}
\hh\step\id\step[1.5]\id\step[1.5]\id\\
\hstr{40}\ffbox{11}{\tau^{4}_t}\\
\hh\step\id\hstep\id\hstep\id\hstep\id\step[1.5]\id\step\id\\
{\hstr{40}\ffbox{7}{\gamma^3_t}}\hstep\id\step\id\\
\hh\hstep\id\hstep\id\hstep\id\hstep\id\step\x\step\id\\
\hh\hstep\id\hstep\id\hstep\id\hstep\cu\step\cu
\end{tangle}
\quad,\quad
\tau^{3}_t=
\begin{tangle}
\hh\step[1.5]\id\step[1.5]\id\step[1.5]\id\\
\ffbox6{(\tau^3_t)^{\mathrm{red}_1}}\\
\hh\step\id\step\id\step\id\step[1.5]\id\step\id\\
\ffbox4{\tau^4_t}\hstep\id\step\id\\
\hh\hstep\id\hstep\id\hstep\id\hstep\id\hstep\id\step\x\step\id\\
\hh\hstep\id\hstep\id\hstep\id\hstep\id\hstep\cu\step\cu
\end{tangle}
\hstretch 100
\end{equation}
The first equation in \eqref{gamma-derive2} results from the
algebra-coalgebra compatibility,
the right comodule-algebra compatibility, the left module-algebra
compatibility, \eqref{mod-alg-rel}, \eqref{rel-cocycle-assoc1}
and the right module algebra property of $B_2$.
The second identity can be obtained by subsequent application of
the right comodule-coalgebra compatibility, the left module-algebra
compatibility, \eqref{mod-alg-rel}, the right module-coalgebra property
of $B_2$, and the entwining property of $\varphi_{2,2}$.
The first relation of \eqref{gamma-34} can now be derived by recursive
substitution of the identities \eqref{gamma-derive2} into the second
identity of \eqref{gamma-derive}.\end{proof}
\abs
Eventually we define
\begin{equation}\mathlabel{beta-4}
\hstretch 100 \vstretch 100
\divide\unitlens by 2
\def5{4}
\beta^4_\ell :=
\begin{tangle}
\hh\id\Step\id\step{\hstr{550}\cd}\step[3.25]{\hstr{125}\cd*}\Step\id\\
\hh{\hstr{200}\hrd}\step\x\step[2.5]{\hstr{600}\hld}\step[2.25]\sw1
\step[1.25]\nw1\step\id\\
\hh\id\step\x\step{\hstr{250}\x}\step[3]\id\step[1.5]{\hstr{150}\cd}\step
{\hstr{300}\hld}\step\id\\
\hh{\hstr{200}\hru}\step\x\step{\hstr{300}\cd}\step\cd\step\id\hstep
{\hstr{200}\hld\hstep\hrd}\hstep\id\step\id\\
\hh\id\step\sw1\step\id\step\id\step[3]\x\step\id\step\id\hstep\id\step\x
\step\id\hstep\id\step\id\\
\hh\id\step\id\Step\id\step\id\Step\sw1\step{\hstr{200}\hru}\step\id\hstep
\cu\step\cu\hstep\x\\
\hh\id\step\id\Step\id\step\id\Step\id\Step{\hstr{200}\x}\step\id\Step\x
\step\id\\
\hh\id\step\id\Step\id\step\id\Step{\hstr{200}\x}\Step\nw1{\hstr{200}\cu}
\step\id\step\id\\
\hh\id\step\id\Step\id\step{\hstr{200}\x}\Step\nw1\Step\x\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\cd\hstep\cd}\step\id\step
\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\hrd}\step\x\step
{\hstr{200}\hld}\step\id\step\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step\id\step\x\step\x\step\id\step\id
\step\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\hru}\step\x\step
{\hstr{200}\hlu}\step\id\step\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\cu\hstep\cu}\step\id\step
\id\Step\id\step\id\\
\hh\id\step\id\Step\x\Step\nw1\Step{\hstr{200}\x}\step\id\Step\id\step\id\\
\hh\id\step\id\step{\hstr{200}\cd}\nw1\Step{\hstr{200}\x}\Step\id\step\id
\Step\id\step\id\\
\hh\id\step\x\Step\id\step{\hstr{200}\x}\Step\id\Step\id\step\id\Step\id
\step\id\\
\hh\x\hstep\cd\step\cd\hstep\id\step{\hstr{200}\hld}\step\sw1\Step\id\step
\id\Step\id\step\id\\
\hh\id\step\id\hstep\id\step\x\step\id\hstep\id\step\id\step\x\step[3]\id
\step\id\step\sw1\step\id\\
\hh\id\step\id\hstep{\hstr{200}\hlu\hstep\hru}\hstep\id\step\cu\step
{\hstr{300}\cu}\step\x\step\hstr{200}\hld\\
\hh\id\step{\hstr{300}\hru}\step{\hstr{150}\cu\step}\id\step[3]
{\hstr{250}\x}\step\x\step\id\\
\hh\id\step\nw1\step[1.25]\sw1\step[2.25]{\hstr{600}\hru}\step[2.5]\x
\step{\hstr{200}\hlu}\\
\hh\id\Step{\hstr{125}\cu*}\step[3.25]{\hstr{550}\cu}\step\id\Step\id
\end{tangle}
\quad\text{and}\quad
\beta^4_r :=
\def5{5}
\begin{tangle}
\hh\id\step[3.25]\id\step[5.25]{\hstr{300}\cd}\step[3.5]\id\step[3]\id\\
\hh\id\step[3.25]\id\step[4.75]\dd\Step{\hstr{200}\hld}\step[3]\cd*
\step[2.5]\id\\
\hh\id\step[3.25]\id\step[4.25]\dd\step[1.5]\sw1\step{\hstr{300}\x}\step
\d\Step\id\\
\hh\id\step[3.25]\id\step[3.75]\dd\step\sw1\step\sw1\step[3]\id
\step[1.5]\d\step[1.5]\id\\
\hh\id\step[2.5]{\hstr{150}\cd}\step[2.5]\dd\step\dd\step\sw1\hstep
{\hstr{350}\ld}\step[-1.5]{\hstr{300}\hld}\step{\hstr{200}\hld}
\step[1.5]\id\\
\hh{\hstr{300}\hrd}\step\id\step[1.5]\id\Step\dd\step[1.5]\id
\step\dd\step[1.5]\id\Step\id\step[1.5]\x\step\id\step[1.5]\id\\
\hh\id\hstep{\hstr{200}\hld\hstep\hrd}\hstep\id\step[1.5]
\dd\Step\id\step\id\Step\id\Step\id\step\cd\hstep\id\step
{\hstr{150}\x}\\
\hh\id\hstep\id\step\x\step\id\hstep\id\step\dd\step[2.5]\id
\step\id\Step\id\Step\x\step\id\hstep\x\step[1.5]\id\\
\hh\id\hstep\cu\step\cu\hstep\x\step[2.5]\dd\step\id\step
{\hstr{200}\cd\hstep\hrd}\cu\hstep\id\step\d\step\id\\
\hh\id\step\id\Step\x\step{\hstr{250}\x}\step\cd\hstep
{\hstr{200}\hrd}\step\x\step\id\hstep\x\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\dd\step\x\step[2.5]\x\step\id\hstep\id
\step\x\step{\hstr{200}\hlu}\hstep\id\step\id\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\id\step[1.5]\id\step{\hstr{250}\x}
\step\x\hstep\cu\step{\hstr{200}\cu}\dd\step\id\step[1.5]\id
\step\id\\
\hh\id\step\id\step[1.5]\id\step[1.5]\x\step[2.5]\x\step\x
\step[2.5]\x\step[1.5]\id\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\id\step\dd{\hstr{200}\cd}\step\cd
\hstep\x\step{\hstr{250}\x}\step\id\step[1.5]\id\step[1.5]\id
\step\id\\
\hh\id\step\id\step[1.5]\id\step\id\hstep{\hstr{200}\hrd}\step
\x\step\id\hstep\id\step\x\step[2.5]\x\step\dd\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\x\hstep\id\step\x\step{\hstr{200}\hlu}
\hstep\cu\step{\hstr{250}\x}\step\x\Step\id\step\id\\
\hh\id\step\d\step\id\hstep\cd{\hstr{200}\hlu\hstep\cu}
\step\id\step\dd\step[2.5]\x\hstep\cd\step\cd\hstep\id\\
\hh\id\step[1.5]\x\hstep\id\step\x\Step\id\Step\id\step\id
\step[2.5]\dd\step\id\hstep\id\step\x\step\id\hstep\id\\
\hh{\hstr{150}\x}\step\id\hstep\cu\step\id\Step\id\Step\id
\step\id\Step\dd\step[1.5]\id\hstep{\hstr{200}\hlu\hstep\hru}
\hstep\id\\
\hh\id\step[1.5]\id\step\x\step[1.5]\id\Step\id\step[1.5]\dd\step\id
\step[1.5]\dd\Step\id\step[1.5]\id\step{\hstr{300}\hlu}\\
\hh\id\step[1.5]{\hstr{200}\hru}\step{\hstr{300}\hru}\step[-1.5]
{\hstr{350}\ru}\hstep\sw1\step\dd\step\dd\step[2.5]{\hstr{150}\cu}
\step[2.5]\id\\
\hh\id\step[1.5]\d\step[1.5]\id\step[3]\sw1\step\sw1\step\dd
\step[3.75]\id\step[3.25]\id\\
\hh\id\Step\d\step{\hstr{300}\x}\step\sw1\step[1.5]\dd\step[4.25]\id
\step[3.25]\id\\
\hh\id\step[2.5]\cu*\step[3]{\hstr{200}\hru}\Step\dd\step[4.75]\id
\step[3.25]\id\\
\hh\id\step[3]\id\step[3.5]{\hstr{300}\cu}\step[5.25]\id\step[3.25]\id\\
\end{tangle}
\end{equation}
In what follows we will tacitly use the various forms of (weak)
(co\n-)associativity of $\m_1$, $\m_2$, $\Delta_1$, $\Delta_2$,
$\mu_l$, $\mu_r$, $\nu_l$ and $\nu_r$ which are given by Lemma
\ref{strong-hd3} in particular. The subsequent graphical calculation yields
$\beta^3_\ell=T^{(4)}(\beta^4_\ell)$.
\begin{equation*}
\beta^3_\ell =\;
\divide\unitlens by 2
\hstretch 100 \vstretch 100
\def5{5}
\begin{tangle}
\hh\step[4.5]\id\step{\hstr{400}\hld}\step[2.5]\id\step[1.5]\id\\
\hh\step[4.5]\x\Step\id\step[2.5]\id\step[1.5]\id\\
\hh\step[3.5]\sw1\hstep\cd\step\cd\step[1.5]\cd*\step\id\\
\hh\step[2.5]\sw1\hstep\sw1\step\x\hstep\dd\step\dd\hstep\hld\step
{\hstr{200}\hrd}\\
\hh\step[1.5]\sw1\hstep\sw1\step[1.5]\dd\step\hru\hstep\sw1\hstep\dd\hstep
\x\step\nw1\\
\hh\hstep\sw1\step\dd\Step\dd\step[1.5]\x\step[1.5]\cu\step\nw1\step\nw1\\
\hh\dd\step[1.5]\dd\step[2.5]\id\Step\id\step{\hstr{200}\x}\step[2.5]\d
\step[1.5]\d\\
\hh\hstr{200}\id\hstep\cd\hstep\cd\hstep\id\hstep\id\hstep\cd\hstep\cd
\hstep\id\\
\hh\id\step{\hstr{200}\hrd}\step\x\step{\hstr{200}\hld}\step\id\step
\id\step{\hstr{200}\hrd}\step\x\step{\hstr{200}\hld}\step\id\\
\hh\id\step\id\step\x\step\x\step\id\step\id\step\id\step\id
\step\x\step\x\step\id\step\id\\
\hh\id\step{\hstr{200}\hru}\step\x\step{\hstr{200}\hlu}\step\id\step
\id\step{\hstr{200}\hru}\step\x\step{\hstr{200}\hlu}\step\id\\
\hh\hstr{200}\id\hstep\cu\hstep\cu\hstep\id\hstep\id\hstep\cu\hstep\cu
\hstep\id\\
\hh\d\step[1.5]\d\step[2.5]{\hstr{200}\x}\step\id\Step\id\step[2.5]\dd
\step[1.5]\dd\\
\hh\hstep\nw1\step\nw1\step\cd\step[1.5]\x\step[1.5]\dd\Step\dd\step
\sw1\\
\hh\step[1.5]\nw1\step\x\hstep\dd\hstep\sw1\hstep\hld\step\dd\step[1.5]
\sw1\hstep\sw1\\
\hh\step[2.5]{\hstr{200}\hlu}\step\hru\hstep\dd\step\dd\hstep\x\step
\sw1\hstep\sw1\\
\hh\step[3.5]\id\step\cu*\step[1.5]\cu\step\cu\hstep\sw1\\
\hh\step[3.5]\id\step[1.5]\id\step[2.5]\id\Step\x\\
\hh\step[3.5]\id\step[1.5]\id\step[2.5]{\hstr{400}\hlu}\step\id
\end{tangle}
\multiply\unitlens by 2
\;=\;
\divide\unitlens by 3
\multiply\unitlens by 2
\hstretch 80 \vstretch 80
\def5{3}
\begin{tangle}
\hh\id\Step{\hstr{600}\hld}\step[4.25]\id\Step\id\\
\hh{\hstr{160}\x}\step[3.75]\id\step[4.25]\id\Step\id\\
\hh\id\step{\hstr{160}\cd}\step{\hstr{280}\cd}\step[2.5]\id\Step\id\\
\hh\id\step{\hstr{160}\hrd}\step\x\step[1.5]{\hstr{320}\hld}
\step[2.5]\id\Step\id\\
\hh\id\step\id\step\x\step{\hstr{120}\x}\Step\id\step[2.5]\id\Step\id\\
\hh\id\step{\hstr{160}\hru}\step\x\step\cd\step\cd\Step\id\Step\id\\
\hh\id\step{\hstr{160}\cu}\step\id\step\id\step\x\step\id\step
{\hstr{160}\cd*}\step\id\\
\hh\id\Step\id\Step\id\step\id\step\id\step{\hstr{160}\hru}\sw1\Step\id
\step\id\\
\hh\id\Step\id\Step\id\step\id\step\id\step\x\step[3]\id\step\id\\
\hh\id\Step\id\Step\id\step\id\step\x\step\nw1\step{\hstr{160}\hld}
\step{\hstr{320}\hrd}\\
\hh\id\Step\id\Step\id\step\x\step\nw1\step\id\step\id\step\x\Step\id\\
\id\Step\id\Step\hx\step\ox{\sstyle 21}\step\id\step\hddcu\step\id\Step\id\\
\id\Step\id\step\hddcd\step\id\step\ox{\sstyle 12}\step\hx\Step\id\Step\id\\
\hh\id\Step\x\step\id\step\id\step\nw1\step\x\step\id\Step\id\Step\id\\
\hh{\hstr{320}\hlu}\step{\hstr{160}\hru}\step\nw1\step\x\step\id\step\id
\Step\id\Step\id\\
\hh\Step\id\step\id\step[3]\x\step\id\step\id\step\id\Step\id\Step\id\\
\hh\Step\id\step\id\Step\sw1{\hstr{160}\hld}\step\id\step\id\step\id
\Step\id\Step\id\\
\hh\Step\id\step{\hstr{160}\cu*}\step\id\step\x\step\id\step\id\step
{\hstr{160}\cd}\step\id\\
\hh\Step\id\Step\id\Step\cu\step\cu\step\x\step{\hstr{160}\hld}\step\id\\
\hh\Step\id\Step\id\step[2.5]\id\Step{\hstr{120}\x}\step\x\step\id
\step\id\\
\hh\Step\id\Step\id\step[2.5]{\hstr{320}\hru}\step[1.5]\x\step
{\hstr{160}\hlu}\step\id\\
\hh\Step\id\Step\id\step[2.5]{\hstr{280}\cu}\step{\hstr{160}\cu}\step\id\\
\hh\Step\id\Step\id\step[4.25]\id\step[3.75]{\hstr{160}\x}\\
\hh\Step\id\Step\id\step[4.25]{\hstr{600}\hru}\Step\id
\end{tangle}
\;=\;
\def5{2}
\begin{tangle}
\hh\id\Step{\hstr{680}\hld}\step[6]\id\step[2.75]\id\\
\hh{\hstr{160}\x}\step[4.25]\id\step[6]\id\step[2.75]\id\\
\hh\id\step{\hstr{160}\cd}\step{\hstr{360}\cd}\step[3.75]\id
\step[2.75]\id\\
\hh\id\step{\hstr{160}\hrd}\step\x\step[2.5]{\hstr{320}\hld}\step[3.75]
\id\step[2.75]\id\\
\hh\id\step\id\step\x\step{\hstr{200}\x}\Step\id\step[3.75]
\id\step[2.75]\id\\
\hh\id\step{\hstr{160}\hru}\step\x\Step\cd\step\cd\step[1.5]
{\hstr{280}\cd*}\step\id\\
\hh\id\step\id\step\sw1\step\id\Step\id\step\x\step\id\hstep
{\hstr{160}\cd}\step{\hstr{240}\hld}\step{\hstr{160}\hrd}\\
\hh\id\step\id\step\id\Step\id\Step\id\step\id\step{\hstr{160}\hru}\dd
\step{\hstr{160}\hld\hstep\hrd}\hstep\id\step\id\step\id\\
\hh\id\step\id\step\id\Step\id\Step\id\step\id\step\x\step[1.5]\id
\step\x\step\id\hstep\id\step\id\step\id\\
\hh\id\step\id\step\id\Step\id\Step\id\step\x\step\nw1\hstep\cu\step
\cu\hstep\x\step\id\\
\hh\id\step\id\step\id\Step\id\Step\x\step\nw1\step\id\step\id\Step
\x\step\id\step\id\\
\id\step\id\step\id\Step\x\step\ox{\sstyle 21}\step\id\step\cu\step\id
\step\id\step\id\\
\id\step\id\step\id\step\cd\step\id\step\ox{\sstyle 12}\step\x\Step\id\step\id
\step\id\\
\hh\id\step\id\step\x\Step\id\step\id\step\nw1\step\x\Step\id\Step\id
\step\id\step\id\\
\hh\id\step\x\hstep\cd\step\cd\hstep\nw1\step\x\step\id\Step\id\Step
\id\step\id\step\id\\
\hh\id\step\id\step\id\hstep\id\step\x\step\id\step[1.5]\x\step\id
\step\id\Step\id\Step\id\step\id\step\id\\
\hh\id\step\id\step\id\hstep{\hstr{160}\hlu\hstep\hru}\step\dd
{\hstr{160}\hld}\step\id\step\id\Step\id\Step\id\step
\id\step\id\\
\hh{\hstr{160}\hlu}\step{\hstr{240}\hru}\step{\hstr{160}\cu}\hstep\id
\step\x\step\id\Step\id\step\sw1\step\id\step\id\\
\hh\step\id\step{\hstr{280}\cu*}\step[1.5]\cu\step\cu\Step\x\step
{\hstr{160}\hld}\step\id\\
\hh\step\id\step[2.75]\id\step[3.75]\id\Step{\hstr{200}\x}\step\x
\step\id\step\id\\
\hh\step\id\step[2.75]\id\step[3.75]{\hstr{320}\hru}\step[2.5]\x\step
{\hstr{160}\hlu}\step\id\\
\hh\step\id\step[2.75]\id\step[3.75]{\hstr{360}\cu}\step{\hstr{160}\cu}
\step\id\\
\hh\step\id\step[2.75]\id\step[6]\id\step[4.25]{\hstr{160}\x}\\
\hh\step\id\step[2.75]\id\step[6]{\hstr{680}\hru}\Step\id
\end{tangle}
\;=T^{(4)}(\beta^4_\ell)
\end{equation*}
where we used the module-comodule compatibility
and \eqref{mod-alg-rel} in the first equation. In the second
identity we use again \eqref{mod-alg-rel}. To get the third equation in the
graphic we applied the right module-algebra and the left comodule-coalgebra
compatibilities of Definition \ref{hp} as well as \eqref{rel-cocycle-assoc2}.
With the help of the module-comodule compatibility the fourth identity
can be derived from the definition of $\beta^4_\ell$ in
\eqref{beta-4} and the definition of $T^{(4)}$ in \eqref{dressing}.
Next we will prove the identity $\beta^3_r =
T^{(4)}\left(\beta^4_r\right)$. And therefore all
relations in both rows of the diagram in Proposition \ref{b-bialg}
are satisfied.
\begin{equation*}
\beta^3_r=\;
\hstretch 60 \vstretch 60
\def5{5}
\begin{tangle}
\hh{\hstr{120}\cd}\step{\hstr{270}\cd}\Step\id\step[1.5]\id\\
\hh{\hstr{120}\hrd}\step\x\Step{\hstr{300}\hld}\step[1.5]\cd*\step\id\\
\hh\id\step\x\step\id\Step\id\step[2.5]{\hstr{90}\x}\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step{\hstr{120}\cd}\step\cd\step
\cu\step\id\\
\hh\id\step\id\step\id\step\id\step{\hstr{120}\hrd}\step\x\step\id
\step[1.5]\id\step[1.5]\id\\
\hh\id\step\id\step\id\step\id\step\id\step\x\step{\hstr{120}\hlu}
\hstep\ld\step[1.5]\id\\
\hh\id\step\id\step\id\step\nw1{\hstr{120}\hlu}\step{\hstr{120}\cu}
\hstep\id\step\d\step\id\\
\hh\id\step\id\step\nw1\step\x\step\sw1\step\cd\step\x\\
\hh\id\step\id\Step\x\step\x\step\sw1\step\x\step\id\\
\hh\id\step\id\step\sw1\step\x\step\x\step\sw1\step\id\step\id\\
\hh\id\step\x\step\sw1\step\x\step\x\Step\id\step\id\\
\hh\x\step\cu\step\sw1\step\x\step\nw1\step\id\step\id\\
\hh\id\step\d\step\id\hstep{\hstr{120}\cd\hstep\hrd\d}\step\id
\step\id\step\id\\
\hh\id\step[1.5]\ru\hstep{\hstr{120}\hrd}\step\x\step\id
\step\id\step\id\step\id\step\id\\
\hh\id\step[1.5]\id\step[1.5]\id\step\x\step{\hstr{120}\hlu}
\step\id\step\id\step\id\step\id\\
\hh\id\step\cd\step\cu\step{\hstr{120}\cu}
\step\id\step\id\step\id\step\id\\
\hh\id\step\id\step{\hstr{90}\x}\step[2.5]\id\Step\id\step\x\step\id\\
\hh\id\step\cu*\step[1.5]{\hstr{300}\hru}\Step\x\step\hstr{120}\hlu\\
\hh\id\step[1.5]\id\Step{\hstr{270}\cu}\step\hstr{120}\cu
\end{tangle}
\;=\;
\def5{3}
\begin{tangle}
\hh\id\step[5]{\hstr{180}\hld}\step[5.5]\id\step[1.5]\id\\
\hh\id\step[4]\sw1\hstep{\hstr{120}\cd}\step[4.5]\id\step[1.5]\id\\
\hh\id\step[3]\sw1\step[1.5]\id\hstep{\hstr{180}\hld}\step[3.5]
{\hstr{120}\cd*}\hstep\id\\
\hh\nw1\step\sw1\step[2.5]\id\hstep\id\step[1.5]{\hstr{210}\x}\Step\id
\hstep\id\\
\hh\step\x\step[3.5]\id\hstep\id\step\cd\step[1.5]{\hstr{180}\hld}
\step{\hstr{120}\hld}\hstep\hrd\\
\hh\sw1{\hstr{120}\cd}\Step\dd\hstep\x\step\id\step\dd\step[1.5]
\x\step\id\hstep\id\hstep\id\\
\hh\id\step{\hstr{120}\hrd}\step{\hstr{120}\x}\step\id\step\lu\hstep
\cd\step\cd\hstep\cu\hstep\id\hstep\id\\
\hh\id\hstep\dd\step\x\Step\x\Step\id\hstep\id\step\x\step\id\step\x
\hstep\id\\
\hh\id\hstep\id\step[1.5]\id\step{\hstr{120}\x\hstep\x}\hstep\cu\step\cu
\hstep\dd\step\id\hstep\id\\
\hh\id\hstep\id\step[1.5]\x\Step\x\Step\x\Step\x\step[1.5]\id\hstep\id\\
\hh\id\hstep\id\step\dd\hstep\cd\step\cd\hstep{\hstr{120}\x\hstep\x}
\step\id\step[1.5]\id\hstep\id\\
\hh\id\hstep\x\step\id\step\x\step\id\hstep\id\Step\x\Step\x\step\dd
\hstep\id\\
\hh\id\hstep\id\hstep\cd\hstep\cu\step\cu\hstep\rd\step\id\step
{\hstr{120}\x\hstep\hlu}\step\id\\
\hh\id\hstep\id\hstep\id\step\x\step[1.5]\dd\step\id\step\x\hstep\dd
\Step{\hstr{120}\cu}\sw1\\
\hh\hlu\hstep{\hstr{120}\hru}\step{\hstr{180}\hru\hstep}\cu\step\id
\hstep\id\step[3.5]\x\\
\hh\hstep\id\hstep\id\Step{\hstr{210}\x}\step[1.5]\id\hstep\id
\step[2.5]\sw1\step\nw1\\
\hh\hstep\id\hstep{\hstr{120}\cu*}\step[3.5]{\hstr{180}\hru}\hstep\id
\step[1.5]\sw1\step[3]\id\\
\hh\hstep\id\step[1.5]\id\step[4.5]{\hstr{120}\cu}\hstep\sw1\step[4]
\id\\
\hh\hstep\id\step[1.5]\id\step[5.5]{\hstr{180}\hru}\step[5]\id
\end{tangle}
\;=\;
\hstretch 100 \vstretch 100
\divide\unitlens by 2
\def5{5}
\begin{tangle}
\hh\step\id\step[6]{\hstr{800}\hld}\step[5]\id\step[3]\id\\
\hh\step\id\step[4]\sw2\step[3.5]{\hstr{300}\cd}\step[3.5]\id\step[3]\id\\
\hh\step\id\Step\sw2\step[5]\dd\Step{\hstr{200}\hld}\step[3]\cd*
\step[2.5]\id\\
\hh\step{\hstr{200}\x}\step[5.5]\dd\step[1.5]\sw1\step{\hstr{300}\x}\step
\d\Step\id\\
\hh\sw1\step{\hstr{200}\cd}\step[4]\dd\step\sw1\step\sw1\step[3]\id
\step[1.5]\d\step[1.5]\id\\
\hh\id\step{\hstr{200}\cd}\step\nw1\step[2.5]\dd\step\dd\step\sw1\hstep
{\hstr{350}\ld}\step[-1.5]{\hstr{300}\hld}\step{\hstr{200}\hld}
\step[1.5]\id\\
\hh\id\step{\hstr{300}\hrd}\hstep\d\step[1.5]\id\Step\dd\step[1.5]\id
\step\dd\step[1.5]\id\Step\id\step[1.5]\x\step\id\step[1.5]\rd\\
\hh\id\step\id\hstep{\hstr{200}\hld\hstep\hrd}\hstep\id\step[1.5]
\dd\Step\id\step\id\Step\id\Step\id\step\cd\hstep\id\step
{\hstr{150}\x}\step\id\\
\hh\id\step\id\hstep\id\step\x\step\id\hstep\id\step\dd\step[2.5]\id
\step\id\Step\id\Step\x\step\id\hstep\x\step[1.5]\id\step\id\\
\hh\id\step\id\hstep\cu\step\cu\hstep\x\step[2.5]\dd\step\id\step
{\hstr{200}\cd\hstep\hrd}\cu\hstep\id\step\d\step\id\step\id\\
\hh\id\step\id\step\id\Step\x\step{\hstr{250}\x}\step\cd\hstep
{\hstr{200}\hrd}\step\x\step\id\hstep\x\step[1.5]\id\step\id\step\id\\
\hh\id\step\id\step\id\step[1.5]\dd\step\x\step[2.5]\x\step\id\hstep\id
\step\x\step{\hstr{200}\hlu}\hstep\id\step\id\step[1.5]\id\step\id
\step\id\\
\hh\id\step\id\step\id\step[1.5]\id\step[1.5]\id\step{\hstr{250}\x}
\step\x\hstep\cu\step{\hstr{200}\cu}\dd\step\id\step[1.5]\id
\step\id\step\id\\
\hh\id\step\id\step\id\step[1.5]\id\step[1.5]\x\step[2.5]\x\step\x
\step[2.5]\x\step[1.5]\id\step[1.5]\id\step\id\step\id\\
\hh\id\step\id\step\id\step[1.5]\id\step\dd{\hstr{200}\cd}\step\cd
\hstep\x\step{\hstr{250}\x}\step\id\step[1.5]\id\step[1.5]\id
\step\id\step\id\\
\hh\id\step\id\step\id\step[1.5]\id\step\id\hstep{\hstr{200}\hrd}\step
\x\step\id\hstep\id\step\x\step[2.5]\x\step\dd\step[1.5]\id\step\id
\step\id\\
\hh\id\step\id\step\id\step[1.5]\x\hstep\id\step\x\step{\hstr{200}\hlu}
\hstep\cu\step{\hstr{250}\x}\step\x\Step\id\step\id\step\id\\
\hh\id\step\id\step\d\step\id\hstep\cd{\hstr{200}\hlu\hstep\cu}
\step\id\step\dd\step[2.5]\x\hstep\cd\step\cd\hstep\id\step\id\\
\hh\id\step\id\step[1.5]\x\hstep\id\step\x\Step\id\Step\id\step\id
\step[2.5]\dd\step\id\hstep\id\step\x\step\id\hstep\id\step\id\\
\hh\id\step{\hstr{150}\x}\step\id\hstep\cu\step\id\Step\id\Step\id
\step\id\Step\dd\step[1.5]\id\hstep{\hstr{200}\hlu\hstep\hru}
\hstep\id\step\id\\
\hh\lu\step[1.5]\id\step\x\step[1.5]\id\Step\id\step[1.5]\dd\step\id
\step[1.5]\dd\Step\id\step[1.5]\d\hstep{\hstr{300}\hlu}\step\id\\
\hh\step\id\step[1.5]{\hstr{200}\hru}\step{\hstr{300}\hru}\step[-1.5]
{\hstr{350}\ru}\hstep\sw1\step\dd\step\dd\step[2.5]\nw1\step
{\hstr{200}\cu}\step\id\\
\hh\step\id\step[1.5]\d\step[1.5]\id\step[3]\sw1\step\sw1\step\dd
{\hstr{200}\Step\cu}\step\sw1\\
\hh\step\id\Step\d\step{\hstr{300}\x}\step\sw1\step[1.5]\dd\step[5.5]
{\hstr{200}\x}\\
\hh\step\id\step[2.5]\cu*\step[3]{\hstr{200}\hru}\Step\dd\step[4]\sw2
\step[3]\id\\
\hh\step\id\step[3]\id\step[3.5]{\hstr{300}\cu}\step[2.5]\sw2\step[5]\id\\
\hh\step\id\step[3]\id\step[5]{\hstr{800}\hru}\step[6]\id
\end{tangle}
\;=T^{(4)}(\beta^4_r)
\end{equation*}
The first identity in this graphical calculation has been obtained from
\eqref{rel-mod-coalg} and \eqref{rel-comod-alg} of Lemma \ref{strong-hd8}.
In the second equation we use \eqref{mod-alg-rel2}, Lemma
\ref{strong-hd5} and \eqref{alg-coalg-rel}.
With the help of the left module-algebra and right comodule-coalgebra
compatibilities of Definition~\ref{hp} and the relativizations
\eqref{rel-mod-coalg} and \eqref{rel-comod-alg} of Lemma~\ref{strong-hd8}
we derive the third identity. The fourth equation holds by definition.
In order to complete the proof of the proposition the equation
$\beta^4_\ell=\beta^4_r$ needs to be shown. This will be done in the
following calculation.
\begin{equation*}
\hstretch 100 \vstretch 100
\divide\unitlens by 2
\def5{4}
\beta^4_\ell=
\begin{tangle}
\hh\id\Step\id\step{\hstr{550}\cd}\step[3.25]{\hstr{125}\cd*}\Step\id\\
\hh{\hstr{200}\hrd}\step\x\step[2.5]{\hstr{600}\hld}\step[2.25]\sw1
\step[1.25]\nw1\step\id\\
\hh\id\step\x\step{\hstr{250}\x}\step[3]\id\step[1.5]{\hstr{150}\cd}\step
{\hstr{300}\hld}\step\id\\
\hh{\hstr{200}\hru}\step\x\step{\hstr{300}\cd}\step\cd\step\id\step[1.5]
\id\step{\hstr{200}\hrd}\hstep\id\step\id\\
\hh\id\step\sw1\step\id\step\id\step[3]\x\step\id\step\id\step[1.5]\x
\step\id\hstep\id\step\id\\
\hh\id\step\id\Step\id\step\id\Step\sw1\step{\hstr{200}\hru}\step\id\step
\dd
\step\cu\hstep\x\\
\hh\id\step\id\Step\id\step\id\Step\id\Step{\hstr{200}\x}\step\id\Step\x
\step\id\\
\hh\id\step\id\Step\id\step\id\Step{\hstr{200}\x}\Step\nw1{\hstr{200}\cu}
\step\id\step\id\\
\hh\id\step\id\Step\id\step{\hstr{200}\x}\Step\nw1\Step\x\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\cd\hstep\cd}\step\id\step
\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\hrd}\step\x\step
{\hstr{200}\hld}\step\id\step\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step\id\step\x\step\x\step\id\step\id
\step\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\hru}\step\x\step
{\hstr{200}\hlu}\step\id\step\id\Step\id\step\id\\
\hh\id\step\id\Step\id\step\id\step{\hstr{200}\cu\hstep\cu}\step\id\step
\id\Step\id\step\id\\
\hh\id\step\id\Step\x\Step\nw1\Step{\hstr{200}\x}\step\id\Step\id\step\id\\
\hh\id\step\id\step{\hstr{200}\cd}\nw1\Step{\hstr{200}\x}\Step\id\step\id
\Step\id\step\id\\
\hh\id\step\x\Step\id\step{\hstr{200}\x}\Step\id\Step\id\step\id\Step\id
\step\id\\
\hh\x\hstep\cd\step\dd\step\id\step{\hstr{200}\hld}\step\sw1\Step\id\step
\id\Step\id\step\id\\
\hh\id\step\id\hstep\id\step\x\step[1.5]\id\step\id\step\x\step[3]\id
\step\id\step\sw1\step\id\\
\hh\id\step\id\hstep{\hstr{200}\hlu}\step\id\step[1.5]\id\step\cu\step
{\hstr{300}\cu}\step\x\step\hstr{200}\hld\\
\hh\id\step{\hstr{300}\hru}\step{\hstr{150}\cu\step}\id\step[3]
{\hstr{250}\x}\step\x\step\id\\
\hh\id\step\nw1\step[1.25]\sw1\step[2.25]{\hstr{600}\hru}\step[2.5]\x
\step{\hstr{200}\hlu}\\
\hh\id\Step{\hstr{125}\cu*}\step[3.25]{\hstr{550}\cu}\step\id\Step\id
\end{tangle}
=
\def5{2}
\begin{tangle}
\hh\id\step[4]\id\step[3.5]{\hstr{700}\cd}\step[5]\id\step[3.5]\id\\
\hh\id\step[4]\id\step[3.5]\id\step[1.5]{\hstr{1100}\hld}\step[5]\id
\step[3.5]\id\\
\hh\id\step[4]\id\step[3.5]\id\step[1.5]\id\step[5]\cd\Step{\hstr{500}\cd*}
\step\id\\
\hh\id\step[3]{\hstr{200}\cd}\Step\cd\step{\hstr{200}\hrd}\step
{\hstr{600}\hld}\step\id\step[1.5]\cd\step[1.5]{\hstr{600}\hld}
\step\id\\
\hh\Rd\step\id\Step\d\step[1.5]\id\step\x\step\x\step{\hstr{400}\hld}
\step\id\step\dd\hstep\cd\step{\hstr{200}\hrd}\Step\id\step\id\\
\hh\id\step{\hstr{200}\hld}\step\rd\step\cd\step\id\step\id\step\x\step\id
\step\id\step{\hstr{200}\hld}\step\id\step\id\step\id\step\x\step\id
\step\ld\step\id\\
\hh\id\step\id\step\x\step\id\step\id\step\x\step\id\step\id\step\id
\step\id\step\id\step\id\step\id\step\x\step\id\step\id\step\id\step
\id\step\id\step\x\\
\hh\id\step\id\step\id\step\id\step\id\step\x\step\x\step\id\step\id
\step\id\step\id\step\id\step\x\step\x\step\id\step\id\step\id\step
\x\step\id\\
\hh\id\step\id\step\id\step\id\step\x\step\x\step\x\step\id\step\id
\step\id\step\x\step\x\step\x\step\id\step\x\step\id\step\id\\
\hh\id\step\id\step\id\step\x\step\x\step\x\step\x\step\id\step\x\step
\x\step\x\step\x\step\id\step\id\step\id\step\id\\
\hh\id\step\id\step\x\step\x\step\x\step\x\step\x\step\id\step\x\step
\x\step\x\step\x\step\id\step\id\step\id\\
\hh\id\step\x\step\x\step\x\step\x\step\x\step\x\step\id\step\x\step
\id\step\id\step\x\step\x\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\x\step\x\step\x\step
\x\step\id\step\id\step\id\step\x\step\id\step\id\step\id\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\id\step\x\step
\x\step\x\step\x\step\id\step\x\step\id\step\id\step\id\step\id
\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\id\step\id\step\x
\step\x\step\id\step\id\step\id\step\x\step\id\step\id\step\id
\step\id\step\id\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\id\step\x\step\id
\step\id\step\id\step\x\step\x\step\id\step\id\step\id\step\id
\step\id\step\id\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\id\step\x\step\id\step
\x\step\x\step\x\step\x\step\id\step\id\step\id\step\id\step\id
\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\id\step\x\step\id\step\id\step\id\step
\x\step\x\step\x\step\x\step\id\step\id\step\id\step\id\step\id
\step\id\\
\hh\id\step\id\step\x\step\x\step\id\step\id\step\x\step\id\step\x\step
\x\step\x\step\x\step\x\step\x\step\id\\
\hh\id\step\id\step\id\step\x\step\x\step\x\step\x\step\id\step\x\step\x
\step\x\step\x\step\x\step\id\step\id\\
\hh\id\step\id\step\id\step\id\step\x\step\x\step\x\step\x\step\id\step\x
\step\x\step\x\step\x\step\id\step\id\step\id\\
\hh\id\step\id\step\x\step\id\step\x\step\x\step\x\step\id\step\id\step
\id\step\x\step\x\step\x\step\id\step\id\step\id\step\id\\
\hh\id\step\x\step\id\step\id\step\id\step\x\step\x\step\id\step\id
\step\id\step\id\step\id\step\x\step\x\step\id\step\id\step\id\step
\id\step\id\\
\hh\x\step\id\step\id\step\id\step\id\step\id\step\x\step\id\step\id\step
\id\step\id\step\id\step\id\step\id\step\x\step\id\step\id\step\x
\step\id\step\id\\
\hh\id\step\ru\step\id\step\x\step\id\step\id\step\id\step{\hstr{200}\hru}
\step\id\step\id\step\x\step\id\step\id\step\cu\step\lu\step
{\hstr{200}\hru}\step\id\\
\hh\id\step\id\Step{\hstr{200}\hlu}\step\cu\hstep\dd\step\id\step
{\hstr{400}\hru}\step\x\step\x\step\id\step[1.5]\d\Step\id\step\Lu\\
\hh\id\step{\hstr{600}\hru}\step[1.5]\cu\step[1.5]\id\step{\hstr{600}\hru}
\step{\hstr{200}\hlu}\step\cu\Step{\hstr{200}\cu}\step[3]\id\\
\hh\id\step{\hstr{500}\cu*}\Step\cu\step[5]\id\step[1.5]\id\step[3.5]\id
\step[4]\id\\
\hh\id\step[3.5]\id\step[5]{\hstr{1100}\hru}\step[1.5]\id\step[3.5]\id
\step[4]\id\\
\hh\id\step[3.5]\id\step[5]{\hstr{700}\cu}\step[3.5]\id\step[4]\id
\end{tangle}
=
\def5{5}
\begin{tangle}
\hh\id\step[3.25]\id\step[5.25]{\hstr{300}\cd}\step[3.5]\id\step[3]\id\\
\hh\id\step[3.25]\id\step[4.75]\dd\Step{\hstr{200}\hld}\step[3]\cd*
\step[2.5]\id\\
\hh\id\step[3.25]\id\step[4.25]\dd\step[1.5]\sw1\step{\hstr{300}\x}\step
\d\Step\id\\
\hh\id\step[3.25]\id\step[3.75]\dd\step\sw1\step\sw1\step[3]\id
\step[1.5]\d\step[1.5]\id\\
\hh\id\step[2.5]{\hstr{150}\cd}\step[2.5]\dd\step\dd\step\sw1\hstep
{\hstr{350}\ld}\step[-1.5]{\hstr{300}\hld}\step{\hstr{200}\hld}
\step[1.5]\id\\
\hh{\hstr{300}\hrd}\step\id\step[1.5]\id\Step\dd\step[1.5]\id
\step\dd\step[1.5]\id\Step\id\step[1.5]\x\step\id\step[1.5]\id\\
\hh\id\hstep{\hstr{200}\hld\hstep\hrd}\hstep\id\step[1.5]
\dd\Step\id\step\id\Step\id\Step\id\step\cd\hstep\id\step
{\hstr{150}\x}\\
\hh\id\hstep\id\step\x\step\id\hstep\id\step\dd\step[2.5]\id
\step\id\Step\id\Step\x\step\id\hstep\x\step[1.5]\id\\
\hh\id\hstep\cu\step\cu\hstep\x\step[2.5]\dd\step\id\step[1.5]\dd
\step\sw1\step\cu\hstep\id\step\d\step\id\\
\hh\id\step\id\Step\x\step{\hstr{250}\x}\step\cd\hstep\cd\step
{\hstr{200}\hrd}\step[1.5]\x\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\dd\step\x\step[2.5]\x\step\id\hstep\id
\step\x\step\id\step[1.5]\id\step\id\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\id\step[1.5]\id\step{\hstr{250}\x}
\step\x\hstep\cu\step{\hstr{200}\hlu}\step\dd\step\id\step[1.5]\id
\step\id\\
\hh\id\step\id\step[1.5]\id\step[1.5]\x\step[2.5]\x\step\x
\step[2.5]\x\step[1.5]\id\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\id\step\dd\step{\hstr{200}\hrd}\step\cd
\hstep\x\step{\hstr{250}\x}\step\id\step[1.5]\id\step[1.5]\id
\step\id\\
\hh\id\step\id\step[1.5]\id\step\id\step[1.5]\id\step
\x\step\id\hstep\id\step\x\step[2.5]\x\step\dd\step[1.5]\id\step\id\\
\hh\id\step\id\step[1.5]\x\step[1.5]{\hstr{200}\hlu}
\step\cu\hstep\cu\step{\hstr{250}\x}\step\x\Step\id\step\id\\
\hh\id\step\d\step\id\hstep\cd\step\sw1\step\dd\step[1.5]\id\step\dd
\step[2.5]\x\hstep\cd\step\cd\hstep\id\\
\hh\id\step[1.5]\x\hstep\id\step\x\Step\id\Step\id\step\id
\step[2.5]\dd\step\id\hstep\id\step\x\step\id\hstep\id\\
\hh{\hstr{150}\x}\step\id\hstep\cu\step\id\Step\id\Step\id
\step\id\Step\dd\step[1.5]\id\hstep{\hstr{200}\hlu\hstep\hru}
\hstep\id\\
\hh\id\step[1.5]\id\step\x\step[1.5]\id\Step\id\step[1.5]\dd\step\id
\step[1.5]\dd\Step\id\step[1.5]\id\step{\hstr{300}\hlu}\\
\hh\id\step[1.5]{\hstr{200}\hru}\step{\hstr{300}\hru}\step[-1.5]
{\hstr{350}\ru}\hstep\sw1\step\dd\step\dd\step[2.5]{\hstr{150}\cu}
\step[2.5]\id\\
\hh\id\step[1.5]\d\step[1.5]\id\step[3]\sw1\step\sw1\step\dd
\step[3.75]\id\step[3.25]\id\\
\hh\id\Step\d\step{\hstr{300}\x}\step\sw1\step[1.5]\dd\step[4.25]\id
\step[3.25]\id\\
\hh\id\step[2.5]\cu*\step[3]{\hstr{200}\hru}\Step\dd\step[4.75]\id
\step[3.25]\id\\
\hh\id\step[3]\id\step[3.5]{\hstr{300}\cu}\step[5.25]\id\step[3.25]\id\\
\end{tangle}
=\beta^4_r
\end{equation*}
where the first equation has been derived with the help of
\eqref{act-coact-triv}. In the second and the third equation we use
\eqref{mod-alg-rel} and \eqref{mod-alg-rel2}.
Finally we apply \eqref{strong-hopf-combo} to obtain the fourth
identity.
Hence Proposition \ref{b-bialg} and therefore Theorem \ref{hp-cp} have been
proven.\end{proof}
|
\section{Introduction}
It is known that there are eight double coverings of the orthogonal group
$O(p,q)$ \cite{Dab88,BD89}:
\[
\rho^{a,b,c}:\;\pin^{a,b,c}(p,q)\longrightarrow O(p,q),
\]
where $a,b,c\in\{+,-\}$. The group $O(p,q)$ consists of four connected
components: identity connected component $O_0(p,q)$, and three components
corresponding to parity reversal $P$, time reversal $T$, and the
combination of these two $PT$, i.e., $O(p,q)=O_0(p,q)\cup P(O_0(p,q))\cup
T(O_0(p,q))\cup PT(O_0(p,q))$. Further, since the four element group
(reflection group) $\{1,P,T,PT\}$ is isomorphic to the finite group
$\dZ_2\otimes\dZ_2$ (Gauss-Klein group) \cite{Sal81a,Sal84}, then $O(p,q)$
may be represented by a semidirect product $O(p,q)\simeq O_0(p,q)\odot
(\dZ_2\otimes\dZ_2)$.
The signs of $a,b,c$ correspond to the signs of the
squares of the elements in $\pin^{a,b,c}(p,q)$ which cover space reflection,
time reversal and a combination of these two: $P^2=a,\,T^2=b,\,(PT)^2=c$.
An explicit form of the group $\pin^{a,b,c}(p,q)$ is given by the following
semidirect product
\[
\pin^{a,b,c}(p,q)\simeq\frac{(\spin_0(p,q)\odot C^{a,b,c})}{\dZ_2},
\]
where $C^{a,b,c}$ are the four double coverings of $\dZ_2\otimes\dZ_2$.
The all eight double coverings of the orthogonal group $O(p,q)$ are given
in the following table:
\begin{center}{\renewcommand{\arraystretch}{1.3}
\begin{tabular}{|c|l|l|}\hline
$a$ $b$ $c$ & $C^{a,b,c}$ & Remark\\ \hline
$+$ $+$ $+$ & $\dZ_2\otimes\dZ_2\otimes\dZ_2$ & $PT=TP$\\
$+$ $-$ $-$ & $\dZ_2\otimes\dZ_4$ & $PT=TP$\\
$-$ $+$ $-$ & $\dZ_2\otimes\dZ_4$ & $PT=TP$\\
$-$ $-$ $+$ & $\dZ_2\otimes\dZ_4$ & $PT=TP$\\ \hline
$-$ $-$ $-$ & $Q_4$ & $PT=-TP$\\
$-$ $+$ $+$ & $D_4$ & $PT=-TP$\\
$+$ $-$ $+$ & $D_4$ & $PT=-TP$\\
$+$ $+$ $-$ & $D_4$ & $PT=-TP$\\ \hline
\end{tabular}
}
\end{center}
Here $Q_4$ is a quaternion group, and $D_4$ is a dihedral group. According
to \cite{Dab88} the group $\pin^{a,b,c}(p,q)$ satisfying to condition
$PT=-TP$ is called {\it Cliffordian}, and respectively {\it non-Cliffordian}
when $PT=TP$.
The $\pin$ and $\spin$ groups (Clifford-Lipschitz groups) widely used in
algebraic topology \cite{BH,Hae56,AtBSh,Kar68,Kar,KT89},
in the definition of pinor and spinor structures on the riemannian
manifolds \cite{Mil63,Ger68,Ish78,Wh78,DT86,DP87,LM89,DR89,Cru91,AlCh94,
Ch94a,Ch94,CGT95,AlCh96},
spinor bundles \cite{RF90,RO90,FT96,Fr98,FT99},
and also have great importance in the theory of the Dirac operator on
manifolds \cite{Bau81,Bar91,Tr92,Fr97,Amm98}.
The Clifford-Lipschitz groups also intensively used in theoretical
physics \cite{CDD82,DW90,FRO90a,RS93,DWGK,Ch97}.
In essence, the D\c{a}browski group is a `detailed' (correct to a group of
discrete transformations) Clifford-Lipschitz group.
In the present paper the D\c{a}browski groups considered in the real and
complex spaces $\R^{p,q}$ and $\C^n$, respectively. The finite group
$\{1,P,T,PT\}$ is associated with an automorphism group $\{\Id,\star,
\widetilde{\phantom{cc}},\widetilde{\star}\}$ of the Clifford algebras
$C\kern -0.2em \ell_{p,q}$ and $\C_{p+q}$. At first, consideration carried out for
even-dimensional algebras $C\kern -0.2em \ell_{p,q},\,\C_{p+q}\,(p+q=2m)$ and associated
spaces $\R^{p,q}$ and $\C^{p+q}$. A relation between signatures of the
Clifford algebras $(p,q)=(\underbrace{++\ldots+}_{p\,\text{times}},
\underbrace{--\ldots-}_{q\,\text{times}})$ and signatures of the
D\c{a}browski groups $(a,b,c)$ is established in section 4. It is shown
that there exist eight non-isomorphic relations between $(p,q)$ and
$(a,b,c)$ over the field $\F=\R$ and only two over the field $\F=\C$.
Further, odd-dimensional spaces are considered in section 5, at this point
the Clifford algebra is understood as a direct sum of two even-dimensional
subalgebras. It is shown (Theorem \ref{t11}) that in this case
there exist two quotient groups $\pin^b$ and $\pin^{b,c}$, the latter
group exists only over the field $\F=\C$: $\pin^{b,c}(p+q-1,\C)$ if
$p+q\equiv 1,5\pmod{8}$. The set of discrete transformations of the
quotient group $\pin^{b,c}(p+q-1,\C)$ does not form a group. It allows
to relate this group with some chiral field in Physics. In conclusion,
by way of example, the algebra $\C_3$ and the quotient group
$\pin^{b,c}(p+q-1,\C)$ are related with a Dirac-Hestenes spinor
field \cite{Hest66,Hest90},
which has a broad application both in Physics and Geometry
\cite{Lou93,RVR93,RRSV95,RSVL,Var98a,Var98b}.
\section{Algebraic Preliminaries}
Clifford algebras play a key role in the definition of the $\pin$ groups.
Thus, in this section we will consider some basic facts about Clifford
algebras which relevant to definition and construction of the $\pin$ groups.
Let $\F$ be a field of characteristic 0 $(\F=\R,\,\F=\Om,\,\F=\C)$, where
$\Om$ is a field of double numbers $(\Om=\R\oplus\R)$, and $\R,\,\C$ are
the fields of real and complex numbers, respectively. A Clifford algebra
over a field $\F$ is an algebra with $2^n$ basis elements: $\mbox{\bf e}_0$
(unit of the algebra), $\mbox{\bf e}_1,\mbox{\bf e}_2,\ldots,\mbox{\bf e}_n$ and the products of the
one-index elements $\mbox{\bf e}_{i_1i_2\ldots i_k}=\mbox{\bf e}_{i_1}\mbox{\bf e}_{i_2}\ldots\mbox{\bf e}_{i_k}$.
Over the field $\F=\R$ the Clifford algebra is denoted as $C\kern -0.2em \ell_{p,q}$, where
the indices $p,q$ correspond to the indices of the quadratic form
\[
Q=x^2_1+\ldots+x^2_p-\ldots-x^2_{p+q}
\]
of a vector space $V$ associated with $C\kern -0.2em \ell_{p,q}$. A multiplication law
of $C\kern -0.2em \ell_{p,q}$ is defined by the following rule:
\begin{equation}\label{e1}
\mbox{\bf e}^2_i=\sigma(q-i)\mbox{\bf e}_0,\quad\mbox{\bf e}_i\mbox{\bf e}_j=-\mbox{\bf e}_j\mbox{\bf e}_i,
\end{equation}
where
\begin{equation}\label{e2}
\sigma(n)=\begin{cases}
-1 & \text{if $n\leq 0$},\\
+1 & \text{if $n>0$}.
\end{cases}
\end{equation}
The square of the volume element
$\omega=\mbox{\bf e}_{12\ldots n}$, $n=p+q$,
plays an important role
in the theory of Clifford algebras:
\begin{equation}\label{e3}
\omega^2=\begin{cases}
-1 & \text{if $p-q\equiv 1,2,5,6\pmod{8}$},\\
+1 & \text{if $p-q\equiv 0,3,4,7\pmod{8}$}.
\end{cases}
\end{equation}
A center ${\bf Z}_{p,q}$ of $C\kern -0.2em \ell_{p,q}$ consists of the unit $\mbox{\bf e}_0$ and the
volume element $\omega$. The element $\omega=\mbox{\bf e}_{12\ldots n}$ is belong
to a center when $n$ is odd. Indeed,
\begin{eqnarray}
\mbox{\bf e}_{12\ldots n}\mbox{\bf e}_i&=&(-1)^{n-i}\sigma(q-i)\mbox{\bf e}_{12\ldots i-1 i+1\ldots n},
\nonumber\\
\mbox{\bf e}_i\mbox{\bf e}_{12\ldots n}&=&(-1)^{i-1}\sigma(q-i)\mbox{\bf e}_{12\ldots i-1 i+1\ldots n},
\nonumber
\end{eqnarray}
therefore, $\omega\in{\bf Z}_{p,q}$ if and only if $n-i\equiv i-1\pmod{2}$, that is,
$n$ is odd. Using (\ref{e3}) we have
\begin{equation}\label{e4}
{\bf Z}_{p,q}=\begin{cases}
\phantom{1,}1 & \text{if $p-q\equiv 0,2,4,6\pmod{8}$},\\
1,\omega & \text{if $p-q\equiv 1,3,5,7\pmod{8}$}.
\end{cases}
\end{equation}
Moreover, when $n$ is odd a center ${\bf Z}_{p,q}$ is isomorphic
to the fields $\C$ and $\Om$, respectively:
\[
{\bf Z}_{p,q}\simeq\begin{cases}
\R\oplus i\R & \text{if $p-q\equiv 1,5\pmod{8}$},\\
\R\oplus e\R & \text{if $p-q\equiv 3,7\pmod{8}$},
\end{cases}
\]
where $e$ is a double unit $(e^2=1)$.
Further, let $\C_n=\C\otimesC\kern -0.2em \ell_{p,q}$ and $\Om_{p,q}=\Om\otimesC\kern -0.2em \ell_{p,q}$
be the Clifford algebras over the fields $\F=\C$ and $\F=\Om$, respectively.
\begin{theorem}\label{t1}
If $n=p+q$ is odd, then
\begin{eqnarray}
C\kern -0.2em \ell_{p,q}&\simeq&\C_{p+q-1}\quad\text{if $p-q\equiv 1,5\pmod{8}$},\nonumber\\
C\kern -0.2em \ell_{p,q}&\simeq&\Om_{p-1,q}\nonumber\\
&\simeq&\Om_{p,q-1}\quad\text{if $p-q\equiv 3,7\pmod{8}$}.\nonumber
\end{eqnarray}
\end{theorem}
\begin{proof} The structure of $C\kern -0.2em \ell_{p,q}$ allows to identify the Clifford
algebras over the different fields. Indeed, transitions $C\kern -0.2em \ell_{p-1,q}
\rightarrowC\kern -0.2em \ell_{p,q},\;C\kern -0.2em \ell_{p,q-1}\rightarrowC\kern -0.2em \ell_{p,q}$ may be represented
as transitions from the real coordinates in $C\kern -0.2em \ell_{p-1,q},\,C\kern -0.2em \ell_{p,q-1}$
to complex coordinates of the form $a+\omega b$, where $\omega$ is
an additional basis element $\mbox{\bf e}_{12\ldots n}$ (volume element). Since
$n=p+q$ is odd, then the volume element $\omega$ in accordance with
(\ref{e4}) belongs to ${\bf Z}_{p,q}$. Therefore, we can to identify it
with imaginary unit $i$ if $p-q\equiv 1,5\pmod{8}$ and with a double unit
$e$ if $p-q\equiv 3,7\pmod{8}$. The general element of the algebra $C\kern -0.2em \ell_{p,q}$
has a form $\mathcal{A}=\mathcal{A}^{\prime}+\omega\mathcal{A}^{\prime}$, where $\mathcal{A}^{\prime}$ is a general
element of the algebras $C\kern -0.2em \ell_{p-1,q},\,C\kern -0.2em \ell_{p,q-1}$.
\end{proof}
{\bf Example}. Let us consider the algebra $C\kern -0.2em \ell_{0,3}$. According to the
theorem \ref{t1} we have $C\kern -0.2em \ell_{0,3}\simeq\Om_{0,2}$, where $\Om_{0,2}$
is an algebra of elliptic biquaternions (it is a first so-called
Grassmann's extensive algebra introduced by Clifford in 1878 \cite{3}).
Since $\Om=\R\oplus\R$ and $\Om_{p,q}=\Om\otimesC\kern -0.2em \ell_{p,q}$, we have
$C\kern -0.2em \ell_{0,3}\simeqC\kern -0.2em \ell_{0,2}\oplusC\kern -0.2em \ell_{0,2}\simeq\BH\oplus\BH$, where $\BH$ is
a quaternion algebra.\\[0.4cm]
Generalizing this example we obtain
\begin{eqnarray}
C\kern -0.2em \ell_{p,q}&\simeq&C\kern -0.2em \ell_{p-1,q}\oplusC\kern -0.2em \ell_{p-1,q}\nonumber\\
&\simeq&C\kern -0.2em \ell_{p,q-1}\oplusC\kern -0.2em \ell_{p,q-1}\quad\text{if $p-q\equiv 3,7\pmod{8}$}.
\label{e5'}
\end{eqnarray}
Over the field $\F=\C$ there is the analogous result \cite{Rash}.
\begin{theorem}\label{t2}
When $p+q\equiv 1,3,5,7\pmod{8}$ the Clifford algebra over the field
$\F=\C$ decomposes into a direct sum of two subalgebras:
\[
\C_{p+q}\simeq\C_{p+q-1}\oplus\C_{p+q-1}.
\]
\end{theorem}
In Clifford algebra $C\kern -0.2em \ell_{p,q}$
there exist four fundamental
automorphisms \cite{Sch49,Rash}:
1) An automorphism $\mathcal{A}\rightarrow\mathcal{A}$.\\
This automorphism, obviously, is an identical automorphism of the algebra
$C\kern -0.2em \ell_{p,q}$, $\mathcal{A}$ is an arbitrary element of $C\kern -0.2em \ell_{p,q}$.
2) An automorphism $\mathcal{A}\rightarrow\mathcal{A}^{\star}$.\\
In more details, for arbitrary element $\mathcal{A}\inC\kern -0.2em \ell_{p,q}$ there exists a
decomposition
\[
\mathcal{A}=\mathcal{A}^{\prime}+\mathcal{A}^{\prime\p},
\]
where $\mathcal{A}^{\prime}$ is an element consisting of homogeneous odd elements, and
$\mathcal{A}^{\prime\p}$ is an element consisting of homogeneous even elements,
respectively.
Then the automorphism $\mathcal{A}\rightarrow\mathcal{A}^{\star}$ is such that the element
$\mathcal{A}^{\prime\p}$ is not changed, and the element $\mathcal{A}^{\prime}$ changes sign:
\[
\mathcal{A}^{\star}=-\mathcal{A}^{\prime}+\mathcal{A}^{\prime\p}.
\]
If $\mathcal{A}$ is a homogeneous element, then
\begin{equation}\label{e5}
\mathcal{A}^{\star}=(-1)^{k}\mathcal{A},
\end{equation}
where $k$ is a degree of element. It is easy to see that the automorphism
$\mathcal{A}\rightarrow\mathcal{A}^\star$ may be expressed via the volume element $\omega$:
\begin{equation}\label{e6}
\mathcal{A}^\star=\omega\mathcal{A}\omega^{-1},
\end{equation}
where $\omega^{-1}=(-1)^{\frac{n(n-1)}{2}}\omega$, $\mathcal{A}$ is an arbitrary
element of $C\kern -0.2em \ell_{p,q}$. When $k$ is odd,
for the basis elements $\mbox{\bf e}_{i_1i_2\ldots
i_k}$ the sign changes, and when $k$ is even the sign is not changed. Over
the field $\F=\C$ we can multiply $\omega$ by $\varepsilon=\pm i^{\frac{n(n-1)}
{2}}$, then the equality (\ref{e6}) is not changed. At this point
we have always
$(\varepsilon\omega)^2=1$. Therefore,
\begin{equation}\label{e7}
\mathcal{A}^\star=(\varepsilon\omega)\mathcal{A}(\varepsilon\omega).
\end{equation}
3) An antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$.\\
The antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is a reversion
of the element $\mathcal{A}$, that is, the substitution of the each basis element
$\mbox{\bf e}_{i_{1}i_{2}\ldots i_{k}}\in\mathcal{A}$ by
the element $\mbox{\bf e}_{i_{k}i_{k-1}\ldots i_{1}}$:
\[
\mbox{\bf e}_{i_{k}i_{k-1}\ldots i_{1}}=(-1)^{\frac{k(k-1)}{2}}\mbox{\bf e}_{i_{1}i_{2}\ldots
i_{k}}.
\]
Therefore, for any $\mathcal{A}\inC\kern -0.2em \ell_{p,q}$ we have
\begin{equation}\label{e8}
\widetilde{\mathcal{A}}=(-1)^{\frac{k(k-1)}{2}}\mathcal{A}.
\end{equation}
4) An antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^{\star}}$.\\
This antiautomorphism is a composition of the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ with the automorphism
$\mathcal{A}\rightarrow\mathcal{A}^{\star}$. In the case of homogeneous element from
formulae (\ref{e5}) and (\ref{e8}) it follows
\begin{equation}\label{e9}
\widetilde{\mathcal{A}^{\star}}=(-1)^{\frac{k(k+1)}{2}}\mathcal{A}.
\end{equation}
It is obvious that\hspace{1mm} $\widetilde{\!\!\widetilde{\mathcal{A}}}=\mathcal{A},\;
(\mathcal{A}^{\star})^{\star}=
\mathcal{A},$ and $\widetilde{(\widetilde{\mathcal{A}^{\star}})^{\star}}=\mathcal{A}$.
The Lipschitz group $\boldsymbol{\Gamma}_{p,q}$, also called the Clifford group, introduced
by Lipschitz in 1886 \cite{Lips}, may be defined as the subgroup of
invertible elements $s$ of the algebra $C\kern -0.2em \ell_{p,q}$:
\[
\boldsymbol{\Gamma}_{p,q}=\left\{s\inC\kern -0.2em \ell^+_{p,q}\cupC\kern -0.2em \ell^-_{p,q}\;|\;\forall x\in\R^{p,q},\;
s{\bf x} s^{-1}\in\R^{p,q}\right\}.
\]
The set $\boldsymbol{\Gamma}^+_{p,q}=\boldsymbol{\Gamma}_{p,q}\capC\kern -0.2em \ell^+_{p,q}$ is called {\it special
Lipschitz group} \cite{Che1}.
Let $N:\;C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p,q},\;N({\bf x})={\bf x}\widetilde{{\bf x}}$.
If ${\bf x}\in\R^{p,q}$, then $N({\bf x})={\bf x}(-{\bf x})=-{\bf x}^2=-Q({\bf x})$. Further, the
group $\boldsymbol{\Gamma}_{p,q}$ has a subgroup
\begin{equation}\label{e10}
\pin(p,q)=\left\{s\in\boldsymbol{\Gamma}_{p,q}\;|\;N(s)=\pm 1\right\}.
\end{equation}
Analogously, {\it a spinor group} $\spin(p,q)$ is defined by the set
\begin{equation}\label{e11}
\spin(p,q)=\left\{s\in\boldsymbol{\Gamma}^+_{p,q}\;|\;N(s)=\pm 1\right\}.
\end{equation}
It is obvious that
\[
\spin(p,q)=\pin(p,q)\capC\kern -0.2em \ell^+_{p,q}.
\]
The group $\spin(p,q)$ contains a subgroup
\begin{equation}\label{e12}
\spin_+(p,q)=\left\{s\in\spin(p,q)\;|\;N(s)=1\right\}.
\end{equation}
It is easy to see that the groups $O(p,q),\,SO(p,q)$ and $SO_+(p,q)$ are
isomorphic correspondngly to the following quotient groups
\begin{eqnarray}
&&O(p,q)\simeq\pin(p,q)/\dZ_2,\nonumber\\
&&SO(p,q)\simeq\spin(p,q)/\dZ_2,\nonumber\\
&&SO_+(p,q)\simeq\spin_+(p,q)/\dZ_2,\nonumber
\end{eqnarray}
\begin{sloppypar}\noindent
where a kernel $\dZ_2=\{1,-1\}$. Thus, the groups $\pin(p,q)$, $\spin(p,q)$
and $\spin_+(p,q)$ are the double coverings of the groups $O(p,q),\,SO(p,q)$
and $SO_+(p,q)$, respectively.\end{sloppypar}
Further, since $C\kern -0.2em \ell^+_{p,q}\simeqC\kern -0.2em \ell^+_{q,p}$, then
\[
\spin(p,q)\simeq\spin(q,p).
\]
In contrast with this, the groups $\pin(p,q)$ and $\pin(q,p)$ are
non-isomorphic. Denote $\spin(n)=\spin(n,0)\simeq\spin(0,n)$.
\begin{theorem}[{\rm\cite{Cor84}}]\label{t3}
The spinor groups
\[
\spin(2),\;\;\spin(3),\;\;\spin(4),\;\;\spin(5),\;\;\spin(6)
\]
are isomorphic to the unitary groups
\[
U(1),\;\;Sp(1)\sim SU(2),\;\;SU(2)\times SU(2),\;\;Sp(2),\;\;SU(4).
\]
\end{theorem}
In accordance with Theorem \ref{t1} and decompositions (\ref{e5'})
over the field $\F=\R$ the algebra $C\kern -0.2em \ell_{p,q}$ is isomorphic to a direct
sum of two mutually annihilating simple ideals $\frac{1}{2}(1\pm\omega)
C\kern -0.2em \ell_{p,q}$: $C\kern -0.2em \ell_{p,q}\simeq\frac{1}{2}(1+\omega)C\kern -0.2em \ell_{p,q}\oplus\frac{1}{2}
(1-\omega)C\kern -0.2em \ell_{p,q}$, where $\omega=\mbox{\bf e}_{12\ldots p+q},\,p-q\equiv 3,7
\pmod{8}$. At this point, each ideal is isomorpic to $C\kern -0.2em \ell_{p-1,q}$ or
$C\kern -0.2em \ell_{p,q-1}$. Therefore, for the Clifford-Lipschitz groups we have the
following isomorphisms
\begin{eqnarray}
\pin(p,q)&\simeq&\pin(p-1,q)\cup\pin(p-1,q)\nonumber\\
&\simeq&\pin(p,q-1)\cup\pin(p,q-1).
\end{eqnarray}
Or, since $C\kern -0.2em \ell_{p-1,q}\simeqC\kern -0.2em \ell^+_{p,q}\subsetC\kern -0.2em \ell_{p,q}$, then
according to (\ref{e11})
\[
\pin(p,q)\simeq\spin(p,q)\cup\spin(p,q)
\]
if $p-q\equiv 3,7\pmod{8}$.
Further, when $p-q\equiv 1,5\pmod{8}$ from Theorem \ref{t1} it follows
that $C\kern -0.2em \ell_{p,q}$ is isomorphic to a complex algebra $\C_{p+q-1}$. Therefore,
for the $\pin$ groups we obtain
\begin{eqnarray}
\pin(p,q)&\simeq&\pin(p-1,q)\cup\mbox{\bf e}_{12\ldots p+q}\pin(p-1,q)\nonumber\\
&\simeq&\pin(p,q-1)\cup\mbox{\bf e}_{12\ldots p+q}\pin(p,q-1)\label{e13}
\end{eqnarray}
if $p-q\equiv 1,5\pmod{8}$ and correspondingly
\begin{equation}\label{e14}
\pin(p,q)\simeq\spin(p,q)\cup\mbox{\bf e}_{12\ldots p+q}\spin(p,q).
\end{equation}
In case $p-q\equiv 3,7\pmod{8}$ we have isomorphisms which are analoguos
to (\ref{e13})-(\ref{e14}), since $\omegaC\kern -0.2em \ell_{p,q}\simC\kern -0.2em \ell_{p,q}$.
Generalizing we obtain the following
\begin{theorem}\label{t4}
Let $\pin(p,q)$ and $\spin(p,q)$ be the Clifford-Lipschitz groups of the
invertible elements of the algebras $C\kern -0.2em \ell_{p,q}$ with odd dimensionality,
$p-q\equiv 1,3,5,7\pmod{8}$. Then
\begin{eqnarray}
\pin(p,q)&\simeq&\pin(p-1,q)\cup\omega\pin(p-1,q)\nonumber\\
&\simeq&\pin(p,q-1)\cup\omega\pin(p,q-1)\nonumber
\end{eqnarray}
and
\[
\pin(p,q)\simeq\spin(p,q)\cup\omega\spin(p,q),
\]
where $\omega=\mbox{\bf e}_{12\ldots p+q}$ is a volume element of $C\kern -0.2em \ell_{p,q}$.
\end{theorem}
In case of low dimensionalities from Theorem \ref{t3} and Theorem
\ref{t4} it immediately follows
\begin{theorem}\label{t5}
For $p+q\leq 5$ and $p-q\equiv 3,5\pmod{8}$,
\begin{eqnarray}
\pin(3,0)&\simeq&SU(2)\cup iSU(2),\nonumber\\
\pin(0,3)&\simeq&SU(2)\cup eSU(2),\nonumber\\
\pin(5,0)&\simeq&Sp(2)\cup eSp(2),\nonumber\\
\pin(0,5)&\simeq&Sp(2)\cup iSp(2).\nonumber
\end{eqnarray}
\end{theorem}
\begin{proof}\begin{sloppypar}\noindent
Indeed, in accordance with Theorem \ref{t4} $\pin(3,0)\simeq\spin(3)
\cup\mbox{\bf e}_{123}\spin(3)$. Further, from Theorem \ref{t3} we have
$\spin(3)\simeq SU(2)$, and a square of the element $\omega=\mbox{\bf e}_{123}$ is
equal to $-1$, therefore $\omega\sim i$. Thus, $\pin(3,0)\simeq SU(2)\cup
iSU(2)$. For the group $\pin(0,3)$ a square of $\omega$ is equal to $+1$,
therefore $\pin(0,3)\simeq SU(2)\cup eSU(2)$, $e$ is a double unit.
As expected, $\pin(3,0)\not\simeq\pin(0,3)$. The isomorphisms for the
groups $\pin(5,0)$ and $\pin(0,5)$ are analogously proved.\end{sloppypar}
\end{proof}
Further, let $E$ be a vector space, then a homomorphism
\[
\rho:\;C\kern -0.2em \ell_{p,q}\longrightarrow\End E,
\]
which maps the unit element of the algebra $C\kern -0.2em \ell_{p,q}$ to $\Id_E$, is
called {\it a representation} of $C\kern -0.2em \ell_{p,q}$ in $E$ ($\End E$ is an
endomorphism algebra of the space $E$). The dimensionality of $E$ is called
a degree of the representation. The addition in $E$ together with the
mapping $C\kern -0.2em \ell_{p,q}\times E\rightarrow E,\;(a,x)\mapsto\rho(a)x,\,a\inC\kern -0.2em \ell_{p,q},
\,x\in E$, turns $E$ in $C\kern -0.2em \ell_{p,q}$-module, {\it a representation module}.
The representation $\rho$ is faithful if its kernel is zero, that is,
$\rho(a)x=0,\,\forall x \in E\Rightarrow a=0$. If the representation
$\rho$ has only two invariant subspaces $E$ and $\{0\}$, then $\rho$ is said
to be simple or irreducible. On the contrary case, $\rho$ is said to be
semi-simple, that is, it is a direct sum of simple modules, and in this case
$E$ is a direct sum of subspaces which are globally invariant under $\rho(a),
\;\forall a\inC\kern -0.2em \ell_{p,q}$. The representation $\rho$ of $C\kern -0.2em \ell_{p,q}$
induces a representation of the group $\pin(p,q)$ which we will denote
by the same symbol $\rho$, and also induces a representation of the group
$\spin(p,q)$ which we will denote by $\Delta_{p,q}$. In so doing, we have
the following \cite{Che1,Rash}
\begin{theorem}\label{t6}
If $p+q=2m$ and $p-q\equiv 0,2,4,6\pmod{8}$, then
\[
C\kern -0.2em \ell_{p,q}\simeq\End_{\F}(I_{p,q})\simeq{\bf\sf M}_{2^m}(\F)
\]
and
\[
C\kern -0.2em \ell_{p,q}\simeq\End_{\F}(I_{p,q})\simeq{\bf\sf M}_{2^m}(\F)\oplus{\bf\sf M}_{2^m}(\F)
\]
if $p+q=2m+1$ and $p-q\equiv 1,3,5,7\pmod{8}$, where $\F=\R,\C,\Om,\BH$,
$I_{p,q}$ is a minimal left ideal of $C\kern -0.2em \ell_{p,q}$, $\End_{\F}(I_{p,q})$ is
an algebra of linear transformations in $I_{p,q}$ over the field $\F$,
${\bf\sf M}_{2^m}(\F)$ is a matrix algebra.
\end{theorem}
Let us consider matrix representations of the fundamental automorphisms
of $C\kern -0.2em \ell_{p,q}$ over the field $\F$ when $p+q$ is even \cite{Sch49,Rash}.
We start with the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$. In
accordance with Theorem \ref{t6} in the matrix representation the
antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ corresponds to an
antiautomorphism of the matrix algebra ${\bf\sf M}_{2^m}(\F)$:
\[
{\sf A}\longrightarrow {\sf A}^T,
\]
in virtue of the well-known relation $({\sf A}{\sf B})^T={\sf B}^T{\sf A}^T$, where $T$
is a symbol
of transposition. On the other hand, in the matrix representation of the
elements $\mathcal{A}\inC\kern -0.2em \ell_{p,q}$ for the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ we have
\[
{\sf A}\longrightarrow\widetilde{{\sf A}}.
\]
The composition of the two antiautomorphisms ${\sf A}^T\rightarrow {\sf A}\rightarrow
\widetilde{{\sf A}}$ gives an automorphism ${\sf A}^T\rightarrow\widetilde{{\sf A}}$
which is an internal automorphism of the algebra $M_{2^m}(\F)$:
\begin{equation}\label{e15}
\widetilde{{\sf A}}={\sf E}{\sf A}^T{\sf E}^{-1},
\end{equation}
where ${\sf E}$ is a matrix, by means of which the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is expressed in the matrix representation
of the algebra $C\kern -0.2em \ell_{p,q}$.
Further, for the automorphism $\mathcal{A}\rightarrow\mathcal{A}^\star$, defined
by the formula (\ref{e6}), in the matrix representation we have
\begin{equation}\label{e16}
{\sf A}^\star={\sf W}{\sf A}{\sf W}^{-1},
\end{equation}
where ${\sf A}$ is a matrix representing an arbitrary element of $C\kern -0.2em \ell_{p,q}$,
${\sf W}$ is a matrix of the volume element $\omega=\mbox{\bf e}_{12\ldots n}$.
Over the field $\F=\C$ we can multiply ${\sf W}$ by the factor
$\varepsilon=\pm i^{\frac{(p+q)(p+q-1)}{2}}$, then $(\varepsilon {\sf W})^2=1$.
Therefore, the relation (\ref{e16}) may be rewritten in the form
\begin{equation}\label{e17}
{\sf A}^\star={\sf W}^{\prime}{\sf A}{\sf W}^{\prime},
\end{equation}
where ${\sf W}^{\prime}=\varepsilon {\sf W}$.
Finally, for the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$,
which is the composition of the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ with the automorphism $\mathcal{A}\rightarrow\mathcal{A}^\star$, using
(\ref{e15}) and (\ref{e17}) we obtain a following expression
\[
\widetilde{{\sf A}^\star}={\sf W}^{\prime}{\sf E}{\sf A}^T{\sf E}^{-1}{\sf W}^{\prime},
\]
or
\begin{equation}\label{e18}
\widetilde{{\sf A}^\star}=({\sf E}{{\sf W}^{\prime}}^T){\sf A}^T({\sf E}{{\sf W}^{\prime}}^T)^{-1}.
\end{equation}
Denoting ${\sf E}{{\sf W}^{\prime}}^T={\sf C}$, where ${\sf C}$ is a matrix representation the
antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$, and substituting
into (\ref{e18}) we obtain definitely
\begin{equation}\label{e19}
\widetilde{{\sf A}^\star}={\sf C}{\sf A}^T{\sf C}^{-1}.
\end{equation}
{\bf Example}. Let consider matrix representations of the fundamental
automorphisms of the Dirac algebra $C\kern -0.2em \ell_{4,1}$. In virtue of Theorem
\ref{t1} there is an isomorphism $C\kern -0.2em \ell_{4,1}\simeq\C_4$, and therefore
$C\kern -0.2em \ell_{4,1}\simeq\C_4\simeq{\bf\sf M}_4(\C)$. In the capacity of the matrix
representations of the units $\mbox{\bf e}_i\in\C_4$ $(i=1,2,3,4)$ we take the
well-known Dirac $\gamma$-matrices (so-called canonical representation):
\[
\renewcommand{\arraystretch}{1}
\gamma_1=\begin{pmatrix}
0 & 0 & 0 & -i\\
0 & 0 & -i& 0\\
0 & i & 0 & 0\\
i & 0 & 0 & 0
\end{pmatrix},\quad\gamma_2=\begin{pmatrix}
0 & 0 & 0 & -1\\
0 & 0 & 1 & 0\\
0 & 1 & 0 & 0\\
-1& 0 & 0 & 0
\end{pmatrix},
\]
\begin{equation}\label{e20}\renewcommand{\arraystretch}{1}
\gamma_3=\begin{pmatrix}
0 & 0 & -i & 0\\
0 & 0 & 0 & i\\
i & 0 & 0 & 0\\
0 & -i& 0 & 0
\end{pmatrix},\quad\gamma_4=\begin{pmatrix}
1 & 0 & 0 & 0\\
0 & 1 & 0 & 0\\
0 & 0 &-1 & 0\\
0 & 0 & 0 &-1
\end{pmatrix}.
\end{equation}
$\gamma$-matrices form the only one basis from the set of isomorphic
matrix basises of $C\kern -0.2em \ell_{4,1}\simeq\C_4$. In the basis (\ref{e20}) the element
$\omega=\mbox{\bf e}_1\mbox{\bf e}_2\mbox{\bf e}_3\mbox{\bf e}_4$ is represented by a matrix ${\sf W}=\gamma_5=\gamma_1
\gamma_2\gamma_3\gamma_4$. Since $\gamma^2_5=1$, then $\varepsilon=1\;
({\sf W}^{\prime}={\sf W})$ and the matrix
\begin{equation}\label{e21}\renewcommand{\arraystretch}{1}
{\sf W}^T={\sf W}=\begin{pmatrix}
0 & 0 &-1 & 0\\
0 & 0 & 0 &-1\\
-1& 0 & 0 & 0\\
0 &-1 & 0 & 0
\end{pmatrix}
\end{equation}
in accordance with (\ref{e17}) is a matrix of the automorphism $\mathcal{A}
\rightarrow\mathcal{A}^\star$. Further, in the matrix representation the
antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is defined by the
transformation $\widetilde{{\sf A}}={\sf E}{\sf A}^T{\sf E}^{-1}$. For the $\gamma$-matrices
we have
$\gamma^T_1=-\gamma_1,\,\gamma^T_2=\gamma_2,\,\gamma^T_3=-\gamma_3,
\,\gamma^T_4=\gamma_4$. Further,
\begin{eqnarray}
&&\gamma_1=-{\sf E}\gamma_1{\sf E}^{-1},\quad\gamma_2={\sf E}\gamma_2{\sf E}^{-1},\nonumber\\
&&\gamma_3=-{\sf E}\gamma_3{\sf E}^{-1},\quad\gamma_4={\sf E}\gamma_4{\sf E}^{-1}.\nonumber
\end{eqnarray}
It is easy to verify that a matrix ${\sf E}=\gamma_1\gamma_3$ satisfies the
latter relations and, therefore, the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ in the basis (\ref{e20}) is defined by the matrix
\begin{equation}\label{e22}\renewcommand{\arraystretch}{1}
{\sf E}=\gamma_1\gamma_3=\begin{pmatrix}
0 & -1 & 0 & 0\\
1 & 0 & 0 & 0\\
0 & 0 & 0 & -1\\
0 & 0 & 1 & 0
\end{pmatrix}.
\end{equation}
Finally, for the matrix ${\sf C}={\sf E}{\sf W}^T$ of the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}^\star}$ from (\ref{e21}) and (\ref{e22}) in accordance with
(\ref{e19}) we obtain
\begin{equation}\label{e23}\renewcommand{\arraystretch}{1}
{\sf C}={\sf E}{\sf W}^T=\begin{pmatrix}
0 & 0 & 0 & 1\\
0 & 0 &-1 & 0\\
0 & 1 & 0 & 0\\
-1& 0 & 0 & 0
\end{pmatrix}.
\end{equation}
\section{Fundamental Automorphisms of\protect\newline
Odd-dimensional Clifford Algebras}
Let us consider the fundamental automorphisms of the algebras $C\kern -0.2em \ell_{p,q}$
and $\C_{p+q}$, where $p-q\equiv 1,3,5,7\pmod{8}$, $p+q=2m+1$. In accordance
with (\ref{e5'}) and Theorem \ref{t2} the algebras $C\kern -0.2em \ell_{p,q}$ and
$\C_{p+q}$ are isomorphic to direct sums of two algebras with the even
dimensionality if correspondingly $p-q\equiv 3,7\pmod{8}$ and $p+q\equiv 1,3,5,7
\pmod{8}$. Therefore, matrix representations of $C\kern -0.2em \ell_{p,q},\,\C_{p+q}$
are isomorphic to the direct sums of complete matrix algebras
${\bf\sf M}_{2^m}(\F)\oplus{\bf\sf M}_{2^m}(\F)$, here $\F=\R,\,\F=\C$. On the other hand,
there exists an homomorphic mapping of $C\kern -0.2em \ell_{p,q}$ and $\C_{p+q}$ into
one full matrix algebra ${\bf\sf M}_{2^m}(\F)$ with preservation of addition,
multiplication and multiplication by the number. Besides, in the case
of $\F=\R$ and $p-q\equiv 1,5\pmod{8}$ the algebra $C\kern -0.2em \ell_{p,q}$ is isomorphic
to the full matrix algebra ${\bf\sf M}_{2^m}(\C)$ (Theorem \ref{t1}), therefore,
representations of the fundamental automorphisms of this algebra may be
realized by means of ${\bf\sf M}_{2^m}(\C)$.
\begin{theorem}\label{t7}
If $p+q=2m+1$, then the following homomorphisms take place\\
1) $\F=\R$
\[
\epsilon:\;\;C\kern -0.2em \ell_{p,q}\longrightarrow{\bf\sf M}_{2^m}(\R)\quad\text{if}\;\;
p-q\equiv 3,7\pmod{8}.
\]
2) $\F=\C$
\[
\epsilon^{\prime}:\;\;\C_{p+q}\longrightarrow{\bf\sf M}_{2^m}(\C)\quad\text{if}\;\;
p+q\equiv 1,3,5,7\pmod{8}.
\]
\end{theorem}
\begin{proof} We start the proof with a more general case of $\F=\C$.
According to (\ref{e4}) the volume element $\omega$ belongs to a center
of $\C_{n}$ ($n=p+q$), therefore, $\omega$ commutes with all basis elements
of this algebra and $(\varepsilon\omega)^2=1$. Further, recalling that a
vector complex space $\C^n$ is associated with the algebra $\C_n$, we see
that basis vectors $\{e_{1},e_{2},\ldots,e_{n}\}$ generate a
subspace $C_{n}\subset C_{n+1}$. Thus, the algebra $\C_{n}$ in $C_{n}$ is
a subalgebra of $\C_{n+1}$ and consists of the elements which does not contain
the element $\mbox{\bf e}_{n+1}$.
A decomposition of the each element $\mathcal{A}\in\C_{n+1}$ may be written in the
form
\[
\mathcal{A}=\mathcal{A}^{1}+\mathcal{A}^{0},
\]
where $\mathcal{A}^{0}$ is a set of all elements which contain $\mbox{\bf e}_{n+1}$, and
$\mathcal{A}^{1}$ is a set of all elements which does not contain $\mbox{\bf e}_{n+1}$, therefore
$\mathcal{A}^{1}\in\C_{n}$. If multiply $\mathcal{A}^{0}$ by $\varepsilon\omega$, then the
elements $\mbox{\bf e}_{n+1}$ are mutually annihilate, therefore $\varepsilon\omega
\mathcal{A}^{0}\in\C_{n}$. Denoting $\mathcal{A}^{2}=\varepsilon\omega\mathcal{A}^{0}$
and taking into account $(\varepsilon\omega)^2=1$ we obtain
\[
\mathcal{A}=\mathcal{A}^{1}+\varepsilon\omega\mathcal{A}^{2},
\]
where $\mathcal{A}^{1},\,\mathcal{A}^{2}\in\C_{n}$. Consider now an homomorphism
$\epsilon:\;\C_{n+1}\rightarrow\C_{n}$, an action of which is defined by
the following law
\begin{equation}\label{e24}
\epsilon:\;\mathcal{A}^{1}+\varepsilon\omega\mathcal{A}^{2}\longrightarrow\mathcal{A}^{1}+\mathcal{A}^{2}.
\end{equation}
Obviously, at this point the all operations (addition, multiplication, and
multiplication by the number) are preserved. Indeed, let
\[
\mathcal{A}=\mathcal{A}^{1}+\varepsilon\omega\mathcal{A}^{2},\quad
\mathcal{B}=\mathcal{B}^{1}+\varepsilon\omega\mathcal{B}^{2},
\]
then in virtue of $(\varepsilon\omega)^{2}=1$ and
commutativity of $\omega$ with all elements, we have for multiplication
\begin{multline}
\mathcal{A}\mathcal{B}=(\mathcal{A}^{1}\mathcal{B}^{1}+\mathcal{A}^{2}\mathcal{B}^{2})+\varepsilon\omega(\mathcal{A}^{1}\mathcal{B}^{2}+
\mathcal{A}^{2}\mathcal{B}^{1})\stackrel{\epsilon}{\longrightarrow}\\
(\mathcal{A}^{1}\mathcal{B}^{1}+
\mathcal{A}^{2}\mathcal{B}^{2})+(\mathcal{A}^{1}\mathcal{B}^{2}+\mathcal{A}^{2}\mathcal{B}^{1})=
(\mathcal{A}^{1}+\mathcal{A}^{2})(\mathcal{B}^{1}+\mathcal{B}^{2}).\nonumber
\end{multline}
that is, the image of product equals to the product of factor images in
the same order.
In the particular case of $\mathcal{A}=\varepsilon\omega$ we have $\mathcal{A}^{1}=0$ and
$\mathcal{A}^{2}=1$, therefore
\[
\varepsilon\omega\longrightarrow 1.
\]
Thus, a kernel of the homomorphism $\epsilon$ consists of all elements
of the form $\mathcal{A}^{1}-\varepsilon\omega\mathcal{A}^{1}$, which under action of
$\epsilon$ are mapped into
zero. It is clear that $\Ker\,\epsilon=\{\mathcal{A}^{1}-\varepsilon\omega\mathcal{A}^{1}\}$
is a subalgebra of $\C_{n+1}$. Moreover, the kernel of
$\epsilon$ is a bilateral ideal of $\C_{n+1}$. Therefore, the algebra
$\C_{n}$, which we obtain in the result of the mapping $\epsilon:\;\C_{n+1}
\longrightarrow\C_{n}$, is {\it a quotient algebra}
\[
{}^\epsilon\C_{n}\simeq\C_{n+1}/\Ker\epsilon.
\]
Further, since the algebra $\C_{n}$ ($n=2m$) is isomorphic to the
full matrix
algebra ${\bf\sf M}_{2^{m}}(\C)$, then in virtue of $\epsilon:\;\C_{n+1}
\longrightarrow\C_{n}\subset\C_{n+1}$ we obtain an homomorphic mapping
of $\C_{n+1}$ onto
the matrix algebra ${\bf\sf M}_{2^{m}}(\C)$.
The homomorphism $\epsilon:\;C\kern -0.2em \ell_{p,q}\rightarrow{\bf\sf M}_{2^m}(\R)$ is
analogously proved. In this case a quotient algebra has a form
\[
{}^\epsilonC\kern -0.2em \ell_{p,q}\simeqC\kern -0.2em \ell_{p+1,q}/\Ker\epsilon
\]
or
\[
{}^\epsilonC\kern -0.2em \ell_{p,q}\simeqC\kern -0.2em \ell_{p,q+1}/\Ker\epsilon,
\]
where $\Ker\epsilon=\{\mathcal{A}^1-\omega\mathcal{A}^1\}$, since in accordance with
(\ref{e3}) at $p-q\equiv 3,7\pmod{8}$ we have $\omega^2=1$, therefore
$\varepsilon=1$.
\end{proof}
Let us consider the form which the fundamental automorphisms of $\C_{n+1}$
take after the homomorphic mapping
$\epsilon:\;\C_{n+1}\rightarrow\C_{n}\subset\C_{n+1}$. First of all,
for the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ it is necessary that
elements $\mathcal{A},\,\mathcal{B},\,\ldots\,\in\C_{n+1}$, which are mapped into one
and the same element
$\mathcal{D}\in\C_{n}$ (a kernel of the homomorphism $\epsilon$ if $\mathcal{D}=0$) after the
transformation $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ are must converted to the
elements $\widetilde{\mathcal{A}},\,\widetilde{\mathcal{B}},\,\ldots\,\in\C_{n+1}$,
which are also
mapped into one and the same element $\widetilde{\mathcal{D}}\in\C_{n}$. Otherwise, the
transformation $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is not transferred from
$\C_{n+1}$ into $\C_n$ as an unambiguous transformation. In particular,
it is necessary in order that $\widetilde{\varepsilon\omega}=\varepsilon
\omega$, since $1$ and element $\varepsilon\omega$ under action of the
homomorphism $\epsilon$ are equally mapped into the unit, then $\widetilde{1}$
and $\widetilde{\varepsilon\omega}$ are also must be mapped into one
and the same element in $\C_n$, but
$\widetilde{1}\rightarrow 1$, and $\widetilde{\varepsilon\omega}
\rightarrow\pm 1$ (in virtue of the formula (\ref{e8})).
Therefore we must assume
\begin{equation}\label{e25}
\widetilde{\varepsilon\omega}=\varepsilon\omega.
\end{equation}
The condition (\ref{e25}) is sufficient for the transfer of the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ from $\C_{n+1}$ into $\C_{n}$. Indeed,
in this case we have
\[
\mathcal{A}^{1}-\mathcal{A}^{1}\varepsilon\omega\;\longrightarrow\;\widetilde{\mathcal{A}^{1}}-
\widetilde{\varepsilon\omega}\widetilde{\mathcal{A}^{1}}=\widetilde{\mathcal{A}^{1}}-
\varepsilon\omega\widetilde{\mathcal{A}^{1}}.
\]
Therefore, the elements of the form $\mathcal{A}^{1}-\mathcal{A}^{1}\varepsilon\omega$
(composing, as known, the kernel of $\epsilon$) under action of the
transformation
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ are converted to the elements of the same
form.
The analogous conditions take place for other fundamental automorphisms.
However, for the automorphism $\mathcal{A}\rightarrow\mathcal{A}^{\star}$ a condition
$(\varepsilon\omega)^{\star}=\varepsilon\omega$ is not valid, since
$\omega$ is odd and in accordance with (\ref{e5}) we have
\begin{equation}\label{e26}
\omega^{\star}=-\omega.
\end{equation}
Thus, the automorphism $\mathcal{A}\rightarrow\mathcal{A}^{\star}$ is not transferred
from $\C_{n+1}$ into $\C_{n}$.
Let us return to the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ and
let consider in more details necessary conditions for the transfer of this
transformation from $\C_{n+1}$ to $\C_n$. First of all, the factor
$\varepsilon$ depending upon the condition $(\varepsilon\omega)^2=1$ and
the square of the element $\omega=\mbox{\bf e}_{12\ldots n+1}$ takes the following
values
\begin{equation}\label{e27}
\varepsilon=
\begin{cases}
1 & \text{if $p-q\equiv 3,7\pmod{8}$},\\
i & \text{if $p-q\equiv 1,5\pmod{8}$}.
\end{cases}
\end{equation}
Further, in accordance with (\ref{e8}) for the transformation $\omega
\rightarrow\widetilde{\omega}$ we obtain
\begin{equation}\label{e28}
\widetilde{\omega}=
\begin{cases}
\phantom{-}\omega & \text{if $p-q\equiv 3,7\pmod{8}$},\\
-\omega & \text{if $p-q\equiv 1,5\pmod{8}$}.
\end{cases}
\end{equation}
Therefore, for the algebras over the field $\F=\R$ the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is transfered at the mappings $C\kern -0.2em \ell_{p,q}
\rightarrowC\kern -0.2em \ell_{p-1,q},\,C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p,q-1}$, where
$p-q\equiv 3,7\pmod{8}$. Over the field $\F=\C$ the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is transfered in any case, since the
algebras $\C_{n+1}$ with signatures $p-q\equiv 3,7\pmod{8}$ and
$p-q\equiv 1,5\pmod{8}$ are isomorphic. In so doing, the condition (\ref{e25})
takes a form
\begin{eqnarray}
\widetilde{\omega}&=&\omega\quad\phantom{i}\text{if $p-q\equiv 3,7\pmod{8}$},
\nonumber\\
\widetilde{i\omega}&=&i\omega\quad\text{if $p-q\equiv 1,5\pmod{8}$}.\nonumber
\end{eqnarray}
Besides, each of these equalities satisfies the condition $(\varepsilon
\omega)^2=1$.
Let us consider now the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}^{\star}}$.
It is obvious that for the transfer of
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}^{\star}}$ from
$\C_{n+1}$ to $\C_{n}$ it is necessary that
\begin{equation}\label{e29}
\widetilde{(\varepsilon\omega)^{\star}}=\varepsilon\omega.
\end{equation}
It is easy to see that the mapping $\C_{p+q}\rightarrow\C_{p+q-1}$, where
$p+q\equiv 1,5\pmod{8}$, in virtue of (\ref{e26}) and the second equality
of (\ref{e28}), satisfies the condition (\ref{e29}), since in this case
\[
\widetilde{(\varepsilon\omega)^{\star}}=\varepsilon\widetilde{\omega^\star}=
-\varepsilon\omega^{\star}=\varepsilon\omega.
\]
Hence it immediately follows that over the field $\F=\R$ the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$ at the mappings $C\kern -0.2em \ell_{p,q}\rightarrow
C\kern -0.2em \ell_{p-1,q},\,C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p,q-1}$ ($p-q\equiv 3,7\pmod{8}$)
is not transferred.
Summarizing obtained above results we come to the following
\begin{theorem}\label{t8}
1) If $\F=\C$ and $\C_{p+q}\simeq\C_{p+q-1}\oplus\C_{p+q-1}$, where
$p+q\equiv 1,3,5,7\pmod{8}$, then the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ at the homomorphic mapping $\epsilon:\,\C_{p+q}\rightarrow
\C_{p+q-1}$ is transferred into a quotient algebra ${}^\epsilon\C_{p+q-1}$
in any case, the automorphism $\mathcal{A}\rightarrow\mathcal{A}^\star$ is not transferred,
and the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$ is transferred
in the case of $p+q\equiv 1,5\pmod{8}$.\\
2) If $\F=\R$ and $C\kern -0.2em \ell_{p,q}\simeqC\kern -0.2em \ell_{p-1,q}\oplusC\kern -0.2em \ell_{p-1,q},\,C\kern -0.2em \ell_{p,q}
\simeqC\kern -0.2em \ell_{p,q-1}\oplusC\kern -0.2em \ell_{p,q-1}$, where $p-q\equiv 3,7\pmod{8}$, then
at the homomorphic mappings $\epsilon:\,C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p-1,q}$
and $\epsilon:\,C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p,q-1}$ the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is transferred correspondingly into quotient
algebras ${}^\epsilonC\kern -0.2em \ell_{p-1,q}$ and ${}^\epsilonC\kern -0.2em \ell_{p,q-1}$ in any case,
and the automorphism $\mathcal{A}\rightarrow\mathcal{A}^\star$ and antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$ are not transferred.
\end{theorem}
\section{Automorphism Groups of $C\kern -0.2em \ell_{p,q}$, $\C_{p+q}$ and Discrete
Transformations of $O(p,q)$, $O(p+q,\C)$}
As noted above, there exists a close relation between D\c{a}browski groups
$\pin^{a,b,c}(p,q)$ and discrete tansformations of the orthogonal group
$O(p,q)$ (in particular, Lorentz group $O(1,3)$) \cite{DWGK,Ch97,Ch94,AlCh94,
AlCh96}. On the other hand, discrete transformations of the group $O(p,q)$
which acting in the space $\R^{p,q}$ associated with the algebra $C\kern -0.2em \ell_{p,q}$,
may be realized via the fundamental automorphisms of $C\kern -0.2em \ell_{p,q}$. In essence,
the group $\pin(p,q)$ is an intrinsic notion of $C\kern -0.2em \ell_{p,q}$, since
in accordance with (\ref{e10}) $\pin(p,q)\subsetC\kern -0.2em \ell_{p,q}$. Let us show
that the D\c{a}browski group $\pin^{a,b,c}(p,q)$ is also completely defined
in the framework of the algebra $C\kern -0.2em \ell_{p,q}$, that is, there is an equivalence
between $\pin^{a,b,c}(p,q)$ and the group $\pin(p,q)\subsetC\kern -0.2em \ell_{p,q}$
complemented by the transformations $\mathcal{A}\rightarrow\mathcal{A}^\star,\,\mathcal{A}\rightarrow
\widetilde{\mathcal{A}},\,\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$ (in connection with
this it should be noted that the Gauss-Klein group $\dZ_2\otimes\dZ_2$
is a finite group
corresponded to the algebra $C\kern -0.2em \ell_{1,0}=\Om$ \cite{Sal81a,Sal84}).
\begin{prop}\label{p1}
Let $C\kern -0.2em \ell_{p,q}$ ($p+q=2m$) be a Clifford algebra over the field $\F=\R$ and
let $\pin(p,q)$ be a double covering of the orthogonal group $O(p,q)=O_0(p,q)
\odot\{1,P,T,PT\}\simeq O_0(p,q)\odot(\dZ_2\otimes\dZ_2)$ of transformations
of the space $\R^{p,q}$, where $\{1,P,T,PT\}\simeq\dZ_2\otimes\dZ_2$ is a
group of discrete transformations of $\R^{p,q}$, $\dZ_2\otimes\dZ_2$ is the
Gauss-Klein group. Then there is an isomorphism between the group
$\{1,P,T,PT\}$ and an automorphism group $\{\Id,\star,\widetilde{\phantom{cc}},
\widetilde{\star}\}$ of the algebra $C\kern -0.2em \ell_{p,q}$. In this case, parity
reversal $P$, time reversal $T$ and combination $PT$ are correspond
respectively to the fundamental automorphisms $\mathcal{A}\rightarrow\mathcal{A}^\star,\,
\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ and $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$.
\end{prop}
\begin{proof} As known, the transformations $1,P,T,PT$ at the conditions
$P^2=T^2=(PT)^2=1,\;PT=TP$ form an abelian group with the following
multiplication table
\begin{center}{\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|c||c|c|c|c|}\hline
& $1$ & $P$ & $T$ & $PT$\\ \hline\hline
$1$ & $1$ & $P$ & $T$ & $PT$\\ \hline
$P$ & $P$ & $1$ & $PT$& $T$\\ \hline
$T$ & $T$ & $PT$& $1$ & $P$\\ \hline
$PT$& $PT$& $T$ & $P$ & $1$\\ \hline
\end{tabular}
}
\end{center}
Analogously, for the automorphism group $\{\Id,\star,\widetilde{\phantom{cc}},
\widetilde{\star}\}$ in virtue of the commutativity $\widetilde{(\mathcal{A}^\star)}=
(\widetilde{\mathcal{A}})^\star$ and the conditions $(\star)^2=(\widetilde{\phantom{cc}
})^2=\Id$ a following multiplication table takes place
\begin{center}{\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|c||c|c|c|c|}\hline
& $\Id$ & $\star$ & $\widetilde{\phantom{cc}}$ & $\widetilde{\star}$\\ \hline\hline
$\Id$ & $\Id$ & $\star$ & $\widetilde{\phantom{cc}}$ & $\widetilde{\star}$\\ \hline
$\star$ & $\star$ & $\Id$ & $\widetilde{\star}$ & $\widetilde{\phantom{cc}}$\\ \hline
$\widetilde{\phantom{cc}}$ & $\widetilde{\phantom{cc}}$ &$\widetilde{\star}$
& $\Id$ & $\star$ \\ \hline
$\widetilde{\star}$ & $\widetilde{\star}$ & $\widetilde{\phantom{cc}}$ &
$\star$ & $\Id$\\ \hline
\end{tabular}
}
\end{center}
The identity of the multiplication tables proves the isomorphism of the
groups $\{1,P,T,PT\}$ and $\{\Id,\star,\widetilde{\phantom{cc}},\widetilde{
\star}\}$.
\end{proof}
Further, in the case of anticommutativity $PT=-TP$ and $P^2=T^2=(PT)^2=\pm 1$
an isomorphism between the group $\{1,P,T,PT\}$ and an automorphism group
$\{{\sf I},{\sf W},{\sf E},{\sf C}\}$, where ${\sf W},{\sf E}$ and ${\sf C}$ in accordance with (\ref{e17}),
(\ref{e15}) and (\ref{e19}) are the matrix representations of the
automorphisms $\mathcal{A}\rightarrow\mathcal{A}^\star,\,\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$
and $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$, is analogously proved.\\[0.4cm]
{\bf Example}. According to (\ref{e20}), (\ref{e21}), (\ref{e22}) and
(\ref{e23}) the matrix representation of the fundamental automorphisms
of the Dirac algebra $\C_4$ is defined by the following expressions:
${\sf W}=\gamma_1\gamma_2\gamma_3\gamma_4,\,{\sf E}=\gamma_1\gamma_3,\,
{\sf C}=\gamma_2\gamma_4$.
The multiplication table of the group $\{{\sf I},{\sf W},{\sf E},{\sf C}\}\sim
\{I,\gamma_1\gamma_2
\gamma_3\gamma_4,\gamma_1\gamma_3,\gamma_2\gamma_4\}$ has a form
\begin{multline}\label{e30}{\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|c||c|c|c|c|}\hline
& $I$ & $\gamma_1\gamma_2\gamma_3\gamma_4$ & $\gamma_1\gamma_3$ &
$\gamma_2\gamma_4$ \\ \hline\hline
$I$ & $I$ & $\gamma_1\gamma_2\gamma_3\gamma_4$ & $\gamma_1\gamma_3$ &
$\gamma_2\gamma_4$ \\ \hline
$\gamma_1\gamma_2\gamma_3\gamma_4$ & $\gamma_1\gamma_2\gamma_3\gamma_4$ &
$I$ & $\gamma_2\gamma_4$ & $\gamma_1\gamma_3$\\ \hline
$\gamma_1\gamma_3$ & $\gamma_1\gamma_3$ & $\gamma_2\gamma_4$ & $-I$ &
$-\gamma_1\gamma_2\gamma_3\gamma_4$\\ \hline
$\gamma_2\gamma_4$ & $\gamma_2\gamma_4$ & $\gamma_1\gamma_3$ &
$-\gamma_1\gamma_2\gamma_3\gamma_4$ & $-I$ \\ \hline
\end{tabular}
}
\;\;\sim\;\;\\{\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|c||c|c|c|c|}\hline
& ${\sf I}$ & ${\sf W}$ & ${\sf E}$ & ${\sf C}$\\ \hline\hline
${\sf I}$ & ${\sf I}$ & ${\sf W}$ & ${\sf E}$ & ${\sf C}$\\ \hline
${\sf W}$ & ${\sf W}$ & ${\sf I}$ & ${\sf C}$ & ${\sf E}$\\ \hline
${\sf E}$ & ${\sf E}$ & ${\sf C}$ & $-{\sf I}$& $-{\sf W}$\\ \hline
${\sf C}$ & ${\sf C}$ & ${\sf E}$ & $-{\sf W}$& $-{\sf I}$\\ \hline
\end{tabular}.
}
\end{multline}
However, in this representation we cannot directly to identify
${\sf W}=\gamma_1\gamma_2\gamma_3\gamma_4$ with the parity reversal $P$, since
in this case the Dirac equation $(i\gamma_4\frac{\partial}{\partial x_4}-
i\boldsymbol{\gamma}\frac{\partial}{\partial{\bf x}}-m)\psi(x_4,{\bf x})=0$ to be not
invariant with respect to $P$. On the other hand, for the canonical basis
(\ref{e20}) there exists a standard representation $P=\gamma_4,\,T=\gamma_1
\gamma_3$ \cite{BLP89}. The multiplication table of a group $\{1,P,T,PT\}\sim
\{I,\gamma_4,\gamma_2\gamma_3,\gamma_4\gamma_1\gamma_3\}$ has a form
\begin{equation}\label{e31}{\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|c||c|c|c|c|}\hline
& $I$ & $\gamma_4$ & $\gamma_1\gamma_3$ & $\gamma_4\gamma_1\gamma_3$
\\ \hline\hline
$1$ & $I$& $\gamma_4$ & $\gamma_1\gamma_3$ & $\gamma_4\gamma_1\gamma_3$
\\ \hline
$\gamma_4$ & $\gamma_4$ & $I$ & $\gamma_4\gamma_1\gamma_3$ & $\gamma_1\gamma_3$
\\ \hline
$\gamma_1\gamma_3$ & $\gamma_1\gamma_3$ & $\gamma_4\gamma_1\gamma_3$ &
$-I$ & $-\gamma_4$ \\ \hline
$\gamma_4\gamma_1\gamma_3$ & $\gamma_4\gamma_1\gamma_3$ & $\gamma_1\gamma_3$
& $-\gamma_4$ & $-I$ \\ \hline
\end{tabular}
}
\;\;\sim\;\;{\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|c||c|c|c|c|}\hline
& $1$ & $P$ & $T$ & $PT$\\ \hline\hline
$1$ & $1$ & $P$ & $T$ & $PT$\\ \hline
$P$ & $P$ & $1$ & $PT$& $T$\\ \hline
$T$ & $T$ & $PT$& $-1$& $-P$\\ \hline
$PT$& $PT$& $T$ & $-P$& $-1$\\ \hline
\end{tabular}.
}
\end{equation}
It is easy to see that the tables (\ref{e30}) and (\ref{e31}) are equivalent,
therefore we have an isomorphism $\{{\sf I},{\sf W},{\sf E},{\sf C}\}\simeq\{1,P,T,PT\}$. Besides,
each of these groups is isomorphic to the group $\dZ_4$.
\begin{theorem}\label{t9}
Let ${\bf\sf A}=\{{\sf I},\,{\sf W},\,{\sf E},\,{\sf C}\}$ be the automorphism group of the algebras
$C\kern -0.2em \ell_{p,q},\;
\C_{p+q}$ $(p+q=2m)$, where
${\sf W}=\mathcal{E}_1\mathcal{E}_2\cdots\mathcal{E}_m\mathcal{E}_{m+1}\mathcal{E}_{m+2}\cdots\mathcal{E}_{p+q}$,
and ${\sf E}=\mathcal{E}_1\mathcal{E}_2\cdots\mathcal{E}_m$, ${\sf C}=\mathcal{E}_{m+1}\mathcal{E}_{m+2}\cdots\mathcal{E}_{p+q}$ if
$m\equiv 1\pmod{2}$, and ${\sf E}=\mathcal{E}_{m+1}\mathcal{E}_{m+2}\cdots\mathcal{E}_{p+q}$,
${\sf C}=\mathcal{E}_1\mathcal{E}_2\cdots
\mathcal{E}_m$ if $m\equiv 0\pmod{2}$. Let ${\bf\sf A}_-$ and ${\bf\sf A}_+$ be the automorphism
groups, in which the all elements respectively commute
$(m\equiv 0\pmod{2})$ and anticommute $(m\equiv 1\pmod{2})$.
Then there are the following isomorphisms between finite groups and
automorphism groups with different signatures
$(a,\,b,\,c)$, where $a,b,c\in\{-,+\}$:\\
1) $\F=\R$. $\bA_-\simeq\dZ_2\otimes\dZ_2$ for the signature $(+,\,+,\,+)$
if $p-q\equiv 0,4\pmod{8}$. $\bA_-\simeq\dZ_4$ for $(+,\,-,\,-)$ if
$p-q\equiv 0,4\pmod{8}$ and for $(-,\,+,\,-),\;(-,\,-,\,+)$ if
$p-q\equiv 2,6\pmod{8}$. $\bA_+\simeq Q_4/\dZ_2$ for $(-,\,-,\,-)$ if
$p-q\equiv 2,6\pmod{8}$. $\bA_+\simeq D_4/\dZ_2$ for $(-,\,+,\,+)$ if
$p-q\equiv 2,6\pmod{8}$ and for $(+,\,-,\,+),\,(+,\,+,\,-)$ if
$p-q\equiv 0,4\pmod{8}$.\\
2) Over the field $\F=\C$ there are only two non-isomorphic groups:
$\bA_-\simeq\dZ_2\otimes\dZ_2$ for the signature $(+,\,+,\,+)$ if
$p-q\equiv 0,4\pmod{8}$ and
$\bA_+\simeq Q_4/\dZ_2$ for
$(-,\,-,\,-)$ if
$p-q\equiv 2,6\pmod{8}$.
\end{theorem}
\begin{proof} First of all, since $\omega^2=+1$ if
$p-q\equiv 0,4\pmod{8}$ and $\omega^2=-1$ if $p-q\equiv 2,6\pmod{8}$,
then in the case of $\F=\R$ for the matrix of the automorphism
$\mathcal{A}\rightarrow\mathcal{A}^\star$ we have
\[
{\sf W}=\begin{cases}
+{\sf I}, & \text{if $p-q\equiv 0,4\pmod{8}$};\\
-{\sf I}, & \text{if $p-q\equiv 2,6\pmod{8}$}.
\end{cases}
\]
Over the field $\F=\C$ we can always to suppose ${\sf W}^2=1$. Further, let us
find now the matrix ${\sf E}$ of the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ at any $n=2m$, and elucidate the conditions at which the
matrix ${\sf E}$ commutes with ${\sf W}$, and also define a square of the matrix
${\sf E}$. Follows to \cite{Rash} let introduce along with the algebra $\C_{p+q}$
an auxiliary algebra $\C_m$ with basis elements
\[
1,\;\varepsilon_\alpha,\;\varepsilon_{\alpha_1\alpha_2}\;(\alpha_1<\alpha_2),\;
\varepsilon_{\alpha_1\alpha_2\alpha_3}\;(\alpha_1<\alpha_2<\alpha_3),\;
\ldots\;\varepsilon_{12\ldots m}.
\]
In so doing, linear operators $\hat{\mathcal{E}}_i$ acting in the space
$\C^m$ associated with the algebra $\C_m$, are defined by a following rule
\begin{eqnarray}
&&\hat{\mathcal{E}}_j\phantom{m+}\;:\;\Lambda\longrightarrow\Lambda\varepsilon_j,\nonumber\\
&&\hat{\mathcal{E}}_{m+j}\;:\;\Lambda^1\longrightarrow-i\varepsilon_j\Lambda^1,\;
\Lambda^0\longrightarrow i\varepsilon_j\Lambda^0,\label{e32}
\end{eqnarray}
where $\Lambda$ is a general element of the auxiliary algebra $\C_m$,
$\Lambda^1$ and
$\Lambda^0$ are correspondingly odd and even parts of $\Lambda$,
$\varepsilon_j$ are units of the auxiliary algebra, $j=1,2,\ldots m$.
Analogously,
in the case of matrix representations of $C\kern -0.2em \ell_{p,q}$ we have
\begin{eqnarray}
&&\hat{\mathcal{E}}_j\phantom{m+}\;:\;\Lambda\longrightarrow\Lambda\beta_j\varepsilon_j,
\nonumber \\
&&\hat{\mathcal{E}}_{m+j}\;:\;\Lambda^1\longrightarrow-\varepsilon_j\beta_{m+j}
\Lambda^1,\;
\Lambda^0\longrightarrow\varepsilon_j\beta_{m+j}\Lambda^0,\label{e33}
\end{eqnarray}
where $\beta_i$ are arbitrary complex numbers.
It is easy to verify that transposition of the matrices of so defined
operators gives
\begin{equation}\label{e34}
\mathcal{E}^T_j=\mathcal{E}_j,\quad \mathcal{E}^T_{m+j}=-\mathcal{E}_{m+j}.
\end{equation}
Further, for the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}:\;
\widetilde{{\sf A}}=
{\sf E}{\sf A}^T{\sf E}^{-1}$, since in this case $\mbox{\bf e}_i\rightarrow\mbox{\bf e}_i$, it is sufficient
to select the matrix ${\sf E}$ so that
\[
{\sf E}\mathcal{E}^T_i{\sf E}^{-1}=\mathcal{E}_i
\]
or taking into account (\ref{e34})
\begin{equation}\label{e35}
{\sf E}\mathcal{E}^T_j{\sf E}^{-1}=\mathcal{E}_j,\quad {\sf E}\mathcal{E}^T_{m+j}{\sf E}^{-1}=-\mathcal{E}_{m+j}.
\end{equation}
Therefore, if $m$ is odd, then the matrix ${\sf E}$ has a form
\begin{equation}\label{e36}
{\sf E}=\mathcal{E}_1\mathcal{E}_2\ldots \mathcal{E}_m,
\end{equation}
since in this case a product $\mathcal{E}_1\mathcal{E}_2\ldots \mathcal{E}_m$ commutes
with all elements $\mathcal{E}_j\;(j=1,\ldots m)$ and anticommutes with all
elements $\mathcal{E}_{m+j}$.
Analogously, if $m$ is even, then
\begin{equation}\label{e37}
{\sf E}=\mathcal{E}_{m+1}\mathcal{E}_{m+2}\ldots \mathcal{E}_{p+q}.
\end{equation}
As required according to (\ref{e35}) in this case a product (\ref{e37})
commutes with $\mathcal{E}_j$ and anticommutes with
$\mathcal{E}_{m+j}$.
Let us consider now the conditions at which the matrix ${\sf E}$ commutes or
anticommutes with ${\sf W}$. Let ${\sf E}=\mathcal{E}_1\mathcal{E}_2\ldots \mathcal{E}_m$, where $m$ is odd,
since
${\sf W}=\mathcal{E}_1\ldots \mathcal{E}_m\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q}$, then
\begin{eqnarray}
\mathcal{E}_1\ldots \mathcal{E}_m\mathcal{E}_1\ldots \mathcal{E}_m\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q}
&=&(-1)^{\frac{m(m-1)}{2}}
\sigma_1\sigma_2\ldots\sigma_m\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q},\nonumber\\
\mathcal{E}_1\ldots \mathcal{E}_m\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q}\mathcal{E}_1\ldots \mathcal{E}_m&=&
(-1)^{\frac{m(3m-1)}{2}}
\sigma_1\sigma_2\ldots\sigma_m\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q},\nonumber
\end{eqnarray}
where $\sigma_i$ are the functions of the form (\ref{e2}). It is easy to see
that in this case the elements
${\sf W}$ and ${\sf E}$ are always anticommute. Indeed, a comparison
$\frac{m(3m-1)}{2}\equiv\frac{m(m-1)}{2}\pmod{2}$ is equivalent to
$m^2\equiv 0,1\pmod{2}$, and since $m$ is odd, then we have always
$m^2\equiv 1\pmod{2}$. At $m$ is even and ${\sf E}=\mathcal{E}_{m+1}\ldots
\mathcal{E}_{p+q}$ it is easy to see that the matrices ${\sf W}$ and ${\sf E}$
are always commute $(m\equiv 0\pmod{2})$.
Further, let $r$ be a quantity of the elements $\mathcal{E}_j$ of the product (\ref{e36})
whose squares equal to $+{\sf I}$, and let $s$ be a quantity of the elements
$\mathcal{E}_{m+j}$ of the product (\ref{e36}) whose squares equal to $-{\sf I}$.
Then a square
of the matrix (\ref{e36})
at $m$ is odd equals to $+{\sf I}$ if $r-s\equiv 3\pmod{4}$ and respectively
$-{\sf I}$ if $r-s\equiv 1\pmod{4}$. Analogously, a square of the matrix (\ref{e37})
at $m$ is even equals to $+{\sf I}$ if $k-t\equiv
0\pmod{4}$ and respectively $-{\sf I}$ if $k-t\equiv 2\pmod{4}$. It is obvious that
over the field $\F=\C$ we can to suppose $\mathcal{E}^2_{1\ldots m}=\mathcal{E}^2_{m+1\ldots
p+q}={\sf I}$.
Let us find now the matrix ${\sf C}$ of the antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}^\star}$: $\widetilde{{\sf A}^\star}={\sf C}{\sf A}^T{\sf C}^{-1}$. Since
in this case $\mbox{\bf e}_i\rightarrow-\mbox{\bf e}_i$, then it is sufficient to select the
matrix ${\sf C}$ so that
\[
{\sf C}\mathcal{E}^T_i{\sf C}^{-1}=-\mathcal{E}_i
\]
or taking into account (\ref{e34})
\begin{equation}\label{e38}
{\sf C}\mathcal{E}^T_j{\sf C}^{-1}=-\mathcal{E}_j,\quad {\sf C}\mathcal{E}^T_{m+j}{\sf C}^{-1}=\mathcal{E}_{m+j}.
\end{equation}
where $j=1,\ldots m$.
In comparison with (\ref{e35}) it is easy to see that in
(\ref{e38}) the matrices $\mathcal{E}_j$ and $\mathcal{E}_{m+j}$ are changed by the roles.
Therefore, if $m$ is odd, then
\begin{equation}\label{e39}
{\sf C}=\mathcal{E}_{m+1}\mathcal{E}_{m+2}\ldots \mathcal{E}_{p+q},
\end{equation}
and if $m$ is even, then
\begin{equation}\label{e40}
{\sf C}=\mathcal{E}_1\mathcal{E}_2\ldots \mathcal{E}_m.
\end{equation}
Permutation conditions of the matrices ${\sf C}$ and ${\sf W}$ are analogous to the
permutation conditions of ${\sf E}$ with ${\sf W}$, that is, the matrix ${\sf C}$ of the form
(\ref{e39})
always anticommutes with ${\sf W}$ ($m\equiv 1\pmod{2}$), and the matrix ${\sf C}$ of the
form
(\ref{e40}) always commutes with ${\sf W}$ ($m\equiv 0\pmod{2}$). Correspondingly,
a square of the matrix (\ref{e39}) equals to $+{\sf I}$ if $k-t\equiv 3\pmod{4}$ and
$-{\sf I}$ if $k-t\equiv 1\pmod{4}$. Analogously, a square of the matrix (\ref{e40})
equals to $+{\sf I}$ if $r-s\equiv 0\pmod{4}$ and $-{\sf I}$ if $r-s\equiv 2\pmod{4}$.
Obviously, over the field $\F=\C$ we can suppose ${\sf C}^2={\sf I}$.
Finally, let us find permutation conditions of the matrices ${\sf E}$ and ${\sf C}$.
First of all, at $m$ is odd ${\sf E}=\mathcal{E}_1\ldots \mathcal{E}_m$,
${\sf C}=\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q}$, alternatively, at $m$ is even
${\sf E}=\mathcal{E}_{m+1}\ldots
\mathcal{E}_{p+q}$, ${\sf C}=\mathcal{E}_1\ldots \mathcal{E}_m$. Therefore,
\[
\mathcal{E}_1\ldots \mathcal{E}_m\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q}=(-1)^{m^2}\mathcal{E}_{m+1}\ldots \mathcal{E}_{p+q}
\mathcal{E}_1\ldots
\mathcal{E}_m,
\]
that is, the matrices ${\sf E}$ and ${\sf C}$ commute at $m\equiv 0\pmod{2}$ and anticommute
at $m\equiv 1\pmod{2}$.
Now we have all the necessary conditions for the definition and
classification of isomorphisms between finite groups and automorphism groups
of Clifford algebras. Let $\F=\R$ and let $m\equiv 0\pmod{2}$, therefore,
a group ${\bf\sf A}=\{{\sf I},\,{\sf W},\,{\sf E},\,{\sf C}\}$ is Abelian. A condition
${\sf W}^2={\sf E}^2={\sf C}^2={\sf I}$
is equivalent to $p-q\equiv 0,4\pmod{8},\,r-s\equiv 0\pmod{4},
\,k-t\equiv 0\pmod{4}$,
which, clearly, are compatible. In accordance with (\ref{e36})--(\ref{e38})
and (\ref{e39})--(\ref{e40}) at $m\equiv 0\pmod{2}$ ${\sf W}={\sf C}{\sf E}$ and
${\sf W}^2=({\sf C}{\sf E})^2=
{\sf C}{\sf E}{\sf C}{\sf E}={\sf C}\sC{\sf E}\sE={\sf I}$. Therefore,
${\bf\sf A}_-\simeq\dZ_2\otimes\dZ_2$ for the signature
$(+,\,+,\,+)$ if $p-q\equiv 0,4\pmod{8}$.
The isomorphism ${\bf\sf A}_-\simeq\dZ_4$ for the signatures $(+,\,-,\,-)$
($p-q\equiv 0,4\pmod{8}$) and $(-,\,+,\,-),\;(-,\,-,\,+)$
($p-q\equiv 2,6\pmod{8}$) is analogously proved. It is easy to see that
for $m\equiv 0\pmod{2}$
there are only four isomorphisms considered previously. Further, for
$m\equiv 1\pmod{2}$ all the elements of the group ${\bf\sf A}$ anticommute and in
this case ${\sf W}={\sf E}{\sf C}$. The signature $(-,\,-,\,-)$ is equivalent to conditions
$p-q\equiv 2,6\pmod{8},\,r-s\equiv 1\pmod{4},\,k-t\equiv 1\pmod{4}$,
here ${\sf W}^2=({\sf E}{\sf C})^2={\sf E}{\sf C}{\sf E}{\sf C}=-{\sf E}\sE{\sf C}\sC=-{\sf I}$
and we have an isomorphism ${\bf\sf A}_+\simeq
Q_4/\dZ_2$, where $Q_4/\dZ_2=\{1,\,{\bf i},\,{\bf j},\,{\bf k}\}$, ${\bf i},\,{\bf j},\,{\bf k}$
are the quaternion units. It is easy to verify that for the signatures
$(-,\,+,\,+)$ at $p-q\equiv 2,6\pmod{8}$ and $(+,\,-,\,+),\;
(+,\,+,\,-)$ at $p-q\equiv 0,4\pmod{8}$ we have an isomorphism
${\bf\sf A}_+\simeq D_4/\dZ_2$, where $D_4/\dZ_2=\{1,\,\mbox{\bf e}_1,\,\mbox{\bf e}_2,\,\mbox{\bf e}_{12}\}$,
$\mbox{\bf e}_1,\mbox{\bf e}_2$ are the units of the algebra $C\kern -0.2em \ell_{1,1}$ or $C\kern -0.2em \ell_{2,0}$.
The eight automorphism groups considered previously, each of which is
isomorphic to one from the four finite groups
$\dZ_2\otimes\dZ_2,\,\dZ_4,\,
Q_4/\dZ_2,\,D_4/\dZ_2$, are the only possible over the field
$\F=\R$. In contrast with this, over the field $\F=\C$ we can suppose
${\sf W}^2={\sf E}^2={\sf C}^2={\sf I}$. At $m\equiv 0\pmod{2}$ we have only one signature
$(+,\,+,\,+)$ and an isomorphism ${\bf\sf A}_-\simeq
\dZ_2\otimes\dZ_2$ if $p+q\equiv 0,4\pmod{8}$, since over the field $\C$ the
signatures
$(+,\,-,\,-),\;(-,\,+,\,-)$ and $(-,\,-\,-)$ are isomorphic to $(+,\,+,\,+)$.
Correspondingly, at $m\equiv 1\pmod{2}$ we have an isomorphism
${\bf\sf A}_+\simeq Q_4/\dZ_2$ for the signature $(-,\,-,\,-)$ if
$p+q\equiv 2,6\pmod{8}$. It should be noted that the signatures
$(+,\,+,\,+)$ and
$(-,\,-,\,-)$ are non-isomorphic, since there exists no a group ${\bf\sf A}$ with the
signature $(+,\,+,\,+)$ in which all the elements anticommute, and also
there exists no a group ${\bf\sf A}$ with $(-,\,-,\,-)$ in which the all elements
commute. Thus, over the field $\C$ we have only two
non-isomorphic automorphism groups: ${\bf\sf A}_-\simeq\dZ_2\otimes\dZ_2,\;
{\bf\sf A}_+\simeq Q_4/\dZ_2$.
\end{proof}
The following Theorem is a direct consequence of the previous Theorem.
Here we establish a relation between signatures $(a,b,c)$ of the
D\c{a}browski groups and signatures $(p,q)$ of the Clifford algebras with
even dimensionality.
\begin{theorem}\label{t10}
Let $\pin^{a,b,c}(p,q)$ be a double covering of the orthogonal group
$O(p,q)$ of the space $\R^{p,q}$ associated with the algebra
$C\kern -0.2em \ell_{p,q}$ and let $\pin^{a,b,c}(p+q,\C)$ be a double covering of the complex
orthogonal group $O(p+q,\C)$ of the space $\C^{p+q}$ associated with the
algebra $\C_{p+q}$. Dimensionalities of the algebras
$C\kern -0.2em \ell_{p,q}$ and $\C_{p+q}$ are even $(p+q=2m)$, squares of the
symbols $a,b,c\in
\{-,+\}$ are correspond to squares of the elements of the finite group
${\bf\sf A}=\{{\sf I},{\sf W},{\sf E},{\sf C}\}:\;a={\sf W}^2,\,b={\sf E}^2,\,c={\sf C}^2$, where ${\sf W},{\sf E}$
and ${\sf C}$
are correspondingly the matrices of the fundamental automorphisms $\mathcal{A}\rightarrow
\mathcal{A}^\star,\,\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ and $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}^\star}$ of $C\kern -0.2em \ell_{p,q}$ and $\C_{p+q}$. Then over the field
$\K=\R$ for the algebra $C\kern -0.2em \ell_{p,q}$ there are eight double coverings
of the group $O(p,q)$ and two non-isomorphic double coverings of the group
$O(p+q,\C)$ for $\C_{p+q}$ over the field $\K=\C$:\\
1) $\F=\R$. Non--Cliffordian groups
\[
\pin^{+,+,+}(p,q)\simeq\frac{(\spin_0(p,q)\odot\dZ_2\otimes\dZ_2\otimes\dZ_2)}
{\dZ_2},
\]
if $p-q\equiv 0,4\pmod{8}$ and
\[
\pin^{a,b,c}(p,q)\simeq\frac{(\spin_0(p,q)\odot(\dZ_2\otimes\dZ_4)}{\dZ_2},
\]\begin{sloppypar}\noindent
if $(a,b,c)=(+,-,-)$ and $p-q\equiv 0,4\pmod{8}$, and also
$(a,b,c)=\{(-,+,-),\,(-,-,+)\}$ if $p-q\equiv 2,6\pmod{8}$.\\
\end{sloppypar}\noindent
Cliffordian groups
\[
\pin^{-,-,-}(p,q)\simeq\frac{(\spin_0(p,q)\odot Q_4)}{\dZ_2},
\]
if $p-q\equiv 2,6\pmod{8}$ and
\[
\pin^{a,b,c}(p,q)\simeq\frac{(\spin_0(p,q)\odot D_4)}{\dZ_2},
\]\begin{sloppypar}\noindent
if $(a,b,c)=(-,+,+)$ and $p-q\equiv 2,6\pmod{8}$, and also if
$(a,b,c)=\{(+,-,+),\,(+,+,-)\}$ and $p-q\equiv 0,4\pmod{8}$.\\
\end{sloppypar}
2) $\F=\C$. A non-Cliffordian group
\[
\pin^{+,+,+}(p+q,\C)\simeq
\frac{(\spin_0(p+q,\C)\odot\dZ_2\otimes\dZ_2\otimes\dZ_2)}
{\dZ_2},
\]
if $p+q\equiv 0,4\pmod{8}$. A Cliffordian group
\[
\pin^{-,-,-}(p+q,\C)\simeq\frac{(\spin_0(p+q,\C)\odot Q_4)}{\dZ_2},
\]
if $p+q\equiv 2,6\pmod{8}$.
\end{theorem}
\section{D\c{a}browski Groups for Odd-dimensional Spaces}
According to Theorem \ref{t8} and Theorem \ref{t4}
in the case of odd-dimensional spaces $\R^{p,q}$ and $\C^{p+q}$
the algebra homomorphisms $C\kern -0.2em \ell_{p,q}\rightarrow
C\kern -0.2em \ell_{p-1,q},\,C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p,q-1}$ and $\C_{p+q}\rightarrow
\C_{p+q-1}$ induce group homomorphisms $\pin(p,q)\rightarrow\pin(p-1,q),\,
\pin(p,q)\rightarrow\pin(p,q-1)$, $\pin(p+q,\C)\rightarrow\pin(p+q-1,\C)$
and correspondingly $\pin(p,q)\rightarrow\spin(p,q),\,\pin(p+q,\C)\rightarrow
\spin(p+q,\C)$.
\begin{theorem}\label{t11}
1) If $\F=\R$ and $\pin^{a,b,c}(p,q)\simeq\pin^{a,b,c}(p-1,q)\cup\omega
\pin^{a,b,c}(p-1,q),\,\pin^{a,b,c}(p,q)\simeq\pin^{a,b,c}(p,q-1)\cup\omega
\pin^{a,b,c}(p,q-1)$ are the D\c{a}browski groups over $\R$, where $p-q\equiv 3,7
\pmod{8}$, then in the result of homomorphic mappings $\pin^{a,b,c}(p,q)
\rightarrow\pin^{a,b,c}(p-1,q)$ and $\pin^{a,b,c}(p,q)\rightarrow
\pin^{a,b,c}(p,q-1)$ take place following quotient groups:
\begin{eqnarray}
\pin^b(p-1,q)&\simeq&\frac{(\spin_0(p-1,q)\odot\dZ_2\otimes\dZ_2)}{\dZ_2},
\nonumber\\
\pin^b(p,q-1)&\simeq&\frac{(\spin_0(p,q-1)\odot\dZ_2\otimes\dZ_2)}{\dZ_2}.
\nonumber
\end{eqnarray}
2) If $\F=\C$ and $\pin^{a,b,c}(p+q,\C)\simeq\pin^{a,b,c}(p+q-1,\C)\cup
\pin^{a,b,c}(p+q-1,\C)$ are the D\c{a}browski groups over $\C$, where $p+q\equiv
1,3,5,7\pmod{8}$, then in the result of an homomorpic mapping $\pin^{a,b,c}
(p+q,\C)\rightarrow\pin^{a,b,c}(p+q-1,\C)$ take place following quotient
groups:
\[
\pin^b(p+q-1,\C)\simeq\frac{(\spin_0(p+q-1,\C)\odot\dZ_2\otimes\dZ_2)}{\dZ_2},
\]
if $p+q\equiv 3,7\pmod{8}$ and
\[
\pin^{b,c}(p+q-1,\C),
\]
if $p+q\equiv 1,5\pmod{8}$, at this point a set of the fundamental automorphisms,
which correspond to the discrete transformations of the space
$\C^{p+q-1}$ associated with a quotient algebra
${}^\epsilon\C_{p+q-1}$, does not form a finite group.
\end{theorem}
\begin{proof} Indeed, over the field $\F=\R$ in accordance with Theorem
\ref{t8} from all the fundamental automorphisms
at the homomorphic mappings $C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p-1,q}$
and $C\kern -0.2em \ell_{p,q}\rightarrowC\kern -0.2em \ell_{p,q-1}$ only the antiautomorphism
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ is transferred into
quotient algebras ${}^\epsilonC\kern -0.2em \ell_{p-1,q}$
and ${}^\epsilonC\kern -0.2em \ell_{p,q-1}$. Further, according to Proposition
\ref{p1} the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$
corresponds to time reversal $T$. Therefore, groups of the discrete
transformations of the spaces $\R^{p-1,q}$ and $\R^{p,q-1}$
associated with the quotient algebras ${}^\epsilonC\kern -0.2em \ell_{p-1,q}$ and
${}^\epsilonC\kern -0.2em \ell_{p,q-1}$ are defined by a two--element group $\{1,T\}\sim
\{{\sf I},{\sf E}\}\simeq\dZ_2$, where $\{{\sf I},{\sf E}\}$ is an automorphism group of the quotient
algebras
${}^\epsilonC\kern -0.2em \ell_{p-1,q},\,{}^\epsilonC\kern -0.2em \ell_{p-1,q}$, ${\sf E}$ is a matrix of the
antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$. Thus, at
$p-q\equiv 3,7\pmod{8}$ there are the homomorphic mappings
$\pin^{a,b,c}(p,q)\rightarrow\pin^b(p-1,q)$ and $\pin^{a,b,c}(p,q)\rightarrow
\pin^b(p,q-1)$, where $\pin^b(p-1,q),\,\pin^b(p,q-1)$ are quotient groups,
$b=T^2={\sf E}^2$. At this point, a double covering of $C^b$ is isomorphic to
$\dZ_2\otimes\dZ_2$.
Analogously, over the field $\F=\C$ at $p+q\equiv 3,7\pmod{8}$ we have
a quotient group $\pin^b(p+q-1,\C)$. Further, according to Theorem
\ref{t8} in the result of the homomorphic mapping $\C_{p+q}\rightarrow
\C_{p+q-1}$ the antiautomorphisms $\mathcal{A}\rightarrow\widetilde{\mathcal{A}},\,\mathcal{A}
\rightarrow\widetilde{\mathcal{A}^\star}$ are transferred into a quotient algebra
${}^\epsilon\C_{p+q-1}$ at $p+q\equiv 1,5\pmod{8}$.
Therefore, a set of the discrete transformations of the space
$\C^{p+q-1}$ associated with the quotient algebra ${}^\epsilon\C_{p+q-1}$
is defined by a three-element set $\{1,T,PT\}\sim\{{\sf I},{\sf E},{\sf C}\}$, where
$\{{\sf I},{\sf E},{\sf C}\}$ is a set of the automorphisms of ${}^\epsilon\C_{p+q-1}$,
${\sf E}$ and ${\sf C}$ are correspondingly the matrices of the antiautomorphisms
$\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ and $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$. It is easy to see
that the set $\{1,T,PT\}\sim\{{\sf I},{\sf E},{\sf C}\}$ does not form a finite group. Thus,
at $p+q\equiv 1,5\pmod{8}$ there is an homomorphism $\pin^{a,b,c}(p+q,\C)
\rightarrow\pin^{b,c}(p+q-1,\C)$, where $\pin^{b,c}(p+q-1,\C)$ is a quotient
group, $b=T^2={\sf E}^2,\,c=(PT)^2={\sf C}^2$.
\end{proof}
{\bf Example}. Let us consider a simplest complex Clifford algebra with
odd dimensionality, $\C_3$. The algebra $\C_3$ may be represented by two
different complexifications:
$\C_3=\C\otimesC\kern -0.2em \ell_{3,0}$
and $\C_3=\C\otimesC\kern -0.2em \ell_{0,3}$, where $C\kern -0.2em \ell_{3,0}$ and $C\kern -0.2em \ell_{0,3}$ are
correspondingly the algebras of hyperbolic and elliptic biquaternions.
In accordance with Theorem \ref{t2} there is a decomposition of $\C_3$
into a direct sum of two subalgebras, which may be represented by a
following scheme:
\[
\unitlength=0.5mm
\begin{picture}(70,50)
\put(35,40){\vector(2,-3){15}}
\put(35,40){\vector(-2,-3){15}}
\put(32.25,42){$\C_{3}$}
\put(16,28){$\lambda_{-}$}
\put(49.5,28){$\lambda_{+}$}
\put(13.5,9.20){$\C_{2}$}
\put(52.75,9){$\C_{2}$}
\put(32.5,10){$\oplus$}
\end{picture}
\]
Here the idempotents
\[
\lambda_{-}=\frac{1-i\omega}{2},\quad
\lambda_{+}=\frac{1+i\omega}{2}
\]
in accordance with \cite{CF97}
may be identified with helicity projection operators. Further, according to
Theorem \ref{t8} at the homomorphic mapping $\epsilon:\,\C_3
\rightarrow\C_2$ the antiautomorphisms
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ and $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$
are transferred into a quotient algebra ${}^\epsilon\C_2$, and the
automorphism $\mathcal{A}\rightarrow\mathcal{A}^\star$ is not transferred. Therefore,
there is an homomorphism $\pin^{a,b,c}(3,\C)\rightarrow\pin^{b,c}(2,\C)$
(Theorem \ref{t11}), where $\pin^{b,c}(2,\C)$ is a quotient group which
double covers the orthogonal group $O(2,\C)$ of the space
$\C^2$ associated with ${}^\epsilon\C_2$.
According to proposition \ref{p1} the automorphism $\mathcal{A}\rightarrow\mathcal{A}^\star$
corresponds to parity reversal $P$ which under action of the
homomorphism $\epsilon$ is not transferred into the quotient algebra
${}^\epsilon\C_2$ and correspondingly quotient group $\pin^{b,c}(2,\C)$.
Thus, we have a `symmetry breaking' of the group of discrete transformations
with excluded operation $P$. In physics there is an analog of this
situation known as a {\it parity violation}. In connection with this it pays
to relate the quotient algebra ${}^\epsilon\C_2$ and quotient group
$\pin^{b,c}(2,\C)$ with some chiral field, for example, neitrino field.
As known, in Nature there exist only left neitrino and right antineitrino and
there exist no right neitrino and left antineitrino, therefore, for the
neitrino field the operation $P$ is violated. In order to proceed this
analogy, at first we must to establish a relation between the
quotient algebra ${}^\epsilon\C_2$ and some spinor field.
Let us show that such a field is a Dirac-Hestenes spinor field
\cite{Hest66,Hest90}. Indeed, in accordance with Theorem \ref{t1} we have
$\C_2\simeqC\kern -0.2em \ell_{3,0}$, further $C\kern -0.2em \ell_{3,0}\simeqC\kern -0.2em \ell^+_{1,3}$, where $C\kern -0.2em \ell_{1,3}$
is a spacetime algebra. Units of the algebra $C\kern -0.2em \ell_{1,3}$ in the matrix
representation have the form
\begin{equation}\label{e41}\renewcommand{\arraystretch}{1}
\Gamma_0=\begin{pmatrix}
I & 0\\
0 & -I
\end{pmatrix},\;\;\Gamma_1=\begin{pmatrix}
0 & \sigma_1\\
-\sigma_1 & 0
\end{pmatrix},\;\;\Gamma_2=\begin{pmatrix}
0 & \sigma_2\\
-\sigma_2 & 0
\end{pmatrix},\;\;\Gamma_3=\begin{pmatrix}
0 & \sigma_3\\
-\sigma_3 & 0
\end{pmatrix},
\end{equation}
where $\sigma_i$ are the Pauli matrices
\[
\renewcommand{\arraystretch}{1}
\sigma_1=\begin{pmatrix}
0 & 1\\
1 & 0
\end{pmatrix},\quad\sigma_2=\begin{pmatrix}
0 & -i\\
i & 0
\end{pmatrix},\quad\sigma_3=\begin{pmatrix}
1 & 0\\
0 &-1
\end{pmatrix},
\]
$I$ is the unit matrix. The Dirac-Hestenes spinor
$\phi$ is an element of the algebra $C\kern -0.2em \ell^+_{1,3}\simeqC\kern -0.2em \ell_{3,0}$
and, therefore, may be represented by the biquaternion number
\begin{equation}
\phi=a^0+a^{01}\Gamma_{01}+a^{02}\Gamma_{02}+a^{03}\Gamma_{03}+
a^{12}\Gamma_{12}+a^{13}\Gamma_{13}+a^{23}\Gamma_{23}+a^{0123}\Gamma_{0123}.
\label{e42}
\end{equation}
Or in the matrix form
\begin{equation}\label{e43}\renewcommand{\arraystretch}{1}
\phi=\begin{pmatrix}
\phi_1 & -\phi^\ast_2 & \phi_3 & \phi^\ast_4 \\
\phi_2 & \phi^\ast_1 & \phi_4 & -\phi^\ast_3\\
\phi_3 & \phi^\ast_4 & \phi_1 & -\phi^\ast_2\\
\phi_4 & -\phi^\ast_3 & \phi_2 & \phi^\ast_1
\end{pmatrix},
\end{equation}
where
\begin{eqnarray}
\phi_1&=&a^0-ia^{12},\nonumber\\
\phi_2&=&a^{13}-ia^{23},\nonumber\\
\phi_3&=&a^{03}-ia^{0123},\nonumber\\
\phi_4&=&a^{01}+ia^{02}.\nonumber
\end{eqnarray}
The spinor $\phi$, defined by the expression (\ref{e42}) or (\ref{e43}),
satisfies to a Dirac-Hestenes equation \cite{Hest90}
\begin{equation}\label{e44}
\partial\phi\Gamma_{21}=\frac{mc}{\hbar}\phi\Gamma_0,
\end{equation}
where $\partial=\Gamma^\nu\frac{\partial}{\partial x^\nu}$. Further, let
${}^\epsilon\phi\in{}^\epsilon\C_2$ be a Dirac-Hestenes 'quotient spinor'.
In virtue of the isomorphism ${}^\epsilon\C_2\simeqC\kern -0.2em \ell_{3,0}\simeqC\kern -0.2em \ell^+_{1,3}$
we have for the spinor ${}^\epsilon\phi$ the representation (\ref{e42}).
It is known \cite{RSVL} that a transformation group of the Dirac-Hestenes
field is a group $\spin_+(1,3)$ which double covers the connected component
of the Lorentz group, since
\[
\renewcommand{\arraystretch}{1}
\spin_+(1,3)\simeq\left\{\begin{pmatrix} a & b \\ c & d\end{pmatrix}\in\C_2:\;
\det\begin{pmatrix} a & b \\ c & d\end{pmatrix}=1\right\}=SL(2,\C).
\]
It is obvious that $\pin(2,\C)\simeq\spin(3,\C)$ and further $\spin(3,\C)\simeq
\spin(1,3)$, whence $\spin_0(3,\C)\simeq\spin_+(1,3)$, therefore,
a transformation group of the quotient spinor ${}^\epsilon\phi$ is also
isomorphic to $\spin_+(1,3)$. The complete transformation group of
${}^\epsilon\phi$ is isomorphic to the quotient group
$\pin^{b,c}(2,\C)$ consisting of
$\spin_+(1,3)$ and the two discrete transformations $T$ and $PT$ (the latter
transformation in virtue of $CPT$-theorem is equivalent to a charge
conjugation $C$). Since the quotient algebra ${}^\epsilon\C_2$ is isomorphic
to a subalgebra of all the even elements of the spacetime algebra $C\kern -0.2em \ell_{1,3}$,
then we can express the automorphisms of ${}^\epsilon\C_2$ via the
automorphisms of $C\kern -0.2em \ell_{1,3}$. In accordance with Theorem \ref{t9}
the automorphism group of $C\kern -0.2em \ell_{1,3}$ is isomorphic to the Abelian group
$\dZ_4$. The non-Cliffordian group double covering the Lorentz group $O(1,3)$
has a form (theorem \ref{t10})
\begin{eqnarray}
\pin^{-,-,+}(1,3)&\simeq&\frac{(\spin_0(1,3)\odot\dZ_2\otimes\dZ_4)}{\dZ_2}
\nonumber\\
&\simeq&\frac{(SL(2,\C)\odot\dZ_2\otimes\dZ_4)}{\dZ_2}.\nonumber
\end{eqnarray}
In the group ${\bf\sf A}_-=\{{\sf I},{\sf W},{\sf E},{\sf C}\}$ the matrices ${\sf E}$ and ${\sf C}$ of the
antiautomorphisms $\mathcal{A}\rightarrow\widetilde{\mathcal{A}}$ and
$\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$
for the basis (\ref{e41}) have the form
\[
\renewcommand{\arraystretch}{1}
{\sf E}=\begin{pmatrix}
0 & 1 & 0 & 0\\
-1& 0 & 0 & 0\\
0 & 0 & 0 & 1\\
0 & 0 & -1& 0
\end{pmatrix},\quad {\sf C}=\begin{pmatrix}
0 & 0 & 0 & -i\\
0 & 0 & i & 0\\
0 & -i& 0 & 0\\
i & 0 & 0 & 0
\end{pmatrix}
\]
Further, in accordance with (\ref{e19}) an action of the
antiautomorphism $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}^\star}$ on the spinor $\phi$, defined by the matrix
representation (\ref{e43}), is expressed as follows
\begin{equation}\label{e45}\renewcommand{\arraystretch}{1}
\widetilde{\phi^\star}=\begin{pmatrix}
\phi^\ast_1 & \phi^\ast_2 & -\phi^\ast_3 & -\phi^\ast_4 \\
-\phi_2 & \phi_1 & -\phi_4 & \phi_3\\
-\phi^\ast_3 & -\phi^\ast_4 & \phi^\ast_1 & \phi^\ast_2\\
-\phi_4 & \phi_3 & -\phi_2 & \phi_1
\end{pmatrix}
\end{equation}
It should be noted that the same result may be obtained via $\Gamma^0
\phi^+\Gamma^0$ (see \cite{VR93}).
Let us assume now that a massless Dirac-Hestenes equation (\ref{e44})
describes the neitrino field
\begin{equation}\label{e46}
\partial{}^\epsilon\phi\Gamma_{21}=0,
\end{equation}
then the antineitrino field is described by an equation
\begin{equation}\label{e47}
\widetilde{\left(\partial{}^\epsilon\phi\Gamma_{21}\right)^\star}=0.
\end{equation}
The equations (\ref{e46}) and (\ref{e47}) are converted into each other under
action of the antiautomorphism $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$ which
in accordance with Proposition \ref{p1} corresponds to the combination $PT$
(charge conjugation by $CPT$--theorem). The fields describing by the equations
(\ref{e46}) and (\ref{e47}) possess a fixed helicity (there exist no right
neitrino and left antineitrino), since the automorphism
$\mathcal{A}\rightarrow\mathcal{A}^\star$ corresponded to parity reversal
$P$ in this case is not defined. On the other hand, in accordance with the
Feynman-Stueckelberg interpretation the antiparticles are considered as
particles moving back in time, therefore a time-reversed equation
(\ref{e46}) describes the antiparticle (antineitrino). Moreover, in the
Feynman--Stueckelberg interpretation time reversal for the chiral field
gives rise to the well--known $CP$--invariance in the theory of weak
interactions. Thus, the actions of the antiautomorphisms $\mathcal{A}\rightarrow
\widetilde{\mathcal{A}}$ and $\mathcal{A}\rightarrow\widetilde{\mathcal{A}^\star}$ and the
corresponding operations $T$ and $PT\sim C$ on the field ${}^\epsilon\phi$
are equivalent. This equivalence immediately follows from (\ref{e8})
and (\ref{e9}) (see also \cite{FRO90b}), namely, for $\mathcal{A}\inC\kern -0.2em \ell^+_{p,q}$
we have always $\widetilde{\mathcal{A}}=\widetilde{\mathcal{A}^\star}$, in our case
${}^\epsilon\phi\inC\kern -0.2em \ell^+_{1,3}$ and $\widetilde{{}^\epsilon\phi}=
\widetilde{{}^\epsilon\phi^\star}$.
\section*{Acknowledgments} I am grateful to Prof. D\c{a}browski for
sending me his interesting works. |
\section{Introduction}
The idea that there may be more than three space dimensions is as old as
Kaluza and Klein's work dating back to the early part of this century.
The advent of superstring theories generated new interest in extra
dimensions since consistent superstring theories exist only in 10 or 26
dimensions. It was conventional to assume that the extra dimensions
are compactified to manifolds of small radii so that they remain hidden
to physics considerations. The value of the small radii was assumed to be
of order $M^{-1}_{P\ell}$ and thus invisible.
Recently the possibility that the extra hidden dimensions may have
radii considerably larger (of order TeV$^{-1}$ or even (milli-eV)$^{-1}$)
has been the subject of intense scrutiny
~\cite{anto,dvali,pheno,dine,ddg1,ddg2,ross,kaku,carone,quiros,frampton}
and has generated considerable amount of excitement in phenomenological
circles. This is related to theoretical developments in string theories
which have made such speculations plausible. It is assumed that the extra
dimensions of string theories are compactified on orbifolds (or other
compact manifolds) of radii $R$. The sizes of these radii depend on
the details of the theory. In the currently popular pictures where it is
assumed that there exist D-branes embedded in the high dimensional space,
if one assumes a scenario where only gravity is in the bulk and all matter
fields are in the brane, $R$ can be as large as a milli meter\cite{dvali},
which is the threshold below which the Newtonian gravitational law has not
been experimentally verified. In other cases, $R$ has to be smaller than
a few TeV$^{-1}$. This latter case may not only have direct
experimental tests in colliders\cite{coll}, but also may have interesting
implications for physics beyond the standard model because its presence
crucially effects the nature of grand unification of forces and matter.
These ideas may also have other theoretical implications such
as a new route to solve the hierarchy problem~\cite{dvali} if it is
assumed that the string scale is in the TeV range\cite{lykk}. Such low
scales also
raise possibilities for new effects in
astrophysical settings, which then lead to new constraints on
them~\cite{pheno}. Clearly a rich new avenue of
particle physics has been opened up by these considerations.
In this article, we explore the effects of extra dimensions on the
unification of gauge couplings. Dienes, Dudas and Gherghetta\cite{ddg1}
began this kind of analysis in a series of papers for the minimal
supersymmetric standard model. They used power law unification, noted
originally by Taylor and Veneziano\cite{taylor} to argue
that indeed MSSM leads to unification even in the presence of extra hidden
dimensions with an arbitrary scale between a TeV$^{-1}$ and the inverse of
the GUT scale. Subsequent papers have addressed various issues related to
the question of unification~\cite{ross,kaku,carone,quiros,frampton}.
For example, it has been noted that in the ``minimal'' versions of the
model considered in \cite{ddg1,ddg2}, the true unification of couplings
predicts a larger value for $\alpha_{s}(M_Z)$ compared to experimental
observations. One way to cure this problem is to include new contributions
to beta functions\cite{ross,kaku,carone,quiros,frampton} by postulating
additional fields at the weak scale.
It is the goal of this paper to continue this investigation further. We
start by discussing briefly the power law variation of couplings
and comparing it with the logarithmic one that is closer to
reality in most unification models\cite{ross}.
We then introduce a new set of variables
constructed out of the beta function coefficients and show how it provides
a different way to look
at the prospects of unification in different cases. We then use these
variables to study both one and two step unification models. In addition
to providing, what we believe is a new way to test for unification, we
find two new results with interesting phenomenological implications which
to the best of our knowledge have not been discussed in the
literature.
(i) The minimal supersymmetric left-right symmetric model
with the seesaw mechanism, which resisted unification in the four
dimenaional case can now be unified if the gauge fields are put in the
bulk and (ii) with non-canonical normalization of gauge couplings, we
find examples where the unification scale is around $10^{11}$ GeV or so
which has been advocated in a recent paper as the possible string scale
from various phenomenological considerations~\cite{quevedo}.
The paper is organized as follows. In section two we present the basic ideas
that introduce the power law running; in section three we compare on
analytical and numerical basis these results with those obtained by the
implementation of the step by step approach which invoke the decoupling theorem
\cite{appel} at each level of the Kaluza Klein tower. Next section is
devoted to discuss
the MSSM and the SM unification. Here our goal is to show that generically,
through the model independent analysis, the compactification scale is fixed by
the experimental accuracy in the gauge coupling constants. Moreover, in the
supersymmetric SU(5) theory the one loop prediction for $\alpha_s$ could be
within the experimental value just by fixing the compactification scale close
below the usual unification mass, without the introduction of extra matter and
assuming that all the standard gauge and scalars and perhaps the fermions
propagate into the bulk. Nevertheless,
the SU(5) unification makes this result unstable under two loop corrections. As
it has been pointed out in references~\cite{kaku,carone,quiros,frampton},
a way to
solve this problem is to modify the bulk content. We present a simple
choice where only the gauge bosons develop excited modes. Section five is
dedicated to discuss the case of two steps models. Here, based on the results
of the analysis, we argue that the excited modes of higher symmetries,
expected to be embedded in the unification theory, as the left right model for
instance, may split the unification scale pushing such symmetries down the
compactification scale. Moreover, as the MSSM particles could be naturally
accommodated in left right representations, we argue that the left right model
must appear below the compactification scale, fixing the unification and
setting lower bounds to the compactification scale. We shall also show that in
some scenarios, this effect could produce consistent results with both the
neutrino physics~\cite{nphys,see-saw} and proton decay.
\section{Power law running}
The evolution of the gauge coupling constants above the compactification
scale $\mu_0$ was derived by Dienes, Dudas and Gherghetta
(DDG)~\cite{ddg1,ddg2}
on the base of an effective (4-dimensional) theory approach.
The general result at one-loop level is given by
\begin{equation}
\alpha_i^{-1}(\mu_0) = \alpha_i^{-1}(\Lambda) + {b_i - \tilde{b}_i\over 2\pi}
\ln\left( {\Lambda\over \mu_0}\right) + {\tilde{b}_i\over 4\pi}\
\int_{r\Lambda^{-2}}^{r\mu_0^{-2}}\! {dt\over t}
\left\{ \vartheta_3\left({it\over \pi R^2}\right)\right\}^\delta ,
\label{exact}
\end{equation}
with $\Lambda$ as the ultraviolet cut-off, $\delta$ the number of extra
dimensions and $R$ the compactification radius identified as $1/\mu_0$. The
Jacobi theta function
\begin{equation}
\vartheta(\tau) = \sum_{n=-\infty}^{\infty} e^{i\pi \tau n^2}
\end{equation}
reflects the sum over the complete (infinite) Kaluza Klein (KK) tower. In Eq.
(\ref{exact}) $b_i$ are the beta functions of the theory below the $\mu_0$
scale, and $\tilde{b}_i$ are the contribution to the beta functions
of the KK states at each excitation
level. Besides, the numerical factor $r$ in the former integral could not be
deduced purely from this approach. Indeed, it is obtained assuming that
$\Lambda\gg \mu_0$ and comparing the limit with the usual renormalization group
analysis, decoupling all the excited states with masses above $\Lambda$, and also
assuming that the number of KK states below certain energy $\mu$ between $\mu_0$
and $\Lambda$ is well approximated by the volume of a $\delta$-dimensional
sphere of radius $\mu/\mu_0$
\begin{equation}
N (\mu,\mu_0) = X_\delta
\left({\mu\over\mu_0}\right)^\delta ;
\label{npl}
\end{equation}
with $X_\delta = \pi^{\delta /2}/\Gamma(1 +\delta /2)$.
The result is a power law behaviour of the gauge coupling constants given by
\begin{equation}
\alpha_i^{-1}(\mu) = \alpha_i^{-1}(\mu_0) - {b_i - \tilde{b}_i\over 2\pi}
\ln\left( {\mu\over \mu_0}\right) - {\tilde{b}_i\over 2\pi}\cdot
{X_\delta\over\delta}\left[ \left({\mu\over\mu_0}\right)^\delta - 1\right] .
\label{ddgpl}
\end{equation}
Nevertheless, as it was pointed out by Ghilencea and Ross~\cite{ross}, in the
MSSM the energy range between $\mu_0$ and $\Lambda$ --identified as the
unification scale-- is relatively small due to the steep behaviour in the
evolution of the couplings. For instance, for a single extra dimension the
ratio $\Lambda/\mu_0$ has an upper limit of the order of 30, which
substantially decreases for higher $\delta$ to be less than 6.
Clearly, this fact seems in conflict with the assumption which justifies Eq.
(\ref{npl}). Moreover, as the number of KK states with masses lesser than
$\mu$ are by definition the number of solutions to the equation
\begin{equation}
\sum_{i=1}^\delta n_i^2 \leq \left({\mu\over \mu_0}\right)^2,
\end{equation}
where $n_i\in\IZ$ corresponds to the component of the
momentum of the fields propagating into the bulk along the $i$-th extra
dimension; we expect
that only the first few levels are relevant in the analysis for $\mu\sim
\mu_0$, restoring the usual logarithmic behaviour rather than the power
behaviour as in Eq. (\ref{ddgpl}). It is then important to know the kind of
errors involved in the power law assumption.
\section{Logarithmic running}
In general, the mass of each KK mode is well approximated by
\begin{equation}
\mu_n^2 = \mu_0^2\sum_{i=1}^\delta n_i^2.
\label{mn}
\end{equation}
Therefore, at each mass level $\mu_n$ there are as many modes as
solutions to Eq. (\ref{mn}).
It means, for instance, that in one extra dimension each
KK level will have 2 KK states that match each other,
with the exception of the zero modes which are
not degenerate and correspond to (some of) the particles
in the original (4-dimensional)
theory manifest below the $\mu_0$ scale. In this particular case,
the mass levels are separated by units of $\mu_0$.
In higher extra dimensions the KK levels are
not regularly spaced any more. Indeed, as it follows from Eq. (\ref{mn}), the
spectra of the excited mass levels correspond to that of the energy in the
$\delta$-dimensional box, where the degeneracy of each energy level is not so
trivial but still computable (see below).
{\it One extra dimension.}
The one-loop renormalization group equations for energies just above the
$n$-th level ($\mu> n\mu_0$) in the simplest case, $\delta = 1$, are
\begin{equation}
{d\over d\ln\mu}\alpha_i^{-1} = -{b_i+ 2n \tilde{b}_i \over 2 \pi},
\end{equation}
for at this level all the low energy particles contribute through
$b_i$ --which already
include the zero modes-- and all the $2n$ excited
states in the first $n$ KK levels, each of one giving a contribution of
$\tilde{b}_i$. Solving this equation requires
boundary conditions at $n\mu_0$, to get
\begin{equation}
\alpha_i^{-1}(\mu) = \alpha_i^{-1}(n\mu_0) - {b_i +2 n \tilde{b}_i\over 2\pi}
\ln\left( {\mu\over n\mu_0}\right),
\end{equation}
while, for the same arguments
\begin{equation}
\alpha_i^{-1}(n\mu_0) = \alpha_i^{-1}\left((n-1)\mu_0\right)
- {b_i + 2 (n-1) \tilde{b}_i\over 2\pi} \ln\left( {n\over n-1}\right),
\end{equation}
and so on, up to
\begin{equation}
\alpha_i^{-1}(2\mu_0) = \alpha_i^{-1}\left(\mu_0\right)
- {b_i + 2 \tilde{b}_i\over 2\pi} \ln\left( 2\right).
\end{equation}
Combining all these equations together is straightforward to get
\begin{equation}
\alpha_i^{-1}(\mu) = \alpha_i^{-1}(\mu_0) - {b_i\over 2\pi}
\ln\left( {\mu\over \mu_0}\right) - {\tilde{b}_i\over 2\pi}\cdot 2\left[
n\ln\left(\mu\over\mu_0\right) - \ln n! \right].
\label{log}
\end{equation}
which explicitly shows a logarithmic behaviour just corrected by the
appearance of the $n$ thresholds below $\mu$.
Using the Stirling's formula $n!\approx n^n e^{-n} \sqrt{2\pi n}$ valid for
large $n$, the last expression takes the form of the power law running
\begin{equation}
\alpha_i^{-1}(\mu) = \alpha_i^{-1}(\mu_0) - {b_i-\tilde{b}_i\over 2\pi}
\ln\left( {\mu\over \mu_0}\right) - {\tilde{b}_i\over 2\pi}\cdot 2\left[
\left(\mu\over\mu_0\right) - \ln \sqrt{2\pi} \right].
\label{limit}
\end{equation}
The last term $\ln\sqrt{2\pi}\approx 0.9189$, thus, the limit is fully
consistent with eq (\ref{ddgpl}). Indeed this small difference could be
absorbed by high energy threshold or even second order corrections.
It is worth pointing out that while in the limiting case of large number
of Kaluza-Klein states Eq. (12) agrees with Eq. (4), for the case of
finite N, there is a difference as can be seen by choosing $\mu =
\mu_0+\epsilon$ in Eq. (4) and Eq. (11).
{\it Higher dimensions.} For the case of two or more extra
dimensions, as in its classical
quantum analogous, each level is characterized by the set of natural numbers
$\{n_1,\dots,n_\delta\}$ which satisfy Eq. (\ref{mn}). It is clear that if
all the $n_i$ numbers are different and non zero, the KK level is $2^\delta
\delta !$-fold degenerated, since there are $\delta !$ ways of distributing
these $\delta$ (absolute) values between the $\delta$ numbers $n_i$, and there
are $2^\delta$ different combinations of the signs for each one of those
combinations. Besides, some of these numbers could be equal or even zero, then,
in general the degeneracy of each level is given by
\begin{equation}
g_N = 2^{\delta-p} {\delta !\over k_1! k_2!\cdots k_l! p!};
\end{equation}
where $k_i$ is the number of times that the value (without sign) of $n_i$
appears in the array $\{n_1,\dots,n_\delta\}$, and $p$ is the number of
zero elements in the same array. The (natural)
index $N$ stands for the label of the level
corresponding to the squared ratio of masses
\begin{equation}
\sum_{i=1}^\delta n_i^2 = \left({\mu_{N-1}\over \mu_0}\right)^2 \equiv N
\qquad ; \quad N \in \relax{\rm I\kern-.18em N} .
\label{N}
\end{equation}
In addition to this ``normal'' degeneracy, there often are
accidental degeneracies due to
certain numerical coincidences: some natural numbers have more than one
non equivalent decompositions of the form (\ref{N}). For instance, for
$\delta=2$, we can write $25= 5^2 + 0 = 4^2 + 3^2$, thus level 25 is 12-fold
degenerated (4 times from the first decomposition plus 8 times from the second
one), while level 5 is just 8-fold degenerated ($5= 2^2 + 1$)
and level 3 does not exist.
Despite this complexity of the spectra, the degeneracy of each level
is always computable and performing a level by level approach of the gauge
coupling running is still possible. In this case,
the renormalization group equations for
energies above the $N$-th level receive contributions from $b_i$ and of
all the KK excited states in the
levels below, in total
\begin{equation}
f_\delta(N) = \sum_{n=1}^N g_\delta(n) ;
\end{equation}
where $g_\delta(n)$ represent the total degeneracy of the level $n$. The
coupling evolution equations then look like
\begin{equation}
\alpha_i^{-1}(\mu) = \alpha_i^{-1}(\mu_{N-1}) -
{b_i + f_\delta(N) \tilde{b}_i\over 2\pi}
\ln\left( {\mu\over \mu_{N-1}}\right).
\end{equation}
Iterating this result for all the first $N$ levels and combining all
them together with Eq. (\ref{N})
we get the logarithmic running given by
\begin{equation}
\alpha_i^{-1}(\mu) = \alpha_i^{-1}(\mu_0) - {b_i\over 2\pi}
\ln\left( {\mu\over \mu_0}\right) - {\tilde{b}_i\over 2\pi} \left[
f_\delta(N)\ln\left(\mu\over\mu_0\right) -
{1\over 2}\sum_{n=1}^N g_\delta(n)\ln n \right],
\label{hdlog}
\end{equation}
where now the correction of the $N$ thresholds appears a little bit more
complex than as before. It is clear that this relationship reduces to Eq.
(\ref{log}) for $\delta=1$.
For large $N$, where the Eq. (\ref{npl}) holds, the former expression
reduces to power law running as we show now.
To prove this, let us take only the last
term in brackets in Eq. (\ref{hdlog}), which, using that $g_\delta(n) =
f_\delta(n) - f_\delta(n-1)$, may be rewritten in the form
\begin{eqnarray}
F_\delta\left({\mu\over\mu_0}\right) &\equiv &
f_\delta(N)\ln\left(\mu\over\mu_0\right) -
{1\over 2}\sum_{n=1}^N g_\delta(n)\ln n \label{f} \\[1ex]
& = &
f_\delta(N)\left[ \ln\left({\mu\over\mu_0}\right) - {1\over 2}\ln N\right]
- {1\over 2} \left[ \sum_{n=1}^N\bigg(f_\delta(n) - f_\delta(n-1)\bigg)\ln n
- f_\delta(N) \ln N\right].
\end{eqnarray}
In the limit when $N$ is large, $N\approx (\mu/\mu_0)^2$. Hence, the first term
vanishes. The remaining terms becomes in the continuum limit
\begin{equation}
F_\delta\left({\mu\over\mu_0}\right) =
-{1\over 2} \left[ \int_1^N dn\ {d f_\delta \over dn}(n) \ln n
- f_\delta (N)\ln N \right ] = \int_{\mu_0}^\mu {d\mu'\over \mu'}
f_\delta(n(\mu')) .
\end{equation}
Assuming in this limit that $f_\delta(n(\mu))
\approx X_{\delta}\left( \frac{\mu}{\mu_0}\right)^\delta-1$, we recover
the
DDG approximation for large $N$:
\begin{equation}
F_\delta\left({\mu\over\mu_0}\right) \approx
{X_\delta\over\delta}\left[ \left({\mu\over\mu_0}\right)^\delta - 1\right]
- \ln\left({\mu\over\mu_0}\right) .
\end{equation}
The explicit difference in the running of the gauge coupling constants governed
by the power law (\ref{ddgpl}) from the logarithmic running (\ref{hdlog}) is
best appreciated from the numerical analysis of the function $F_\delta$, since
$[\alpha_i^{-1}(\mu)]^{DDG} - [\alpha_i^{-1}(\mu)] = \tilde{b}_i
[F_\delta (\mu/\mu_0) - F_{\delta}^{DDG}(\mu/\mu_0)]/{2\pi}$; where
the index $DDG$ stands to identify the power law expressions. Such difference
has been plotted in figure 1 for one and two extra dimensions. We
notice for $\delta=1$ that $F_\delta- F_\delta^{DDG}$ tends to
converge quickly to an asymptotic value, while, for $\delta=2$ it is more
unstable but still convergent. This fact is a consequence of the complexity in
the level degeneracy. From here, is easy to figure out that a more unstable
behaviour arise for higher $\delta$. Indeed, for $\delta=3$ we found that a
more large number of thresholds is required to stabilize the difference into a
small slope of the asymptotic value, which tends to be higher for larger extra
dimensions.
Nevertheless, as $F_\delta$ has a steep evolution, in the limit of large
$N$ those differences becomes to be strongly suppressed compared with the actual
value of $F_\delta$, which dominates the running of
the gauge coupling constants,
as it is depicted in figure 2, where we can observe that for $\mu/\mu_0>10$,
$F_\delta- F_\delta^{DDG}$ represent only less than 2\% of the value of
$F_\delta$. However, for lower ratios the deviation of
$F_\delta$ from the power law remains within 2\% to 50\%.
Before closing the present section, we point out a natural
extension to our analysis to the case where the $\delta$ compactification
radii $R_i$ are not all equal, since
the requirement of their equality is so far unjustified. In this case,
the masses of the excited KK modes are given by
\begin{equation}
m_n^2 = \sum_{i=1}^\delta n_i^2\mu_i^2;
\label{mmn}
\end{equation}
where we have defined $\mu_i=1/R_i$. Assuming that $\mu_0= 1/R_{max}$, the
inverse of the largest radius, the
contributions of the bulk to the renormalization group equations should start
at these mass levels. Hereafter the running will cross a new threshold
each
time that $\mu$ reachs a level in the tower, again characterized, as those of
the non cubic $\delta$ dimensional box, by the squared ratio of masses
\begin{equation}
M_n \equiv \left({m_n\over \mu_0}\right)^2 =
\sum_{i=1}^\delta n_i^2 \left({\mu_i\over \mu_0}\right)^2 .
\end{equation}
As it is clear, our approach will work following the same steps as before, and
then, we will arise to a logarithmic running of the same form as in Eq.
(\ref{hdlog}), but now with $F_\delta$ given by
\begin{equation}
F_\delta(\mu,\mu_0,\cdots\mu_\delta) =
f_\delta(N)\ln\left(\mu\over\mu_0\right) -
{1\over 2}\sum_{n=1}^N g_\delta(n)\ln M_n ,
\end{equation}
for $\mu$ just above the $N$-th level of the tower. In the continuous limit (when
$\mu$ is large compared with whatever $\mu_i$) we may
assume that the number of states below the energy scale $\mu$ is well
approximated by the volume of the $\delta$ dimensional ellipsoid defined by Eq.
(\ref{mmn}) where $m_n\approx \mu$
\begin{equation}
N(\mu,\mu_0,\cdots\mu_\delta)\approx
X_\delta \prod_{i=1}^\delta\left(\mu\over\mu_i\right).
\end{equation}
In this limit
\begin{equation}
F_\delta(\mu,\mu_0,\cdots\mu_\delta) \approx
{X_\delta\over\delta}\left[ \prod_{i=1}^\delta\left({\mu\over\mu_i}\right)
- \prod_{i=1}^\delta\left({\mu_0\over\mu_i}\right) \right]
- \ln\left({\mu\over\mu_0}\right) .
\label{fdeg}
\end{equation}
with the explicit extraction of the zero modes. Clearly, when all the radii are
equal it reproduces the DDG expression.
\section{Unification in extra dimensions}
Let us now analyze the implications of extra dimensions for unification
of the gauge couplings. Many features of unification can be studied
without bothering about the detailed subtleties concerning the logarithmic
vrs power law running. So we will simply use the generic form for the
evolution equation suggested by Eq. (17) i.e.
\begin{equation}
\alpha_i^{-1}(M_Z) = \alpha^{-1} + {b_i\over 2\pi}
\ln \left( {\Lambda\over M_Z}\right) + {\tilde{b}_i\over 2\pi}
F_\delta\left( {\Lambda\over \mu_0}\right),
\label{rgef}
\end{equation}
being $\alpha$ the unified coupling.
It is clear from Eq. (27) that the information that comes from the bulk
can be separated into
two independent parts: all the structure of the spectra of the KK tower,
defined by the compactification scale and the number of extra dimensions is
completely embedded into the $F_\delta$ function, in such a way that its
contribution to the running of the gauge couplings is actually model
independent. The model dependence coming from such questions as to which
representations are in bulk etc are all encoded in the beta functions
$\tilde{b}_i$.
In other words, no matter how the fields of the low energy four dimensional
theory are distributed among the bulk and the wall, the $F_\delta$ function is
not affected and conversely, changes in the KK tower, namely the splitting
in the
compactification radii, affect only the form of $F_\delta$ through the
changes in the internal mass spectra [Eq.(\ref{fdeg})].
Notice that Eq. (\ref{rgef}) is similar to that of the two step unification
model, where a new gauge symmetry appears at an intermediate energy scale.
It was already noted sometime ago~\cite{mohapatra} that
the solutions to the renormalization group equations in those models are
very constrained by the one step unification in the MSSM. The
argument goes as follows: let us define the vectors
${\bf b}= (b_1,b_2,b_3)$;
$\tilde{\bf b}= (\tilde{b}_1,\tilde{b}_2,\tilde{b}_3)$;
${\bf a} = (\alpha_1^{-1}(M_Z),\alpha_2^{-1}(M_Z),\alpha_3^{-1}(M_Z))$ and
${\bf u} = (1,1,1)$ and
note that Eq. (\ref{rgef}) takes the simplest vectorial form
\begin{equation}
{\bf a} = \alpha^{-1} {\bf u} + x {\bf b} + y \tilde{\bf b}
\label{rgev}
\end{equation}
where $x = \ln(\Lambda/M_Z)/2\pi$ and $y = F_\delta/2 \pi$. From here is
easy to eliminate variables, for instance $\alpha$ and $x$
to get
\begin{equation}
\Delta\alpha\equiv ({\bf u}\times {\bf b})\cdot {\bf a} =
[({\bf u}\times {\bf b})\cdot \tilde{\bf b}]\ y.
\label{dalpha}
\end{equation}
In the MSSM ${\bf b} = (33/5,1,-3)$. Now, inserting the experimental
values~\cite{pdg}
\begin{eqnarray}
\alpha^{-1}_1(M_Z) &=& {3\over 5}\alpha^{-1}_Y(M_Z)
= 58.9946 \pm 0.0546, \nonumber\\
\alpha^{-1}_2(M_Z) &=& 29.571\pm 0.043, \quad\mbox{and} \label{alphas}\\
\alpha^{-1}_3(M_Z) &=& \alpha_s^{-1}(M_Z) = 8.396\pm 0.127. \nonumber
\end{eqnarray}
we get $\Delta\alpha_{expt} = 0.928\pm 0.517$. So, $\Delta\alpha_{expt}$
is consistent
with zero within two standard deviations. One step unification condition
is of course given by $\Delta\alpha=0$. Thus, assuming one step
unification, we find that the Eq. (\ref{dalpha}) reduces to the
constraint~\cite{mohapatra}
\begin{equation}
(7\tilde{b}_3 - 12\tilde{b}_2 + 5 \tilde{b}_1)F_\delta= 0;
\end{equation}
since $[({\bf u}\times {\bf b})\cdot \tilde{\bf b}] = -4(7\tilde{b}_3 -
12\tilde{b}_2 + 5 \tilde{b}_1)/5$.
This equation was first written down in Ref.~\cite{mohapatra} for the case
of generic unification with an intermediate scale and has subsequently
been re-used in several recent papers\cite{quiros,frampton}.
There are two solutions to the last equation:
\begin{itemize}
\item[(a)] $F_\delta(\Lambda/\mu_0)=0$, which means $\Lambda = \mu_0$,
bringing us back to the MSSM by pushing up the compactification scale to
the unification scale.
\item[(b)] Assume that the beta coefficients $\tilde{b}$ conspire to eliminate
the term between brackets:
\begin{equation}
(7\tilde{b}_3 - 12\tilde{b}_2 + 5 \tilde{b}_1) = 0 .
\label{cons}
\end{equation}
The immediate consequence of Eq. (\ref{cons}) is the indeterminacy of
$F_\delta$, which means that we may put $\mu_0$ as a free parameter in the
theory. For instance we could choose $\mu_0\sim$ 10 TeV to maximize the
phenomenological impact of such models~\cite{pheno}.
It is compelling to stress that
this conclusion is independent of the explicit form of $F_\delta$.
\end{itemize}
Furthermore, as
$ [({\bf u}\times {\bf b})\cdot \tilde{\bf b}] =
(b_1- b_3)(\tilde{b}_2- \tilde{b}_3) - (b_2- b_3)(\tilde{b}_1- \tilde{b}_3);
$
Eq. (\ref{cons}) is equivalent to the condition obtained by DDG expressed by
\begin{equation}
{B_{12}\over B_{13} } = {B_{13}\over B_{23} } = 1; \qquad \mbox{where}\qquad
B_{ij} = {\tilde{b}_i- \tilde{b}_j\over b_i- b_j} .
\end{equation}
What is clear at this point, is that the DDG minimal model can not satisfy Eq.
(\ref{cons}) by itself, since it contains the MSSM fields plus extra fermionic
and scalar matter not matching that of the MSSM~\cite{ddg2}, for the fields in
the bulk are accommodated in $N=2$ hyperrepresentations, implying a higher
prediction for $\alpha_s$ at low $\mu_0$~\cite{ross}. Indeed, in this case we
have $(7\tilde{b}_3 - 12\tilde{b}_2 + 5 \tilde{b}_1) = -3$. However, as
Carone~\cite{carone} showed, there are some scenarios where option (b) may be
realized. In those cases, the MSSM fields are distributed in a nontrivial
way among
the bulk and the boundaries. There are also other possible extensions to those
scenarios. We may add matter to the MSSM, as long as we do not affect the
constraint (\ref{cons}), in order to get the same MSSM accuracy for
$\alpha_s$. Also, bulk fields with non zero modes could be added to the scheme
to satisfy Eq. (\ref{cons}). Those cases have been considered in
references~\cite{kaku,quiros,frampton}.
In fact looked from this perspective, the embedding of the standard model
into higher dimensional space-time looks worse. The reason is that
since ${\bf b}^{SM} = (41/10,-19/6,-7)$ we have
$\Delta\alpha_{expt}=41.13\pm 0.655\neq 0$;
as is well known this is the reason for the failure of grand unification
of the standard model. Now let us form a
second linear combination obtained from (\ref{rgev}), namely
\begin{equation}
\tilde{\Delta}\alpha\equiv ({\bf u}\times \tilde{\bf b})\cdot {\bf a} =
[({\bf u}\times \tilde{\bf b})\cdot {\bf b}]\ x =
- [({\bf u}\times {\bf b})\cdot \tilde{\bf b}]\ x .
\label{delta2}
\end{equation}
Eq. (34) implies that in order to get a good solution where $x$ and $y$
are positive (as they are by definition since
$\Lambda >M_Z$ and $\Lambda >\mu_0$), the following
constraint must be satisfied
\begin{equation}
Sign(\Delta\alpha)
= Sign[({\bf u}\times {\bf b})\cdot \tilde{\bf b}]
= - Sign(\tilde{\Delta}\alpha) .
\label{cons2}
\end{equation}
However, in the minimal model where all fields are assumed to have KK
excitations $\tilde{\bf b} = (1/10,-41/6,-21/2) + (4/3,4/3,4/3)\eta$, with
$\eta$ the number of generation in the bulk, we get
$\tilde{\Delta}\alpha = 38.973\pm 0.625$; and $({\bf u}\times {\bf b}^{SM})\cdot
\tilde{\bf b}^{SM} = 1/15$. Hence, the constraint (\ref{cons2}) is not
fulfilled and unification does not occur. Moreover, these results mean
that $\Lambda< M_Z$, which clearly is an inconsistent solution. However,
extra matter could of course improve this
situation~\cite{quiros,frampton}.
Note further that strictly speaking $\Delta\alpha_{expt}\neq 0$ in the
MSSM within the
experimental accuracy of one standard deviation and thus the right
condition that must apply over
and above Eq. (\ref{cons2}) is
\begin{equation}
\alpha^{-1} =
{ ({\bf b}\times \tilde{\bf b})\cdot {\bf a} \over
({\bf u}\times {\bf b})\cdot \tilde{\bf b} } >1
\label{cons3}
\end{equation}
obtained also from Eq. (\ref{rgev}), which insures that $\alpha$ remains in
the perturbative regime. Therefore, whenever the conditions (\ref{cons2}) and
(\ref{cons3}) are satisfied by $\tilde{\bf b}$, a unique solution for
$\Lambda$, and $\mu_0$ can be obtained from eqs. (\ref{dalpha}) and
(\ref{delta2}).
Let us now turn to the cases where
the normalization of the gauge coupling differs from that of SU(5)
introduced above. Indeed for a gauge group $G$ with coupling constant
$\alpha$, where the simple group $G_i$ is embedded, the coupling constant
$\alpha_i$ of $G_i$ is related to $\alpha$ by a linear relationship
$\alpha_i = c_i \alpha$, with the (embedding) factor defined by
\begin{equation}
c_i = {Tr\ T^2 \over Tr\ T_i^2} ;
\end{equation}
where $T$ is the generator of $G_i$ normalized over a representation $r$ of
$G$, and $T_i$ is the same generator but normalized over the representations
of $G_i$ contained into $r$; the traces run over the complete representation
$r$~\cite{ponce}.
In Table I we present the embedding $c_i$ factors for several models of
general interest as they were introduced in references~\cite{models}. The
first
entry in that table correspond to most of the models in the literature. In
models with $c_3 = {1\over 2}$ the color group SU(3) is embedded through the
chiral color extension~\cite{2su3} SU(3)$_{cL}\times$SU(3)$_{cR}$. In general
$c_2 = 1/F$, where $F$ is the number of families contained in the same
representation~\cite{comment}. $c_1$ depends in how the hypercharge is embedded
into the group and is given by $c_1^{-1} = {1\over 2} Tr (\frac{Y}{2})^2$.
In SU(5), for
instance, $(c_1,c_2,c_3)= ({3\over 5},1,1)$. These are the canonical values.
They reflect the absence of extra fields in the fermionic representations with
non trivial quantum numbers. The group $[SU(4)]^3\times Z_3$ in Table I is
not the vector-like color version of the two family $SU(4)_{color}$
model, but it
is the one family model introduced in reference~\cite{models}. Last two models
in Table I are three family $SU(4)_{color}$~\cite{patis} models.
We are considering these models as
examples of semisimple unified theories.
The evolution equations are still given by (\ref{rgev}), but with the
$c_i$ factors included in both the gauge coupling constants,
${\bf a} = (c_1\alpha_Y^{-1}(M_Z),
c_2 \alpha_2^{-1}(M_Z),c_3 \alpha_3^{-1}(M_Z))$
and beta functions
${\bf b}= (c_1 b_Y,c_2 b_2,c_3 b_3)$;
$\tilde{\bf b}= (c_1\tilde{b}_Y,c_2 \tilde{b}_2,c_3 \tilde{b}_3)$. The
unnormalized beta functions for the MSSM are
$(b_Y,b_2,b_3) = (11,1,-3)$ and for the SM
$(b^{SM}_Y,b^{SM}_2,b^{SM}_3) = (41/6,-19/6,-7)$. Using these values and
the experimental data (\ref{alphas}), we find that $\Delta\alpha\neq 0 $
for all those models (Table I also). The reason is that in non canonical
models extra scalars at the $M_Z$ scale are required in order to achieve
unification in both the MSSM and SM~\cite{ponce}.
Now, we proceed as follows. We first assume all gauge bosons as high dimensional
fields, then, for models in Table I we explore the scenarios
where the other MSSM (SM) fields propagate into the bulk: Higgs and/or fermion
fields.
In the analysis we always
consider all the three families together. The contribution to the unnormalized
$\tilde{b}_i$ from these particles are given in Table II and they have been
calculated assuming that mirror fields are contained in all chiral $N=2$
hypermultiplets of the KK tower. For those scenarios that fulfill the
conditions above we calculate $\Lambda$, $\mu_0$ and $\alpha$ using both, the
power law and the logarithmic approaches for $\delta=1$ using the central
experimental values of the gauge coupling constants and $M_Z=91.186$ GeV
as inputs. The results are shown in Table III. We have not found solutions for
the SM with this minimal content.
What is important to emphasize from these results is that in the supersymmetric
SU(5) class of models (all models with canonical normalization), the
scenario where all the
fields live in the bulk
predicts a unification scale at $10^{16}$ GeV and a compactification
scale is slightly below this scale. Here the small $\Lambda/\mu_0$ ratio
makes it important to consider logarithmic approach rather than the power
law. As a matter of fact, only the first one or two levels of the tower
contributes to the renormalization group equations. Thus,
it means that at the one loop level only few thresholds near the
unification scale are required to predict the right value for
$\alpha_s\approx 0.119$, bringing single scale MSSM models into better
agreement with experiments.
Comparing Table III with previous results for the SM (see
references~\cite{quiros,frampton}),
we may note that the scenarios where a solution
could be obtained are very different for the MSSM than for the SM (whether
we choose canonical or noncanonical normalization).
The reason is again that the SM by itself require in
general a large number of extra chiral scalars or fermions to get
unified~\cite{ponce}. Part of
these kind of fields are provided by supersymmetry,
but as it follows from Table I, only in the canonical class do the new
fields of supersymmetry bring $\Delta\alpha$ close to zero.
When SM is embedded into higher dimensions, a new class of
contributions corresponding to excited scalar or fermionic
modes of the KK tower emerge\cite{frampton} and without any need for
supersymmetry, they lead to unification of gauge couplings. It must be
pointed out however that in most cases extra chiral fields carrying color
quantum numbers make $\alpha$ run into the non perturbative regime
of the theory. On the other hand, in the MSSM, extra scalars and fermions are
contained in the hypermultiplets of the KK modes; since there are more
particles contributing to beta functions, their contributions should be
pushed up to high mass scales in order to preserve unification.
Let us now comment on the effect of the two loop contributions to the
coupling evolution. Since in this case the evolution equations change, in
order to use our method to understand the results for unification, we
redefine the values of $\alpha^{-1}_i(M_Z)$ by moving the two loop
contributions to the left hand side of the evolution equations.
Now, since two loop corrections tend to increase the one loop predictions for
$\alpha_s$, once we subtract them from $\alpha^{-1}_s(M_Z)_{expt}$, it
tends to make $\Delta\alpha$ negative in the case of MSSM. This upsets
the consistency of Eq. (\ref{cons2}) and makes the possibility of
extra dimensions below the GUT scale less likely~\cite{ross}.
This can be avoided if one changes the scenario by including extra
particles or by picking up some of those in the
bulk~\cite{kaku,carone,quiros,frampton}. Along this line, a
simple possibility is to let the gauge bosons propagate in the bulk,
a case which seems not to have been considered so far.
Our finding in this case including two loop corrections is that
we get the predictions: $\Lambda\sim
1.282\times 10^{14}$ GeV; $\mu_0\sim 2.4 \times 10^{13}$ GeV and $\alpha^{-1} =
29.6122$.
On the other hand, notice that the SM and supersymmetric non canonical
models ( such as SU(5)$\times$SU(5)) are not so sensitive to the
inclusion of the two loop effects, since
$\Delta\alpha$ starts out as a large positive number and retains its sign
when two loop effects are subtracted as indicated above.
This has the implication that our predictions only will change
slightly when two loop corrections are included. A particularly
interesting outcome in the case of models with noncanonical is that
there are several cases where, the unification scale comes out to be
around $10^{11}$ GeV ( see Table III, cases $SU(5)\times SU(5)$,
$[SU(3)]^4$, $[SU(6)]^4$ ). These models fit nicely into the new
intermediate string scale models recently proposed in ~\cite{quevedo}.
Turning this point around, one could presume that the preferred GUT
group in the case of such string models would be the ones such as
$[SU(5)]^2$ etc rather than the canonical SO(10) or SU(5). It is worth
noting that such models have a number of phenomenologically desirable
features such as automatic R-parity conservation, no baryon violation
from the gauge theory etc. In fact, it is the property of automatic
baryon number conservation that makes such low unification scales
phenomenologically acceptable.
\section{Unification in two step models with extra dimensions}
Two steps unification models are of great current interest mainly motivated by
neutrino physics~\cite{nphys,see-saw}. In them
the general picture is as follows: at the $M_Z$ scale we have the MSSM
theory,
which remains valid up to certain intermediate scale $M_I$. Hereafter a new
gauge symmetry $G'$ rules the evolution of the gauge couplings
up to the unification scale $M$. In this framework the one loop
renormalization group equations are given by
\begin{equation}
\alpha_i^{-1}(M_Z) = \alpha^{-1}_M + {b_i\over 2\pi}
\ln \left( {M_I\over M_Z}\right) + {b'_i\over 2\pi}
\ln\left( {M\over M_I}\right),
\label{rg2s}
\end{equation}
where $\alpha_M$ is the unified coupling and $b'_i$ the beta functions of the
$G'$ theory. This equation resembles the case discussed in the previous
section, and we will therefore try the same procedure to solve it.
In terms of $\Delta'\alpha\equiv ({\bf u}\times{\bf b}')\cdot {\bf a}$,
the condition
to get a good solution where the hierarchy $M>M_I>M_Z$ is fulfilled now
reads
\begin{equation}
Sign(\Delta\alpha)
= Sign[({\bf u}\times {\bf b})\cdot {\bf b}']
= - Sign(\Delta'\alpha) .
\label{con2s}
\end{equation}
Again, the unification in the MSSM (canonical models), will imply that
$M_I\sim M$ unless $(7b'_3 - 12b'_2 + 5 b'_1) \approx 0$.
Some examples where the value of $\alpha_s$ was used to make the
$\Delta \alpha_{expt}$ nonzero and thus realize this scenario
were presented in
ref.~\cite{mohapatra} in the context of the left right model (LRM)~\cite{lrm}.
Other possible ways to realize intermediate scales would be to add extra
scalars at the weak scale to MSSM, in which case the vector ${\bf b}$
changes again making $\Delta \alpha_{expt}\neq 0$ and thereby opening up a
way to have an intermediate scale; another way would be to consider
non canonical models, where the change in the normalization again leads to
$\Delta \alpha_{expt}\neq 0$ allowing now the intermediate scale~\cite{ponce2}.
Let us suppose that KK modes of the $G'$ theory appear at certain scale
$\mu_0$
below the unification scale. Once the gauge couplings cross $\mu_0$, the
steepness of the running will imply the change of the unification scale, and
in the worst case the loss of unification. To fix this problem the
intermediate scale may be moved to a proper value $M_I'$ to restore
the unification at a new
scale $\Lambda$. In the presence of extra dimensions, the
renormalization group equations are written as
\begin{equation}
\alpha_i^{-1}(M_Z) = \alpha^{-1}_\Lambda + {b_i\over 2\pi}
\ln \left( {M'_I\over M_Z}\right) + {b'_i\over 2\pi}
\ln\left( {\Lambda\over M'_I}\right) +
{\tilde{b}'_i\over 2\pi} F_\delta\left( {\Lambda\over \mu_0}\right).
\label{rg3s}
\end{equation}
where $\alpha_\Lambda$ is the new unified coupling and $\tilde{b}'$ the beta
function of the excited modes. Now, to understand the role of $\mu_0$,
we proceed as before by defining the
vectors ${\bf u}= (1,1,1)$; ${\bf b} = (b_1,b_2,b_3)$;
${\bf b}'=(b'_1,b'_2,b'_3)$;
$\tilde{\bf b}'=(\tilde{b}'_1,\tilde{b}'_2,\tilde{b}'_3)$.
It then follows from Eq. (40) that the following consistency condition
must be satisfied:
\begin{equation}
\ln\left( {M'_I\over \Lambda}\right) =
\left[ {({\bf u}\times {\bf b} )\cdot \tilde{\bf b}' \over
({\bf u}\times {\bf b})\cdot {\bf b}' }\right]\
F_\delta\left( {\Lambda\over \mu_0}\right)
\label{split}
\end{equation}
Note that the condition for the existence of an intermediate scale
in the presence of extra dimensions is not as strigent as it is for the
case without them. In the latter case, we must have vanishing of
$({\bf u}\times {\bf b})\cdot {\bf b}'$ and this normally means very
precise cancellation among the beta function coefficients that occurs
only rarely. And cases where
an intermediate scale could not occur before can now support such scales.
However, in order to get the right hierarchy $\Lambda > M'_I$ the bulk fields
must then satisfy
\begin{equation}
Sign[({\bf u}\times {\bf b} )\cdot \tilde{\bf b}'] =
- Sign[({\bf u}\times {\bf b})\cdot {\bf b}'].
\label{consplit}
\end{equation}
Lets assume now the left right model
$G_{LR}\equiv SU(3)_c\times SU(2)_L\times
SU(2)_R\times U(1)_{B-L}$ as the intermediate theory. Let us also assume
as the content of the Higgs sector above the $M_I$ scale $n_T$ right
handed pairs of
triplets $\Delta_R(1,1,3,-2)+ \bar{\Delta}_R(1,1,3,2)$, $n_B$ bidoublets
$\phi(1,2,2,0)$ and perhaps $n_L$ pairs of left handed doublets
$\chi_L(1,2,1,1) + \bar\chi_L(1,2,1,-1)$ and $n_R$ pairs of right handed
doublets $\chi_R(1,1,2,1) + \bar\chi_R(1,1,2,-1)$. With this content
\[
{\bf b}'=\left({12\over 5},0,-3\right)+
\left({3\over 5}(10 n_T + 2n_R + n_B+n_L),n_B+n_L,0\right),
\]
and thus
\begin{equation}
({\bf u}\times {\bf b})\cdot {\bf b}' =
{12\over 5} (3 (n_B+n_L)-10 n_T - 2 n_R + 3).
\label{crosslr}
\end{equation}
In the simplest cases where $n_B=1,2$ and $n_T=1$ without doublets, which are
actually the minimal and next to minimal scenarios in the supersymmetric
version of the LRM~\cite{susylr} where the see-saw~\cite{see-saw} mechanism is
naturally implemented, we have $({\bf u}\times {\bf b})\cdot {\bf b}'<0$,
and $\Delta'\alpha>0$; respectively. Eventually, within the
experimental
accuracy this result means that at the one loop order a wrong hierarchy
$M_I>M>M_Z$ obtains.
Clearly this problem could be fixed by adding more scalars, but also two loop
corrections may fix this problem in a natural way, even though we will still
have $M_I\sim M$. What is important to emphasize here, is that the
condition (\ref{consplit}) now means that
$({\bf u}\times {\bf b})\cdot \tilde{\bf b}'$ must be positive.
Now we procced to consider the possible contents of the bulk. The contribution
of the LRM $N=2$ hyperrepresentations to the beta functions $\tilde{\bf
b}'$ and to Eq. (\ref{split}) are shown in Table V, assuming mirror particles
in all fermionic and scalar representations. From the same table we can note
that there are several scenarios with the right positive sign to change the
two step prediction of the model. As an interesting example we have considered
the case where only all the gauge bosons propagate into the bulk for the
minimal and next to minimal case with one extra dimension and plotted in figure
3 the splitting effect produced for these KK modes over the unification scale
as a function of the ratio $\Lambda/\mu_0$ by using the logarithmic approach.
In order to stress this effect we have assumed for the plots that
$\Delta\alpha=0$, meaning the MSSM accuracy on $\alpha_s$. The correction
introduced by assuming the experimental accuracy of $\alpha_s$ only
will
produce an initial splitting, as we argue above. However, the total behaviour
and our conclusions still remain.
In Fig. 4 and 5, we plot the running of the gauge couplings in this model
with one extra dimension and with only gauge bosons propagating in the
bulk. We find that the couplings unify with values of $M_I=M_{W_R}$
anywhere
from a TeV to $10^{16}$ GeV. It is however important to realize that the
values of the unification scale depends on $M_{W_R}$. If we want the
unification group to be SO(10), constraints of proton life time would
require that $\Lambda \geq 10^{15}$ GeV. This requires as we see from
Fig. 4 that $M_{W_R}\simeq 10^{13}$ GeV (for the case of two bidoublets).
This value of the intermediate scale is of course what is required for
understanding the neutrino oscillation phenomena. The low value of
$\Lambda$ also means that $p\rightarrow e^+ +\pi^0$ mode mediated via
the gauge boson exchange is now in the range accessible to the
super-Kamiokande experiment\cite{superK}.
As the value of the $M_{W_R}$ is lowered, the unification scale also goes
down. In these cases, considerations of proton decay would suggest that
the GUT group be some group other than SO(10) and furthermore string
related discrete symmetries be invoked to prevent catastrophic proton
decay.
It is worth mentioning that these
result do not change if we add the three families of fermions to the
bulk; however, the addition of the Higgs fields may produce the wrong sign
and then, this scenario
will split the unification scale with an inconsistent hierarchy.
There is another interesting case where the splitting is produced with the
right hierarchy. Lets assume that only the MSSM fields develop KK modes
(the gauge, Higgs and the matter).
If this is the case, the appearance of extra dimensions (i.e. $\mu_0\ll
\Lambda$), below the unification scale pushes down the left right scale
to intermediate energies. This is clear from Figure 3. It is then clear
that if we accept that the left right scale has a lower bound about
800 GeV~\cite{pdg2}, this does impose a lower bound on the
compactification
scale $\mu_0$ about $10^{11}$ GeV. Moreover, this bound seems to be
independent of the bulk content since we obtain similar results for the cases
considered above. Other less trivial
scenarios that mix the LRM fields may produce a similar splitting but we
do not consider them here.
From figure 3 we also note that for $\delta=1$ only the first few
thresholds are
required to get the right hierarchy $M_Z<M_I<\mu_0<\Lambda$.
In contrast with the analysis performed in the previous section, now the two
loop contributions will not affect our conclusions since so far they will
only add an extra contribution to the splitting on the right way.
On the other hand, as in non canonical models and in those which add extra
matter to the MSSM we may expect to start with an initial and may be more
important splitting as a consequence of the absence of one step unification,
the condition over the bulk matter (\ref{consplit}) will not apply any more,
unless the specific two step model predicts an initial wrong hierarchy, as it
really happens in the minimal LRM as discussed above. Otherwise, the bulk
content is not constrained and the contributions of the KK modes will only
change the initial hierarchy. However, as we expect
that $M_I$ always be smaller than $\mu_0$, whatever be the change, this
phenomenological condition will introduce a lower bound to the compactification
scale, although this bound will be very model dependent.
Finally, we note that keeping $\alpha_\Lambda$ in the
perturbative range also puts some constraints on the models as can be seen
from the following equation:
\begin{equation}
\alpha_\Lambda^{-1} = \alpha_M^{-1} -
\left[ {({\bf b}\times {\bf b}' )\cdot \tilde{\bf b}' \over
({\bf u}\times {\bf b})\cdot {\bf b}' }\right]\
F_\delta\left( {\Lambda\over \mu_0}\right) .
\end{equation}
Now, if the coefficient of $F_\delta$ is negative, $\alpha_\Lambda$ will
becomes smaller and the theory shall remains perturbative. This is actually
the case in the scenarios presented above, as long as fermions do not develop
KK modes. Otherwise, $\alpha_\Lambda$ will increase very quickly, following the
steepness of $F_\delta$, and for small ratios of $\Lambda/\mu_0$ it will
cross to the non perturbative regime.
\section{Conclusion}
In conclusion, we have compared the logarithmic running with power law
running in theories with higher dimension and pointed out by explicit
examples that for realistic cases where, only few KK modes contribute,
the power law running may not accurately reflect the correct situation to
a high precision.
We then analyze the general conditions to achieve unification in
presense of extra dimensions. We find that [Eq.(\ref{cons2})] provides
a generic constraint on the nature of the bulk fields for the
compactification scale to be below the unification scale.
We have considered both the cases where the low energy group is embedded
into the GUT group in a canonical or noncanonical manner and derive the
unification as well as the compactification scales from LEP data on the
low energy couplings. We find that in
supersymmetric SU(5) and other canonical models, the experimental accuracy at
one loop level requires only the first excited modes of the MSSM fields to
appear below the unification scale. However, this model will predict
$\alpha_s$ at the MSSM accuracy at two loop order. A new scenario not
considered before in the literature is the case where only the gauge
bosons propagate into the bulk. In this case the one loop order prediction for
$\alpha_s$ is small enough and is such that the two loop effects push it
within the experimental error with an unification scale $\Lambda\sim
10^{14}$ GeV and $\mu_0\sim 10^{13}$ GeV.
Finally, considering the effect of extra dimensions on the two step models, we
found that in those models with the MSSM as the low energy limit,
the presence of the KK excited modes makes it easier to have an
intermediate scale. We derive the constraint that
the models of this type must satisfy. We provide examples where, the
existence of an
$M_I$ depends on the bulk content. On the other hand,
if only the MSSM fields develop excited modes, as it was assumed
in the DDG minimal model, their contributions can split the unification
scale, setting the supersymmetric left right model at intermediate energies.
We have found several examples of models where we have the
ordering of scales as $M_{W_R} \leq \mu_0 \leq \Lambda $. A particular
scenario where LRM gauge bosons are the only particles propagating in the
bulk (besides the graviton), we obtain the $W_R$ scale at $10^{13}$ GeV
which is of great interest for neutrino masses. Since the unification
scale in this case is of order $10^{15}$ GeV, it predicts proton life time
accessible to the ongoing Super-Kamiokande experiment.
We have also found some scenarios, where it is possible to get a lower
bound on the scale of the extra dimensions from two step unification and
present experimental lower bound on the $W_R$ scale.
\section*{Acknowledgments.}
APL would like to thank the kind hospitality of the members of the Department
of Physics at U. MD. The work of APL is supported in part by CONACyT
(M\'exico). The work of RNM is supported by a grant from the National
Science Foundation under grant number PHY-9802551.
|
\section{Introduction}
In the same way as complex numbers are associated
with the Euclidean geometry, the two other
systems of bidimensional hypercomplex numbers
can be associated with geometries of physical relevance.
The parabolic numbers can be associated with the Galileo group
and the hyperbolic numbers with the Lorentz group of special relativity
\cite{Yag79,Sob95}.
The Hamilton quaternions as well as
the bidimensional systems are included in the more general Clifford
algebras \cite{Hes84,Kel94,Cli73,Cli78}. They can represent the Euclidean geometry because
their invariant quantity is an algebraic quadratic form as well as
the Euclidean and pseudo-Euclidean invariants.
The hyperbolic numbers offer the possibility to represent
the four-compo\-nent Dirac spinor as
a two-component hyperbolic spinor. Hucks has shown \cite{Huc93} that the Lorentz group
is equivalent to the hyperbolic unitary group and that
the operations of C, P, and T on
Dirac spinors are closely related to the three types of complex conjugation
that exist when both hyperbolic and ordinary imaginary
units are present.
Since the relativistic spin group is representable as an unitary group, the special linear
group, which is normally used to represent the relativistic spin,
has to be considered as an unitary group as well. Porteous \cite{Por81,Por95} proves
the unitarity of the special linear group
with the help of the double field, which
corresponds to the null basis representation of the hyperbolic numbers.
There are various applications of hyperbolic numbers.
They have been applied by Reany \cite{Rea93} to
2nd-order linear differential equations. A function theory for hyperbolic numbers
has been presented by Motter and Rosa \cite{Mot98}. Extensions to an n-dimensional
space have been given by several authors \cite{Rea94,Fje98,Fje298,Wum02}, including
an analysis of hyperbolic Fourier transformations \cite{Zhe04}.
The functional calculus of hyperbolic numbers is also covered by the
more general approach of Superanalysis (see e.g. Khrennikov \cite{Khr99}).
Considerations of a non-relativistic hyperbolic Hilbert space with respect to the Born formula
have been given by Kocik \cite{Koc99}, Khrennikov \cite{Khr03,Khr203},
and Rochon and Tremblay \cite{Tre04}.
Xuegang et al. investigated
the Dirac wave equation, Clifford algebraic spinors, a hyperbolic Hilbert space,
and the hyperbolic spherical harmonics in hyperbolic spherical polar coordinates
\cite{Xue00,Xue200,Xue01}.
The hypercomplex numbers,
and the geometries generated
by these numbers, have been investigated by Catoni et al. \cite{Cat05}.
Further applications of hyperbolic numbers, including e.g.
their application to general relativity by Kunstatter et al.,
can be found in \cite{Cap41,Mac69,Kun83,Fje86,Hes91,Kwa92,Yu95,Xue99}.
It has been shown by Baylis and Jones \cite{Bay89} that a $\bfa{R}_{3,0}$ Clifford algebra
has enough structure to describe relativity as well as
the more usual $\bfa{R}_{1,3}$ Dirac algebra or the $\bfa{R}_{3,1}$ Majorana algebra.
Baylis represents relativistic space-time points as paravectors and applies
these paravectors to Electrodynamics \cite{Bay99}.
The approach to relativity in terms of $\bfa{R}_{1,3}$ has been investigated
by Hestenes \cite{Hes84,Hes66,Hes03}, or e.g. by Gull, Lasenby, and Doran \cite{Gul93}.
An overview of the structural differences in the above low-dimensional Clifford algebras
is given by Dimakis \cite{Dim89} in the context of a general spinor
representation within Clifford algebras.
The approach used in this work is congruent to Baylis paravectors.
However, hyperbolic numbers are included in the algebra. This corresponds
actually to a hyperbolic complexification of the Baylis algebra. According to
Porteous \cite{Por95} the resulting algebra is
isomorphic to the three-dimensional universal complex Clifford algebra $\bar{\bfa{C}}_{3,0}$.
The correspondance to the Baylis algebra is given by the restriction to
a subset of the algebra.
This restriction can be justified in the hyperbolic Hilbert space by the hermiticity of
the Poincar\'e mass operator \cite{Ulr052}.
This work is an extension of \cite{Ulr051}. It introduces beside the
Poincar\'e mass operator an analogeous operator for the spin, which is
related to the Pauli-Lubanski vector. In addition, the approach is
extended to spherical symmetries.
The mathematical background of the presented
approach with respect to the classification of Clifford algebras
is not fully explained in \cite{Ulr052,Ulr051}. This will be
improved in this work based on the general overview given by Porteous \cite{Por81,Por95}. The
notation of Porteous has been adopted in many cases.
\section{Hyperbolic numbers}
\label{not}
Vector spaces can be defined over the commutative ring of
hyperbolic numbers $z\in\bfa{H}$
\begin{equation}
\label{beg}
z=x+iy+jv+ijw\;,\hspace{0.5cm}x,y,v,w \in\bfa{R}\;,
\end{equation}
where the complex unit $i$ and the hyperbolic unit $j$ have the properties
\begin{equation}
i^2=-1\;,\hspace{0.5cm}j^2=1\;.
\end{equation}
The hyperbolic numbers defined in this way are a commutative extension of the complex numbers to
include new roots of the polynomial equation $z^2 - 1 = 0$.
In the terminology of Clifford algebras
they are represented by $\bar{\bfa{C}}_{1,0}$, i.e.
they correspond to the universal one-dimensional complex Clifford algebra
(the notation follows Porteous \cite{Por95}).
Beside the grade involution, two anti-involutions play a major role in the
description of Clifford algebras and their structure, conjugation and reversion.
Conjugation changes the sign of the complex and
the hyperbolic unit
\begin{equation}
\label{conj}
\bar{z}=x-iy-jv+ijw\;.
\end{equation}
The hyperbolic numbers with conjugation
are equivalent to the double
field $^2\bar{\bfa{C}}^\sigma$ of Porteous, where $\sigma$ and the bar symbol denote
swap and complex conjugation. The notation of Porteous has the advantage that it can clearly specify, whether
the double field is defined over the real $^2\bfa{R}$, complex $^2\bfa{C}$,
or quaternionic numbers $^2\bfa{Q}$. If necessary the notation used in this work
is extended to $\bfa{H}(\bfa{R})$,
$\bfa{H}(\bfa{C})=\bfa{H}$, or $\bfa{H}(\bfa{Q})$, where $\bfa{H}(\bfa{R})$ and
$\bfa{H}(\bfa{Q})$ correspond to the universal Clifford algebras
$\bfa{R}_{1,0}$ and $\bfa{R}_{0,3}$, respectively.
With respect to the Clifford conjugation the square of the
hyperbolic number can be calculated as
\begin{equation}
\label{square}
z\bar{z}=x^2+y^2-v^2-w^2+2ij(xw-yv)\;,
\end{equation}
i.e. in general the square of a hyperbolic number is not a
real number.
Beside the conjugation, the second important anti-involution is the reversion, which
changes only the sign of the complex unit
\begin{equation}
\label{cconj}
z^{\dagger}=x-iy+jv-ijw\;.
\end{equation}
Anti-involutions reverse the ordering in the multiplication,
e.g. $(ab)^\dagger=b^\dagger a^\dagger$. This becomes
important when non-commuting elements of an algebra are considered.
In physics, reversion is denoted as hermitian conjugation. Note, that in \cite{Ulr052} it has been
suggested to relate hermiticity in the physical sense to the conjugation anti-involution.
\section{Hyperbolic algebra}
\label{vec}
Consider a hyperbolic vector with the coordinates $z^\mu=(z^0,z^i)\in\bar{\bfa{H}}^{\,3,1}$.
The bar symbol indicates that the vector has the signature $(3,1)$ in the hermitian
product with respect to conjugation, i.e. $z_\mu \bar{z}^\mu=z_0\bar{z}^0-z_i\bar{z}^i$.
The vector can be represented in terms of a Clifford algebra as
\begin{equation}
\label{veco}
Z=z^\mu e_\mu\;.
\end{equation}
The basis elements $e_\mu=(e_0,e_i)$ include the
unity and the Pauli algebra
multiplied by the hyperbolic
unit~$j$
\begin{equation}
e_\mu=(1,j\sigma_i)\;.
\end{equation}
The pseudoscalar of the hyperbolic algebra, which will appear thoughout this work,
is defined as
\begin{equation}
I=e_0\bar{e}_1e_2\bar{e}_3=ij\;.
\end{equation}
Important for the further analysis is the behaviour of the above hypercomplex units under
conjugation and reversion. In Table \ref{invo} it is displayed whether the sign
of the unit is changed or not under the considered operation.
For completeness the graduation as an example of an important involution is displayed.
An involution
does not change the ordering in a product, i.e. $\widehat{ab}=\hat{a}\hat{b}$.
The graduation is used to identify the even elements of a Clifford algebra.
A certain subset of these elements defines the spin group of the considered Clifford algebra \cite{Por95}.
\begin{table}
\begin{center}
\begin{tabular}{|c|c|c|c|}
\hline
$a$ & $\bar{a}$ & $a^\dagger$ & $\hat{a}$ \\
\hline
$e_i$ & $-$ & $+$ & $-$\\
\hline
$\sigma_i$ & $+$ & $+$ & $+$\\
\hline
$I$ & $+$ & $-$ & $-$\\
\hline
$i$ & $-$ & $-$ & $+$\\
\hline
$j$ & $-$ & $+$ & $-$\\
\hline
\end{tabular}
\end{center}
\caption{Effect of conjugation, reversion, and graduation on the used hypercomplex units.\label{invo}}
\end{table}
Adding the hyperbolic unit to the Pauli algebra corresponds in fact to a
hyperbolic complexification.
In the terminology of Clifford algebras the complexification is leading to the
isomorphisms
\begin{equation}
\bfa{R}_{3,0}\otimes\bar{\bfa{H}}(\bfa{R})
\simeq\bar{\bfa{H}}(\bfa{R})_{3,0}\simeq\bfa{H}(2)\simeq\bar{\bfa{C}}_{3,0}\;.
\end{equation}
The full structure therefore corresponds to the universal three-dimensional complex Clifford
algebra. The representation of the Clifford algebra in terms of matrices is given by
$\bfa{H}(2)$, the algebra of hyperbolic $2\times 2$ matrices. This representation will
be assumed in some of the equations within this work.
The vector $Z$ of Eq.~(\ref{veco}) has sixteen real dimensions.
In \cite{Ulr052} it has been shown that the four-dimensional
real Minkowski vector can be considered as the magnitude of $Z$,
if the square is restriced to real numbers, i.e. $Z\bar{Z}\in\bfa{R}$.
Based on the assumption that only such vectors are of physical relevance,
the following
investigation is restricted to the real four-dimensional Minkowski space,
and the remaining phase contributions are neglected.
A Minkowski vector $x^\mu\in\bfa{R}^{\,3,1}$ is expressed in the above algebra
as $X=x^\mu e_\mu$.
The spatial vector contributions can be written explicitly in the
following representation
\begin{eqnarray}
X&=&x^0+j\bfa{x}\;,
\end{eqnarray}
where $\bfa{x}=x^i\sigma_i$.
Using the Pauli matrices as the explicit representation of $\sigma_i$,
the vector $X$ can be expressed in terms of a hyperbolic
$2\times 2$ matrix according to
\begin{equation}
X=\left(\begin{array}{cc}
\;x^0+jx^3\;&\;jx^1-ijx^2\;\\
\;jx^1+ijx^2\;&\;x^0-jx^3\;\\
\end{array}\right)\;.
\end{equation}
The scalar product of two vectors
can be defined as
\begin{equation}
\label{scalar}
X\cdot Y
=\frac{1}{2}(X\bar{Y}+Y\bar{X})\;.
\end{equation}
The wedge product is given as
\begin{equation}
\label{outer}
X\wedge Y
=\frac{1}{2}(X\bar{Y}-Y\bar{X})\;.
\end{equation}
The wedge product corresponds to a so-called biparavector, which can be
used e.g. for the description of the electromagnetic field or
the relativistic angular momentum (see also Baylis \cite{Bay99}).
The basis elements of the relativistic $\bar{\bfa{C}}_{3,0}$ paravector algebra can be considered
as the basis vectors of the relativistic vector space.
These basis elements form a non-cartesian orthogonal basis with respect to the
scalar product defined in Eq.~(\ref{scalar})
\begin{equation}
e_{\mu}\cdot e_{\nu}
=g_{\mu\nu}\;,
\end{equation}
where $g_{\mu\nu}$ is the metric tensor of the Minkowski space.
As an example of the paravector algebra
the energy-momentum vector of a free
classical pointlike particle, moving
with velocity
$\bfa{v}$ relative to the observer,
is expressed in the Pauli algebra notation.
The relativistic momentum vector for this
particle can be written as
\begin{equation}
\label{koko}
P=\frac{E}{c}+j\bfa{p}=mc\exp{(j\bfa{\xi})}\;,
\end{equation}
with $c$ denoting the velocity of light,
$\bfa{\xi}$ the rapidity, $E$ the energy and
$\bfa{p}$ the momentum of the particle.
The rapidity is defined as $tanh\xi=v/c=pc/E$,
where $\xi=|\bfa{\xi}|$ and $p=|\bfa{p}|$.
Rapidity and momentum
point into the same direction $\bfa{n}=\bfa{v}/|\bfa{v}|$ as the velocity.
In the following $c$ and $\hbar$ will be set equal to one.
In quantum mechanics the momentum
is replaced in coordinate space by the operators $p^\mu=i\partial^\mu$.
\section{Lorentz transformations}
\label{trafo}
Porteous \cite{Por95} shows that the general linear group can be
considered as an unitary group. The unitarity is related to the double field
together with an appropriate correlation. For the complex linear group this is a $^2\bar{\bfa{C}}^\sigma$-correlation.
The complex double field $^2\bar{\bfa{C}}^\sigma$, with swap $\sigma$
and conjugation, correponds to the null basis representation of the
hyperbolic numbers with conjugation $\bar{\bfa{H}}$ as given
in Eqs.~(\ref{beg}) and (\ref{conj}).
In addition, there is an isomorphism between the general linear group and the hyperbolic
unitary group $GL(n,\bfa{C})\simeq U(n,\bfa{H})$,
and in particular for the special groups in two dimensions, i.e. $SL(2,\bfa{C})\simeq SU(2,\bfa{H})$
(see also Hucks \cite{Huc93}).
The unitarity of the group $U(n,\bfa{H})$ is understood with respect to a $\bar{\bfa{H}}$-correlation.
The group $SU(2,\bfa{H})$ corresponds to the spin group of
$SO(3,1,\bfa{R})$ and its elements can
be used to express rotations and boosts of the
paravectors defined in the last section.
The rotation of a paravector can be expressed as \cite{Bay99}
\begin{equation}
\label{rota}
X\rightarrow X^\prime=R X\, R^\dagger\;.
\end{equation}
For the boosts one finds the transformation rule
\begin{equation}
\label{boost}
X\rightarrow X^\prime= B X B^\dagger\;.
\end{equation}
The rotations and boosts are given as
\begin{equation}
\label{rotmat}
R=\exp{(-i\bfa{\theta}/2)}\;,\hspace{0.5cm}B=\exp{(j\bfa{\xi}/2)}\;.
\end{equation}
Based on the Pauli matrices an explicit matrix representation
of the boost operator $B$ can be given, e.g. for a boost in the direction of the
$x$-axis one finds
\begin{equation}
B_1=\left(\begin{array}{cc}
\;\cosh{\xi_1/2}\;&\;j\sinh{\xi_1/2}\;\\
\;j\sinh{\xi_1/2}\;&\;\cosh{\xi_1/2}\;\\
\end{array}\right)\;.
\end{equation}
The boosts are invariant under reversion
$B^\dagger = B$, whereas the conjugated
boost corresponds to the inverse
$\bar{B}=B^{-1}$. For rotations reversion and conjugation correspond both to the inverse
$R^\dagger=\bar{R}=R^{-1}$. This relationship indicates that in non-relativistic physics
hermitian operators can be defined either
with respect to reversion or conjugation.
Boosts and rotations
can be combined to form the Lorentz transformation
\begin{equation}
\label{lorentz}
X\rightarrow\; X^\prime= L X L^\dagger\;,
\end{equation}
which can be expressed in terms of its
infinitesimal generators as
\begin{equation}
\label{lorentzmat}
L=\exp{\left(-i\theta^iJ_i-i\xi^iK_i\right)}\;.
\end{equation}
From these equations the infinitesimal generators of a Lorentz
transformation can be identified as
\begin{equation}
\label{gener}
\bfa{J}=\bfa{\sigma}/2\;,\hspace{0.5cm}\bfa{K}= ij\bfa{\sigma}/2\;.
\end{equation}
One can show
that the generators satisfy the
Lie algebra of the Lorentz
group $SO(3,1,\bfa{R})$.
The Lorentz transformations can be expressed also with relativistic second
rank tensors
\begin{equation}
\label{relspin}
L=\exp{(-iS_{\mu\nu}\omega^{\mu\nu}/2)}\;,
\end{equation}
where the spin angular momentum is defined in terms of the wedge product as
\begin{equation}
\label{spinwedge}
e_\mu\wedge e_\nu=2S_{\mu\nu}\;.
\end{equation}
The elements of the spin angular momentum therefore represent
planes in space-time that are formed by the basis elements of the
algebra.
The following example shows how coordinate vector, momentum vector, orbital angular momentum
and spin angular momentum can be related to each other in the paravector algebra
\begin{equation}
\label{ospinref}
X\bar{P}=x_\mu p^\mu -iS_{\mu\nu}L^{\mu\nu}\;,
\end{equation}
where $L^{\mu\nu}=x^\mu p^\nu-x^\nu p^\mu$ corresponds to
the relativistic orbital angular momentum.
\section{Poincar\'e mass operator}
\label{secmassop}
With the above vector representation the Poincar\'e mass operator can be introduced as a product of a momentum vector and
its conjugated counterpart
\begin{equation}
\label{pmassop}
M^2=P\bar{P}\;.
\end{equation}
The explicit form of the mass operator is obtained by a multiplication of the basis matrices.
The mass operator can be separated into a spin dependent and a
spin independent contribution
\begin{equation}
\label{spindef}
P\bar{P}=p_\mu p^\mu-i\sigma_{\mu\nu}p^\mu p^\nu\;,
\end{equation}
where the spin term is given by
\begin{equation}
\label{sigspin}
\sigma_{\mu\nu}=
\left(\begin{array}{cccc}
\;0\;&\;-ij\sigma_1\;&\;-ij\sigma_2\;&\;-ij\sigma_3\;\\
\;ij\sigma_1\;&\;0\;&\;\sigma_3\;&\;-\sigma_2\;\\
\;ij\sigma_2\;&\;-\sigma_3\;&\;0\;&\;\sigma_1\;\\
\;ij\sigma_3\;&\;\sigma_2\;&\;-\sigma_1\;&\;0\;\\
\end{array}\right)\;.
\end{equation}
Since the spin contribution is anti-symmetric, the last term in Eq.~(\ref{spindef})
is in this case zero. The spin structure becomes important
when interactions are introduced by the minimal substitution of the momentum operators.
The anti-symmetric contribution $\sigma_{\mu\nu}$ correponds to the wedge product of Eq.~(\ref{spinwedge}),
which gives the analogous relation $\sigma_{\mu\nu}=2S_{\mu\nu}$.
The basic fermion equation is introduced as an eigenvalue equation of the mass operator.
With the hyperbolic algebra defined above the
equation can be written as
\begin{equation}
\label{equat}
M^2\psi(x)=m^2\psi(x)\;,
\end{equation}
The wave function $\psi(x)$ has the general structure
\begin{equation}
\label{Ansatz}
\psi(x)=\varphi(x)+j\chi(x)\;,
\end{equation}
where $\varphi(x)$ and $\chi(x)$ can be represented as two-component
spinor functions (see Hucks \cite{Huc93}). They depend on the four space-time
coordinates $x^\mu$.
\section{Poincar\'e spin operator}
\label{planewave}
In analogy to the mass operator
a spin operator can be introduced.
The spin operator
corresponds to the second Casimir operator of the Poincar\'e group, which
describes a system that is invariant under relativistic
translations and rotations. The basic equation for the spin operator
can be defined as
\begin{equation}
\label{spinequat}
S^2\psi(x)=-s(s+1)\psi(x)\;,
\end{equation}
where the square of the spin operator corresponds to
\begin{equation}
\label{pspinop}
S^2=W\bar{W}/m^2\;.
\end{equation}
The operator $W$ denotes the Pauli-Lubanski vector, which
can be expressed in terms of the complex Clifford algebra as
\begin{equation}
\label{Luba3}
W=-\bfa{J}\cdot\bfa{p}-j\,(\bfa{J}p^0+\bfa{K}\times\bfa{p})\;.
\end{equation}
The spatial spin operator can be derived
from the Pauli-Lubanski vector by projections (see Michel \cite{Mic59} or Wightman \cite{Wig60})
\begin{equation}
\bfa{S}=\frac{1}{m}\,W\cdot\bfa{n}\;.
\end{equation}
The projection vectors $n^{\mu}$ are an arbitrary set of four orthogonal vectors
satisfying the relation $n^{\mu}\cdot n^{\nu}=g^{\mu\nu}$.
The spin operator takes the following form
\begin{equation}
\label{spinoperator}
\bfa{S}=\frac{1}{m}\left(\bfa{J}p^0+\bfa{K}\times\bfa{p}-(\bfa{J}\cdot\bfa{p})
\frac{\bfa{p}}{p^0+m}\right)\;,
\end{equation}
if the set of projection vectors is chosen as
\begin{eqnarray}
n^0&=&m^{\!-1}(p^0,p^k)\;,\nonumber\\%\hspace{0.5cm}
n^i&=&m^{\!-1}(p^i,m\delta^{ik}+p^ip^k/(p^0+m))\;.
\end{eqnarray}
The eigenstates of the spin operator
can be introduced as eigenvectors of the squared spin operator and of the
z-component $S_z=S_3$
\begin{eqnarray}
S^2\;\vert\,s\,m_s\,\rangle
&=&-s(s+1)\;\vert\,s\,m_s\,\rangle\;,
\\
S_z\;\vert\,s\,m_s\,\rangle
&=&m_s\;\vert\,s\,m_s\,\rangle
\nonumber\;.
\end{eqnarray}
To find an explicit representation of these eigenstates one has to consider
that the spin operator in the above form
can be obtained also with a boost acting
on the operator vector $\bfa{J}$
\begin{equation}
\label{shref}
\bfa{S}=B \bfa{J}\bar{B}\;.
\end{equation}
Therefore, the relativistic spinor can be
related to the non-relativistic
Pauli spinor by a boost
\begin{equation}
\label{begin}
\vert\,s\,m_s\,\rangle
=B\chi_{m_s}=u(\bfa{p},m_s)\;.
\end{equation}
If the rapidity in
the boost (see Eq.~(\ref{rotmat}))
is expressed in terms of the particle momentum, one finds that
the boosted Pauli spinor corresponds to the
hyperbolic representation of the Dirac spinor
\begin{equation}
\label{spnra}
u(\bfa{p},m_s)=\sqrt{\frac{p^0+m}{2m}}\left(1
+\frac{j\bfa{p}}
{p^0+m}\right)\chi_{m_s}\;.
\end{equation}
The anti-particle spinor can be derived from the
particle spinor, which is multiplied by the
hyperbolic unit $v(\bfa{p},m_s)=ju(\bfa{p},m_s)$.
The spinors can be combined with the Hilbert space state for the momentum
$\vert p^\mu\rangle$ to form the plane wave
expansion
\begin{eqnarray}
\label{solute}
\psi(x)&=&\sum_{m_s}\int \! \frac{d^3\bfa{p}}{(2\pi)^32p^0}\,\left(
u(\bfa{p},m_s)e^{-ip_\mu x^\mu}\,b(p,m_s)\right.\nonumber\\
&&+\left.v(\bfa{p},m_s)e^{ip_\mu x^\mu}\,\bar{d}(p,m_s)\right)\;,
\end{eqnarray}
which is a general solution of the Poincar\'e mass operator and
the Poincar\'e spin operator~(\ref{spinequat}).
The relativistic wave function is an element of the spinor space,
which is a minimal left ideal in the terminology of Clifford algebras.
The elements of a minimal left ideal
have rank~1 and therefore they can be represented as column vectors
$\psi^i(x)\in \bar{\bfa{H}}^2$. The transformation rule of the spin operator in Eq.~(\ref{shref}) is in
contrast to Eq.~(\ref{boost}). The above rule is used for relativistic operators acting in the
spinor space. The transformation rule is related to the $\bar{\bfa{H}}$-correlation, which maps
the elements of the spinor space to their dual space.
Since the Poincar\'e mass operator and
the Poincar\'e spin operator are operators in the spinor space they have to obey the same transformation law.
This is the reason why e.g. the mass
operator is defined as $P\bar{P}$ and not as $PP^\dagger$.
\section{Massless particles}
\label{massless}
Massless particles like neutrinos are normally described in terms of helicity states.
It is shown that massless particles can also be described with
boosted Pauli spinors.
Representing particles in terms of Pauli spinors corresponds to
a decoupling of the polarization axis and the direction of momentum.
In explicit calculations this decoupling is leading to
ambiguities in the coupling of angular momenta of multiple particles. Therefore, e.g. in
calculations of scattering amplitudes, the
helicity basis is the best choice, even in a non-relativistic scheme.
From the theoretical point of view,
massless particles that are represented by Pauli spinors provide a direct
analogy to the description of the last
section. What has to be shown is that this representation is conform with the
Poincar\'e group in the massless case
\begin{equation}
\label{equat5}
M^2\psi(x)=0\;.
\end{equation}
If the polarization axis for a massive particle is chosen, e.g. as the z-axis, the momentum
can take any direction without any restriction. This is not the case for massless
particles. The spatial momentum can never be perpendicular to the chosen
polarization axis. The spatial symmetry of the momentum is
therefore broken, which will become apparent in the following equations.
The spinor will be defined within the little
group of a standard vector \cite{Tun85}. Since massless particles are moving
with the velocity of light they have no rest frame. Therefore,
the standard frame will be defined as the system in which
the momentum is directed along the polarization
axis. If the particle is polarized in the direction of the z-axis
the positive-energy standard vector is given as
\begin{equation}
\label{standml}
p^\mu=(\vert p^3\vert,0,0, p^3)\;.
\end{equation}
In this description the helicity can be either positive or negative. The states are
always characterized by the two possible polarizations corresponding to the
chosen polarization axis. The components of the Pauli-Lubanski
vector in the standard frame are
\begin{eqnarray}
\label{PLu}
W^0&=&-p^3 J^3\nonumber\;,\\
W^1&=&-\vert p^3\vert(J^1+K^2)\;,\\
W^2&=&-\vert p^3\vert(J^2-K^1)\nonumber\;,\\
W^3&=&-\vert p^3\vert J^3\nonumber\;.
\end{eqnarray}
The operators $W^0$ and $W^3$ are linear dependent and can
be represented by $J^3$.
The three generators $J^3$, $W^1$ and $W^2$ satisfy the
Lie algebra of $E_2$, the Euklidean group in two dimensions,
which defines the little group of
the $m^2=0$ representation.
An arbitrary momentum vector $p^\mu$ is obtained with
a boost perpendicular to the z-axis
\begin{equation}
\label{rapidy}
p^\mu=(\vert p^3\vert \cosh{\xi}\,,\,\vert p^3\vert\, n^1 \sinh{\xi}\,,\,
\vert p^3\vert\, n^2\sinh{\xi}\,,p^3)\;.
\end{equation}
The unit vector $n^i_\perp=(n^1,n^2,0)$
characterizes the direction of the boost.
With the perpendicular momentum vector $p_\perp^i=(p^1,p^2,0)$
the rapidity can be defined by the relation $\tanh{\xi}=p_\perp/p^0$,
where $p_\perp=\vert \bfa{p}_\perp\vert$.
Using $\bfa{\xi}_\perp=\bfa{n}_\perp\xi$ the boost can
be written in the form $B=\exp{(j\bfa{\xi}_\perp/2)}$.
The momentum contribution parallel
to the polarization will be denoted as $p_\parallel^i=(0,0,p^3)$, with $p_\parallel=\vert \bfa{p}_\parallel\vert$.
In the spinor representation the generators $\bfa{J}=\bfa{\sigma}/2$ and $\bfa{K}=ij\bfa{\sigma}/2$
can be inserted into Eq.~(\ref{PLu}).
Then one finds in the standard frame and therefore in all
frames
$W\bar{W}=0$,
i.e.~the spin is given in the
degenerate spin $s=0$ representation of $E_2$.
The basis vectors are chosen as eigenvectors
of $J_3$ with the eigenvalues $m_s=\pm \frac{1}{2}$. They can be represented
with the Pauli spinor $\chi_{m_s}$. A general spinor
can be calculated using Eq.~(\ref{begin}) with the
boost parameters defined above
\begin{equation}
\label{m0spnra}
u(\bfa{p},m_s)=\sqrt{\,\frac{p^0+p_\parallel}{2p_\parallel}\,}\left(1
+\frac{j\bfa{p}_\perp}
{p^0+ p_\parallel }\right)\chi_{m_s}\;.
\end{equation}
The antiparticle spinor is obtained if the
above expression is multiplied
by the hyperbolic unit.
A set of four
orthogonal projection vectors can be introduced to derive the spin operator
from the Pauli-Lubanski vector
\begin{eqnarray}
n^0&=& p_\parallel^{-1}(p^0,p^k_\perp)\;,\nonumber\\%\hspace{0.5cm}
n^i&=& p_\parallel^{-1}(p_{\perp}^i, p_\parallel \delta^{ik}+p_{\perp}^i p^k_\perp/(p^0+ p_\parallel ))\;.
\end{eqnarray}
The spin operator is then defined as
\begin{equation}
\label{nspin}
\bfa{S}=\frac{1}{ p_\parallel }\, W\cdot \bfa{n}\;.
\end{equation}
The spin operator corresponds again to a boosted vector of the
angular momentum generators $\bfa{J}$. Explicitly written one finds
\begin{equation}
\bfa{S}= \frac{1}{ p_\parallel }
\left(\bfa{J}p^0+\bfa{K}\times\bfa{p}_\perp-(\bfa{J}\cdot\bfa{p}_\perp)\frac{\bfa{p}_\perp}{p^0+p_\parallel}\right)\;.
\end{equation}
As mentioned above only the third component $S_z=S_3$ of the spin operator is relevant within $E_2$.
For this component the last term in the above equation is zero.
The action of the spin operators on the spinor can be summarized in the
relations
\begin{eqnarray}
\label{ndef}
S^2\;\vert\,0\,m_s\,\rangle
&=&0\;,
\\
S_z\;\vert\,0\,m_s\,\rangle
&=&m_s\;\vert\,0\,m_s\,\rangle
\nonumber\;,
\end{eqnarray}
where the square of the spin operator corresponds to $S^2=W\bar{W}/p_\parallel^2 $.
The plane wave expansion is formally equivalent to
Eq.~(\ref{solute}), but
the spinors $u(\bfa{p},m_s)$ and $v(\bfa{p},m_s)$
have to be replaced
with the specific $m^2=0$ form
given above. From these equations it follows that
\begin{equation}
S^2 \psi(x)=0\;.
\end{equation}
In fact, massless particles are also spinless particles. However, they can be
described by a non-zero momentum and a polarization comparable to the
massive particles.
\section{Maxwell equations}
\label{electro}
The Maxwell equations can be derived from an eigenvalue equation of the mass operator,
where the mass operator is now acting on a vector field
\begin{equation}
\label{wavb}
M^2 A(x)=0\;.
\end{equation}
The equation can be expressed
with the electromagnetic fields according to
\begin{eqnarray}
\label{maxwell}
P\bar{P}A&=&-\bfa{\nabla}\cdot\bfa{E}-\partial^0 C\nonumber\\
&&+ij\bfa{\nabla}\cdot\bfa{B}\\
&&-j(\bfa{\nabla}\times\bfa{B}-
\partial^0\bfa{E}
-\bfa{\nabla}C)\nonumber\\
&&-i(\bfa{\nabla}\times\bfa{E}+\partial^0\bfa{B})
=0\nonumber\;.
\end{eqnarray}
This expression is obtained if one evaluates $\bar{P}A(x)$, inserts the usual definitions for the
electromagnetic fields, and then multiplies the resulting terms by the operator
$P$. If $P\bar{P}$ is calculated first, Eq.~(\ref{wavb}) reduces to the wave operator acting on the
vector potential giving zero. Both forms are equivalent in the Lorentz gauge.
In Eq.~(\ref{maxwell}) the four homogeneous
Maxwell equations are included.
The calculation provides two additional terms depending on
\begin{equation}
\label{zero}
C(x)=\partial_\mu\, A^\mu(x)\;.
\end{equation}
These terms disappear in the Lorentz gauge.
\section{Photon plane wave states}
\label{photop}
In this section a plane wave expansion for free
photon fields is derived. The techniques
developed for massless fermions will be applied. In this representation
the polarization
vector of the photon corresponds to a generalization of the Pauli spinor. Again it is mentioned that the
helicity basis has advantages in explicit calculations.
The transformation properties of the vector
components $A^\mu(x)$ can be understood in terms of
$4\times 4$ transformation
matrices acting on four-component vectors
\begin{eqnarray}
\label{lorentz2}
x^\mu\rightarrow x^{\mu\prime}= (L)^{\mu}_{\;\;\nu}\,x^\nu\;.
\end{eqnarray}
The transformation matrices can be derived from the generators
$J_i$ and $K_i$ according to Eq.~(\ref{lorentzmat}).
The third components of the generators are
\begin{equation}
\label{photongen}
(J_3)^{\mu}_{\;\;\nu}=\left(\begin{array}{cccc}
\;0\;&\;0\;&\;0\;&\;0\;\\
\;0\;&\;0\;&\;-i\;&\;0\;\\
\;0\;&\;i\;&\;0\;&\;0\;\\
\;0\;&\;0\;&\;0\;&\;0\;\\
\end{array}\right),
\hspace{0.5cm}
(K_3)^{\mu}_{\;\;\nu}=\left(\begin{array}{cccc}
\;0\;&\;0\;&\;0\;&\;i\;\\
\;0\;&\;0\;&\;0\;&\;0\;\\
\;0\;&\;0\;&\;0\;&\;0\;\\
\;i\;&\;0\;&\;0\;&\;0\;\\
\end{array}\right).
\end{equation}
The results of Section \ref{massless} will be used in the following.
The standard frame is the system in which the momentum
is directed along the polarization axis.
The Pauli-Lubanski vector in the standard frame
is given by Eq.~(\ref{PLu}), where the generators now have to
be replaced by the generators of Eq.~(\ref{photongen}), including the
remaining generator components.
The two eigenstates of
$J_3$ with eigenvalues $m_s=\pm 1$ are given as usual as
\begin{equation}
\chi_{\pm}^\mu
=(\mp\, \epsilon_1^\mu - i\epsilon_2^\mu)/\sqrt{2},
\end{equation}
where $\epsilon_1^\mu$ and $\epsilon_2^\mu$ are
unit vectors in the direction of the x- and y-axis.
In the standard frame one finds that the squared spin operator acting on
the eigenstates is zero
\begin{equation}
(W \bar{W})
^{\mu}_{\;\;\nu}\,\chi_{m_s}^\nu=0.
\end{equation}
General eigenstates of the spin operator are obtained again by a boost of the
eigenstate from the standard frame of the particle $\vert\,0\,m_s\,\rangle=e^\mu(\bfa{p},m_s)=(B)^{\mu}_{\;\;\nu}\,\chi_{m_s}^\nu$.
As in section \ref{massless} the direction of the boost is perpendicular
to the polarization axis, i.e. $B=\exp{(-i\xi^i_\perp K_i/2)}$. The boost is
performed with the generators
given in Eq.~(\ref{photongen}). The calculation is leading to the
explicit form of the polarization vector
\begin{equation}
e^\mu(\bfa{p},m_s)=\left(\,\frac{p^{m_s}}{ p_\parallel }\,,
\chi_{m_s}^i+ \frac{p^i_\perp p^{m_s}}{ p_\parallel (p^0+ p_\parallel )}
\right)\;,
\end{equation}
where $p^\pm=(\mp p^1-
ip^2)/\sqrt{2}$.
For the eigenstates one finds the relations
\begin{eqnarray}
S^2\;\vert\,0\,m_s\,\rangle
&=&0\;,
\\
S_z\;\vert\,0\,m_s\,\rangle
&=&m_s\;\vert\,0\,m_s\,\rangle
\nonumber\;.
\end{eqnarray}
The spin operator $S_z=S_3$
is defined as in Eq.~(\ref{nspin}) with the appropriate
generators for $\bfa{J}$ and $\bfa{K}$. The above equations
are equivalent to Eq.~(\ref{ndef}) except
for the different eigenvalues. For photons
one finds $m_s=\pm 1$.
The plane wave expansion of the
free photon field is given as
\begin{equation}
\label{solutex}
A(x)=
\sum_{m_s}\int \! \frac{d^3\bfa{p}}{(2\pi)^32p^0}\,
e^\mu (\bfa{p},m_s)\,e_\mu\!\left(e^{-ip_\mu x^\mu}a(p,m_s)+
e^{ip_\mu x^\mu}\,\bar{a}(p,m_s)\right).
\end{equation}
The plane wave satisfies the relation
\begin{equation}
S^2 A(x)=0\;.
\end{equation}
The coordinate vector $\chi_{m_s}^\mu\in \bar{\bfa{C}}^{\,3,1}$ plays in this context the role
of the Pauli spinor. It can be considered as an element of the minimal left ideal with
respect to the transformations induced by the generators of Eq.~(\ref{photongen}).
This is indicated in the above formulas by the tensor indices.
The indices can be omitted, e.g. $e(\bfa{p},m_s)=B\chi_{m_s}$,
to provide a representation free
notation.
\section{Quadratic Dirac equation}
\label{blab1}
Electromagnetic interactions can be introduced with the minimal substitution of the
momentum operator. The mass operator of Eq.~(\ref{pmassop}) transforms into
\begin{equation}
\label{basic}
M^2=(P-eA(x))(\bar{P}-e\bar{A}(x))\;,
\end{equation}
and is now invariant under local gauge transformations.
This mass operator can be inserted into Eq.~(\ref{equat}). If Pauli matrices
and electromagnetic fields are expressed with the
anti-symmetric tensor $\sigma_{\mu\nu}$ given in Eq.~(\ref{sigspin})
and $F^{\mu\nu}=\partial^\mu A^\nu- \partial^\nu A^\mu$ one finds
\begin{equation}
\label{Pauli1}
\left((p-eA)_\mu(p-eA)^\mu -\frac{e}{2}
\sigma_{\mu\nu} F^{\mu\nu}-m^2\right)\psi(x)=\,0\;.
\end{equation}
This is the quadratic Dirac equation. Using the hyperbolic algebra
it can be represented as a $2\times 2$ matrix equation, whereas conventionally
the quadratic Dirac equation is a $4\times 4$ matrix equation.
The spinor function $\psi(x)$
used for the mass operator has a two-component structure,
whereas in the Dirac equation $\psi(x)$ corresponds to a four-component spinor.
One finds in both cases the same two coupled
differential equations.
For the hyperbolic algebra one can derive from Eq.~(\ref{Pauli1})
\begin{equation}
\label{Pauli}
\left((p-eA)_\mu(p-eA)^\mu -e\,ijE^i\sigma_i
+eB^i\sigma_i -m^2\right)\psi(x)=0\;.
\end{equation}
In the Dirac theory the spin tensor is defined
according to $\sigma_{\mu\nu}=i/2\,[\gamma_\mu,\gamma_\nu]$.
Using this tensor the Dirac form of Eq.~(\ref{Pauli}) can be written as
\begin{equation}
\label{Pauli3}
\left((p-eA)_\mu(p-eA)^\mu -
e\,iE^i\alpha_i
+eB^i\sigma_i -m^2\right)\psi(x)=0\;.
\end{equation}
Comparing this equation with Eq.~(\ref{Pauli}) one observes
that in both cases the term including the electric field is the only
term which couples the components
of the spinor. In the mass operator
equation the coupling term
is proportional to $j\bfa{\sigma}$, in the quadratic Dirac equation
the term corresponds to
$\bfa{\alpha}=\gamma_5\bfa{\sigma}$.
The Dirac
representation of $\gamma_5$ and $j$ have
the same effect on the spinor, a swap of the spinor components.
For the Poincar\'e mass operator one finds
\begin{equation}
j\psi(x)=\chi(x)+j\varphi(x)\;,
\end{equation}
whereas in the Dirac representation the corrsponding relation is given as
\begin{equation}
\gamma_5\psi(x)=
\left(\begin{array}{c}
\;\chi(x)\;\\ \;\varphi(x)
\end{array}\right)\;.
\end{equation}
In the hyperbolic formalism
the terms proportional to the hyperbolic unit include the
differential equation of the lower component, the other terms describe the
differential equation of the upper component. Conventionally,
the coupled differential equations for upper and lower components are separated by
the matrix structure.
\section{Orbital angular momentum and single particle potentials}
\label{orbitan}
The hyperbolic numbers
can be used also for the description of the
orbital angular momentum, which will be shown in this section
(compare with Xuegang \cite{Xue00}).
A spacelike relativistic vector $x^\mu$ can be parametrized
in relativistic spherical coordinates as
\begin{equation}
\label{para}
x^{\mu}=\left(\begin{array}{c}
x^{0}\\
x^{1}\\
x^{2}\\
x^{3}\\
\end{array}\right)=
\left(\begin{array}{c}
\rho\, \sinh{\xi}\\
\rho\, \cosh{\xi}\,\sin{\theta}\,\cos{\phi}\\
\rho\, \cosh{\xi}\,\sin{\theta}\,\sin{\phi}\\
\rho\, \cosh{\xi}\,\cos{\theta}\\
\end{array}\right)\;.
\end{equation}
The time coordinate is given as
$\rho\, \sinh{\xi}$, where $\rho > 0$. In the limit of $\xi\rightarrow 0$ the
vector reduces to a non-relativistic vector in spherical
coordinates.
Based on the Lorentz transformation given
in Eqs.~(\ref{lorentz2}) and (\ref{photongen}) the above vector can be obtained
from a standard vector $x^\mu=(0,0,0,\rho)$ with the
following transformation
\begin{equation}
\label{lform}
L(\theta,\phi,\xi)= \exp{(-iJ_3\,\phi\,)}\,
\exp{(-iJ_2\,\theta\,)}\,\exp{(-iK_3\,\xi\,)}\;.
\end{equation}
For an
irreducible group representation of the orbital angular momentum
the relation between the generators of
boosts and rotations
is assumed to be the same as for the spin $(\frac{1}{2},0)$ representation
of the Lorentz group used earlier in this work
\begin{equation}
\bfa{K}=ij\bfa{J}.
\end{equation}
The boost generator in Eq.~(\ref{lform})
can then be replaced by the third component of the rotation generator
\begin{equation}
\label{lform2}
L(\theta,\phi,\xi)= \exp{(-iJ_3\,\phi\,)}\,
\exp{(-iJ_2\,\theta\,)}\,\exp{(jJ_3\,\xi\,)}\;.
\end{equation}
The orbital angular momentum will be denoted by $\bfa{L}$ in contrast to the
spin angular momentum $\bfa{S}$. In the following equations $\bfa{J}$ will be specialized to
$\bfa{J}=\bfa{L}$.
The $(l,0)$ representation of the Lorentz group provides the following
equations for the irreducible states
\begin{eqnarray}
L^2\;\vert\,l\,m_l\,\rangle
&=&-l(l+1)\;\vert\,l\,m_l\,\rangle\;,
\\
L_z\;\vert\,l\,m_l\,\rangle
&=&m_l\;\vert\,l\,m_l\,\rangle
\nonumber\;.
\end{eqnarray}
Again the third component is denoted as $L_z=L_3$.
In this basis the transformation of Eq.~(\ref{lform}) is represented as
\begin{equation}
D^l_{m_l^\prime m_l}(\theta,\phi,\xi)&=&
\langle\,l\,m_l^\prime\,\vert\,e^{-iL_3\,\phi\,}\,
e^{-iL_2\,\theta\,}\,e^{\,jL_3\,\xi\,}\,\vert\,l\,m_l\,\rangle\nonumber\\
&=&e^{-im_l^\prime\,\phi\,}\,d^{\,l}_{m_l^\prime m_l}
(\theta\,)\,e^{\,jm_l\xi\,}\;.
\end{eqnarray}
Compared to the non-relativis\-tic
case the relativistic rotation matrices
are extended by the additional hyperbolic phase factor
$e^{\,jm_l\xi\,}$.
An application of these functions can be the solution of the Poincar\'e mass operator
with appropriate model potentials.
A relativistic generalization of the $1/ \vert \bfa{x}\vert$
central potential is suggested, which could be used to
describe an electron moving in the potential of a nucleus
\begin{equation}
\label{newpot}
eA^\mu(x)= -\frac{Z\alpha}{\rho}\,\epsilon^\mu(x)\;.
\end{equation}
The polarization vector corrsponds to the unit vector
of the $\xi$-coordinate
\begin{equation}
\epsilon^\mu(x)= \frac{1}{\rho}\,\frac{\partial}
{\partial\,\xi}\,x^\mu(\rho,\theta,\phi,\xi)
=
\left(\begin{array}{c}
\cosh{\xi}\\
\sinh{\xi}\,\sin{\theta}\,\cos{\phi}\\
\sinh{\xi}\,\sin{\theta}\,\sin{\phi}\\
\sinh{\xi}\,\cos{\theta}\\
\end{array}\right)
\;.
\end{equation}
In the static limit $\xi\rightarrow 0$ this potential reduces
to the $1/ \vert \bfa{x} \vert$ potential.
Some general remarks on the solution
procedure will be given here.
The $\bfa{\sigma}$-terms
in Eq.~(\ref{Pauli})
imply a coupling of the orbital angular momentum with the spin
$\bfa{J}=\bfa{L}+\bfa{S}$.
There is also a term proportional to the hyperbolic unit
which has a
different parity. The wave function $\psi(x)$ therefore has a similar
structure as the Dirac spinor
\begin{equation}
\psi(x)=\varphi_l(x)+ j\chi_{l^\prime}(x)\;,
\end{equation}
with
\begin{equation}
l^\prime=
\left\{\begin{array}{c}
l-1\\
l+1\\
\end{array}\right.
\hspace{0.5cm}\mathrm{for}\hspace{0.5cm}
\begin{array}{c}
l=j+1/2\\
l=j-1/2\\
\end{array}\;.
\end{equation}
The eigenvalue of the coupled spin operator should not be confused with
the hyperbolic unit in this equation.
Since relativistic single particle potentials depend on the relative time,
the energy of the single particle states is not a
conserved quantity. Therefore, the following eigenvalue problem has to be considered
\begin{equation}
M^2\;\psi(x)= m^2\;\psi(x)\;,
\end{equation}
where the solutions
$\psi_i(x)$
are eigenstates of the mass operator
with the quantum numbers $i=(njlm_j)$.
Given the model potential of
Eq.~(\ref{newpot}) the eigenvalues $m_i^2$ of this equation should be close to the
spectrum of the Dirac equation with a non-relativistic $1/ \vert \bfa{x}\vert$
central potential. For the ground state $GS=(1\frac{1}{2}0m_j)$
one can therefore expect to find approximately
\begin{equation}
m_{GS}^2\approx 1-Z^2\alpha^2\;.
\end{equation}
A detailed solution of this problem is beyond the scope of this
work.
\section{Summary}
The representation of relativistic quantum physics in terms of mathematical
structures is not unique. The relationships and isomorphisms between different
representations can be understood within Clifford algebras as the common
underlying mathematical framework.
The most popular representation of relativistic physics is based on the Dirac algebra $\bfa{R}_{1,3}$.
However, the Majorana algebra $\bfa{R}_{3,1}$, or the $\bfa{R}_{3,0}$ algebra suggested by Baylis,
provide frameworks that can be used for relativistic calculations as well.
The three-dimensional universal complex Clifford algebra $\bar{\bfa{C}}_{3,0}$ is
proposed for an alternative representation of relativistic quantum physics.
The Baylis algebra is isomorphic to the subset of
$\bar{\bfa{C}}_{3,0}$ considered in this work. The structural difference
appears in the shape of the hyperbolic unit. The full structure of the
complex $\bar{\bfa{C}}_{3,0}$ Clifford algebra provides sixteen real dimensions, the same number as
the Dirac algebra, which is in contrast to
the eight-dimensional Baylis algebra.
|
\section{Introduction}
It is a deep rooted desire to explain complicated experimental
observations with simple and transparent models, understandable
in laymen's terms.
This very acceptable ambition inspired a recent publication ~\cite{Bial},
where an attempt was made to explain the
relations between the multiplicities of different strange baryons
produced in heavy ion reactions with the help of linear
coalescence rehadronization model. This
model was based on
simple quark counting and elementary probability estimates.
However, the considerations in Ref.~\cite{Bial}
lead to
a clear violation of strangeness conservation in strong interactions:
In order to explain the data in laymen's terms, Bialas had to assume
in heavy ion reactions the number produced of strange quarks, $ s $ is not
equal with the number of produced anti-strange quarks, $ {\overline s} $.
We show here that this unacceptable assumption was related to the neglected
requirement that all quarks have to hadronize, as no free quarks are observed
in nature. This necessity leads to a non-linearity
even in the simplest algebraic rehadronization model (see Ref.~\cite{bz}).
Properly taking into account the non-linear competition
for quarks by the various hadron forming coalescence channels
is a natural way to correct the linear coalescence
treatment in the simplest possible manner.
This leads to
the normalized, non-linear {\it al}gebraic {\it co}alescence
{\it r}ehadronization model,
ALCOR \cite{ALCOR},
which automatically takes care of the conservation of strangeness in strong
interactions as well.
\section{Linear coalescence in rehadronization?}
In the linear coalescence rehadronization model Bialas assumed,
that the number of produced particles is proportional
to the product of the numbers of constituent
particles within the reaction volume.
In the following we denote
the {\it number} of particles by the particle symbols,
in particular, $q$ stands for the number of light (up and down)
quarks.
Linear coalescence yields the following relations:
\begin{eqnarray}
{p} &=& a_{p}\, q^3, \label{e:lic1} \\
{\Lambda} | {\Sigma} &=& a_{\Lambda}\, q^2 \, s, \\
{\Xi} &=& a_{\Xi}\, q^{\phantom{2}} \, s^2 ,\\
{\Omega} &=& a_{\Omega}\, s^3,
\end{eqnarray}
where ${\Lambda} | {\Sigma} = \Lambda + \Sigma$ stands for
the total number of strange baryons that contain a single strange
quark.
\newpage
Similarly, for the anti-baryons one obtains
\begin{eqnarray}
{\overline p} &=& a_{\overline p}\,\, {\overline q}^{\, 3}, \\
{\overline \Lambda} | {\overline \Sigma} &=& a_{\overline \Lambda}\,\, {\overline q}^{\, 2}\,\, {\overline s}, \\
{\overline \Xi} &=& a_{\overline \Xi}\,\, {\overline q}^{\phantom{\, 2}}\, {\overline s}^{\, 2}, \\
{\overline \Omega} &=& a_{\overline \Omega}\,\, {\overline s}^{\, 3}. \label{e:licu}
\end{eqnarray}
Furthermore, Ref.~\cite{Bial} also assumed, that
the coefficients of proportionality for particles and their antiparticles
are equal: $ a_{\overline \Omega} = a_{\Omega} $, etc. As the coefficients $a_{\overline \Omega}, a_{\Omega}$ describe
effectively the (particle anti-particle symmetric) likelihood
that a multi-quark bound state is formed once the constituents are given,
this seems to be a very reasonable model at first sight.
This linear model is attractive not only because of its simple formulation
but also because of its simple predictions, namely that the unknown
$a$ coefficients cancel from the anti-baryon to baryon
ratios. This leads to the following equations:
\begin{eqnarray}
{{\overline p} \ove {p} } &=& \left[{{\overline q} \ove q }\right]^3, \label{e:ao1} \\
{{\overline \Lambda} | {\overline \Sigma} \ove {\Lambda} | {\Sigma} } &=& \left[{{\overline q} \ove q }\right]^2 \,
\left[{{\overline s} \ove s }\right], \\
{{\overline \Xi} \ove {\Xi} } &=& \left[{{\overline q} \ove q }\right]^{\phantom{2}} \,
\left[{{\overline s} \ove s }\right]^2, \\
{{\overline \Omega} \ove {\Omega} } &=& \left[{{\overline s} \ove s }\right]^3. \label{e:aow}
\end{eqnarray}
In order to explain the value of
\begin{equation}
{\overline \Omega \ove \Omega} = 0.383 \pm 0.081,
\end{equation}
as measured in Pb +Pb reactions at CERN SPS energies,
Bialas had to assume \ \ $ s \neq {\overline s} $ as a consequence of
eq.~(\ref{e:aow}) of his linear model,
{\it which is a clear violation of strangeness conservation
in strong interactions.}
What is the origin of this contradiction?
The linear coalescence model, as defined in eqs. (3) and (4)
is a good approximation {\it only if
the composite particles use up a small fraction} of the
constituents.
This is the case e.g. in the coalescence treatment of
the deuteron $(pn)$ or triton $(pnn)$ formation from the gas of protons
and neutrons.
In that situation
most particles will remain in nucleon ($p$ or $n$) state. Thus
the competition is negligible among the tritons and the deuterons
for the building block nucleons in that case.
In the case of hadronization of a quark matter {\it all} of the constituent
particles (the quarks) have to be placed into composite particles,
namely into colorless hadrons. No free quarks are observed in Nature.
{\it This is the very essence of color confinement.}
Thus, during the hadronization process,
a fixed number of strange quarks
must be distributed among hyperons and anti-K mesons.
The anti-strange quarks have to be distributed
among anti-hyperons and K mesons. [ Note that K meson is
a $(q {\overline s})$ bound state, while
the anti-K meson is a $ (\overline q s )$ bound state.]
However, in heavy ion collisions we have incoming nuclei in the initial
state, that contain nucleons, formed by constituent quarks, only.
As quarks are produced in quark antiquark pairs,
the number of quarks in the final state is larger than the
number of anti-quarks. Hence we must have more K mesons
than anti-K mesons.
{The processes creating different hadrons
are not independent, they compete with each other,
{\it contrary} to the basic assumption in the linear coalescence
model.}
The redistribution can be counted for by introducing normalization
factors, $b_s, b_{\overline s}, b_q, b_{\overline q}$.
These $b_i$ normalization factors are not free parameters!
They are determined from the requirement that all quarks must be
recombined into hadrons during the hadronization process,
as described first in the { ALCOR} model, Ref.~\cite{ALCOR}.
\section{The ALCOR approach}
The non-linear coalescence equations for the formation of
quark-antiquark clusters during rehadronization read as:
\begin{eqnarray}
{p} &=& C_p \, b_q^3 \, q^3, \label{coal1} \\
{\Lambda} | {\Sigma} &=& C_{\Lambda} \, b_q^2 \, b_s \, q^2 \, s, \\
{\Xi} &=& C_{\Xi} \, b_q^{\phantom{2}} \, b_s^2 \, q^{\phantom{2}} \, s^2, \\
{\Omega} &=& C_{\Omega} \, b_s^3 \, s^3, \\
{\overline p} &=& C_{\overline p} \, b_{\overline q}^3 \, {\overline q}^3, \\
{\overline \Lambda} | {\overline \Sigma} &=& C_{\overline \Lambda} \, b_{\overline q}^2 \, b_{\overline s} \, {\overline q}^2 \, {\overline s}, \\
{\overline \Xi} &=& C_{\overline \Xi} \, b_{\overline q}^{\phantom{2}} \, b_{\overline s}^2 \,
{\overline q}^{\phantom{2}} \, {\overline s}^2,
\label{coal2} \\
{\overline \Omega} &=& C_{\overline \Omega} \, b_{\overline s}^3 \, {\overline s}^3.
\end{eqnarray}
The meson yields are determined similarly,
\begin{eqnarray}
{\pi}^d &=& C_{\pi} \, \BM{q}{\overline q}, \\
{K} &=& C_{K} \, \BM{q}{\overline s}, \\
{\overline K} &=& C_{\overline K} \, \BM{\overline q}{s}, \\
{\eta} &=& C_{\eta} \, \BM{s}{\overline s}. \label{coal3}
\end{eqnarray}
The normalization coefficients $b_q$, $b_s$, $b_{\overline q}$ and
$b_{\overline s}$ are determined uniquely
by the requirement, that {\it the number of the constituent
quarks do not change during the hadronization} ---
which is the basic assumption for all quark counting methods:
\begin{eqnarray}
s &=& 3 \, {\Omega} + 2 \, {\Xi} + {\Lambda} | {\Sigma} +
{\overline K} + {\eta}, \label{e:cons1}\\
{\overline s} &=& 3 \, {\overline \Omega} + 2 \, {\overline \Xi} + {\overline \Lambda} | {\overline \Sigma} +
{K} + {\eta}, \\
q &=& 3 \, {p} +
2 \, {\Lambda} | {\Sigma} +
{\Xi} + {K} + {\pi}^d, \\
{\overline q} &=&
3 \, {\overline p} +
2 \, {\overline \Lambda} | {\overline \Sigma} +
{\overline \Xi} +
{\overline K}
+ {\pi}^d \ . \label{e:cons4}
\end{eqnarray}
Here ${\pi}^d $ is the number of directly produced $(q{\overline q})$ states.
( Note that other particle
symbols also stand for the number of directly produced particles
of a given type, and care must be taken when comparing the ALCOR predictions
with data, especially regarding the corrections of particle numbers
for the feed-down of resonance decay contributions.)
Similar, non-linear coalescence equations with constraints
of using up all the coalescing particles were proposed
in a simpler form first in Ref.~\cite{bz}
in the so called combinatoric break-up model,
assuming a constant coalescence coefficient for all the baryons and another
one for all the mesons. Later the ALCOR model was formulated as presented
above in a simple form. ALCOR equations were
solved numerically in Refs.~\cite{ALCOR,ALC97,ALC99}, using,
as input values to the coalescence equations,
the number of quarks as predicted by Monte Carlo event generators.
The ALCOR equations were derived from a set of rate equations
in Ref.~\cite{ALC96}, in the
sudden approximation.
Substituting eqs.(\ref{coal1} - \ref{coal3}) into the constraints
given by eqs.~(\ref{e:cons1}-\ref{e:cons4}) one
obtains a set of non-linear equations
for the $b_i$ normalization constants.
{\it However, one can predict some relations even without solving these
set of nonlinear equations}.
It turns out that
the particle-antiparticle ratios can be expressed in terms of
the effective number of quarks even in the non-linear ALCOR model,
similarly as it was done by Bialas in the linear coalescence model.
Let us denote by upper case $Q$, $S$ the reduced, effective
number of up+down and strange quarks, in contrast to the lower case
$q$ and $s$, that stand for the total number of (up+down) and
strange quarks before the hadronization:
\begin{eqnarray}
Q &=& b_q \, q,\\
{\overline Q} &=& b_{\overline q} \, {\overline q },\\
S &=& b_s \, s, \\
{\overline S} &=& b_{\overline s} \, {\overline s }
\end{eqnarray}
Furthermore, one could make the usual assumption
that the $C$ coalescence coefficients for baryons
are equal to that of the corresponding antibaryons,
\begin{eqnarray}
C_p &=& C_{\overline p}, \\
C_{\Lambda} &=& C_{\overline \Lambda}, \\
C_{\Xi} &=& C_{\overline \Xi}, \\
C_{\Omega} &=& C_{\overline \Omega}.
\end{eqnarray}
With these notations, the following relations
are obtained in the non-linear ALCOR model:
\begin{eqnarray}
{{\overline p} \ove {p} }
&=&
\left[{{\overline Q} \ove Q }\right]^3, \\
{{\overline \Lambda} | {\overline \Sigma} \ove {\Lambda} | {\Sigma} } &=&
\left[{{\overline Q} \ove Q }\right]^2 \,
\left[{{\overline S} \ove S }\right], \\
{{\overline \Xi} \ove {\Xi} } &=&
\left[{{\overline Q} \ove Q }\right]^{\phantom{2}} \,
\left[{{\overline S} \ove S }\right]^2, \\
{{\overline \Omega} \ove {\Omega} } &=&
\left[{{\overline S} \ove S }\right]^3, \\
{{\overline K} \ove {K} } &=&
{{\overline Q}\, S \ove Q \, {\overline S} }. \label{e:aoa}
\end{eqnarray}
These equations are formally similar to the equations of the linear treatment
displayed in eqs.~(\ref{e:ao1}-\ref{e:aow}), however, they are given
in terms of the effective number of quarks, $Q = b_q q$, that
are complicated, non-linear functions of the number of quarks $q, s$
available before the hadronization.
Thus one easily gets the following interesting relations:
\begin{eqnarray}
{{\overline \Lambda} | {\overline \Sigma} \ove {\Lambda} | {\Sigma} } &=& {{\overline p} \ove {p} } \,
\left[ {{K} \ove {\overline K} } \right], \label{e:i1}\\
{{\overline \Xi} \ove {\Xi} } &=& {{\overline p} \ove {p} } \,
\left[ {{K} \ove {\overline K} } \right]^2, \label{e:i2}\\
{{\overline \Omega} \ove {\Omega} } &=& {{\overline p} \ove {p} } \,
\left[ {{K} \ove {\overline K} } \right]^3.\label{e:i3}
\end{eqnarray}
These are the relations among the ratios of the observable
number of particles that should be satisfied if
the particle production in some reaction proceeds via
algebraic recombination of the independent valence quarks.
This way a direct relation is obtained in a self-consistent manner,
that connects the presence/absence of independent, initial quarks
with the observable yields of hadrons. This relation is to a large
extent model-independent (i.e. independent of the initial quark
content, independent from the values of the coalescence coefficients
$C$ and independent from the values of the non-linear
renormalization factors $b_i$). Only two physical assumptions
enter these model-independent eqs.~(\ref{e:i1}-\ref{e:i3}):
that the rehadronization process is sudden and that
the valence quarks are
available in unbound states before the hadronization.
One may evaluate the following measurable numbers:
\begin{eqnarray}
d_{\Lambda} & = &
{{\overline \Lambda} | {\overline \Sigma} \ove {\Lambda} | {\Sigma} } \, {{p} \ove {\overline p} }, \\
d_{\Xi} & = &
\left[{{\overline \Xi} \ove {\Xi} } \, {{p} \ove {\overline p} }\right]^{1/2}, \\
d_{\Omega} & = &
\left[{{\overline \Omega} \ove {\Omega} } \, {{p} \ove {\overline p} }\right]^{1/3},\\
d & = &
\left[ {{K} \ove {\overline K} } \right].
\end{eqnarray}
If hadronization proceeds via a sudden recombination of quarks.
then all these $d$ numbers should be equal,
$d = d_{\Lambda} = d_{\Xi} = d_{\Omega}$.
The experimental values for the anti-baryon to baryon number ratios
are determined in Refs.~\cite{EXP1,EXP2,EXP3} as follows:
\begin{eqnarray}
\frac {\overline N}{\,\, N\,\, }
&=& 0.070 \pm 0.010 ,\\
\frac {{\overline \Lambda} | {\overline \Sigma}} {\,\, \Lambda | \Sigma \,\,} &=& 0.133 \pm 0.007 ,\\
\frac {\overline \Xi} {\,\, \Xi \,\,} &=& 0.249 \pm 0.019 ,\\
\frac {\overline \Omega} {\,\, \Omega \,\,} &=& 0.383 \pm 0.081 ,\\
\frac {K^+} {\,\, K^-\,\, } &=& \,\, d \,\, = \,\, 1.80 \pm 0.2
\end{eqnarray}
which yields the following values for the $d$ factors of strange
baryons :
\begin{eqnarray}
d_{\Lambda} &=& 1.9 \pm 0.3,\\
d_{\Xi} &=& 1.89 \pm 0.15,\\
d_{\Omega} &=& 1.76 \pm 0.15.
\end{eqnarray}
These numbers show a very good agreement with $d$,
a statistically acceptable, good agreement between the ALCOR model
and the experimental data in case of Pb + Pb reaction at CERN SPS.
This indicates, regardless of the details of the confinement
mechanism, that hadron production in
Pb +Pb reaction at CERN SPS proceeds via a sudden and complete
coalescence of constituent quarks to hadrons.
Taking into account the conservation of strangeness in strong interactions,
which demands ${\overline s} = s, $
from the $ { \overline \Omega } / \Omega $ ratio we arrive at
\begin{equation}
\frac{ {\overline S}} { \,\, S \,\, } =
\frac{ b_{\overline s} } { b_s } =
0.75 \pm 0.06.
\end{equation}
Thus we obtained agreement with the experimental
data without assuming $ s \neq {\overline s}.$
The deviation of $ \overline \Omega / \Omega $
from unity is caused by the difference in the
normalization factors $ b_s$ and $ b_{\overline s} $,
which can easily be understood:
there are more quarks then antiquarks in the initial
system, and thus more $ \overline s $ quarks are
used up in the $ K^+$
production
then $ s $ quarks in the $ K^- $ meson creation.
Thus less $\overline s $ remains for the $ \overline \Omega $
production then $ s $ quarks for the $ \Omega $
production.
In the time-dependent solution of confining rate equations
for the hadronization~\cite{ALCOR2}, conservation laws are ensured
by the structure of the coupled system of differential equations.
The ALCOR model, as presented above, can be reobtained from a set of non-linear
rate equations in the sudden approximation~\cite{ALC96}, in
the limit when the time of hadronization is very short.
Indeed, a detailed analysis of single particle spectra and two-particle
correlation data indicates a short duration, $\Delta \tau \approx 1.5$ fm/c,
for the production of final state hadrons at CERN SPS Pb + Pb reactions
~\cite{scl99}. This result excludes a long-lived, evaporative mixture of
quark-gluon plasma and hadronic phase, that would produce pions over
an order of magnitude larger period. Thus the experimental data prefer
a sudden production of hadrons, a process that may happen to be
out of local thermal and chemical equilibrium.
Finally, let us point out, that the strangeness conservation
leads to the following relation:
\begin{eqnarray}
s & =& {\overline s},\\
\null\hspace{-0.4cm}
3\, \frac{{\Omega}}{\,\, {\Xi} \,\, } \, \frac{{\Xi}}{\,\,{\Lambda} | {\Sigma}\,\,}
+ 2\,\frac{{\Xi}}{\,\,{\Lambda} | {\Sigma}\,\,}
+ \frace{{\overline K} }{{\Lambda} | {\Sigma}}
+ 1
\!\! & = & \!\!
\frace{{\overline \Lambda} | {\overline \Sigma}}{{\Lambda} | {\Sigma}}\, \left[3\,\frace{{\overline \Omega}}{{\overline \Xi}}\,\frace{{\overline \Xi}}{{\overline \Lambda} | {\overline \Sigma}}
+2 \, \frace{{\overline \Xi}}{{\overline \Lambda} | {\overline \Sigma}}
+ \frace{K}{{\overline \Lambda} | {\overline \Sigma}}
+ 1
\right]. \label{totals}
\end{eqnarray}
With the SPS Pb+Pb data \cite{EXP1,EXP2,EXP3}
the left hand side of this equation is $2.57$,
while the right hand side is $2.66$. Thus the strangeness
conservation equation is also well fulfilled for this case.
The validity of eq.~(\ref{totals}) is an absolute measure of
strangeness conservation in strong interactions.
If the experimental values do not satisfy eq.~(\ref{totals}),
then the detectors miss important parts of the particle
momentum distributions, or they have a serious
systematic error.
Further observations of Ref. \cite{Bial}, namely
$ d_{\Lambda} = d_{\Xi} $ for the SPS
$^{32}S+ ^{23}S$ reactions, but $ d_{\Lambda} \neq d_{\Xi} $
for the SPS p+Pb reaction remain unchanged
with our analysis.
\section{ Conclusions}
We clarified the
reason why the linear coalescence
model should not be applied to the hadroni\-za\-tion of
the quark matter: it leads to a violation
of strangeness conservation.
We have shown, however, that this shortcoming can be corrected
if a a nonlinear coalescence model is introduced,
that takes into account the conservation of strangeness in
strong interactions and the fact that no free quarks are observed
in Nature. In particular, we have shown that
the solution of the ALCOR hadronization model
yields simple, parameter independent relations
in a consistent manner.
These relations were surprisingly simple and similar to those obtained from
the linear coalescence picture. Data on strange particle production
in Pb+Pb reactions at CERN SPS satisfy the coalescence model predictions,
while they are known to be not satisfied by data from p + Pb
reactions at CERN SPS. Based on considerations of the
production ratios of strange hadrons,
we have obtained an elementary,
self-consistent and model independent proof,
that {\it quark degrees of freedom are
liberated} {\it in Pb + Pb reactions at CERN SPS}
before the onset of hadron formation.
\section*{Acknowledgement}
This work was supported by the OTKA Grant No. T029158.
|
\section{Introduction}
The use of atmospheric \v{C}erenkov telescopes to make observations at
large zenith angles was suggested by Sommers and Elbert
\cite{kn:sommers1987} as a potentially efficient method of measuring
gamma rays at the highest energy. The underlying principle is that at
large zenith angles, the footprint of an air shower as seen in optical
\v{C}erenkov light (which is a penetrating component of the cascade)
becomes large. This means that, although the threshold energy of a
telescope will increase as the zenith angle increases, the accompanying
increase in collecting area results in a useful gamma ray detection
rate. Sommers and Elbert discussed this large zenith angle technique in
the context of first-generation (non-imaging) gamma ray telescopes.
Markarian 501 (Mrk 501) is a relatively close BL Lac object (at a red
shift $z$ of 0.034) which shows strong, variable, non-thermal emission
from radio to X-ray wavelengths. Since its discovery as a VHE gamma ray
source \cite{kn:quinn1996}, Mrk 501 has been extensively monitored in
the VHE waveband. In 1997, the object went into a prolonged active state
at all wavebands. The VHE activity was detected by a number of Northern
hemisphere facilities employing imaging techniques (e.g.
\cite{kn:catanese1997,kn:aharonian1997,kn:hayashida1998,kn:barrau1997}).
This state of high activity lasted from 1997 March -- October and the
source exhibited a complex time history with flaring activity on a time
scale as short as a few hours \cite{kn:aharonian1998a}. Multiwavelength
studies of Mrk 501 during this time have been reported
\cite{kn:catanese1997} and the energy spectrum at TeV energies has been
established \cite{kn:aharonian1998a,kn:samuelson1998,kn:krennrich1998}.
The upper bound to the TeV energy spectrum of an AGN is of considerable
interest as it, together with the distance to the source, leads to a
value for the density of intergalactic infra red photons (see e.g.
\cite{kn:stecker1992,kn:biller1998,kn:funk1998}). The highest energy
photons observed from this source to date have been detected by the
HEGRA group and are in excess of 25 TeV \cite{kn:aharonian1998b}.
We report here observations of Mrk 501 made from the Southern hemisphere
at zenith angles in excess of $70^{\circ}$ using the University of
Durham Mark 6 imaging telescope. Our measurements of cascades which
develop through three atmospheres demonstrate that imaging techniques
may be applied at these large zenith angles with the prospect of
improving the significance of the gamma ray signal at high energies.
\section{Observations}
\subsection{The Mark 6 Telescope}
The University of Durham Mark 6 telescope is located at Narrabri NSW
Australia and has been described in detail \cite{kn:armstrong1998}. It
uses the established imaging technique to separate gamma rays from
background cosmic rays and incorporates a robust, noise-free trigger
involving signals from three parabolic flux collectors of diameter 7 m
mounted on a single alt-azimuth platform. The imaging camera comprises
109 pixels (91 with $0.25^{\circ}$ diameter and 18 with $0.5^{\circ}$
diameter) and is mounted at the focus of the central flux detector. It
combines with two low resolution cameras (19 pixels each of diameter
$0.5^{\circ}$) which are mounted at the foci of the left and right flux
collectors to provide a trigger demanding a simultaneous temporal (10 ns
gate) and spatial ($0.5^{\circ}$) coincidence of the \v{C}erenkov light
detected in the three cameras. This trigger system allows the telescope
to detect small light flashes, and hence low energy gamma rays, free of
the triggers due to light produced by individual local muons. This is
particularly important for observations at very large zenith angles when
the background from local muons mimicking the signatures of gamma rays
can become important.
\subsection{Observations of Mrk 501}
Mrk 501 was observed in 1997 July and August under clear and moonless
skies during an interval of extreme activity when detection rates for
gamma rays $> 300$ GeV (using conventional imaging telescopes) exceeded
10 photons per min (see e.g. Bradbury \cite{kn:bradbury1997b}).
Data were recorded at zenith angles of $70^{\circ} - 73^{\circ}$ using
the standard Mark 6 telescope procedure which involves recording data in
15 min segments. OFF source observations were taken by alternately
observing regions of the sky which differ by $\pm 15$ min in right
ascension from the position of Mrk 501 so ensuring that the ON and OFF
segments possess identical azimuth and zenith profiles. This observing
pattern, which involves reversal of the order of the ON and OFF source
measurements, eliminates any first order changes in telescope
performance due to residual secular changes in atmospheric clarity,
temperature etc. An observing log is shown in table
\ref{table:observing_log}.
The background count rate of the telescope at $70^{\circ}$ to the zenith
was about 70 cpm, some 10\% of that at the directions close to the
zenith where the telescope would normally be operated. A preliminary
estimate of the threshold for gamma ray detection with the Mark 6
telescope at $70^{\circ}$ to the zenith is $\sim 15$ TeV which may be
compared with the value of $\sim 300$ GeV for zenith angles $ \leq
30^\circ$. We note that Sommers and Elbert \cite{kn:sommers1987}
estimate a factor of $\sim \times 70$ in threshold energy difference
between zenith angles of $30^\circ$ and $70^\circ$ compared with the
value of $\sim \times 50$ from our estimate.
\begin{table}
\caption{Observing log for our observations of Markarian 501 during
1997.\label{table:observing_log}}
\begin{indented}
\item[]\begin{tabular}{@{}lc}
\br
Date & No. of scans\\ \ns
& ON source \\
\mr
1997 July 4 & 3\\
1997 July 5 & 5\\
1997 July 7 & 4\\
1997 July 9 & 3\\
1997 July 23 & 1\\
1997 July 28 & 5\\
1997 July 30 & 5\\
1997 July 31 & 3\\
1997 August 1 & 4\\
1997 August 5 & 4\\
\br
\end{tabular}
\end{indented}
\end{table}
Data were accepted for analysis only if the sky was clear and stable
according to the telescope background count rate and a boresighted FIR
radiometer which provides a measure of sky clarity
\cite{kn:buckley1998}. The gross counting rates in each pair of ON-OFF
segments used were consistent at the $2.5 \sigma$ level. A total of 9.25
hrs of data for ON-source observations and an equal amount OFF source
meet these criteria.
\section{Results}
The reduction and analysis of accepted data follows a well established
routine \cite{kn:chadwick99}. The gains and pedestals of all 147 PMTs
and digitizer electronics are calibrated within each 15 min segment
using embedded laser and false coincidence events. PMT noise is
equalised for the ON and OFF source segments using the software padding
technique \cite{kn:cawley1993} prior to identifying the accurate
location of the source in the camera's field of view using the axial CCD
camera. Events confined to within $1.1^{\circ}$ of the centre of the
camera and which have in excess of 600 digital counts (to ensure
reliable image reconstruction) are considered suitable for further
analysis. Finally, the spatial moments of each image relative to the
source position are evaluated and those events are rejected which are
unlikely to have been initiated by gamma rays.
In addition to the background rejection based upon the image analysis, a
measure of the fluctuations between the centroids of the samples
recorded by the left and right collectors of the Mark 6 telescope
($D_{\rm dist}$) provide an additional discriminant
\cite{kn:chadwick1998a}. Gamma rays are identified on the basis of image
shape and left/right fluctuation and then plotting the number of events
as a function of the pointing parameter {\it ALPHA}. Gamma ray events
from a point source appear as an excess at small values of {\it ALPHA}.
The selection criteria appropriate to observations at large zenith
angles ($> 70^{\circ}$) are summarised in table \ref{select_table}.
Small changes have been allowed from the standard set of parameters
identified for the analysis of data taken with the Mark 6 telescope at
zenith angles $< 45^{\circ}$ to accommodate e.g. the narrower images of
cascades propagated through three atmospheres.
\begin{table}
\caption{The image parameter selections applied to the Markarian 501
data recorded at zenith angles between $70^\circ$ and $73^\circ$.
\label{select_table}}
\begin{indented}
\item[]\begin{tabular}{@{}lccccc}
\br
Parameter&Ranges&Ranges&Ranges&Ranges\\
\mr
{\it SIZE} (d.c.)&$600-800$&$800-1200$&$1200-1500$&$1500-5000$\\
{\it DISTANCE}&$0.35^{\circ}-0.85^{\circ}$&$0.35^{\circ}-0.85^{\circ}$&$0.35^{\circ}-0.85^{\circ}$&$0.35^{\circ}-0.85^{\circ}$\\
{\it ECCENTRICITY}&$0.35-0.85$&$0.35-0.85$&$0.35-0.85$&$0.35-0.85$\\
{\it WIDTH}&$ < 0.14^{\circ}$&$ < 0.18^{\circ}$&$ < 0.22^{\circ}$&$ < 0.22^{\circ}$\\
{\it CONCENTRATION}&$ < 0.70$&$ < 0.70$&$ < 0.70$&$ < 0.30$\\
$D_{\rm dist}$&$ < 0.20^{\circ}$&$ < 0.14^{\circ}$&$ < 0.12^{\circ}$&$ < 0.10^{\circ}$\\
\br
\end{tabular}
\end{indented}
\end{table}
The number of events in the ON and OFF samples after the application of
the selections described above are summarised in table
\ref{result_table}. The {\it ALPHA} plot of the differences of the ON
and OFF distributions is shown in figure \ref{fig:alphaplot}. The excess
of events at small values of {\it ALPHA} suggest the detection of a
source with an excess of events with {\it ALPHA} $< 30^{\circ}$ at a
significance of $5.6 \sigma$. The width of the {\it ALPHA} distribution,
with excess events with {\it ALPHA} of up to $30^{\circ}$, may be a
consequence of the effects of magnetic field of $~ 0.4$ G, acting on
the cascade over unusually large linear distances \cite{kn:chadwick99a}.
\begin{table}
\caption{The results of various event selections for the Markarian 501
data. \label{result_table}}
\begin{indented}
\item[]\begin{tabular}{@{}lrrrr}
\br
& On & Off & Difference & Significance \\
\mr
Number of events & 40869 & 41232 & $-363$ & $-1.3~\sigma$ \\
\\
Number of size and & 21570 & 21286 & 284 & $1.4~\sigma$ \\ \ns
distance selected events & & & & \\
\\
Number of shape & 1935 & 1750 & 185 & $3.1~\sigma$ \\ \ns
selected events & & & & \\
\\
Number of shape and & 647 & 475 & 172 & $5.6~\sigma$ \\ \ns
{\it ALPHA} selected events & & & & \\
\br
\end{tabular}
\end{indented}
\end{table}
\begin{figure}[tb]
\centerline{\psfig{file=chadwickb01.EPS,height=10cm}}
\caption{The difference in the distribution of values of {\it ALPHA} for
the shape selected data recorded ON and OFF source from Mrk 501.
\label{fig:alphaplot}}
\end{figure}
The ON and OFF source data have been re-analysed using a matrix of
assumed source positions to provide a ``false source'' analysis. In
figure \ref{fig:mapleplot} we show that the excess of gamma ray events
originates from the source which is located in the right ascension,
declination plot at a position consistent with Mrk 501. (The camera
centre does not coincide with the source position.)
\begin{figure}[tb]
\centerline{\psfig{file=chadwickb02.eps,height=12cm}}
\caption{False source analysis: excess events with {\it ALPHA} $ <
30^\circ$. The contours are spaced at $1.0 \sigma$ intervals, with the
grey scale being such that black indicates a probability of $>
6\sigma$ for a gamma ray originating from that direction. The position
of Mrk 501 is marked by *.} \label{fig:mapleplot}
\end{figure}
\section{Conclusion}
Observations of Mrk 501 during outburst made at zenith angles $>
70^\circ$ using a Southern hemisphere imaging telescope have
demonstrated the efficacy of the large zenith angle ACT technique
suggested by Sommers and Elbert \cite{kn:sommers1987}.
The selection criteria used to reject background in observations with
our imaging telescope at large zenith angles are similar to those
employed for observations at small zenith angles. The main difference is
in the maximum value of the width of images accepted, which are smaller
than those in observations near the zenith due to the increased distance
to cascade maximum.
In the absence of a network of gamma ray observatories giving full sky
coverage, this detection of Mrk 501 has demonstrated that observations
at large zenith angles may allow more continuous monitoring of strong
TeV gamma ray sources.
Pending completion of a simulation study of the detailed response of the
Mark 6 telescope at zenith angles of $\sim 70^\circ$ it is not possible
to ascribe a reliable value to the highest energy gamma rays detected in
this short exposure observation. Thus it has not yet been possible to
extend the spectrum beyond the limiting energy available from more
extensive observations using atmospheric \v{C}erenkov telescope
techniques at smaller zenith angles from Northern hemisphere
observatories.
\ack
We are grateful to the UK Particle Physics and Astronomy Research
Council for support of the project and the University of Sydney for the
lease of the Narrabri site. The Mark 6 telescope was designed and
constructed with the assistance of the staff of the Physics Department,
University of Durham. The efforts of Mrs S E Hilton and Mr K
Tindale are acknowledged with gratitude.
\section*{References}
|
\section*{
References}\list
{\arabic{enumi}.}{\settowidth\labelwidth{#1}\leftmargin\labelwidth
\advance\leftmargin\labelsep
\usecounter{enumi}}
\def\hskip .11em plus .33em minus .07em{\hskip .11em plus .33em minus .07em}
\sloppy\clubpenalty4000\widowpenalty4000
\sfcode`\.=1000\relax}
\let\Large=\large
\def\op#1{\mathop{\fam0 #1}\limits}
\newcommand{{\rm Id\,}}{{\rm Id\,}}
\newcommand{{\rm pr}}{{\rm pr}}
\newcommand{{\rm Id\,}}{{\rm Id\,}}
\newcommand{{\rm Ker\,}}{{\rm Ker\,}}
\newcommand{{\rm Im\, }}{{\rm Im\, }}
\newcommand{\nm}[1]{\mid {#1}\mid}
\newcommand{\begin{equation}}{\begin{equation}}
\newcommand{\end{equation}}{\end{equation}}
\newcommand{\begin{eqnarray}}{\begin{eqnarray}}
\newcommand{\end{eqnarray}}{\end{eqnarray}}
\newcommand{\begin{eqnarray*}}{\begin{eqnarray*}}
\newcommand{\end{eqnarray*}}{\end{eqnarray*}}
\newcommand{\begin{eqalph}}{\begin{eqalph}}
\newcommand{\end{eqalph}}{\end{eqalph}}
\newcommand{{\cal A}}{{\cal A}}
\newcommand{{\cal T}}{{\cal T}}
\newcommand{{\leftarrow}}{{\leftarrow}}
\newcommand{{\rm Ann\,}}{{\rm Ann\,}}
\newcommand{{\cal O}}{{\cal O}}
\newcommand{{\cal T}}{{\cal T}}
\newcommand{{\cal P}}{{\cal P}}
\newcommand{{\cal L}}{{\cal L}}
\newcommand{{\cal V}}{{\cal V}}
\newcommand{{\cal E}}{{\cal E}}
\newcommand{{\cal N}}{{\cal N}}
\newcommand{{\cal R}}{{\cal R}}
\newcommand{{\cal H}}{{\cal H}}
\newcommand{{\cal F}}{{\cal F}}
\newcommand{{\cal C}}{{\cal C}}
\newcommand{{\cal S}}{{\cal S}}
\newcommand{{\cal D}}{{\cal D}}
\newcommand{{\cal B}}{{\cal B}}
\newcommand{{\cal J}}{{\cal J}}
\newcommand{{\bf L}}{{\bf L}}
\newcommand{{\bf Q}}{{\bf Q}}
\newcommand{{\bf R}}{{\bf R}}
\newcommand{{\bf C}}{{\bf C}}
\newcommand{{\bf Z}}{{\bf Z}}
\newcommand{\alpha}{\alpha}
\newcommand{\beta}{\beta}
\newcommand{\delta}{\delta}
\newcommand{\lambda}{\lambda}
\newcommand{\Lambda}{\Lambda}
\newcommand{\phi}{\phi}
\newcommand{\pi}{\pi}
\newcommand{\psi}{\psi}
\newcommand{\xi}{\xi}
\newcommand{\omega}{\omega}
\newcommand{\Omega}{\Omega}
\newcommand{\mu}{\mu}
\newcommand{\nu}{\nu}
\newcommand{\gamma}{\gamma}
\newcommand{\Gamma}{\Gamma}
\newcommand{\epsilon}{\epsilon}
\newcommand{\varepsilon}{\varepsilon}
\newcommand{\theta}{\theta}
\newcommand{\Theta}{\Theta}
\newcommand{\rho}{\rho}
\newcommand{\upsilon}{\upsilon}
\newcommand{\vartheta}{\vartheta}
\newcommand{\widehat\cL}{\widehat{\cal L}}
\newcommand{\sigma}{\sigma}
\newcommand{\Sigma}{\Sigma}
\newcommand{\zeta}{\zeta}
\newcommand{\flat}{\flat}
\newcommand{\sharp}{\sharp}
\newcommand{{\bf\Omega}}{{\bf\Omega}}
\newcommand{{\bf\Theta}}{{\bf\Theta}}
\newcommand{{\bf T}}{{\bf T}}
\newcommand{\wedge}{\wedge}
\newcommand{\widetilde}{\widetilde}
\newcommand{\widehat}{\widehat}
\newcommand{\overline}{\overline}
\newcommand{\partial}{\partial}
\newcommand{\op\longleftarrow}{\op\longleftarrow}
\newcommand{\op\longrightarrow}{\op\longrightarrow}
\newcommand{\op\hookrightarrow}{\op\hookrightarrow}
\newcommand{\longleftrightarrow}{\longleftrightarrow}
\newcommand{\otimes}{\otimes}
\newcommand{\approx}{\approx}
\newcommand{{\ol q}}{{\overline q}}
\renewcommand{\arabic{section}.\arabic{equationa}\alph{equation}}{\thesection.\arabic{equation}}
\let\ssection=\section
\renewcommand{\section}{\setcounter{equation}{0}\ssection}
\newcounter{eqalph}[section]
\newcounter{equationa}[section]
\newcounter{example}[section]
\newcounter{remark}[section]
\newcounter{theorem}[section]
\newcounter{proposition}[section]
\newcounter{lemma}[section]
\newcounter{corollary}[section]
\newcounter{definition}[section]
\setcounter{example}{1}
\setcounter{remark}{1}
\setcounter{theorem}{1}
\setcounter{proposition}{1}
\setcounter{lemma}{1}
\setcounter{corollary}{1}
\setcounter{definition}{1}
\def\arabic{section}.\arabic{remark}{\arabic{section}.\arabic{remark}}
\def\arabic{section}.\arabic{remark}{\arabic{section}.\arabic{remark}}
\def\arabic{section}.\arabic{definition}{\arabic{section}.\arabic{definition}}
\def\arabic{section}.\arabic{definition}{\arabic{section}.\arabic{definition}}
\def\arabic{section}.\arabic{definition}{\arabic{section}.\arabic{definition}}
\def\arabic{section}.\arabic{definition}{\arabic{section}.\arabic{definition}}
\def\arabic{section}.\arabic{definition}{\arabic{section}.\arabic{definition}}
\newenvironment{proof}{\noindent {\it Proof.}}{$\Box$
\medskip }
\newenvironment{ex}{\refstepcounter{remark} \medskip\noindent{\it Example
\arabic{section}.\arabic{remark}.}}{ \medskip }
\newenvironment{rem}{\refstepcounter{remark} \medskip\noindent{\it Remark
\arabic{section}.\arabic{remark}.}}{ \medskip }
\newenvironment{theo}{\refstepcounter{definition} \medskip\noindent{\bf
Theorem \arabic{section}.\arabic{definition}}.\it}{\medskip }
\newenvironment{prop}{\refstepcounter{definition} \medskip\noindent{\bf
Proposition \arabic{section}.\arabic{definition}}.\it}{\medskip }
\newenvironment{lem}{\refstepcounter{definition} \medskip\noindent{\bf Lemma
\arabic{section}.\arabic{definition}}.\it }{\medskip }
\newenvironment{cor}{\refstepcounter{definition} \medskip\noindent{\bf
Corollary \arabic{section}.\arabic{definition}}.\it }{\medskip }
\newenvironment{defi}{\refstepcounter{definition} \medskip\noindent{\bf
Definition \arabic{section}.\arabic{definition}}.\it }{\medskip }
\def\arabic{section}.\arabic{equationa}\alph{equation}{\arabic{section}.\arabic{equation}}
\newenvironment{eqalph}{\stepcounter{equation}
\setcounter{equationa}{\value{equation}}
\setcounter{equation}{0}
\def\arabic{section}.\arabic{equationa}\alph{equation}
{\arabic{section}.\arabic{equationa}\alph{equation}}
\begin{eqnarray}}{\end{eqnarray}
\setcounter{equation}{\value{equationa}}}
\newcommand{\nw}[1]{[{#1}]}
\newcommand{\rm Der}{\rm Der}
\hyphenation{ma-ni-fold La-gran-gi-ans di-men-si-o-nal
-di-men-si-o-nal La-gran-gi-an Ha-mil-to-ni-an multi-symplec-tic}
\begin{document}
\hbox{}
\noindent
{\large \bf Constraints in Hamiltonian time-dependent mechanics}
\newline\newline
{\sc Giovanni Giachetta$^1$, Luigi Mangiarotti$^1$ \newline and Gennadi
Sardanashvily$^2$}
\newline{\small
$^1$ Department of Mathematics and Physics, Camerino University, 62032
Camerino, Italy\newline
E-mail: [email protected]\newline
$^2$ Department of Theoretical Physics, Moscow State University, 117234
Moscow, Russia} \newline
E-mail: [email protected]\newline\newline
{\bf Abstract}
In Hamiltonian time-dependent mechanics, the
Poisson bracket does not define dynamic equations, that implies the
corresponding peculiarities of describing time-dependent holonomic
constraints. As in conservative mechanics, one can consider the Poisson
bracket of constraints, separate them in first and second class constraints,
construct the Koszul--Tate resolution and a BRST complex. However, the Poisson
bracket of constraints and a Hamiltonian makes no sense.
Hamiltonian vector fields for first class constraints are not
generators of gauge transformations. In the case of Lagrangian constraints, we
state the comprehensive relations between solutions of the Lagrange equations
for an almost regular Lagrangian and solutions of the Hamilton equations for
associated Hamiltonian forms, which live in the Lagrangian constraint space.
Degenerate quadratic Lagrangian systems are studied in details. We construct
the Koszul--Tate resolution for Lagrangian constraints of these systems in an
explicit form.
\tableofcontents
\section{Introduction}
The
technique of Poisson and symplectic manifolds is well known to provide the
adequate Hamiltonian formulation of classical and quantum conservative
mechanics. This is also the case of presymplectic Hamiltonian
systems. Since every presymplectic form can be represented as a pull-back of a
symplectic form by a coisotropic imbedding \cite{got82,book98},
a presymplectic Hamiltonian system can be seen as a Dirac
constraint system \cite{cari85,book98}. An autonomous Lagrangian system
also exemplifies a presymplectic Hamiltonian system where a
presymplectic form is the exterior differential of the Poincar\'e--Cartan
form, while a Hamiltonian is the energy function
\cite{cari93a,leon96,book98,mun}.
A generic example of conservative Hamiltonian mechanics
is a regular Poisson manifold $(Z,w)$ where a Hamiltonian is a real
function ${\cal H}$ on $Z$. Given the corresponding Hamiltonian vector field
$\vartheta_{\cal H}=w^\sharp(df)$, the closed subbundle $\vartheta_{\cal H}(Z)$ of the tangent
bundle
$TZ$ is an
autonomous first order dynamic equation on a manifold $Z$, called the Hamilton
equations. The evolution equation on the Poisson algebra $C^\infty(Z)$ is the
Lie derivative $ {\bf L}_{\vartheta_{\cal H}}f= \{{\cal H},f\}$,
expressed into the Poisson bracket of the Hamiltonian ${\cal H}$ and
functions $f$ on $Z$. This
description, however, cannot be extended in a straightforward manner to
time-dependent mechanics subject to time-dependent transformations.
The existent formulations of time-dependent mechanics
imply usually a preliminary splitting
of a configuration space $Q={\bf R}\times M$ and a momentum phase space
$\Pi={\bf R}\times Z$,
where $Z$ is a Poisson manifold \cite{cari93,chinea,eche,ham,mora,leon93}.
From the physical viewpoint, this means that a certain reference frame is
chosen. In this case, the momentum phase space $\Pi$ is endowed with the
Poisson product of the zero Poisson structure on
${\bf R}$ and the Poisson structure on $Z$. A
Hamiltonian is defined as a real function ${\cal H}$ on
$\Pi$. The corresponding Hamiltonian vector field $\vartheta_{\cal H}$ on
$\Pi$ is vertical with respect to
the fibration $\Pi\to{\bf R}$. Due to the canonical imbedding
\begin{equation}
\Pi\op\times_{\bf R} T{\bf R}\to T\Pi, \label{mm6}
\end{equation}
one
introduces the vector field
\begin{equation}
\gamma_{\cal H}=\partial_t +\vartheta_{\cal H}, \label{mm0}
\end{equation}
where $\partial_t$ is the standard vector field on ${\bf R}$ \cite{ham}. The first order
dynamic equation $\gamma_{\cal H}(\Pi)\subset T\Pi$ on the manifold $\Pi$ plays the
role of Hamilton equations. The evolution equation on the Poisson algebra
$C^\infty(\Pi)$ is given by the Lie derivative
\begin{equation}
{\bf L}_{\gamma_{\cal H}}f= \partial_t f +\{{\cal H},f\}. \label{mm7}
\end{equation}
This is not the case of mechanical systems subject to time-dependent
transformations. These transformations, including canonical and
inertial frame transformations, violate the splitting
${\bf R}\times Z$. As a consequence, there is no
canonical imbedding (\ref{mm6}), and the vector field (\ref{mm0}) is not well
defined. At the same time, one can treat the imbedding (\ref{mm6}) as a
trivial connection on the bundle $\Pi\to{\bf R}$, while $\gamma_{\cal H}$
(\ref{mm0}) is the sum of the horizontal lift onto $\Pi$ of the vector field
$\partial_t$ by this connection and of the vertical vector field
$\vartheta_{\cal H}$. This observation make us to think of non-relativistic
time-dependent mechanics as being a particular field theory on fibre bundles
over ${\bf R}$, where the time axis ${\bf R}$ is parameterized by the
Cartesian coordinates
$t$ with the transition functions
$t'=t+$const. Then ${\bf R}$ is
provided with the above mentioned standard vector field $\partial_t$ and the
standard 1-form
$dt$. Every fibre bundle over ${\bf R}$ is obviously trivial, but
its trivialization is not necessarily canonical.
\begin{rem}
The following peculiarity of bundles over ${\bf R}$ is important. Let
$Y\to{\bf R}$ be a fibre bundle coordinated by $(t,y^A)$, and
$J^1Y$ its first order jet manifold, equipped with the adapted coordinates
$(t,y^A,y^A_t)$. There is the canonical imbedding
\begin{equation}
\lambda=\partial_t+y^A_t\partial_A: J^1Y\op\hookrightarrow_Y TY \label{mm8}
\end{equation}
onto the affine subbundle
of $TY\to Y$ of elements $\upsilon\in TY$ such that
$\upsilon\rfloor dt=1$. This subbundle is modelled over the vertical tangent bundle
$VY\to Y$. As a consequence, there is one-to-one correspondence between the
connections $\Gamma$ on the fibre bundle $Y\to {\bf R}$, treated as sections of the
affine jet bundle $\pi^1_0:J^1Y\to Y$ \cite{book99}, and the nowhere vanishing
vector fields $\Gamma=\partial_t +\Gamma^A\partial_A$ on $Y$, called
horizontal vector fields,
such that $\Gamma\rfloor dt=1$ \cite{book98,book99}. The corresponding
covariant differential reads
\begin{eqnarray*}
D_\Gamma=\lambda-\Gamma:J^1Y\op\to_Y VY, \qquad \dot y^A\circ D_\Gamma=y^A_t-\Gamma^A.
\end{eqnarray*}
Let us also recall the total derivative
$d_t =
\partial_t +y^A_t\partial_A +\cdots$ and the exterior algebra
homomorphism
\begin{equation}
h_0:\phi dt+ \phi_A dy^A \mapsto (\phi+\phi_A y^A_t) dt \label{cmp100}
\end{equation}
which sends exterior forms on $Y\to{\bf R}$ onto the horizontal forms on
$J^1Y\to {\bf R}$,
and vanishes on contact forms $\theta^A=dy^A
-y^A_tdt$.
\end{rem}
Lagrangian time-dependent mechanics follows
directly Lagrangian field theory
\cite{giach92,krupkova,leon97a,book98,massa}. It implies the
existence of a configuration space $Q\to{\bf R}$ of a mechanical system,
and a Lagrangian is
defined as a horizontal density
\begin{equation}
L={\cal L} dt, \qquad {\cal L}: J^1Q\to{\bf R}, \label{mm43}
\end{equation}
on the velocity phase space $J^1Q$.
However, there is the essential difference
between field
theory and time-dependent mechanics.
The curvature of any connection $\Gamma$ on
a configuration bundle $Q\to{\bf R}$ vanishes
identically, and these
connections fail to be dynamic variables, but characterize reference frames.
The horizontal vector
field $\Gamma$ sets a tangent vector at
each point of the configuration space
$Q$, which can be seen as the velocity of an "observer" at this point
\cite{book98,massa,sard98}. There is the correspondence
between the connections $\Gamma$ on the configuration bundle $Q\to{\bf R}$ and the
trivializations of $Q\to{\bf R}$ such that $\Gamma=\partial_t$ in the adapted coordinates
(see Section 2).
A generic momentum phase space of time-dependent mechanics
is a fibre bundle $\Pi\to {\bf R}$ endowed with a regular Poisson structure whose
characteristic distribution belongs to the vertical tangent bundle $V\Pi$ of
$\Pi\to {\bf R}$ \cite{ham}. Such a Poisson structure however cannot provide
dynamic equations. A first order dynamic equation on
$\Pi\to{\bf R}$, by definition, is a section of the affine jet bundle $J^1\Pi\to
\Pi$, i.e., a connection on $\Pi\to{\bf R}$. Being a horizontal vector
field, such a
connection cannot be a Hamiltonian vector field with respect to the
above mentioned Poisson structure on $\Pi$.
One can overcome this difficulty as follows.
Let $Q\to{\bf R}$ be a configuration bundle of time-dependent mechanics.
The corresponding momentum phase space is the vertical cotangent bundle
$\Pi=V^*Q\to {\bf R}$, called the Legendre bundle, while the cotangent bundle
$T^*Q$ is the homogeneous momentum phase
space. $T^*Q$ admits the canonical Liouville form
$\Xi$ and the symplectic form
$d\Xi$, together with the corresponding non-degenerate Poisson
bracket $\{,\}_T$ on the ring $C^\infty(T^*Q)$.
Let us consider the subring of $C^\infty(T^*Q)$ which comprises the
pull-backs $\zeta^*f$ onto $T^*Q$ of functions $f$ on the vertical
cotangent bundle $V^*Q$ by the canonical fibration
\begin{equation}
\zeta:T^*Q\to V^*Q. \label{mm5}
\end{equation}
This subring is closed under the Poisson bracket
$\{,\}_T$, and
$V^*Q$ is provided with the regular Poisson structure $\{,\}_V$ such that
\begin{equation}
\zeta^*\{f,g\}_V=\{\zeta^*f,\zeta^*g\}_T\label{m72'}
\end{equation}
\cite{vais}. Its characteristic
distribution coincides with the
vertical tangent bundle $VV^*Q$ of $V^*Q\to {\bf R}$.
Given a section $h$ of the bundle (\ref{mm5}), let us consider the
pull-back forms
\begin{equation}
{\bf\Theta} =h^*(\Xi\wedge dt), \qquad {\bf\Omega}=h^*(d\Xi\wedge dt) \label{z401}
\end{equation}
on $V^*Q$, but these forms are independent of a section
$h$ and are canonical exterior forms on $V^*Q$. The pull-backs $h^*\Xi$ are
called the Hamiltonian forms.
With ${\bf\Omega}$,
the Hamiltonian vector field
$\vartheta_f$ for a function $f$ on $V^*Q$ is given by the relation
\begin{equation}
\vartheta_f\rfloor{\bf\Omega} = -df\wedge dt, \label{z406}
\end{equation}
while the Poisson bracket (\ref{m72'}) is written as
\begin{eqnarray*}
\{f,g\}_Vdt=\vartheta_g\rfloor\vartheta_f\rfloor{\bf\Omega}.
\end{eqnarray*}
Note that a generic momentum
phase space
$\Pi\to{\bf R}$ of time-dependent mechanics can be seen locally as
the Poisson product over ${\bf R}$ of the Legendre bundle $V^*Q\to{\bf R}$ and
a fibre bundle over
${\bf R}$, equipped with the zero Poisson structure.
The pair $(V^*Q,{\bf\Omega})$ is the particular
$(n=1)$-polysymplectic phase space of the covariant Hamiltonian field theory
(see
\cite{cari,book,got91,sard95} for a survey). Following its general scheme, we
can formulate the Hamiltonian time-dependent mechanics as follows
\cite{book98,sard98}.
A connection $\gamma$ on the Legendre bundle
$V^*Q\to{\bf R}$ is called canonical if the corresponding horizontal vector
field is canonical for the Poisson structure on $V^*Q$, i.e., the form
$\gamma\rfloor{\bf\Omega}$ is closed. We will prove that such a form is necessarily exact.
A canonical connection $\gamma$ is a said to be a Hamiltonian connection if
\begin{equation}
\gamma\rfloor{\bf\Omega}= dH \label{z405}
\end{equation}
where $H$ is a Hamiltonian form on $V^*Q$. We show that every Hamiltonian form
admits a unique Hamiltonian connection $\gamma_H$, and that any canonical
connection is locally a Hamiltonian one. Given a Hamiltonian form $H$,
the kernel of the covariant differential $D_{\gamma_H}$, associated with the
Hamiltonian connection $\gamma_H$, is a closed imbedded subbundle of the jet
bundle
$J^1V^*Q\to {\bf R}$, and so is the system of first order PDEs on
the Legendre bundle
$V^*Q\to {\bf R}$. These are
the Hamilton equations in time-dependent mechanics, while the Lie derivative
\begin{equation}
{\bf L}_{\gamma_H}f=\gamma_H\rfloor df
\label{m59}
\end{equation}
defines the evolution equation on $C^\infty(V^*Z)$. As in
the polysymplectic case \cite{book,sard94,sard95},
this Hamiltonian dynamics is equivalent to the Lagrangian one for
hyperregular Lagrangians, while a degenerate
Lagrangian involves a set of associated Hamiltonian forms in
order to exhaust solutions of the Lagrange equations.
The main peculiarity of Hamiltonian time-dependent mechanics
lies in the fact that, since $\gamma_H$ is not a
vertical vector field, the right-hand side of the evolution equation
(\ref{m59}) is not expressed into the Poisson bracket in a canonical way, but
contains a frame-dependent term.
Every connection $\Gamma$ on the
configuration bundle $Q\to{\bf R}$ is an affine section of
the bundle (\ref{mm5}), and defines the Hamiltonian form $H_\Gamma=\Gamma^*\Xi$ on
$V^*Q$. The corresponding Hamiltonian connection is the canonical lift
$V^*\Gamma$ of $\Gamma$ onto the Legendre bundle $V^*Q$ \cite{book,book99}.
Then any Hamiltonian form $H$ on $V^*Q$ admits splittings
\begin{eqnarray}
&& H=H_\Gamma -\widetilde{\cal H}_\Gamma dt, \label{m46'}\\
&& \gamma_H=V^*\Gamma + \vartheta_{\widetilde{\cal H}_\Gamma}, \nonumber
\end{eqnarray}
where $\vartheta_{\widetilde{\cal H}_\Gamma}$ is the vertical Hamiltonian field for the function
$\widetilde{\cal H}_\Gamma$, which the energy
function with respect to the reference frame $\Gamma$ (see Section 4). With the
splitting (\ref{m46'}), the evolution equation (\ref{m59}) takes the form
\begin{equation}
{\bf L}_{\gamma_H}f= V^*\Gamma\rfloor H +\{\widetilde{\cal H}_\Gamma,f\}_V. \label{m96}
\end{equation}
Let the configuration bundle $Q\to{\bf R}$ with an $m$-dimensional typical
fibre $M$ be coordinated by
$(t,q^i)$. Then Legendre bundle $V^*Q$ and the cotangent bundle
$T^*Q$ are provided with holonomic coordinates $(t,q^i,p_i=\dot q_i)$ and
$(t,q^i,p_i,p)$, respectively. Relative to these
coordinates, a Hamiltonian form
$H$ on $V^*Q$ reads
\begin{equation}
H=h^*\Xi= p_i dq^i -{\cal H} dt. \label{b4210}
\end{equation}
It is the well-known integral
invariant of Poincar\'e--Cartan, where ${\cal H}$ is a Hamiltonian
in time-dependent mechanics. A glance at the expression (\ref{b4210}) shows
that ${\cal H}$ fails to be a scalar under time-dependent
transformations.
Accordingly, the evolution equation (\ref{m96}) takes the local form
\begin{equation}
{\bf L}_{\gamma_H}=\partial_tf +\{{\cal H},f\}_V, \label{mm15}
\end{equation}
but one should bear in mind that the terms in its right-hand side,
taken separately, are not well-behaved objects under time-dependent
transformations. In particular, the equality
$\{{\cal H},f\}_V=0$ is not preserved under time-dependent
transformations.
The above peculiarities of Hamiltonian time-dependent mechanics
imply the corresponding peculiarities of describing time-dependent holonomic
constraints. As in conservative mechanics, one can consider the Poisson
bracket of constraints, separate them in first and second class constraints,
construct the Koszul--Tate resolution and BRST complex. However, the Poisson
bracket of constraints and a Hamiltonian makes no sense in time-dependent
mechanics. Hamiltonian vector fields for first class constraint functions are
not generators of gauge transformations. We will pay a special attention to
Lagrangian constraints. Every Lagrangian
$L$ defines the Legendre map
\begin{equation}
\widehat L:J^1Q \op\to_Q V^*Q, \qquad p_i\circ\widehat L =\pi_i,\label{a303}
\end{equation}
whose image $N_L=\widehat L(J^1Q)\subset V^*Q$ is called the Lagrangian constraint
space. We state the comprehensive relationship between solutions of the
Lagrange equations for an almost regular Lagrangian $L$ and solutions in $N_L$
of the Hamilton equations for associated Hamiltonian forms. The detailed
analysis of degenerate quadratic Lagrangian systems in Section 7 is
appropriate for application to many physical models. In Section 9, we
construct the Koszul--Tate resolution for Lagrangian constraints of such a
degenerate system in an explicit form.
\section{Interlude I. Non-relativistic reference frames}
As was mentioned above, a reference frame in non-relativistic mechanics is
identified with a connection
$\Gamma$ on the configuration bundle
$Q\to{\bf R}$. Being flat, every
connection $\Gamma$ on $Q\to{\bf R}$
yields an integrable
horizontal distribution on $Q$, whose integral manifolds
are integral curves of the horizontal vector field $\Gamma$ which are
transversal to the fibres of the bundle $Q\to{\bf R}$.
\begin{prop} \label{gn1} {\rm \cite{book,book99}.}
Each connection $\Gamma$ on a
bundle $Q\to{\bf R}$ defines an atlas of local constant trivializations of
$Q\to{\bf R}$ such that the associated bundle coordinates $(t,{\ol q}^i)$ on $Q$
possess the transition function
${\ol q}^i\to {\ol q}'^i({\ol q}^j)$ independent of $t$, and
$\Gamma=\partial_t$
with respect to these coordinates. Conversely,
every atlas of local constant trivializations of the bundle
$Q\to{\bf R}$ sets a connection on $Q\to{\bf R}$ which is $\partial_t$
relative to this atlas.
\end{prop}
\begin{prop} {\rm \cite{book98,sard98}.}
Every bundle trivialization
\begin{equation}
\psi: Q\cong {\bf R}\times M \label{m33}
\end{equation}
yields a complete
horizontal vector field $\Gamma$ on this bundle. Conversely, every complete
connection
$\Gamma$ on $Q\to{\bf R}$ defines its trivialization (\ref{m33})
such that $\Gamma=\partial_t$.
\end{prop}
One can think
of the atlas of
local constant trivializations and the bundle coordinates $(t,{\ol q}^i)$ in
Proposition \ref{gn1} as being also a reference frame corresponding to the
connection $\Gamma$.
These coordinates are said to be adapted to the reference frame $\Gamma$. In
particular, the Hamiltonian form $H_\Gamma$ relative to the adapted coordinates
reduces to the pure kinematic term $H_\Gamma=p_idy^i$. Therefore, we will call
$H_\Gamma$ a frame Hamiltonian form.
Unless otherwise stated, by a reference frame will be meant
a complete reference frame.
Given a trivialization (\ref{m33}) of the
configuration bundle $Q\to{\bf R}$, we have the corresponding trivializations of
velocity and momentum phase spaces
\begin{eqnarray*}
J^1Q\cong {\bf R}\times TM, \qquad V^*Q\cong {\bf R}\times T^*M.
\end{eqnarray*}
\section{Interlude II. Lagrangian time-dependent dynamics}
To obtain a complete
picture of the relations between Lagrangian and Hamiltonian time-dependent
mechanics in Section 6, we will refer to the following three types of PDEs in
the first order calculus of variations. These are Lagrange, Cartan and
Hamilton--De Donder equations.
Given a Lagrangian $L$
on the velocity phase space $J^1Q$,
we follow the first variational formula of the
calculus of variations \cite{book,book98,sard97}, which provides the
canonical decomposition of the Lie derivative ${\bf L}_{J^1u}L= (J^1u\rfloor{\cal L})dt$
of $L$ along a projectable vector field $u$ on $Q\to{\bf R}$. We have
\begin{equation}
J^1u\rfloor{\cal L}= u_V\rfloor {\cal E}_L + d_t(u\rfloor H_L), \label{C30}
\end{equation}
where $u_V=(u\rfloor\theta^i)\partial_i$,
\begin{equation}
H_L=L +\pi_i\theta^i, \quad \pi_i=\partial^t_i{\cal L},
\label{303}
\end{equation}
is the Poincar\'e--Cartan form and
\begin{equation}
{\cal E}_L=
(\partial_i- d_t\pi_i){\cal L} \overline dq^i: J^2Q\to V^*Q \label{305}
\end{equation}
is the Euler--Lagrange operator associated with $L$. The kernel
${\rm Ker\,}{\cal E}_L\subset J^2Q$ of ${\cal E}_L$ defines the Lagrange equations on
$Q$, given by the coordinate relations
\begin{equation}
(\partial_i- d_t\pi_i){\cal L}=0. \label{b327}
\end{equation}
On-shell,
the first variational formula (\ref{C30}) leads to
the weak identity
\begin{eqnarray*}
{\bf L}_{J^1u}L\approx d_t(u\rfloor H_L)dt,
\end{eqnarray*}
and then, if ${\bf L}_{J^1u}L=0$,
to the weak conservation law
\begin{equation}
0\approx dt(u\rfloor H_L)= - d_t {\cal T} \label{K4}
\end{equation}
of the symmetry current
\begin{equation}
{\cal T} =-(u\rfloor H_L)=-\pi_i(u^t q^i_t-u^i
)-u^t{\cal L}. \label{Q30}
\end{equation}
Being the Lepagean equivalent of
the Lagrangian $L$ on $J^1Q$ (i.e., $L=h_0(H_L)$ where $h_0$ is the morphism
(\ref{cmp100})), the Poincar\'e--Cartan form $H_L$ (\ref{303}) is
also the Lepagean equivalent of the Lagrangian
\begin{equation}
\overline L = \widehat h_0(H_L) = ({\cal L} + (\widehat q_t^i - q_t^i)\pi_i)dt, \qquad \widehat
h_0(dy^i)=\widehat y^i_t dt,
\label{cmp80}
\end{equation}
on the repeated jet manifold $J^1J^1Q$, coordinated by $(t,q^i,q^i_t,\widehat
q^i_t, q^i_{tt})$. The
Euler--Lagrange operator ${\cal E}_{\overline L}:J^1J^1Q\to V^*J^1Q$ for $\overline L$ reads
\begin{eqnarray}
&& {\cal E}_{\overline L} = (\partial_i{\cal L} - \widehat d_t\pi_i
+ \partial_i\pi_j(\widehat q_t^j - q_t^j))\overline dq^i + \partial_i^t\pi_j(\widehat
q_t^j - q_t^j) \overline dq_t^i, \label{2237} \\
&&\widehat d_t=\partial_t +\widehat q^i_t\partial_i +q^i_{tt}\partial_i^t. \nonumber
\end{eqnarray}
Its kernel
${\rm Ker\,}{\cal E}_{\overline L}\subset J^1J^1Y$ defines the Cartan equations
\begin{equation}
\partial_i^t\pi_j(\widehat q_t^j - q_t^j)=0, \qquad
\partial_i {\cal L} - \widehat d_t\pi_i
+ (\widehat q_t^j - q_t^j)\partial_i\pi_j=0. \label{b336}
\end{equation}
Since ${\cal E}_{\overline L}\mid_{J^2Q}={\cal E}_L$, the Cartan equations (\ref{b336}) are
equivalent to the Lagrange equations (\ref{b327}) on integrable
sections $\overline c=\dot c$ of
$J^1Q\to {\bf R}$. These equations are equivalent
in the case of regular Lagrangians.
On sections $\overline c: {\bf R}\to J^1Q$, the Cartan equations (\ref{b336}) are
equivalent to the relation
\begin{equation}
\overline c^*(u\rfloor dH_L)=0 \label{C28}
\end{equation}
which is assumed to hold for all vertical vector fields $u$ on $J^1Q\to {\bf R}$.
With the Poincar\'e--Cartan form $H_L$ (\ref{303}),
we have the Legendre morphism
\begin{eqnarray*}
\widehat H_L: J^1Q\op\to_Q T^*Q, \qquad
(p_i, p)\circ\widehat H_L =(\pi_i, {\cal L}-\pi_i q^i_t ).
\end{eqnarray*}
Let $Z_L=\widehat H_L(J^1Q)$ be an imbedded subbundle
$i_L:Z_L\hookrightarrow T^*Q$ of $T^*Q\to Q$.
It is provided with the pull-back
De Donder form $i^*_L\Xi$. We have
\begin{equation}
H_L=\widehat H_L^*\Xi_L=\widehat H_L^*(i_L^*\Xi). \label{cmp14}
\end{equation}
By analogy with the Cartan equations
(\ref{C28}), the Hamilton--De Donder equations for sections $\overline r$ of
$T^*Q\to {\bf R}$ are written as
\begin{equation}
\overline r^*(u\rfloor d\Xi_L)=0 \label{N46}
\end{equation}
where $u$ is an arbitrary vertical vector field on
$T^*Q\to {\bf R}$.
\begin{theo}\label{ddd} {\rm\cite{got91}.} Let the Legendre morphism
$\widehat H_L:J^1Q\to Z_L$ be a submersion. Then a section $\overline c$ of $J^1Q\to {\bf R}$
is a solution of the Cartan equations (\ref{C28}) iff $\widehat
H_L\circ\overline c$ is a solution of the Hamilton--De Donder equations
(\ref{N46}), i.e., Cartan and Hamilton--De Donder equations are
quasi-equivalent.
\end{theo}
\section{Hamiltonian time-dependent dynamics}
Let the Legendre bundle $V^*Q\to{\bf R}$ be provided with the holonomic
coordinates $(t,q^i,q^i_t)$. Relative to these coordinates,
the canonical 3-form ${\bf\Omega}$ (\ref{z401}) and
the canonical
Poisson structure (\ref{m72'}) on $V^*Q$ read
\begin{eqnarray}
&& {\bf\Omega} = dp_i\wedge dq^i\wedge dt, \label{z401'} \\
&&\{f,g\}_V=\partial^if\partial_ig-\partial^ig\partial_if, \qquad f,g\in C^\infty{V^*Q}.
\label{m72}
\end{eqnarray}
The corresponding symplectic foliation
coincides with the fibration $V^*Q\to {\bf R}$. The symplectic forms on the
fibres of
$V^*Q\to{\bf R}$ are the pull-backs $\Omega_t=dp_i\wedge dq^i$
of the canonical symplectic form on the typical fibre $T^*M$ of the
Legendre bundle $V^*Q\to {\bf R}$ with respect to trivialization morphisms
\cite{cari89,ham,sard98}. Given such a trivialization,
the Poisson structure (\ref{m72}) is isomorphic to the product of the
zero Poisson structure on ${\bf R}$ and the canonical symplectic structure
on $T^*M$.
\begin{rem}
It is easily seen that an automorphism $\rho$ of the Legendre bundle
$V^*Q\to{\bf R}$ is a canonical transformation of the Poisson structure
(\ref{m72}) iff it preserves the canonical 3-form ${\bf\Omega}$ (\ref{z401'}).
Let us emphasize that canonical transformations are compatible with
the fibration $V^*Q\to{\bf R}$, but not necessarily with the fibration
$\pi_Q:V^*Q\to Q$. We will restrict ourselves to the
holonomic coordinates on $V^*Y$ and holonomic transformations which are
obviously canonical.
\end{rem}
With respect to the Poisson bracket (\ref{m72}), the Hamiltonian vector field
$\vartheta_f$ for a function $f$ on the momentum phase space $V^*Q$ is
\begin{equation}
\vartheta_f = \partial^if\partial_i- \partial_if\partial^i. \label{m73}
\end{equation}
A Hamiltonian vector field, by definition, is canonical.
A converse is the following.
\begin{prop} \label{ex0}
Every vertical canonical
vector field on the Legendre bundle
$V^*Q\to{\bf R}$ is locally a Hamiltonian vector field.
\end{prop}
The proof is based on the following facts.
\begin{lem}\label{ex1}
Let $\sigma$ be a 1-form on $V^*Q$. If $\sigma\wedge dt$ is closed form, it is exact.
\end{lem}
\begin{proof}
Since $V^*Q$ is diffeomorphic to ${\bf R}\times T^*M$, we have the De
Rham cohomology group
\begin{eqnarray*}
H^2(V^*Q)= H^0({\bf R})\otimes H^2(T^*M) \oplus H^1({\bf R})\otimes H^1(T^*M).
\end{eqnarray*}
The form $\sigma\wedge dt$ belongs to its second item which is zero.
\end{proof}
\begin{lem}\label{ex2}
If the 2-form $\sigma\wedge dt$ is exact, then $\sigma\wedge dt= dg\wedge dt$ locally.
\end{lem}
\begin{proof}
The proof is based on the relative Poincar\'e lemma \cite{book}.
\end{proof}
Let $\gamma=\partial_t +\gamma^i\partial_i +\gamma_i\partial^i$ be a canonical connection on the
Legendre bundle $V^*Q\to{\bf R}$. Its components obey the relations
\begin{equation}
\partial^i\gamma^j-\partial^j\gamma^i=0, \qquad
\partial_i\gamma_j- \partial_j\gamma_i=0,\qquad
\partial_j\gamma^i+\partial^i\gamma_j=0.\label{d025}
\end{equation}
Canonical connections constitute an affine space modelled
over the vector space of vertical canonical vector fields on $V^*Q\to {\bf R}$.
\begin{prop}\label{ex3}
If $\gamma$ is a canonical connection, then the form
$\gamma\rfloor{\bf\Omega}$ is exact.
\end{prop}
\begin{proof}
Every connection $\Gamma$
on $Q\to {\bf R}$ gives rise to the connection
\begin{equation}
V^*\Gamma=\partial_t +\Gamma^i\partial_i -p_i\partial_j\Gamma^i \partial^j \label{m38}
\end{equation}
on $V^*Q\to{\bf R}$
which is a Hamiltonian connection for the frame Hamiltonian form
\begin{equation}
V^*\Gamma\rfloor{\bf\Omega}=dH_\Gamma, \qquad H_\Gamma=p_idq^i -p_i\Gamma^idt. \label{m61}
\end{equation}
Let us consider the decomposition $\gamma=V^*\Gamma +\vartheta$,
where $\Gamma$ is a connection on $Q\to {\bf R}$. The assertion follows from the
relation (\ref{m61}) and Proposition \ref{ex0}.
\end{proof}
Thus, every canonical connection $\gamma$ on $V^*Q$ defines an
exterior 1-form $H$ modulo closed forms so that
$dH=\gamma\rfloor{\bf\Omega}$. Such a form is called a locally Hamiltonian form.
\begin{prop}\label{germ}
Every locally Hamiltonian form on the momentum phase space $V^*Q$ is locally a
Hamiltonian form modulo closed forms.
\end{prop}
\begin{proof}
Given locally Hamiltonian forms $H_\gamma$ and $H_{\gamma'}$, their difference
$\sigma=H_\gamma -H_{\gamma'}$ is a 1-form on $V^*Q$ such that the 2-form $\sigma\wedge dt$ is
closed. By virtue of Lemmas \ref{ex1} and \ref{ex2}, the form $\sigma\wedge dt$
is exact and $\sigma=fdt + dg$ locally. Put $H_{\gamma'}=H_\Gamma$ where $\Gamma$ is
a connection on $V^*Q\to{\bf R}$. Then
$H_\gamma$ modulo closed forms takes the local form $H_\gamma= H_\Gamma + fdt$, and
coincides with the pull-back of the Liouville form
$\Xi$ on $T^*Q$ by the local section $p=-p_i\Gamma^i+f$
of the fibre bundle (\ref{mm5}).
\end{proof}
\begin{prop} \label{gm452'} Conversely, each Hamiltonian form $H$
on the momentum phase space
$V^*Q$ admits a unique canonical connection $\gamma_H$ on $V^*Q\to {\bf R}$
such that the relation (\ref{z405}) holds.
\end{prop}
\begin{proof} Given a Hamiltonian form $H$, its exterior differential
\begin{equation}
dH=h^*d\Xi=(dp_i+\partial_i{\cal H} dt)\wedge(dq^i-\partial^i{\cal H} dt) \label{z408}
\end{equation}
is a presymplectic form of
constant rank $2m$ since the form
\begin{equation}
(dH)^m =(dp_i\wedge dq^i)^m -m(dp_i\wedge dq^i)^{m-1}\wedge d{\cal H}\wedge dt \label{m90}
\end{equation}
is nowhere vanishing. It is also seen that $(dH)^m\wedge dt\neq 0$.
It follows that the kernel of $dH$ is a 1-dimensional
distribution. Then the desired Hamiltonian connection
\begin{equation}
\gamma_H=\partial_t +\partial^i{\cal H}\partial_i -\partial_i{\cal H}\partial^i \label{m57}
\end{equation}
is a unique vector field $\gamma_H$ on $V^*Q$ such that $\gamma_H\rfloor dH=0$,
$\gamma_H\rfloor dt=1$.
\end{proof}
\begin{rem}
Hamiltonian forms constitute an affine space
modelled over the vector space of horizontal densities $fdt$ on $V^*Q\to{\bf R}$,
i.e., over $C^\infty(V^*Q)$. Accordingly Hamiltonian
connections $\gamma_H$ form an affine space modelled over the vector
space of Hamiltonian vector fields.
Every Hamiltonian form $H$ defines the associated Hamiltonian map
\begin{equation}
\widehat H=J^1\pi_Q\circ\gamma_H:\partial_t + \partial^i{\cal H}:V^*Q\to J^1Q. \label{mm41}
\end{equation}
With the Hamiltonian
map (\ref{mm41}), we have another Hamiltonian form
\begin{equation}
H_{\widehat H}= -\widehat H\rfloor{\bf\Theta}= p_idq^i -p_i\partial^i{\cal H}. \label{mm42}
\end{equation}
It is readily observed that $H_{\widehat H}=H$ iff $H$ is a frame Hamiltonian
form.
\end{rem}
Given a Hamiltonian connection $\gamma_H$ (\ref{m57}), the corresponding Hamilton
equations
$D_{\gamma_H}=0$ take the coordinate form
\begin{eqalph}
&& q^i_t =\partial^i{\cal H}, \label{m41a}\\
&& p_{ti} =-\partial_i{\cal H}. \label{m41b}
\end{eqalph}
Their classical solutions are integral sections of the
Hamiltonian connection $\gamma_H$, i.e.,
$\dot r=\gamma_H\circ r$.
On sections
$r$ of the Legendre bundle $V^*Q\to {\bf R}$, the Hamilton equations (\ref{m41a})
-- (\ref{m41b})
are equivalent to the relation
\begin{equation}
r^*(u\rfloor dH)=0 \label{y5}
\end{equation}
which is assumed to hold for any vertical vector field $u$ on $V^*Q\to{\bf R}$.
The following two constructions are useful.
It is readily observed that a Hamiltonian form $H$
(\ref{b4210}) is the Poincar\'e--Cartan form (\ref{303}) for the Lagrangian
\begin{equation}
L_H=h_0(H) = (p_iq^i_t - {\cal H})\omega \label{Q3}
\end{equation}
on the jet manifold $J^1V^*Q$. Given
a projectable vector field $u$ on the configuration bundle
$Q\to{\bf R}$ and its lift
\begin{equation}
\widetilde u=u^t\partial_t + u^i\partial_i - \partial_i u^j p_j\partial^i \label{cmp90}
\end{equation}
onto the Legendre bundle $V^*Q\to{\bf R}$, we have
\begin{equation}
{\bf L}_{\widetilde u}H= {\bf L}_{J^1\widetilde u}L_H. \label{b4180}
\end{equation}
It is easily seen that the
Hamilton equations (\ref{m41a}) -- (\ref{m41b}) for $H$
are exactly the Lagrange equations
for $L_H$, i.e., they
characterize the
kernel of the Euler--Lagrange operator
\begin{equation}
{\cal E}_H= (q^i_t-\partial^i{\cal H})\overline dp_i -(p_{ti}+\partial_i{\cal H})\overline dq^i: J^1V^*Q\to
V^*V^*Q \label{mm40}
\end{equation}
for the Lagrangian $L_H$, called the Hamilton
operator for $H$.
Using the relation (\ref{b4180}), let us obtain the Hamiltonian conservation
laws in time-dependent mechanics. As in field theory, by gauge
transformations in time-dependent mechanics are meant automorphism of the
configuration bundle $Q\to {\bf R}$, but only over translations of
the base ${\bf R}$. Then, projectable vector fields
\begin{equation}
u=u^t\partial_t +u^i\partial_i, \qquad u\rfloor dt=u^t={\rm const.}, \label{m223}
\end{equation}
on $V^*Q\to{\bf R}$ can be seen as generators of local 1-parameter groups
of local gauge transformations.
Given a Hamiltonian form $H$ (\ref{b4210}), its Lie derivative (\ref{b4180})
reads
\begin{equation}
{\bf L}_{\widetilde u}H= {\bf L}_{J^1\widetilde u}L_H=(-u^t\partial_t{\cal H}+p_i\partial_tu^i -u^i\partial_i{\cal H} +\partial_j
u^ip_i\partial^j{\cal H})dt. \label{mm24}
\end{equation}
The first variational formula (\ref{C30}) applied to the Lagrangian
$L_H$ (\ref{Q3}) leads to the weak identity ${\bf L}_{\widetilde u}H\approx d_t(u\rfloor
H)dt$. If the Lie derivative (\ref{mm24}) vanishes, we have the conserved
symmetry current
\begin{equation}
J_u= u\rfloor dH= p_iu^i - u^t{\cal H}, \label{b4306}
\end{equation}
along $u$. Every vector field (\ref{m223}) is a
superposition of a vertical vector field and a reference frame on $Q\to{\bf R}$.
If $u$ is a vertical vector field, $J_u$ is the N\"other current
\begin{equation}
J_u(q)= u\rfloor q=p_iu^i, \qquad q=p_i\overline dq^i\in V^*Q. \label{z606}
\end{equation}
The symmetry current along a reference frame $\Gamma$
\begin{equation}
J_\Gamma=p_i\Gamma^i-{\cal H}=-\widetilde{\cal H}_\Gamma \label{mm25}
\end{equation}
is the energy function
with respect to the reference frame $\Gamma$, taken with the sign minus
\cite{eche95,book98,sard98}. It is readily observed that, given a Hamiltonian
form $H$, the energy functions $\widetilde{\cal H}_\Gamma$ constitute an affine space modelled
over the vector space of N\"other currents.
\begin{prop} \label{mm30}
Given a Hamiltonian form $H$, the conserved currents (\ref{b4306})
form a Lie algebra with respect to the Poisson bracket
\begin{equation}
\{J_u,J_{u'}\}_V=J_{[u,u']}. \label{c2}
\end{equation}
\end{prop}
The second of the above mentioned constructions enables us to represent
the right-hand side of the evolution equation (\ref{mm15}) as a pure Poisson
bracket. Given a Hamiltonian form $H=h^*\Xi$, let us consider its pull-back
$\zeta^*H$ onto the cotangent bundle $T^*Q$. It is readily observed that the
difference
$\Xi-\zeta^*H$ is a horizontal 1-form on $T^*Q\to{\bf R}$, while
\begin{equation}
{\cal H}^*=\partial_t\rfloor(\Xi-\zeta^*H)=p+{\cal H} \label{mm16}
\end{equation}
is a function on $T^*Q$. Then the relation
\begin{equation}
\zeta^*({\bf L}_{\gamma_H}f)=\{{\cal H}^*,\zeta^*f\}_T \label{mm17}
\end{equation}
holds for every function $f\in C^\infty(V^*Q)$.
In particular, given a projectable vector
field $u$ (\ref{m223}), the symmetry current $J_u$ (\ref{b4306}) is conserved
if and only if
\begin{equation}
\{{\cal H}^*,\zeta^*J_u\}_T=0. \label{mm31}
\end{equation}
Moreover, let $\vartheta_{{\cal H}^*}$
be the Hamiltonian vector field for the function ${\cal H}^*$ (\ref{mm16}) with
respect to the canonical Poisson structure $\{,\}_T$ on $T^*Q$. Then
\begin{equation}
T\zeta(\vartheta_{{\cal H}^*})=\gamma_H. \label{mm18}
\end{equation}
\section{Time-dependent constraints}
As was mentioned above, an algebra of time-dependent constraints on
the momentum phase space $V^*Q$ can be described similarly to that in
conservative Hamiltonian mechanics.
Let $N$ be a closed imbedded subbundle $i_N:N\op\hookrightarrow V^*Q$ of the
Legendre bundle
$V^*Q\to{\bf R}$, treated as a constraint space. Let us consider the ideal
$I_N$ of real functions
$f$ on
$V^*Q$ which vanish on $N$, i.e.,
$i_N^*f=0$. Its elements are said to be constraints. There is the isomorphism
\begin{equation}
C^\infty(V^*Q)/I_N\cong C^\infty(N)\label{gm93}
\end{equation}
of associative commutative algebras. $N$ cannot be
neither Lagrangian nor symplectic submanifold with respect to the Poisson
structure on $V^*Q$.
By the normalize $\overline I_N$ of the ideal
$I_N$ is meant the subset of functions of $C^\infty(V^*Q)$ whose Hamiltonian
vector fields restrict to vector fields on $N$, i.e.,
\begin{equation}
\overline I_N=\{f\in C^\infty(V^*Q):\,\, \{f,g\}_V\in I_N, \,\, \forall g\in I_N\},
\label{gm95}
\end{equation}
\cite{kimura}. It follows from the Jacobi identity that the normalizer
(\ref{gm95}) is a Poisson subalgebra of
$C^\infty(V^*Q)$. Put
\begin{equation}
I'_N= \overline I_N\cap I_N. \label{gm96}
\end{equation}
It is naturally a Poisson subalgebra of $\overline I_N$. Its elements are called
the first class constraints, while
the remaining elements of $I_N$ are the second class constraints. It is
readily observed that $I^2_N\subset I'_N$, i.e., the products of
second class constraints are first class constraints.
\begin{rem}
Let $N$ be a coisotropic
submanifold of $V^*Q$, i.e., $w^\sharp({\rm Ann\,} TN)\subset TN$. Then $I_N\subset
\overline I_N$ and $I_N=I'_N$, i.e., all constraints are of the first class.
\end{rem}
The relation (\ref{mm17}) enables us to extend the constraint algorithm of
conservative mechanics and time-dependent mechanics on a product ${\bf R}\times
M$ (see \cite{chinea,leon93}) to mechanical systems subject to time-dependent
transformations.
Let $H$ be a Hamiltonian form on the momentum phase space $V^*Q$.
In accordance with the relation (\ref{mm17}), a constraint $f\in I_N$ is
preserved if the bracket (\ref{mm17}) vanishes. It follows that the
solutions of the Hamilton equations (\ref{m41a}) -- (\ref{m41b}) do not leave
the constraint space $N$ if
\begin{equation}
\{{\cal H}^*,\zeta^*I_N\}_T\subset \zeta^*I_N. \label{mm20}
\end{equation}
If the relation (\ref{mm20}) fails to hold, let us introduce
secondary constraints $\{{\cal H}^*,\zeta^*f\}_T$, $f\in I_N$, which belong to
$\zeta^* C^\infty(V^*Q)$. If the collection of primary and secondary
constraints is not closed with respect to the relation (\ref{mm20}), let us
add the tertiary constraints $\{{\cal H}^*,\{{\cal H}^*,\zeta^*f_a\}_T\}_T$ and so
on.
Let us assume that $N$ is a final constraint space for a Hamiltonian form
$H$. If a Hamiltonian form $H$ satisfies the relation (\ref{mm20}), so is a
Hamiltonian form
\begin{equation}
H_f=H-fdt \label{mm21}
\end{equation}
where $f\in I'_N$ is a first class constraint.
Though Hamiltonian forms $H$ and $H_f$ coincide
with each other on the constraint space $N$, the corresponding Hamilton
equations have different solutions on the constraint space $N$ because
$dH\mid_N\neq dH_f\mid_N$. At the same time, $d(i_N^*H)=d(i_N^*H_f)$.
Therefore, let us introduce the constrained Hamiltonian form
\begin{equation}
H_N=i_N^*H_f \label{mm23}
\end{equation}
which is the same for all $f\in I'_N$.
Note that $H_N$ (\ref{mm23}) is not a true Hamiltonian form on
$N\to {\bf R}$ in general.
On sections $r$ of the fibre bundle $N\to{\bf R}$, we can write the equations
\begin{equation}
r^*(u_N\rfloor dH_N) =0, \label{N44}
\end{equation}
where $u_N$ is an arbitrary vertical vector field on $N\to {\bf R}$. They are
called the constrained
Hamilton equations.
\begin{prop} \label{cmp22} For any Hamiltonian form $H_f$
(\ref{mm21}), every solution
of the Hamilton equations which lives in the constraint space
$N$ is a solution of the constrained Hamilton equations (\ref{N44}).
\end{prop}
\begin{proof} The constrained Hamilton
equations can be written as
\begin{equation}
r^*(u_N\rfloor di^*_NH_f)=r^*(u_N\rfloor dH_f\mid_N) =0. \label{N44'}
\end{equation}
They differ from the Hamilton equations (\ref{y5}) for $H_f$ restricted to $N$
which read
\begin{equation}
r^*(u\rfloor dH_f\mid_N) =0, \label{cmp10}
\end{equation}
where $r$ is a section of $N\to {\bf R}$ and $u$ is an arbitrary vertical vector
field on $V^*Q\to {\bf R}$. A solution $r$ of the equations
(\ref{cmp10}) satisfies obviously the weaker condition
(\ref{N44'}).
\end{proof}
\begin{rem}
One also can consider the problem of constructing a generalized Hamiltonian
system, similar to that for Dirac constraint system in conservative mechanics
\cite{book98}. Let $H$ satisfies the condition
$\{{\cal H}^*,\zeta^*I'_N\}_T\subset I_N$, whereas
$\{{\cal H}^*,\zeta^*I'_N\}_T\not\subset I_N$. The goal is to find
a constraint $f\in I_N$ such that the modified Hamiltonian $H -fdt$ would
satisfy both the conditions
\begin{eqnarray*}
\{{\cal H}^* +\zeta^*f ,\zeta^* I'_N\}_T\subset \zeta^*I_N, \qquad
\{{\cal H}^* +\zeta^*f ,\zeta^* I_N\}_T\subset \zeta^*I_N.
\end{eqnarray*}
The first of them is fulfilled for any $f\in I_N$,
while the latter is an equation for a second-class constraint
$f$.
\end{rem}
It should be emphasized that, in contrast with the conservative
case, the Hamiltonian vector fields $\vartheta_f$ for the first class constraints
$f\in I'_N$ in time-dependent mechanics are not generators of gauge symmetries
of a Hamiltonian form in general. At the same time, generators of gauge
symmetries define an ideal of constraints as follows.
The above construction, except the isomorphism
(\ref{gm93}), can be applied to any ideal $I$ of $C^\infty(V^*Q)$. Then one
says that the Poisson algebra $\overline I/ I'$ is the reduction of the Poisson
algebra
$C^\infty(V^*Q)$ via the ideal $I$ \cite{kimura}.
In particular, an ideal $I$ is said to be
coisotropic if it is a Poisson algebra. In this case, $I$ is a Poisson
subalgebra of the normalize
$\overline I$ (\ref{gm95}), and coincides with $I'$ (\ref{gm96}).
Let ${\cal A}$ be a Lie algebra of generators $u$ of gauge symmetries of a
Hamiltonian form $H$. In accordance with the relation (\ref{c2}), the
corresponding symmetry currents $J_u$ (\ref{b4306}) on
$V^*Q$ constitute a Lie algebra with
respect to the canonical Poisson bracket on $V^*Q$. Let $I_{\cal A}$ denotes the
ideal of $C^\infty(V^*Q)$ generated by these symmetry currents.
It is readily observed that this ideal is coisotropic. Then one can think of
$I_{\cal A}$ as being an ideal of first class constraints compatible with the
Hamiltonian form $H$, i.e.,
\begin{equation}
\{{\cal H}^*,\zeta^*I_{\cal A}\}_T\subset \zeta^*I_{\cal A}. \label{mm33}
\end{equation}
Note that any Hamiltonian form $H_u=H-J_udt$,
$u\in{\cal A}$, obeys the same relation (\ref{mm33}), but other currents $J_{u'}$
are not conserved with respect $H_u$ if $[u,u']\neq 0$.
Let now ${\cal A}$ be an arbitrary Lie algebra of vertical vector fields $u$ on the
configuration bundle $Q\to{\bf R}$. The relation (\ref{c2}) remains true, while
the corresponding symmetry currents $J_u$ (\ref{z606}) on
$V^*Q$ constitute a Lie algebra and generate the corresponding coisotropic
ideal $I_{\cal A}$ of $C^\infty(V^*Q)$ with respect to the canonical Poisson bracket
on $V^*Q$.
\begin{prop}
Let ${\cal A}$ be a finite-dimensional Lie algebra of vertical vector fields on
the configuration bundle $Q\to{\bf R}$. If there exists a reference frame $\Gamma$ on
$Q\to{\bf R}$ such that $[\Gamma,{\cal A}]=0$, then there exists a non-frame Hamiltonian
form $H$ on the Legendre bundle $V^*Q$ such that ${\cal A}$ is the algebra of gauge
symmetries of $H$.
\end{prop}
\begin{proof}
Let $\overline{\cal A}$ be the universal enveloping algebra of the Lie algebra of the
symmetry currents $J_u$, $u\in {\cal A}$, (\ref{z606}). Then each non-zero
element $C$ of its center of order $>1$ can be written as a polynomial in
$J_u$, and defines the desired Hamiltonian form $H=H_\Gamma -Cdt$.
\end{proof}
\section{Lagrangian constraints}
Let us consider the Hamiltonian description of Lagrangian mechanical systems
on a configuration bundle $Q\to{\bf R}$. If a Lagrangian is degenerate, we have
the Lagrangian constraint subspace of the Legendre bundle $V^*Q$ and a set of
Hamiltonian forms associated with the same Lagrangian.
Given a Lagrangian $L$ (\ref{mm43}) on the velocity phase space $J^1Q$,
a Hamiltonian form $H$ on the momentum phase space $V^*Q$ is said to be
associated with $L$ if $H$ satisfies the relations
\begin{eqalph}
&&\widehat L\circ\widehat H\circ \widehat L=\widehat L,\label{2.30a} \\
&&H=H_{\widehat H}+\widehat H^*L \label{2.30b}
\end{eqalph}
where $\widehat H$ and $\widehat L$ are the Hamiltonian morphism (\ref{mm41})
and the Legendre map (\ref{a303}), respectively. A glance
at the relation (\ref{2.30a}) shows that $\widehat L\circ\widehat H$ is the projector
\begin{equation}
p_i(z)=\pi_i(t,q^i,\partial^j{\cal H}(z)), \qquad z\in N_L,
\label{b481'}
\end{equation}
from $\Pi$ onto the Lagrangian constraint space $N_L=\widehat L( J^1Y)$.
Accordingly, $\widehat H\circ\widehat L$ is the projector from $J^1Y$ onto $\widehat
H(N_L)$. A Hamiltonian form is called weakly associated with a Lagrangian $L$
if the condition (\ref{2.30b}) holds on the Lagrangian constraint space $N_L$.
\begin{prop} \label{jp} {\rm \cite{book}.}
If a bundle morphism $\Phi:V^*Q\op\to_Q J^1Q$ obeys the relation
(\ref{2.30a}), then the Hamiltonian form $H=-\Phi\rfloor{\bf\Theta}+\Phi^*L$ is
weakly associated with the Lagrangian $L$. If $\Phi=\widehat H$, then $H$ is
associated with $L$.
\end{prop}
\begin{lem} \label{cmp110}
Any Hamiltonian form $H$ weakly associated with a Lagrangian $L$ obeys
the relation
\begin{equation}
H\mid_{N_L}=\widehat H^*H_L\mid_{N_L}, \label{4.9}
\end{equation}
where $H_L$ is the Poincar\'e--Cartan form (\ref{303}).
\end{lem}
\begin{proof}
The relation (\ref{2.30b}) takes the coordinate form
\begin{equation}
{\cal H}(z)=p_i\partial^i{\cal H}-{\cal L}(t,q^i,\partial^j{\cal H}(z)), \qquad z\in N_L. \label{b481}
\end{equation}
Substituting (\ref{b481'}) and (\ref{b481}) in (\ref{b4210}), we
obtain the relation (\ref{4.9}).
\end{proof}
The difference between associated and weakly associated Hamiltonian forms
lies in the following. Let $H$ be an associated Hamiltonian form, i.e., the
equality (\ref{b481}) holds everywhere on $V^*Q$.
The exterior differential of this
equality leads to the relations
\begin{eqnarray*}
&& \partial_t{\cal H}(z) =-(\partial_t{\cal L})\circ \widehat H(z), \qquad
\partial_i{\cal H}(z) =-(\partial_i{\cal L})\circ \widehat H(z), \qquad
z\in N_L, \\
&&
(p_i-(\partial^t_i{\cal L})(t,q^i,\partial^j_t{\cal H}))\partial^i_t\partial^a_t{\cal H}=0.
\end{eqnarray*}
The last of them shows that the Hamiltonian form is not regular outside
the Lagrangian constraint space $N_L$. In particular, any Hamiltonian form is
weakly associated with the Lagrangian $L=0$, while the associated Hamiltonian
forms are only $H_\Gamma$.
Here we restrict our consideration to almost regular Lagrangians $L$, i.e.,
if: (i) the Lagrangian constraint space $N_L$ is a closed imbedded subbundle
$i_N:N_L\to V^*Q$ of the bundle $V^*Q\to Q$, (ii) the Legendre map $\widehat
L:J^1Q\to N_L$ is a fibred manifold, and (iii) the pre-image $\widehat L^{-1}(z)$
of any point $z\in N_L$ is a connected submanifold of $J^1Q$.
\begin{prop} \label{mm71}
As an immediate consequence of the above conditions (i), (ii) and
Proposition
\ref{jp}, a Hamiltonian form $H$ weakly associated with an almost regular
Lagrangian $L$ exists iff the fibred manifold $J^1V^*Q\to N_L$ admits a global
section.
\end{prop}
The condition (iii) leads to the following property.
\begin{lem} \label{3.22} {\rm \cite{book,book98}.}
The
Poincar\'e--Cartan form $H_L$ for an almost regular
Lagrangian $L$ is constant on the connected
pre-image $\widehat L^{-1}(z)$ of any point $z\in N_L$.
\end{lem}
An immediate consequence of this fact is the following assertion.
\begin{prop} \label{3.22'} {\rm \cite{book}.} All Hamiltonian forms
weakly associated with an almost regular Lagrangian $L$ coincide with each
other on the Lagrangian constraint space $N_L$,
and the Poincar\'e--Cartan form $H_L$ (\ref{303})
for $L$ is the pull-back
\begin{equation}
H_L=\widehat L^*H, \qquad
\pi_iq^i_t-{\cal L}={\cal H}(t,q^j,\pi_j), \label{2.32}
\end{equation}
of any such a Hamiltonian form $H$.
\end{prop}
It follows that, given Hamiltonian forms $H$ an $H'$ weakly associated with an
almost regular Lagrangian $L$, their difference is $fdt$, $f\in I_N$.
However, $\widehat H\mid_{N_L}\neq \widehat H'\mid_{N_L}$ in general. Therefore, the
Hamilton equations for $H$ and $H'$ do not coincide necessarily on the
Lagrangian constraint space $N_L$. Their solutions can
leave $N_L$, i.e., the relation (\ref{mm20}) fails to hold in general.
Proposition \ref{3.22'} enables us to connect Lagrange and Cartan
equations for an almost regular Lagrangian $L$ with the Hamilton
equations for Hamiltonian forms weakly associated with $L$ \cite{book}.
\begin{theo}\label{3.23}
Let a section $r$ of $V^*Q\to {\bf R}$
be a solution of the Hamilton equations (\ref{m41a}) -- (\ref{m41b})
for a Hamiltonian form $H$ weakly associated with an almost regular
Lagrangian $L$. If $r$ lives in the constraint space $N_L$, the
section $c=\pi_Q\circ r$
of $Q\to {\bf R}$ satisfies the Lagrange
equations (\ref{b327}), while $\overline c=\widehat H\circ r$ obeys the Cartan equations
(\ref{b336}).
\end{theo}
The proof is based on the relation
\begin{equation}
{\cal E}_{\overline L}=(J^1\widehat L)^*{\cal E}_H \label{b4.1000}
\end{equation}
or on the
equivalent relation $\overline L=(J^1\widehat L)^*L_H$ which are derived from the
equality (\ref{2.32}). The converse assertion is more intricate.
\begin{theo}\label{3.24} Given an almost regular Lagrangian $L$,
let a section $\overline c$ of the jet bundle
$J^1Q\to {\bf R}$ be a solution of the
Cartan equations (\ref{b336}).
Let $H$ be a Hamiltonian form weakly associated with $L$, and let $H$ satisfy
the relation
\begin{equation}
\widehat H\circ \widehat L\circ \overline c=J^1(\pi^1_0\circ\overline c).\label{2.36}
\end{equation}
Then, the section $r=\widehat L\circ \overline c$
of the Legendre bundle $V^*Q\to {\bf R}$ is a solution of the
Hamilton equations (\ref{m41a}) -- (\ref{m41b}) for $H$.
\end{theo}
\begin{rem} \label{cmp9}
Since $\widehat H\circ \widehat L$ in Theorem (\ref{3.24}) is a projection operator,
the condition (\ref{2.36}) implies that the solution $\overline s$ of the Cartan
equations is actually an integrable section $\overline c=\dot c$ where $c$ is a
solution of the Lagrange equations.
In fact, the
relation (\ref{b4.1000}) gives more than it is needed for proving Theorem
\ref{3.23}. Using this relation, one can justify that, if $\gamma$ is a Hamiltonian
connection for a Hamiltonian form $H$ weakly associated with an almost regular
Lagrangian
$L$, then the composition $J^1\widehat H\circ\gamma\circ\widehat L$ takes its values in
${\rm Ker\,} {\cal E}_{\overline L}\cap J^2Y$, i.e., this is a local holonomic Lagrangian
connection on
$\widehat H(N_L)$ \cite{book}. A converse of this assertion, however, fails to
be true in the case of degenerate Lagrangians.
Let a Lagrangian $L$ be
hyperregular, i.e., the Legendre map $\widehat L$
is a diffeomorphism. Then $\widehat L^{-1}$ is a Hamiltonian map, and there is a
unique Hamiltonian form $H=H_{\widehat L^{-1}}+\widehat L^{-1*}L$
weakly associated with $L$. In this case, both the relation (\ref{b4.1000}) and
the converse one ${\cal E}_H=(J^1\widehat H)^*{\cal E}_{\overline L}$
hold. It follows that the Lagrange equations for $L$ and the Hamilton
equations for $H$ are equivalent.
\end{rem}
We will say that a set of Hamiltonian forms
$H$ weakly associated with an almost regular Lagrangian $L$ is
complete if, for each
solution
$c$ of the Lagrange equations, there exists a solution
$r$ of the Hamilton equations for a Hamiltonian form $H$ from this
set such that $c=\pi_Q\circ r$.
By virtue of Theorem \ref{3.24} and Remark \ref{cmp9}, a set
of weakly associated Hamiltonian forms
is complete if, for every solution $c$ on ${\bf R}$ of the Lagrange
equations for $L$, there is a Hamiltonian form $H$ from this
set which fulfills the relation
\begin{equation}
\widehat H\circ \widehat L\circ \dot c=\dot c. \label{2.36'}
\end{equation}
In accordance with Proposition \ref{mm71}, on an open neighbourhood in $V^*Q$
of each point $z\in N_L$, there exists a complete set of local Hamiltonian
forms weakly associated with an almost regular Lagrangian
$L$. Moreover, one can always construct a complete set of associated local
Hamiltonian forms \cite{sard95,zak}
Given a Hamiltonian form $H$ weakly associated with an almost regular
Lagrangian $L$, let us consider the corresponding constrained Hamiltonian
form $H_N$ (\ref{mm23}). By virtue of
Proposition (\ref{3.22'}),
$H_N$ is the same for all Hamiltonian forms weakly associated with $L$, and
$H_L=\widehat L^* H_N$. The first of these facts leads to the assertion proved
similarly to Proposition \ref{cmp22}.
\begin{prop} \label{mm72} For any Hamiltonian form $H$ weakly
associated with an almost regular Lagrangian $L$, every solution
of the Hamilton equations which lives in the Lagrangian constraint space
$N_L$ is a solution of the constrained Hamilton equations (\ref{N44}).
\end{prop}
Using the equality $H_L=\widehat L^* H_N$, one can show that the constrained
Hamilton equations (\ref{N44}) are equivalent to the
Hamilton--De Donder equations (\ref{N46}) and, by virtue of Theorem
\ref{ddd}, are quasi-equivalent to the Cartan equations
(\ref{C28}) \cite{book,book98}.
\section{Quadratic degenerate systems}
Let us study the important case of almost regular
quadratic Lagrangians. We show that, in this case, there always exist both a
complete set of associated Hamiltonian forms and a complete set of
non-degenerate weakly associated Hamiltonian forms. The latter is important
for quantization.
Given a configuration bundle
$Q\to {\bf R}$, let us consider a quadratic Lagrangian $L$ which has the
coordinate expression
\begin{equation}
{\cal L}=\frac12 a_{ij} q^i_t q^j_t +
b_i q^i_t + c, \label{N12}
\end{equation}
where $a$, $b$ and $c$ are local functions on $Q$. This property is
coordinate-independent due to the affine transformation law of the coordinates
$q^i_t$. The associated Legendre map
\begin{equation}
p_i\circ\widehat L= a_{ij} q^j_t +b_i \label{N13}
\end{equation}
is an affine morphism over $Q$. It defines the corresponding linear
morphism
\begin{equation}
\overline L: VQ\op\to_Q V^*Q,\qquad p_i\circ\overline
L=a_{ij}\dot q^j. \label{N13'}
\end{equation}
Let the Lagrangian $L$ (\ref{N12}) be almost regular, i.e.,
the matrix function $a_{ij}$ is of constant rank. Then
the Lagrangian constraint space $N_L$
(\ref{N13}) is an affine subbundle of the bundle $V^*Q\to Q$, modelled
over the vector subbundle $\overline N_L$ (\ref{N13'}) of $V^*Q\to Q$.
Hence, $N_L\to Q$ has a global section. For the sake of simplicity, let us
assume that it is the canonical
zero section $\widehat 0(Q)$ of $V^*Q\to Q$. Then $\overline N_L=N_L$.
Accordingly, the kernel
of the Legendre map (\ref{N13}) is an affine
subbundle of the affine jet bundle $J^1Q\to Q$, modelled over the kernel of
the linear morphism $\overline L$ (\ref{N13'}). Then there exists a connection
\begin{eqnarray}
&&\Gamma: Q\to {\rm Ker\,}\widehat L\subset J^1Q, \label{N16}\\
&& a_{ij}\Gamma^j_\mu + b_i =0, \label{250}
\end{eqnarray}
on $Q\to {\bf R}$.
Connections (\ref{N16}) constitute an affine space modelled over the linear
space of vertical vector fields $\upsilon$ on $Q\to {\bf R}$, satisfying the conditions
\begin{equation}
a_{ij}\upsilon^j =0 \label{cmp21}
\end{equation}
and, as a consequence, the conditions $\upsilon^i b_i=0$.
If the Lagrangian (\ref{N12}) is regular, the
connection (\ref{N16}) is unique.
The matrix $a$ in the Lagrangian $L$ (\ref{N12}) can be seen as a degenerate
fibre metric of constant rank in
$VQ\to Q$. Then it satisfies the following Lemma.
\begin{lem} \label{mm45}
Given a $k$-dimensional vector bundle $E\to Z$, let $a$ be a section of rank
$r$ of the tensor bundle $\op\vee^2E^*\to Z$. There is a splitting
\begin{equation}
E= {\rm Ker\,} a\op\oplus_Z E' \label{mm50}
\end{equation}
where $E'=E/{\rm Ker\,} a$ is the quotient bundle, and $a$ is a non-degenerate
fibre metric in $E'$.
\end{lem}
\begin{proof}
Since $a$ exists, the structure group $GL(k,{\bf R})$ of the vector bundle $E\to
Z$ is reducible to the subgroup $GL(r,k-r;{\bf R})$ of general linear
transformations of ${\bf R}^k$ which keep its $r$-dimensional subspace, and to
its subgroup $GL(r,{\bf R})\times GL(k-r,{\bf R})$.
\end{proof}
\begin{theo}\label{04.2} There exists a linear bundle
map
\begin{equation}
\sigma: V^*Q\op\to_Q VQ, \qquad
\dot q^i\circ\sigma =\sigma^{ij}p_j, \label{N17}
\end{equation}
such that $\overline L\circ\sigma\circ i_N= i_N$.
\end{theo}
\begin{proof}
The map (\ref{N17}) is a solution of the algebraic equations
\begin{equation}
a_{ij}\sigma^{jk}a_{kb}=a_{ib}. \label{mm100}
\end{equation}
By virtue of Lemma \ref{mm45}, there exist the bundle slitting
\begin{equation}
VQ={\rm Ker\,} a\op\oplus_Q E' \label{mm46}
\end{equation}
and a (non-holonomic) atlas of this bundle such that transition
functions of
${\rm Ker\,} a$ and $E'$ are independent. Since $a$ is a non-degenerate fibre metric
in $E'$, there exists an atlas of $E'$ such
that $a$ is brought into a diagonal matrix with non-vanishing
components
$a_{AA}$. Due to the splitting (\ref{mm46}), we have the corresponding
bundle splitting
\begin{equation}
V^*Q=({\rm Ker\,} a)^*\op\oplus_Q {\rm Im\, } a. \label{mm46'}
\end{equation}
Then the desired map $\sigma$ is represented by a direct sum $\sigma_1\oplus\sigma_0$
of an arbitrary section $\sigma_1$ of the bundle
$\op\vee^2{\rm Ker\,} a^*\to Q$
and the section
$\sigma_0$ of the bundle $\op\vee^2E'\to Q$,
which has non-vanishing components $\sigma^{AA}=(a_{AA})^{-1}$ with respect to
the above mentioned atlas of $E'$. Moreover,
$\sigma$ satisfies the particular relations
\begin{equation}
\sigma_0=\sigma_0\circ\overline L\circ\sigma_0, \quad a\circ\sigma_1=0, \quad \sigma_1\circ a=0.
\label{N21}
\end{equation}
\end{proof}
\begin{cor}
The splitting (\ref{mm46}) leads to the splitting
\begin{eqalph}
&& J^1Q={\cal S}(J^1Q)\op\oplus_Q {\cal F}(J^1Q)={\rm Ker\,}\widehat L\op\oplus_Q{\rm Im}(\sigma\circ
\widehat L), \label{N18} \\
&& q^i_t={\cal S}^i+{\cal F}^i= [q^i_t
-\sigma^{ik}_0 (a_{kj}q^j_t + b_k)]+
[\sigma^{ik}_0 (a_{kj}q^j_t + b_k)], \label{b4122}
\end{eqalph}
while the splitting (\ref{mm46'})
can be written as
\begin{eqalph}
&& V^*Q={\cal R}(V^*Q)\op\oplus_Q{\cal P}(V^*Q)={\rm Ker\,}\sigma_0 \op\oplus_Q N_L, \label{N20} \\
&& p_i = {\cal R}_i+{\cal P}_i= [p_i -
a_{ij}\sigma^{jk}_0p_k] +
[a_{ij}\sigma^{jk}_0p_k]. \label{N20'}
\end{eqalph}
\end{cor}
It is readily observed that, with respect to the coordinates ${\cal S}^i_\lambda$
and ${\cal F}^i_\lambda$ (\ref{b4122}), the Lagrangian (\ref{N12}) reads
\begin{equation}
{\cal L}=\frac12 a_{ij}{\cal F}^i{\cal F}^j +c', \label{cmp31}
\end{equation}
while the Lagrangian constraint space is given by the reducible constraints
\begin{equation}
{\cal R}_i= p_i -
a_{ij}\sigma^{jk}_0p_k=0. \label{zzz}
\end{equation}
Given the linear map $\sigma$ (\ref{N17}) and the connection $\Gamma$
(\ref{N16}), let us consider the affine Hamiltonian map
\begin{equation}
\Phi=\widehat\Gamma+\sigma:V^*Q \op\to J^1Q, \qquad
\Phi^i = \Gamma^i + \sigma^{ij}p_j, \label{N19}
\end{equation}
and the Hamiltonian form
\begin{eqnarray}
&& H=H_\Phi +\Phi^*L= p_idq^i - [p_i\Gamma^i
+\frac12 \sigma_0{}^{ij}p_ip_j
+\sigma_1{}^{ij}p_ip_j -c']dt=
\label{N22}\\
&& \qquad ({\cal R}_i+{\cal P}_i)dq^i - [({\cal R}_i+{\cal P}_i)\Gamma^i
+\frac12
\sigma_0^{ij}{\cal P}_i{\cal P}_j
+\sigma_1^{ij}p_ip_j -c']dt.\nonumber
\end{eqnarray}
In particular, if $\sigma_1$ is non-degenerate, so is the Hamiltonian form
(\ref{N22}).
\begin{theo} \label{cmp30} The Hamiltonian forms (\ref{N22})
parameterized by connections $\Gamma$ (\ref{N16}) are weakly associated with the
Lagrangian (\ref{N12}) and constitute a complete set.
\end{theo}
\begin{proof}
By the very definitions of $\Gamma$ and $\sigma$, the Hamiltonian map (\ref{N19})
satisfies the condition (\ref{2.30a}). Then $H$ is weakly associated with $L$
(\ref{N12}) in accordance with Proposition \ref{jp}.
Let us write the corresponding Hamilton equations (\ref{m41a}) for
a section $r$ of the Legendre bundle $V^*Q\to {\bf R}$. They are
\begin{equation}
\dot c= (\widehat\Gamma+\sigma)\circ r, \qquad c=\pi_Q\circ r. \label{N29}
\end{equation}
Due to the surjections ${\cal S}$ and ${\cal F}$ (\ref{N18}),
the Hamilton equations (\ref{N29}) break in two parts
\begin{eqnarray}
&&{\cal S}\circ \dot c=\Gamma\circ c, \qquad \dot r^i-
\sigma^{ik} (a_{kj}\dot r^j + b_k)=\Gamma^i\circ c,
\label{N23} \\
&&{\cal F} \circ \dot c=\sigma\circ r, \qquad
\sigma^{ik} (a_{kj}\dot r^j + b_k)=
\sigma^{ik}r_k.\label{N28}
\end{eqnarray}
Let $c$ be an arbitrary section of $Q\to {\bf R}$,
e.g., a solution of the Lagrange
equations. There exists a connection $\Gamma$ (\ref{N16}) such
that the relation (\ref{N23}) holds, namely, $\Gamma={\cal S}\circ\Gamma'$ where
$\Gamma'$ is a
connection on $Q\to {\bf R}$ which has $c$ as an integral section.
It is easily seen that, in this case, the Hamiltonian map (\ref{N19})
satisfies the relation (\ref{2.36'}) for $c$.
Hence, the Hamiltonian forms (\ref{N22}) constitute
a complete set.
\end{proof}
It is readily observed that, if $\sigma_1=0$, then $\Phi=\widehat H$ and the
Hamiltonian forms (\ref{N22}) are associated with the Lagrangian (\ref{N12})
in accordance with Proposition \ref{jp}. Thus, for different $\sigma_1$, we have
different complete sets of Hamiltonian forms (\ref{N22}). Hamiltonian forms $H$
(\ref{N22}) of such a complete set differ from each other in the term
$\upsilon^i{\cal R}_i$, where $\upsilon$ are vertical vector fields
(\ref{cmp21}). If follows from the splitting (\ref{N20}) that this term
vanishes on the Lagrangian constraint space. The corresponding constrained
Hamiltonian form
$H_N=i_N^*H$ and the constrained Hamilton equations (\ref{N44}) can be
written. In the case of quadratic Lagrangians, we can improve Proposition
\ref{mm72} as follows.
\begin{prop} \label{cmp23} For every Hamiltonian
form $H$ (\ref{N22}),
the Hamilton equations (\ref{m41b}) and (\ref{N28}) restricted to the
Lagrangian constraint space $N_L$ are equivalent to the constrained Hamilton
equations.
\end{prop}
\begin{proof} Due to the splitting (\ref{N20}), we have the corresponding
splitting
of the vertical tangent bundle $V_QV^*Q$ of the bundle $V^*Q\to Q$.
In particular, any
vertical vector field
$u$ on
$V^*Q\to {\bf R}$ admits the decomposition
\begin{eqnarray*}
u= [u-u_{TN}] + u_{TN}, \qquad u_{TN}=u^i\partial_i +a_{ij}\sigma^{jk}_0u_k\partial^i,
\end{eqnarray*}
such that $u_N=u_{TN}\mid_{N_L}$ is a vertical vector field on the Lagrangian
constraint space $N_L\to {\bf R}$. Let us consider the equations
\begin{equation}
r^*(u_{TN}\rfloor dH)=0 \label{cmp15}
\end{equation}
where $r$ is a section of $V^*Q\to {\bf R}$ and $u$ is an arbitrary vertical vector
field on $V^*Q\to {\bf R}$. They are equivalent to the pair of equations
\begin{eqalph}
&& r^*(a_{ij}\sigma^{jk}_0\partial^i\rfloor dH)=0,
\label{b4125a} \\
&& r^*(\partial_i\rfloor dH)=0. \label{b4125b}
\end{eqalph}
The equations (\ref{b4125b}) are obviously the Hamilton equations
(\ref{m41b}) for $H$. Bearing in mind the relations (\ref{250}) and
(\ref{N21}), one can easily show that the equations (\ref{b4125a})
coincide with the Hamilton equations (\ref{N28}). The proof is
completed by observing that, restricted to the Lagrangian constraint space
$N_L$, the equations (\ref{cmp15}) are exactly the constrained Hamilton
equations (\ref{N44'}).
\end{proof}
Proposition \ref{cmp23} shows that, restricted to the Lagrangian constraint
space, the Hamilton equations for different Hamiltonian forms (\ref{N22})
associated with the same quadratic Lagrangian (\ref{N12}) differ from each
other in the equations (\ref{N23}). These equations are independent of
momenta and play the role of gauge-type conditions.
We aim to obtain the Koszul--Tate resolution for the constraints
(\ref{zzz}) Since these constraints are not necessarily irreducible, we need
an infinite number of ghosts and antighosts \cite{fisch,kimura}.
\section{Simple BRST manifolds}
Let $E=E_0\oplus E_1\to Z$ be the Whitney sum of vector bundles $E_0\to
Z$ and $E_1\to Z$ over a paracompact manifold $Z$. One can think of $E$ as
being a bundle of vector superspaces with a typical fibre $V=V_0\oplus V_1$
where transition functions of $E_0$ and $E_1$ are independent. Let us consider
the exterior bundle
\begin{equation}
\wedge E^*=\op\bigoplus^\infty_{k=0} (\op\wedge^k_Z E^*), \
\label{mm80}
\end{equation}
which is the tensor bundle $\otimes E^*$ modulo elements
\begin{eqnarray*}
e_0e'_0 - e'_0e_0, \quad e_1e'_1 + e'_1e_1, \quad e_0e_1 - e_1e_0\quad
e_0,e'_0\in E_{0z}^*,
\quad e_1,e'_1\in E_{1z}^*, \quad z\in Z.
\end{eqnarray*}
$\wedge E^*$ is the bundle of commutative superalgebras $\wedge V$ which is the tensor
product $\vee E_0^*\otimes\wedge E_1^*$ modulo elements
\begin{eqnarray*}
e_0e_1 - e_1e_0\quad
e_0\in E_{0z}^*,
\quad e_1\in E_{1z}^*, \quad z\in Z.
\end{eqnarray*}
The global sections of $\wedge E^*$ constitute a commutative superalgebra ${\cal A}(Z)$
over the free
$C^\infty(Z)$-module $E_0^*(Z)\oplus E_1^*(Z)$ of global sections of $E^*$.
This is the product of the commutative algebra ${\cal A}_0(Z)$ of global sections of
$\vee E_0^*\to Z$ and the graded algebra ${\cal A}_1(Z)$ of global sections of the
Grassman bundle
$\wedge E_1^*\to Z$. We use the notation $\nw .$ for the Grassman parity.
\begin{rem}
Let ${\cal A}_1$ be the sheaf of sections of the Grassman bundle $\wedge E_1^*$. The
pair $(Z,{\cal A}_1)$ is a graded manifold \cite{bart}.
By the well-known Batchelor theorem, every graded manifold is isomorphic to a
sheaf of sections of some Grassman bundle, but not in a canonical way.
Therefore, the construction below can be extended to an arbitrary commutative
superalgebra over a free $C^\infty(Z)$-module ${\cal A}={\cal A}_1\oplus{\cal A}_2$ of finite
rank. We call $(Z,{\cal A})$ a BRST manifold, while sections of $\wedge E^*$ are
said to be BRST functions.
\end{rem}
Let us study the ${\cal A}(Z)$-module $\rm Der {\cal A}(Z)$ of graded derivations of
${\cal A}(Z)$. Recall that by a graded derivation of the commutative superalgebra
${\cal A}(Z)$ is meant an endomorphism of ${\cal A}(Z)$ such that
\begin{equation}
u(ff')=u(f)f'+(-1)^{\nw u\nw f}fu (f') \label{mm81}
\end{equation}
for the homogeneous elements $u\in \rm Der{\cal A}(Z)$ and $f,f'\in {\cal A}(Z)$.
\begin{prop}
Graded derivations (\ref{mm81}) are represented by
sections of a vector bundle.
\end{prop}
\begin{proof}
Let $\{c^a\}$ be the holonomic bases for $E^*\to Z$ with respect to some bundle
atlas $(z^A,v^i)$ of $E\to Z$ with transition functions $\{\rho^a_b\}$, i.e.,
$c'^a=\rho^a_b(z)c^b$. Then BRST functions read
\begin{equation}
f=\op\sum_{k=0} \frac1{k!}f_{a_1\ldots
a_k}c^{a_1}\cdots c^{a_k}, \label{z785}
\end{equation}
where $f_{a_1\cdots
a_k}$ are local functions on $Z$, and we omit the symbol of an exterior product
of elements $c$. The coordinate transformation law of BRST functions
(\ref{z785}) is obvious.
Due to the canonical splitting
$VE= E\times E$, the vertical tangent bundle
$VE\to E$ can be provided with the fibre bases $\{\partial_a\}$ dual of $\{c^a\}$.
These are fibre bases for ${\rm pr}_2VE=E$. Then
any derivation $u$ of ${\cal A}(U)$ on a trivialization domain $U$ of $E$ reads
\begin{equation}
u= u^A\partial_A + u^a\partial_a, \label{mm83}
\end{equation}
where $u^A, u^a$ are local BRST functions and $u$ acts on $f\in {\cal A}(U)$ by
the rule
\begin{equation}
u(f_{a\ldots b}c^a\cdots c^b)=u^A\partial_A(f_{a\ldots b})c^a\cdots c^b +u^a
f_{a\ldots b}\partial_a\rfloor (c^a\cdots c^b). \label{cmp50'}
\end{equation}
This rule implies the corresponding
coordinate transformation law
\begin{equation}
u'^A =u^A, \qquad u'^a=\rho^a_ju^j +u^A\partial_A(\rho^a_j)c^j \label{lmp2}
\end{equation}
of derivations (\ref{mm83}).
Let us consider
the vector bundle
${\cal V}_E\to Z$ which is locally isomorphic to the vector bundle
\begin{eqnarray*}
{\cal V}_E\mid_U\approx\wedge E^*\op\otimes_Z({\rm pr}_2VE\op\oplus_Z TZ)\mid_U,
\end{eqnarray*}
and has the transition functions
\begin{eqnarray*}
&& z'^A_{i_1\ldots i_k}=\rho^{-1}{}_{i_1}^{a_1}\cdots
\rho^{-1}{}_{i_k}^{a_k} z^A_{a_1\ldots a_k}, \\
&& v'^i_{j_1\ldots j_k}=\rho^{-1}{}_{j_1}^{b_1}\cdots
\rho^{-1}{}_{j_k}^{b_k}\left[\rho^i_jv^j_{b_1\ldots b_k}+ \frac{k!}{(k-1)!}
z^A_{b_1\ldots b_{k-1}}\partial_A(\rho^i_{b_k})\right]
\end{eqnarray*}
of the bundle coordinates $(z^A_{a_1\ldots a_k},v^i_{b_1\ldots b_k})$,
$k=0,\ldots$. These transition functions
fulfill the cocycle relations. It is readily observed that, for any
trivialization domain $U$, the
${\cal A}$-module $\rm Der{\cal A}(U)$ with the transition functions (\ref{lmp2}) is
isomorphic to the ${\cal A}$-module of local sections of ${\cal V}_E\mid_U\to U$.
One can show that, if $U'\subset U$ are open
sets, there is the restriction morphism $\rm Der{\cal A}(U)\to
\rm Der{\cal A}(U')$. It follows that, restricted to an open subset $U$, every
derivation $u$ of
${\cal A}(Z)$ coincides with some local section $u_U$ of ${\cal V}_E\mid_U\to U$, whose
collection $\{u_U, U\subset Z\}$ defines uniquely a global section of
${\cal V}_E\to Z$, called a BRST vector field on $Z$. BRST vector
fields constitute a Lie
superalgebra with respect to the bracket
\begin{eqnarray*}
[u,u']=uu' + (-1)^{\nw u\nw{u'}+1}u'u.
\end{eqnarray*}
\end{proof}
\begin{cor}
The sheaf of sections of ${\cal V}_E\to Z$ is isomorphic to the sheaf of graded
derivations of the sheaf ${\cal A}$.
\end{cor}
There is the exact sequence over $Z$ of vector
bundles
\begin{equation}
0\to \wedge E^*\op\otimes_Z{\rm pr}_2VE\to{\cal V}_E\to \wedge E^*\op\otimes_Z TZ\to 0. \label{cmp92}
\end{equation}
Its splitting
\begin{equation}
\widetilde\gamma:\dot z^A\partial_A \mapsto \dot z^A(\partial_A +\widetilde\gamma_A^a\partial_a) \label{cmp70}
\end{equation}
transforms every vector field $\tau$ on $Z$
into a BRST vector field
\begin{eqnarray*}
\tau=\tau^A\partial_A\mapsto \nabla_\tau=\tau^A(\partial_A +\widetilde\gamma_A^a\partial_a),
\end{eqnarray*}
which is the derivation $\nabla_\tau$ of ${\cal A}(Z)$ such that
\begin{eqnarray*}
\nabla_\tau(sf)=(\tau\rfloor ds)f +s\nabla_\tau(f), \quad f\in{\cal A}(Z),\quad s\in
C^\infty(Z).
\end{eqnarray*}
Thus, one can think of the splitting (\ref{cmp70}) as being a
BRST connection on $Z$.
For instance, every linear connection
\begin{eqnarray*}
\gamma=dz^A\otimes (\partial_A +\gamma_A{}^a{}_bv^b \partial_a)
\end{eqnarray*}
on the vector bundle $E\to Z$ yields the BRST connection
\begin{equation}
\gamma_S=dz^A\otimes (\partial_A +\gamma_A{}^a{}_bc^b\partial_a) \label{cmp73}
\end{equation}
on $Z$ such that, for any vector field $\tau$ on $Z$ and any BRST function $f$,
the graded derivation $\nabla_\tau(f)$ is exactly the covariant derivative
of $f$ relative to the connection $\gamma$.
The $\wedge E^*$-dual ${\cal V}^*_E$ of ${\cal V}_E$ is a vector bundle over $Z$
which is locally isomorphic to the vector bundle
\begin{eqnarray*}
{\cal V}^*_E\mid_U\approx \wedge E^*\op\otimes_Z({\rm pr}_2VE^*\op\oplus_Z T^*Z)\mid_U,
\end{eqnarray*}
and has the transition functions
\begin{eqnarray*}
&& v'_{j_1\ldots j_kj}= \rho^{-1}{}_{j_1}^{a_1}\cdots
\rho^{-1}{}_{j_k}^{a_k} \rho^{-1}{}_j^a v_{a_1\ldots a_ka}, \nonumber\\
&& z'_{i_1\ldots i_kA}=
\rho^{-1}{}_{i_1}^{b_1}\cdots
\rho^{-1}{}_{i_k}^{b_k}\left[z_{b_1\ldots b_kA}+ \frac{k!}{(k-1)!}
v_{b_1\ldots b_kj}\partial_A(\rho^j_{b_k})\right]
\end{eqnarray*}
of the bundle coordinates $(z_{a_1\ldots a_kA},v_{b_1\ldots b_kj})$,
$k=0,\ldots$, with respect to the dual bases $\{dz^A\}$ for $T^*Z$ and
$\{dc^b\}$ for ${\rm pr}_2V^*E=E^*$.
Global sections of this vector bundle
constitute the ${\cal A}(Z)$-module of exterior BRST
1-forms $\phi=\phi_A dz^A + \phi_adc^a$
on $Z$, which have the coordinate transformation law
\begin{eqnarray*}
\phi'_a=\rho^{-1}{}_a^b\phi_b, \qquad \phi'_A=\phi_A
+\rho^{-1}{}_a^b\partial_A(\rho^a_j)\phi_bc^j.
\end{eqnarray*}
Then
the morphism $\phi:u\to {\cal A}(Z)$ can be seen as the interior product
\begin{equation}
u\rfloor \phi=u^A\phi_A + (-1)^{\nw{\phi_a}}u^a\phi_a. \label{cmp65}
\end{equation}
There is the exact sequence
\begin{equation}
0\to \wedge E^*\op\otimes_ZT^*Z\to{\cal V}^*_E\to \wedge E^*\op\otimes_Z {\rm pr}_2VE^*\to 0.
\label{cmp72}
\end{equation}
Any BRST connection $\widetilde\gamma$ (\ref{cmp70}) yields the
splitting of the exact sequence (\ref{cmp72}), and defines the corresponding
decomposition of BRST 1-forms
\begin{eqnarray*}
\phi=\phi_A dz^A + \phi_adc^a =(\phi_A+\phi_a\widetilde\gamma_A^a)dz^A +\phi_a(dc^a
-\widetilde\gamma_A^adz^A).
\end{eqnarray*}
BRST $k$-forms $\phi$ can be defined as sections
of the graded exterior bundle $\overline\wedge^k_Z{\cal V}^*_E$ such that
\begin{eqnarray*}
\phi\overline\wedge\sigma =(-1)^{\nm\phi\nm\sigma +\nw\phi\nw\sigma}\sigma\overline\wedge \phi.
\end{eqnarray*}
The interior product (\ref{cmp65})
is extended to higher BRST forms by the rule
\begin{eqnarray*}
u\rfloor (\phi\overline\wedge\sigma)=(u\rfloor \phi)\overline\wedge \sigma
+(-1)^{\nm\phi+\nw\phi\nw{u}}\phi\overline\wedge(u\rfloor\sigma).
\end{eqnarray*}
The graded exterior differential
$d$ of BRST functions is introduced by the condition
$u\rfloor df=u(f)$
for an arbitrary BRST vector field $u$, and is
extended uniquely to higher BRST forms by the rules
\begin{eqnarray*}
d(\phi\overline\wedge\sigma)= (d\phi)\overline\wedge\sigma +(-1)^{\nm\phi}\phi\overline\wedge(d\sigma), \qquad d\circ d=0.
\end{eqnarray*}
It takes the coordinate form
\begin{eqnarray*}
d\phi= dz^A \overline\wedge \partial_A(\phi) +dc^a\overline\wedge \partial_a(\phi),
\end{eqnarray*}
where the left derivatives
$\partial_A$, $\partial_a$ act on the coefficients of BRST forms by the rule
(\ref{cmp50'}), and they are graded commutative with the forms $dz^A$, $dc^a$.
The Lie
derivative of a BRST form $\phi$ along a BRST vector field $u$ is given by
the familiar formula
\begin{eqnarray*}
{\bf L}_u\phi= u\rfloor d\phi + d(u\rfloor\phi).
\end{eqnarray*}
\section{The Koszul--Tate resolution}
To construct the vector bundle $E$ of antighosts, let
us consider the vertical tangent bundle $V_Q(V^*Q)$ of $V^*Q\to Q$.
Let us chose the bundle $E$ as the Whitney sum of the
bundles $E_0\oplus E_1$ over $V^*Q$ which are the infinite Whitney sum
over
$V^*Q$ of the copies of
$V_Q(V^*Q)$. We have
\begin{equation}
E= V_Q(V^*Q)\op\oplus_{V^*Q}V_Q(V^*Q)\oplus\cdots.
\label{mm84}
\end{equation}
This bundle is provided with the holonomic coordinates $(t,q^i,p_i,\dot
p_i^{(k)})$, $k=0,1,\ldots$, where $(t,q^i,p_i,\dot
p_i^{(2r)})$ are coordinates on $E_0$, while
$(t,q^i,p_i,\dot p_i^{(2r+1)})$ are those on $E_1$. We call
$k$ the antighost number, while $k\,{\rm mod2}$ is the Grassman parity. The
dual of
$E\to V^*Q$ is
\begin{eqnarray*}
E^*= V^*_Q(V^*Q)\op\oplus_{V^*Q}V^*_Q(V^*Q)\oplus\cdots.
\end{eqnarray*}
It is endowed with the
associated fibre bases
$\{c_i^{(k)}\}$, $k=1,2,\ldots$, such that
$c_i^{(k)}$ have the same linear coordinate
transformation law as the coordinates $p_i$. The corresponding
BRST vector fields and BRST forms are introduced on $V^*Q$ as sections of the
vector bundles ${\cal V}_E$ and ${\cal V}^*_E$, respectively.
The $C^\infty(V^*Q)$-module ${\cal A}(V^*Q)$ of BRST functions is graded by the
antighost number as
\begin{eqnarray*}
{\cal A}(V^*Q)=\op\oplus_{k=0}^\infty {\cal N}^k, \qquad {\cal N}^0=C^\infty(V^*Q).
\end{eqnarray*}
Its terms ${\cal N}^k$ constitute a complex
\begin{equation}
0\op\longleftarrow C^\infty(V^*Q) \op\longleftarrow {\cal N}^1\op\longleftarrow \cdots \label{mm90}
\end{equation}
with respect to the Koszul--Tate differential
\begin{eqnarray}
&& \delta: C^\infty(V^*Q)\to 0, \nonumber \\
&& \delta(c^{2r}_i)= a_{ij}\sigma^{jk}c^{2r-1}_k, \qquad r>0, \label{mm91}\\
&& \delta(c^{2r+1}_i)=(\delta_i^k- a_{ij}\sigma^{jk})c^{2r}_k, \qquad, r>0, \nonumber\\
&& \delta(c^1_i)=(\delta_i^k- a_{ij}\sigma^{jk})p_k. \nonumber
\end{eqnarray}
The nilpotency property $\delta\circ\delta=0$ of this differential is the corollary
of the relations (\ref{mm100}) and (\ref{N21}).
\begin{prop}
The complex (\ref{mm90}) with respect to the differential (\ref{mm91}) is the
Koszul--Tate resolution, i.e., its homology groups are
\begin{eqnarray*}
H_{k>1}=0, \qquad H_0=C^\infty(V^*Q)/I_N=C^\infty(N_L).
\end{eqnarray*}
\end{prop}
Note that, in particular cases of the degenerate quadratic Lagrangian
(\ref{N12}), the complex (\ref{mm90}) may have a subcomplex, which is also
the Koszul--Tate resolution. For instance, if the fibre metric $a$ in $VQ\to
Q$ is diagonal with respect to a holonomic atlas of $VQ$, the constraints
(\ref{zzz}) are irreducible and the complex (\ref{mm90}) contains a
subcomplex which consists only of the antighosts
$c_i^{(1)}$.
Now let us construct the BRST
charge ${\bf Q}$ such that
\begin{eqnarray*}
\delta(f)=\{{\bf Q}, f\}, \qquad f\in {\cal A}(V^*Q)
\end{eqnarray*}
with respect to some Poisson bracket. The problem is to find the Poisson
bracket such that $\{f,g\}=0$ for all $f,g\in C^\infty(V^*Q)$. To overcome
this difficulty, one can consider the vertical extension of Hamiltonian
formalism onto the configuration bundle $VQ\to{\bf R}$ \cite{giach99,book98}.
\begin{lem} \label{mm81'}
Given a fibre bundle $Y\to X$, there is the isomorphism
\begin{eqnarray*}
VV^*Y \op\cong_{VY} V^*VY, \quad p_i\longleftrightarrow\dot v_i,
\quad \dot p_i\longleftrightarrow\dot y_i.
\end{eqnarray*}
\end{lem}
\begin{proof}
The proof is based on inspection of the
transformation laws of the holonomic
coordinates $(x^\lambda, y^i, p_i)$ on
$V^*Y$ and $(x^\lambda, y^i, v^i)$ on $VY$.
\end{proof}
Given a configuration bundle $Q\to{\bf R}$, let us consider the vertical
tangent bundle $VQ\to{\bf R}$, seen as a configuration bundle of the above
mentioned vertical extension of Hamiltonian formalism. By
virtue of Lemma
\ref{mm81'}, the corresponding Legendre bundle $V^*(VQ)$ is isomorphic to
$V(V^*Q)$, and is provided with the holonomic coordinates $(t,q^i,p_i,\dot
q^i,\dot p_i)$ such that $(q^i,\dot p_i)$ and $(\dot q^i, p_i)$ are
conjugate pairs of canonical coordinates. The momentum phase space $V(V^*Q)$ is
endowed with the canonical exterior 3-form
\begin{equation}
{\bf\Omega}_V=\partial_V{\bf\Omega}=[d\dot p_i\wedge dq^i +dp_i\wedge d\dot q^i]\wedge dt,
\label{m145}
\end{equation}
where we use the compact notation
\begin{eqnarray*}
\dot\partial_i=\frac{\partial}{\partial\dot q^i}, \quad \dot\partial^i=\frac{\partial}{\partial\dot
p_i}, \quad \partial_V=\dot q^i\partial_i +\dot p_i\partial^i.
\end{eqnarray*}
The corresponding Poisson bracket on $V(V^*Q)$ reads
\begin{eqnarray*}
\{f,g\}_{VV} =\dot\partial^if\partial_ig +\partial^if\dot\partial_ig -\partial^ig\dot\partial_if
-\dot\partial^ig\partial_if.
\end{eqnarray*}
To extend this bracket to BRST functions, let us consider the following
graded extension of Hamiltonian formalism \cite{gozz,book98}.
We will assume that $Q\to{\bf R}$ is a vector bundle, and will further denote
$\Pi=V^*Q$.
Let us consider the vertical tangent bundle $VV\Pi$. It
admits the canonical decomposition
\begin{equation}
VV\Pi=V\Pi\op\oplus_{\bf R} V\Pi\op\longrightarrow^{{\rm pr}_1} V\Pi. \label{cmp68}
\end{equation}
Let choose the bundle $E$ as the
Whitney sum of the bundles $E_0\oplus E_1$ over $V\Pi$ which are the
infinite Whitney sum over
$V\Pi$ of the copies of
$VV\Pi$. In view of the decomposition (\ref{cmp68}), we have
\begin{eqnarray*}
E=V\Pi\op\oplus_{\bf R} V\Pi\oplus\cdots\op\to^{{\rm pr}_1} V\Pi.
\end{eqnarray*}
This bundle is provided with the holonomic coordinates $(t,q^i,p_i,\dot
q^i_{(k)},\dot p_i^{(k)})$, $k=0,1,\ldots$, where $(t,q^i,p_i,\dot
q^i_{(2r)},\dot p_i^{(2r)})$ are coordinates on $E_0$ and
$(t,q^i,p_i,\dot
q^i_{(2r+1)},\dot p_i^{(2r+1)})$ are those on $E_1$. The dual
of
$E\to V\Pi$ is
\begin{eqnarray*}
E^*=V\Pi\op\oplus_{\bf R} V\Pi^*
\oplus\cdots.
\end{eqnarray*}
It is endowed with the
associated fibre bases
$\{\overline c^i_{(k)},\overline c_i^{(k)},c^i_{(k)},c_i^{(k)}\}$, $k=1,\ldots$. The
corresponding BRST vector fields and BRST forms are introduced on $V\Pi$ as
sections of the vector bundles ${\cal V}_E$ and ${\cal V}^*_E$, respectively. Let us
complexify these bundles as ${\bf C}\op\otimes_{\bf R}{\cal V}_{VV\Pi}$ and
${\bf C}\op\otimes_{\bf R}{\cal V}^*_{VV\Pi}$.
The BRST extension of the form (\ref{m145}) on $V^*Q$ is the 3-form
\begin{equation}
\Omega_S=[d\dot p_i\wedge dq^i +dp_i\wedge d\dot q^i +i\op\sum_{k=1}^\infty(d\overline
c_i^{(k)}\wedge dc^i_{(k)}- dc_i^{(k)}\wedge d\overline c^i_{(k)})]\wedge dt
\end{equation}
The corresponding bracket of BRST functions on $V^*Q$ reads
\begin{eqnarray}
&&\{f,g\}_S=\{f,g\}_{VV} +i\op\sum_{k=1}^\infty (-1)^{k[f]}[
\frac{\partial f}{\partial \overline c_i^{(k)}}\frac{\partial g}{\partial c^i_{(k)}} + (-1)^k
\frac{\partial f}{\partial \overline c^i_{(k)}}\frac{\partial g}{\partial c_i^{(k)}}- \label{mm103}\\
&&\qquad \frac{\partial f}{\partial c_i^{(k)}}\frac{\partial g}{\partial \overline c^i_{(k)}} - (-1)^k
\frac{\partial f}{\partial c^i_{(k)}}\frac{\partial g}{\partial \overline c_i^{(k)}}]. \nonumber
\end{eqnarray}
It satisfies the condition $\{f,g\}_S=-(-1)^{[f][g]}\{g,f\}_S$.
Then the desired BRST charge takes the form
\begin{eqnarray*}
{\bf Q}=i[\overline c^i_{(1)}(\delta_i^k - a_{ij}\sigma^{jk})p_k + \op\sum_{r=1}^\infty (\overline
c^i_{(2r)}a_{ij}\sigma^{jk}c_k^{(2r-1)} + \overline
c^i_{(2r+1)}(\delta_i^k-a_{ij}\sigma^{jk})c_k^{(2r)})].
\end{eqnarray*}
Using the bracket (\ref{mm103}), one can extend this charge in order to obtain
the BRST complex for antighosts
$c_i^{(k)}$ and ghosts $\overline c^i_{(k)}$.
|
\section{Introduction}
The collimated, highly supersonic Herbig-Haro (HH) jets that emerge from protostars in star forming regions propagate through a complex ambient medium composed
of many dense cloud cores and may eventually collide with some of these
objects.
Although these outflows
are usually observed to propagate away from their sources
in approximately straight lines (see e.g., Reipurth 1997 for a
recent review on protostellar outflows), there are some
cases where the outflow is observed to be deflected.
Among these,
there are some examples, like
HH 270/110 (Reipurth et al. 1996, Rodriguez et al. 1998),
HH 30 (L\'opez et al. 1995), and the
molecular outflow in L 1221 (Umemoto et al. 1991)
in which the deflection seems unlikely to
be caused by peculiar changes in the direction of motion of the central
source. Particularly in the first
case, there seems to be strong
evidence that the jet deflection is caused by the encounter with a dense
cloud.
Previous analytical and numerical work was performed
to investigate the hydrodynamics of interactions
of shock-fronts and supersonic winds with interstellar clouds
(see, e.g., McKee \& Cowie 1975,
Rozyczka \& Tenorio-Tagle 1987, Klein, McKee, \& Colella 1994,
Xu \& Stone 1995, Raga et al. 1998), but all those studies
have focused on the effects of the interaction on the
structure of the cloud.
The problem of the interaction of an astrophysical jet with a
rigid surface has
been investigated analytically with some detail by
Canto, Tenorio-Tagle \& Rozyczka
(1988, hereafter CTR).
In the context of extragalactic jets, the well known correlation and
spatial alignment between radio and
optical structures in extended extragalactic radio-sources and Seyfert
galaxies (e.g., McCarthy et al. 1987, Viegas \& de Gouveia Dal Pino 1992)
have led to a series of analytical
and numerical studies involving the interaction of
light jets with ambient clouds (e.g., Higgins at al. 1995,
Fedorenko \& Courvoisier 1996, Steffen et al. 1997).
In the context of
radiatively cooling, heavy jets (i.e., denser than their surroundings)
$~-~$ a picture believed to be consistent with protostellar jets,
the problem of jet/cloud encounters
and their effects on the jet structure has been discussed by
Raga \& Canto (1995) who focused in the early stages of
the interaction using a simple analytical model and
two-dimensional simulations involving slab jets impacting a large,
flat surface
of high density at an arbitrary angle of incidence.
Subsequently, Canto \& Raga (1996) and Raga \& Canto (1996)
have discussed a
steady-state solution
based on Bernoulli's theorem for adiabatic
and radiatively cooling jets penetrating into a cloud with
a plane-parallel
pressure stratification with exponential profile
(Canto \& Raga 1996) and a spherically symmetric
pressure stratification
with power-law profile (Raga \& Canto 1996). Also,
de Gouveia Dal Pino,
Birkinshaw \& Benz (1996; hereafter GBB96)
and de Gouveia Dal Pino \& Birkinshaw (1996;
hereafter GB96)
have examined numerically
the structure and evolution of jets normally
propagating through
stratified environments with different power-law
pressure distributions.
In the present work, with the help of fully three-dimensional simulations
(which naturally retain sensitivity to asymmetric effects)
we attempt to extend these prior studies by examining
the structure and evolution of radiatively cooling and adiabatic
jets undergoing frontal and off-axis collisions with compact clouds of
finite density.
Although the typical sizes of the cloud cores in
star forming regions are usually much higher than the jet radius ($R_j$),
with the assumption here of more compact clumps (with radius
$R_c \gtrsim R_j$) we are able to follow the whole evolution of the
interacting
system and the deflected beam (whose lifetime time must not much
exceed
the time it takes for the shock that develops
with the impact to travel over the cloud diameter).
Besides, with this assumption we can more realistically
examine, for example,
the effects of the
impact of a beam
with the $edges$ of an ambient cloud in an off-axis collision, and also the
effects of the curvature of the cloud in the beam deflection
$-$ effects that are clearly absent in plane-parallel analyses.
In \S
2 of this paper, we outline the numerical method and the initial conditions.
In \S 3, we present the
results of the simulations, and in \S
4 we address the conclusions and the possible
implications of our results for protostellar jets.
\section{The Numerical Model and Setup}
To simulate the jet/cloud interactions, we have
employed a modified version of our three-dimensional hydrodynamical
code based on the smoothed particle
hydrodynamics (SPH) technique (de Gouveia Dal Pino \& Benz 1993;
see also Benz, 1990, 1991, Monaghan 1992, and Steinmetz \& Mueller 1993 for
an overview of the method and a
discussion of the capabilities and limitations of the SPH).
SPH is a Lagrangean, gridless approach to fluid dynamics in
which particles
track the flow and move with it. The code solves
the hydrodynamics equations of continuity, momentum, and energy
explicitly in time, in Cartesian coordinates.
Originally developed to investigate interactions
between planets and
planatesimals (e.g., Benz, Cameron, \& Melosh 1989),
and stellar encounters (e.g., Benz, Bowers, Cameron, \& Press 1990,
Davies, Benz, \& Hills 1992),
the code was successively implemented to study
supernova explosions (e.g., Herant, Benz,
\& Colgate 1992), and the structure and evolution of overdense jets
propagating into initially homogeneous ambient media
(de Gouveia Dal Pino \& Benz 1993, hereafter GB93).
The later version was subsequently modified to investigate
pulsed jets (de Gouveia Dal Pino \& Benz 1994, hereafter GB94);
molecular outflows and related processes of momentum
transfer between the jet and the molecular environment
(Chernin, Masson, de Gouveia Dal Pino, \& Benz 1994);
jet propagation into stratified environments in star
formation regions (GBB96 and GB96); and more recently,
the
effects of magnetic
fields on the structure of overdense radiatively cooling jets
(e.g., Cerqueira, de Gouveia Dal Pino, \& Herant 1997, Cerqueira
\& de Gouveia Dal Pino 1999, hereafter CG99).
A series of validation tests of the code
and direct comparisons with results of
standard Eulerian, grid-based calculations in two and three-dimensions
were successfully performed $-$ our SPH calculations
produce similar results to those of grid-based schemes,
even considering
a small number of particles, although the spatial resolution
may be inferior (with the production of $smoother $ interfaces)
due to intrinsic numerical diffusion
(GB93, GB94, Chernin et al. 1994, GB96, CG99, Benz 1990; see also
Davies et al. 1993 for a detailed comparison of SPH calculations with
those produced by different finite-difference methods).
In the
present work, we have modified the pure hydrodynamical version of the
code above (see GB93 and GB96) by introducing
a cloud in the homogeneous ambient domain.
Following is a short summary of the key features of the code that
are relevant to this work (for more detailed discussion
of the basic assumptions see the references above).
The computational domain is a 3-D rectangular box
which represents the ambient medium and has
dimensions
$-16R_j \le$ x $\le 16R_j$, $-6R_j \le$ y,z $\le 6R_j$,
where $R_j$ is
the
initial jet radius (and $R_j$ is the code distance unit).
The Cartesian
coordinate system has its origin at the center of the box
and the jet flows through the x-axis, and is continuously
injected into the
bottom of the box [at
$\vec{r}=(-16R_j,0,0)$]. Inside the box, the particles are
initially distributed on a Cartesian grid.
Outflow boundary conditions are
assumed for the boundaries of the box (GB96).
The particles are
smoothed out by a spherically symmetric kernel function of width $h$.
As in previous work (GB93, GB94, GB96, CG99), the initial resolution,
as characterized by the initial value of $h$,
was chosen to be $0.4 R_j$ and $0.2
R_j$ for the ambient, and the jet and cloud particles (see below), respectively.
With this initial particle spacing, the calculations are started with about
74,000 particles.
Former validation tests (e.g., GB93) have shown that the above choice of
initial values of $h$ is
more than appropriate to reveal (with accuracy and reasonably good definition)
the physical properties of the structure of overdense jets.
(The decrease of these initial values by a
factor two, for example, enormously increase the number of particles in the
system and thus the required amount of computer time and space, without significant
improvement to the resolution.) During the evolution of the flow, $h$
is consistently allowed to vary (Benz et al. 1990)
and the resolution is naturally established by the particle
distribution $-$ high-density regions (like shock zones)
have higher resolution because of the larger concentration of particles,
while low-density regions (like the cocoon that envelopes the
beam) retain a lower resolution (see below).
(For a discussion on the basic criteria to monitor accuracy in our SPH code and also in SPH schemes in general,
see also
Benz 1990, 1991, Monaghan 1992, and references therein.)
As in previous work, the jet, the cloud, and the ambient
gas are treated as a fully ionized fluid with an adiabatic index
$\gamma=5/3$ and an ideal equation of state.
The
radiative cooling, which is due to collisional excitation and
recombination, is
implicitly calculated using a time-independent cooling function for a
gas of cosmic abundances cooling from $T \simeq 10^6$ to $10^4$ K (the
cooling is set to zero for $T ~{\rm <} ~ 10^4$ K; see GB93, GB96).
The time integration is done using a second order Runge-Kutta-Fehlberg
integrator. The shock waves which arise in the flow are handled by the
usual
Newmann-Ritchmyer artificial viscosity and a link-list
method is used to find the particle's neighbors (e.g., Monaghan 1992).
An initially isothermal cloud with a Gaussian density (and pressure) profile
$$ n_{cl} = n_a \, + \, n_c \, \exp ^{-(\vec{r} - \vec {r_c})^2/\sigma^2},
\eqno(1)$$
is
placed near the center of the computational domain,
where $n_a$ is the ambient number density,
$n_c$ is the number density in the center of the cloud,
$ r_c^2 = x_c^2 + y_c^2 + z_c^2 $ is its central coordinate, and
$\sigma$ is the width of the Gaussian profile which
we assume to be
$\sigma$ = 0.75
$R_c$ in all the simulations, where
$R_c$ is the initial radius of the cloud.
For simplicity, the initial temperature of the cloud is
assumed to be the same as that of the surrounding ambient medium
($T_{cl} = T_a$).
During the simulations, the cloud
is held steady by the application of an
appropriate gravitational potential upon its particles.
Similarly to previous studies of jet propagation
in stratified environments (Hardee et al. 1992, GBB96, GB96),
this external potential is simply evaluated by assuming that the
cloud is initially in hydrostatic equilibrium so that
$\vec {\nabla } p_{cl} = \rho_{cl} \, \vec {g}$,
where $p_{cl} = n_{cl} \, K \, T_a$ is the thermal pressure in the cloud,
$\rho_{cl} = \bar m \, n_{cl}$ is the mass density,
and
$\vec {g}$ is the gravitational acceleration. Substituting eq. (1) into
the equation above and performing the derivation, it yields
$$ \vec {g} \, = \, - {2 K T_a \over \bar m \sigma^2} \,
\, (\vec {r} - \vec {r_c} ) \, { n_c \exp ^{-(\vec{r}-\vec{r_c})^2/\sigma^2} \over n_a \,
+ \, n_c \exp ^{-(\vec{r}-\vec{r_c})^2/\sigma^2} }
\eqno(2)$$
where $\bar m \simeq 0.5 m_H$ is the mean mass per particle,
$m_H$ is the hydrogen mass, and $K$ is the
Boltzmann constant.
Since in the SPH scheme it is trivial to distinguish between particles
in the jet and those in the ambient medium or in the cloud,
it is easy to have the
external force above computed only over the cloud particles, and
once a particle is swept from the cloud by the impinging jet this force is
no longer computed on it. During the jet/cloud interaction, this
force becomes negligible with respect to
the impact forces. Its effects
on the jet dynamics
are also negligible since the jet is assumed to be highly supersonic
(see below).
The models are parameterized by the dimensionless numbers:
{\it i}) the density ratio between the jet and the ambient medium,
$\eta=n_j/n_a$; {\it ii}) the
ambient Mach number, $M_a=v_j/c_a$ (where $v_j$
is the jet velocity and $c_a$ is the ambient sound speed);
{\it iii}) the jet to
the ambient medium pressure ratio at the jet inlet, $\kappa=p_j/p_a$,
that we assume to be equal to unit;
{\it iv}) the square root of the density ratio between the jet and the center
of the
cloud $\beta = (n_j/n_c)^{1/2}$ (see, e.g., Canto \& Raga 1995);
{\it v}) the ratio between the cloud and the jet radius
$R_c/R_j$;
and {\it vi}) the ratio of
the cooling length in the post-shocked gas behind the bow shock to the
jet radius $q_{bs} = d_{cool}/R_j$ (see, e.g., GB93).
The major advantages and limitations of our SPH calculations have
been addressed in previous work (GB93, Chernin et al. 1994, GB96);
in particular, two points
should be remarked.
Firstly, as we mentioned
above, the
properties of low-density regions are more poorly sampled
because they do not contain as many particles as a denser region.
However, not having to calculate properties of $ empty $
regions is actually one of the advantages of SPH codes over
fixed grid based codes. Secondly,
turbulent effects which should exist in these flows
(since the expected Reynolds numbers are very high, $Re > 10^4$; GB93),
are more difficult to examine because the numerical viscosity of the
code may be too dissipative and the initial particle spacing
is large relative to the size of the eddies that may develop in
the flow ($\sim d_{cool}/Re < R_j/10^4$).
Thus, for this work, we can only consider the
bulk properties of the jet/cloud interaction, i.e., over
a size scale larger than that of most of the
eddies, at which the internal turbulent motions are averaged
out.
\section{The Simulations}
We have carried out a series of numerical experiments of jet/cloud encounters
involving clouds with $R_c = (1 - 2) R_j$ and
initial jet/cloud density parameter
$\beta \simeq 3.5 \times 10^{-2} - 2 \times 10^{-1}$.
The parameters of the simulations were chosen to resemble typical
conditions found in protostellar jets and their environment.
We have adopted an initial
number density ratio between the jet and the ambient medium
$\eta=n_j/n_a=3$, $n_j=600$ cm$^{-3}$, ambient Mach numbers $M_a =
v_j/c_a = 12 - 24 $ (with $v_j \simeq 200 - 400$ km s$^{- 1}$
and
$c_a = (\gamma K T_a /\bar m)^{1/2} \simeq 16.6 $ km s$^{-1}$
is the ambient
sound speed), and $R_j = 2 \times
10^{15}$ cm. The corresponding initial jet Mach numbers are
numbers $M_j = v_j/c_j \simeq 20.8 - 41.7 $
and
$c_j = (\gamma K T_j /\bar m)^{1/2}= c_a/(\eta)^{1/2} \simeq 9.6 $ km s$^{-1}$
(this last condition on $c_j$ is due to the assumed pressure equilibrium
at the jet inlet; see \S 2).
The paragraphs below present the results of the simulations performed
for both off-axis and frontal collisions involving radiatively
cooling and adiabatic jets, and Table 1 summarizes the values of the
input parameters.
\subsection {Radiatively Cooling Jet/Cloud Off-Axis Interactions}
Figure 1 shows the results of a jet/cloud off-axis
interaction in the x-y plane. It depicts the density
contour in the mid-plane section
and the velocity field distribution
evolution of a
radiatively cooling jet which impacts a dense cloud
with an initial radius
$R_c = R_j$,
central coordinates (0, 1.2 $R_j$, 0), and a square root of the
jet to central cloud density ratio
$\beta \approx 4 \times 10^{-2}$.
The incident jet has initial
$\eta =3$ and
ambient Mach number $M_a=12$ (see Table 1), and
the cloud has
central coordinates (0, 1.2 $R_j$, 0).
We find that a double shock pattern develops in the region of impact
at $t/t_d=2.0$ (where $t_d=R_j/c_a \simeq $ 38 yr corresponds to the
transverse jet dynamical time). The incident beam is deflected by
a shock nearly parallel to the surface of the cloud and the
pressure behind this induces a second shock
which slowly propagates into the dense cloud
causing an increase in
its central density by a factor $\sim$ 1.5, from
1850 $n_a$ before the impact to $\sim$ 2800 $n_a$ at the quasi-steady regime
(see below)
which is attained after the impact (at $t/t_d >$ 4.0).
As expected from previous analytical study (CTR),
due to the highly radiatively cooling regime (which is expected in protostellar
outflows),
the angle between the two shocks is very small and they are
both effectively parallel to the cloud surface at the impact zone.
Due to the impact, the beam is initially deflected by an angle
$\theta \simeq 40^o$ but
this deflection angle tends to decrease with time as the beam partially
penetrates the cloud and describes a C-shaped trajectory around the
curved jet/cloud contact discontinuity. After $t/t_d = 4.5$, the
bow shock leaves the computational domain and the deflected
beam propagates at an apparently $quasi-steady$ $state$
regime with nearly constant
deflection angle $\theta \simeq 30^o$ and velocity field
distribution. Later on, however, as the jet slowly excavates a
way through the dense cloud this regime must have an end
and the jet will continue
its propagation without significant deflection (see below).
Weak internal knots develop along the deflected beam as indicated
in Fig. 1 and also by the
density
and pressure profiles across the flow depicted in Fig. 2, for
$t/t_d =$ 4 and 6.5. In the first stages of the interaction,
the velocity field along the deflected beam is somewhat
complex with velocity fluctuations along the
ejected parts of the beam which
are correlated with the positions of the knots.
At $t/t_d =$ 4, it varies from $\sim$ 11.8 $c_a$ near the region of
impact, to $\sim$ 12.3 $c_a$ at x = 4 $R_j$,
$\sim$ 10.0 $c_a$ at x = 5 $R_j$, and
$\sim$ 10.5 $c_a$ at x = 10 $R_j$.
Later on, the velocity
field tends to become more uniform and slightly
decreases with distance from the impact region. At $t/t_d=$ 6.5,
it decreases from
$\sim$ 11.8 $c_a$ near the region of impact to $\sim$ 10.5 $c_a$
in the
middle and $\sim$ 10 $c_a$ at the end of the outflow.
This corresponds
to an average velocity for the deflected beam
$v_{j}^{\prime} \simeq 10.8$ $c_a$, which is
compatible with
the expected value (see e.g., CTR)
$$v_{j}^{\prime} \simeq v_j cos{\theta},
\eqno(3)$$
which gives $v_{j}^{\prime} \simeq$ 10.4 $c_a$
(for $v_j =$ 12 $c_a$ and $\theta \simeq 30^o$).
Before the impact, the bow shock propagates downstream with a
speed (GB93)
$v_{bs} \simeq v_j/[1 + {(\eta \alpha)}^{-1/2}] \simeq 8$,
and after the impact
$v_{bs}^{\prime} \simeq 5.5 $,
which roughly agrees with the predicted estimate
$v_{bs}^{\prime} \simeq v_j^{\prime}/[1 + {(\eta \alpha)}^{-1/2}]\simeq 5$,
for a measured
$\alpha = (R_j/R_h)^2 \simeq 1/4$ from the simulation (where
$R_h$ is the radius at the
jet head which is initially equal to $R_j$ and increases
by a factor about two
after the impact, see Fig. 1).
The impact reduces the jet collimation, as expected
(e.g., CTR, Raga \& Canto 1995), and the reflected
beam has an initial opening angle $\psi \simeq 30^o$
which reduces to $\psi\simeq 20^o $, as
indicated by the density contour maps of Fig. 1,
after $t/t_d$= 4.
We note that the deflected jet fades as it propagates downstream at a
distance
$\sim 10 R_j$ and this fading can be testified by the density
and pressure profiles across the flow depicted in Fig. 2, for
$t/t_d =$ 6.5.
This fading is in part caused by a drastic spreading of the jet material
in the other directions mainly due to the more frontal interaction
that
the beam experiences with the cloud in the x-z plane (see Fig. 3).
For comparison, Figure 4 shows the result of a jet/cloud
interaction for a system
with the same initial conditions as those in Fig. 1 except that
the cloud is positioned off-axis also in the x-z plane, i.e.,
it has central coordinates (0, 1.2 $R_j$, 1.2 $R_j$) and is located
in one of the quadrants of the y-z plane. In this case, the jet
effectively impacts a smaller section of the cloud and the collision
is therefore much
weaker $-$ since the density in the cloud decreases from the center to
the edges according to Eq. (1), the effective density parameter $\beta$
is larger by almost an order of magnitude than that of Fig. 1
($\beta \simeq 2 \times 10^{-1}$). As a consequence,
the deflection angle of the
beam is smaller (at the time depicted,
$\theta \simeq 16^o$
against
$\theta \simeq 30^o$ in Fig. 1). In Fig. 4, we also note that
the interaction between the beam and the cloud is almost completed.
The jet has
already
penetrated most of the cloud extension and is resuming
its original direction of propagation. Consistently, the deflected jet
of Fig. 4 has a propagation velocity
($v_{j}^{\prime} \, \simeq 11.5 c_a$
$ \simeq \, v_j \, cos {\theta} $)
which is larger than that
of Fig. 1, and is fading earlier as indicated by the density
contour map (the jet density decreases to less than $n_a$ above
5 $R_j$).
We can estimate a lower limit for the survival
time $t_c$ of the jet/cloud interaction and the
complete depletion of the deflected beam
by evaluating the time that the second shock
takes to travel over the cloud diameter $d_c = 2 R_c = f R_j$
(where $f > 0$ is a multiple factor of the jet radius). One-dimension
momentum flux conservation argument and the resulting shock
geometry for a radiatively cooling jet interacting with a
cloud of constant density and a density parameter
$\beta = (n_j/n_c)^{1/2} \ll 1$ yields a cloud shock speed
$v_{cs} \simeq \beta v_j \sin {\theta} $
(see e.g., CTR, Raga \& Canto 1995),
and $t_c \gtrsim d_c/v_{cs}$
or
$$\frac{t_c}{t_d } \gtrsim \frac{f}{M_a \beta \sin {\theta}}
\eqno(4)$$
For the deflected jet of Fig. 1 this equation gives
$t_c/t_d \gtrsim $ 7.5 (for an average $\theta = 34^o$ and $f = $2),
so that after this time,
we would expect the incident beam to have
penetrated the cloud almost completely thus suffering no more deflection.
This time is, however, just a lower limit because
as the jet penetrates the cloud both the
deflection angle and the density parameter $\beta$ decrease
(due to the compression of the cloud) thus
increasing $t_c$ in the equation above.
For the jet in Fig. 4 we estimate
$t_c/t_d >$ 3.
Figure 5 depicts the mid-plane density contour and the
velocity field distribution of five radiative cooling
jets interacting with clouds with different initial values of
jet/cloud density parameter $\beta$.
All the inicident jets have the same initial conditions as in
Fig. 1.
The top jet is propagating into a homogeneous
medium and thus suffers no deflection. The four other
systems have
$\beta \simeq 2 \times 10^{-1}$, $7 \times 10^{-2}$,
$4 \times 10^{-2}$ (as in Fig. 1), and
$3.5 \times 10^{-2}$
from top to bottom, respectively.
As expected, the angle of deflection increases
with the increase of the density of the cloud (smaller $\beta$)
and the
larger the angle the slower the propagation speed of the
deflected beam.
Likewise, the interaction time increases with
decreasing $\beta$ (Eq. 4) and the jet with
$\beta \simeq 2 \times 10^{-1}$ (Fig. 5, second panel), for example,
has already practically
resumed its initial direction of propagation after partially
destroying the cloud, while the jet with
$3.5 \times 10^{-2}$ (bottom) is still interacting
with the cloud.
Equation (3) gives the following lower limit interaction times:
$t_c/t_d >$ 0, 2.5, 5, 7.5, and 8.5, respectively. (We note,
however, that Eq. (3) is valid essentially for
$\beta \ll 1$ so that the estimated values above for
$t_c$ for the
systems with larger $\beta$ are less reliable).
Figure 6 shows the evolution of an interacting system
with initial conditions similar to those of Figs. 1 and 5
but the cloud has now a radius twice as bigger
($R_c = 2 R_j$) and an effective density parameter
$\beta \approx 6 \times 10^{-2}$ (which is comparable to that of the
third jet of
Figure 5).
The larger cross section of the obstacle causes the deflection angle of
the beam to be larger and the interaction time much longer
(Eq. 4 gives
$t_c/t_d >$ 9.3, for a
cloud diameter $d_c = 4 R_j$, or f = 4).
The compressing shock increases the central density of the
cloud to a maximum factor $\simeq$ 1.4 and
the deflection angle of the beam
varies from
$\theta \simeq 45^o$ (at the initial impact at $t/t_d =$ 2)
to $\theta \simeq 40^o$ after $t/t_d = $ 3.5
(against $\theta \simeq 25^o$ in the third jet of Fig. 5, and
$\theta \simeq 35^o$ in the jet of Fig. 1 at the same
evolution time). Consistent with these deflection angles (Eq. 3), the
average velocity of the deflected beam varies from
$v_{j}^{\prime} \simeq$ 9 to 7 $c_a$,
and the bow shock velocity decreases from $\simeq$ 8 $c_a$ before the
impact to $\simeq$ 5 $c_a$ after it.
The opening angle of the deflected
beam varies from $\psi \simeq 35^o$
to $\psi\simeq 25^o $, as
indicated by the density contour maps of Fig. 6.
The density contour maps also indicate the formation of
some bright knots along the deflected beam and bow shock.
Those knots have densities $n_k/n_a \simeq$ 5 to 15 and
were originally produced in the region of impact at the
contact discontinuity and carried downstream by
the deflected jet.
At $t/t_d =$ 7, the density in the deflected beam decreases
from
$ n_j^{\prime}/n_a \simeq 5$ near the impact to less than 1 above
$\sim 5 R_j$.
\subsection {Jet/Cloud Off-Axis Interactions for Jets with Different
$M_a$}
Figure 7 depicts two interacting systems with
different initial ambient Mach numbers,
$M_a =$ 12 (top) and 18 (bottom). Both jets have
an initial density parameter
$\beta \simeq 3.5 \times 10^{-2}$ and
the other initial conditions are the same as in Figs. 1 and 5.
(Note that the jet with $M_a$ = 12 is the same jet
in the bottom panel
of Fig. 5.)
The jet with larger $M_a$ interacts faster with the cloud (Eq. 4)
and thus fades
earlier too
($t_c/t_d >$ 6 and 8.5, for the $M_a =$ 18 and 12, respectively).
At
the time depicted in Fig. 7, the $M_a = 18$ jet is already
starting resuming
its original direction.
Initially, both systems have similar deflection angles
($\theta \simeq 40^o$) but because of the larger interaction rate of
the $M_a=$ 18 jet, at the time depicted its angle has become
smaller ($\theta \simeq 35^o$ in the
$M_a =$ 18, against
$\theta \simeq 40^o$ in the $M_a= $12
jet).
The density in the deflected beam decreases from
$n_j^{\prime}/n_a \simeq $ 8 near the impact region, to
less than unity above a distance $\sim 5 R_j$ in the
$M_a =$ 18 jet, while in the $M_a =$12 it decreases
from $n_j^{\prime}/n_a \simeq $ 11 near the impact region, to
$\simeq $ 2.4 at $\sim 5 R_j$, and
less than unity above $\sim 8 R_j$.
Figure 8 shows an example of an even higher Mach
number jet ($M_a = $ 24) after it had
impacted a cloud (in an off-axis collision)
with a relatively large density parameter
($\beta \simeq 2 \times 10^{-1}$
which is comparable to
that of the second jet in Fig. 5).
(The counterpart of this jet propagating
into an initial homogeneous environment can
be found, e.g., in GBB96, and Cerqueira, de Gouveia
Dal Pino \& Herant 1997.)
In the time depicted in the figure, the interaction has already
finished and the jet has completely resumed its
original propagation direction but the interaction left some
interesting signatures
in the system. We see that the remains of the cloud are still
present at x = 0 and have been involved by the bow
shock structure. Also,
part
of the dense shell that developed at the
head of the beam from the cooling of the shocked jet material
has been
detached from the head by the collision
and left behind in the cocoon causing a
remarkable asymmetry in the jet head region.
\subsection {Frontal Jet/Cloud Interactions}
Figure 9 shows an example of a strong frontal impact of a jet with
a dense cloud with a density
parameter $\beta \simeq 7 \times 10^{-2}$. The
initial conditions in this system are the same as in the
third jet of Fig. 5 except for the location of the
cloud which now has coordinates (0, 0, 0).
The frontal collision is obviously much stronger and
the compression increases the density of the cloud
by a factor $\sim$ 2.5
(from $n_c/n_a \simeq$ 700 before the impact, to
$n_c/n_a \simeq$ 1800 after it).
The jet splits into two beams on either side of the cloud
with equal deflection angle
$\theta \simeq 45^o$ and a double bow shock structure (or double lobes) develops.
By the time the bow shocks leave the computational
domain ($t/t_d \simeq $ 6) most of the cloud has been destroyed by the interaction
and the double deflected beam has almost faded, although the jet
has not yet completely resumed its original direction.
The propagation velocity of the deflected beams varies
from 7 to 10 $c_a$ along the flow,
and the average density in the post-impact beam
is $\sim$ 4 $n_a$.
Simulations involving weaker frontal interactions with less dense clouds
have shown that the beam easily sweeps the cloud material to
the working surface causing it to become more knotty and wider.
\subsection {Adiabatic Jet/Cloud Interactions}
Figure 10 shows the density in the mid-plane section and the
velocity field distribution of an adiabatic jet which is
interacting with a (adiabatic) cloud in an off-axis collision
after it had propagated over a distance $\sim 30 R_j$.
The initial conditions are the same as in Figure 1.
The interaction
causes a compression in the interacting cloud of the same
amount as that produced by the radiatively cooling jet
(by a factor $\sim $ 1.5) but the adiabatic jet
penetrates less deeply into the cloud describing a more
pronounced C-shaped trajectory around the contact discontinuity.
The deflection angle is in turn larger
($\theta \simeq 40^o$ against
$\theta \simeq 35^o $ in the radiatively cooling jet at
$t/t_d =$4) and so the opening angle
(as indicated by the density contour maps $\psi \simeq 35^o$
in the adiabatic jet and
$\psi \simeq 16^o$ in the radiatively cooling counterpart in Fig. 1).
This result is consistent with previous analytical
and two-dimensional numerical studies of the interaction of
adiabatic and radiatively cooling jets with
plane-parallel and spherically stratified clouds (Canto \& Raga 1996,
Raga \& Canto 1996).
At the time depicted, the adiabatic jet is trying to resume
its original direction causing the fading of the deflected beam
(the density in the deflected beam decreases from
$n_j^{\prime}/n_a \simeq$ 9.5 near the impact region to less than unity at a
distance
$\sim 4 R_j$ in the adiabatic jet, while it decreases from
$n_j^{\prime}/n_a \simeq$ 11.5 to 1 at a distance
$\sim 10 R_j$ in the cooling jet).
The average velocity in the adiabatic and radiatively cooling
deflected beams
are of the same order
$v_j^{\prime} \simeq$ 11 $c_a$.
\section{Discussion and Conclusions}
We have presented the results of fully three-dimensional simulations of
overdense, radiatively cooling and adiabatic jets colliding
with dense, compact
ambient clouds. Frontal and off-axis collisions were examined.
Evaluated for a set of parameters which are particularly appropriate
to protostellar jets [with initial density ratios between the jet and the
ambient medium $\eta \approx 3$,
ambient Mach numbers $M_a \approx 12-24$,
and jet/cloud density parameters
$\beta = (n_j/n_c)^{1/2} \simeq 3.5 \times 10^{-2} - 2 \times 10^{-1} $],
our results indicate
that important transient and also permanent effects
may occur on the jet as a
consequence of the interaction.
Our main results can be summarized as follows.
1. As in previous analytical and
two-dimensional numerical study (CTR, Raga \& Canto 1995),
we find that the primary effect of a radiatively cooling jet/cloud
collision is to deflect the beam by
a shock to a direction
initially nearly parallel to the surface of the cloud at the
impact region.
A secondary shock, which is induced by the increased pressure behind the
first,
slowly propagates into the dense cloud
causing an overall increase in
its density.
2. The deflected beam initially describes a C-shaped trajectory around the
curved jet/cloud contact discontinuity but the
deflected angle
tends to decrease with time as the beam slowly
penetrates the cloud.
Later, when the jet has
penetrated most of the cloud extension the
deflected beam fades and the jet tends to resume
its original direction of propagation. Due to the
small size of the clouds
[with radius $R_c \simeq (1-2) R_j$], the lifetimes of
the
interactions deduced from the simulations
are only $\sim$ few 10 to $\sim$ few 100 yr (for jets
with $M_a= $ 24 - 12) but they are longer than the predicted
time from analytical
modeling (Eq.3;
see also Raga \& Canto 95).
3. During the interaction, weak internal knots develop along the deflected beam.
The
velocity field initially has a
complex structure with variations along the flow
that later evolves to a more uniform
distribution. At the region of impact the velocity
is the order of the incident jet velocity, but the average
velocity of the deflected beam is compatible with the predicted value
$v_{j}^{\prime} \simeq v_j cos {\theta}$, where
$\theta$ is the deflection angle, and $v_j$ is
the velocity of the incident beam (e.g., CTR).
The impact also increases the jet opening angle, as expected.
4. Jets in off-axis collisions with clouds with different density
parameter $\beta$ result different deflection angles
and interacting times $-$
the larger the density of the cloud (the smaller the $\beta$)
the
larger the angle and the longer the interaction time (Fig. 5).
The increase in the radius of the cloud also
makes the deflection angle and the interaction time larger,
while the
increase in the incident jet Mach number, $M_a$,
naturally decreases the time of the interaction.
5. Frontal collisions with very dense clouds,
although they are expected to be even rarer,
may also produce peculiar transient features.
They are naturally stronger and faster than off-axis
interactions with similar initial conditions.
Such interactions cause the splitting of the jet
into two beams
which produce a double bow shock structure on either side of the cloud.
This is, however, a very transient feature and thus highly unlikely to
be observed. Weaker frontal interactions, on the other hand, do
not produce a jet splitting and most of the
cloud material is simply swept to the working surface at the jet head.
6. Adiabatic jets interacting with clouds in off-axis collisions
penetrate less deeply into the cloud and describe a more
pronounced C-shaped trajectory around the contact discontinuity.
The deflection angle is in turn larger
than that in its radiatively cooling counterpart.
This result is consistent with previous analytical
and two-dimensional numerical studies involving jet
interactions with plane-parallel and spherically
stratified obstacles
(Canto \& Raga 1996, Raga \& Canto 1996).
The basic features found above in the deflected beam,
such as the decrease in the
jet velocity and the increase in the opening angle
with respect to the incident beam,
have been detected in the HH 110 jet (Reipurth et al. 1996,
Rodriguez et al. 1998) which is
possibly the most convincing example,
among the protostellar jets, of beam
deflection by interaction with an ambient cloud.
As stressed by those authors, no driving source has been
detected at the apex of the HH 110 jet and it seems to be
the deflected part of the fainter HH 270 jet.
They have proposed that the deflection is caused
by an interaction of the jet with a dense cloud core
with $\beta \simeq 0.03 - 0.3$. Since this
system seems to lie close to the plane of the sky we may
directly compare its morphology with the results of the simulations.
Although the measured deflected angle ($\sim 58^o$)
is larger than those obtained in our simulations,
the $head-neck$ bright structure we see in the density contour
maps of the off-axis simulations at the region of impact
in strong interactions (see Figs. 1, 5, 6, and 7),
is remarkably similar to the morphology of the HH 110 knot A
located at the apex of the HH 110 in the region
where the deflection of the HH 270 jet is believed to occur
(see Fig. 6 of Reipurth et al. 1996).
Besides, the HH 110 flow is observed to have a well collimated
chain of knots near the apex and then to widen in a cone of
large opening angle ($\psi \simeq 12^o$), until it fades in a
final curve (Reipurth et al. 1996).
All those morphological characteristics are
clearly detected in our off-axis simulations, and
the wide range of proper motions measured for the knots
in HH 110 is compatible with the complex velocity
structure found in the deflected beams.
Furthermore, the estimated average velocity of HH 110 is consistent
with the $v_{j}^{\prime} \simeq v_j \, cos {\theta}$ relation (Reipurth et al.
1996).
All these similarities
strongly support the proposed jet/cloud interaction interpretation
for the HH 110/HH 270 system.
The fact that the deflection angles derived from the simulations
are smaller and the opening angles are larger than those observed
in the HH 270/HH 110 system [even for the simulations
involving an incident jet with $v_j \simeq$ 300 km s$^{-1}$
(or $M_a = $ 18) which is the order of that inferred from observations]
and the fact that the jet/cloud interaction is probably
still taking place (thus requiring an interacting time $\simeq$
the dynamical time of the deflected jet),
indicate that the interacting
cloud in that system must have a radius $R_c \gg R_j$,
as suggested by Reipurth et al. (1996), and
a density parameter $\beta \lesssim$ 0.1, as
indicated by the simulations.
Applied to the general context of
the protostellar jets,
the interactions examined here are essentially transient
processes and thus the probability of they being observed
must be very small,
since the typical dynamical lifetimes of the observed outflows
($\tau \gtrsim 10^{4} $ yr; e.g., Bally \& Devine 1997)
are much larger than the inferred survival times of the jet/cloud
interactions.
Nonetheless these interactions may leave a variety of interesting
more permanent features imprinted in the remaining outflow.
For example, we have found that after a weak interaction
of only few decades
with a compact cloud (with density parameter $\beta \simeq $ 0.2)
the $M_a =$ 24 jet has retained in
its working surface
the remains of the cloud, and
some fragments of the
dense
shell have been detached from the head during the collision
producing remarkable asymmetries in the beam (Fig. 8).
Weak interactions are also able to produce some
wiggling in the deflected beam (Figs. 5 and 8) but this feature
will last not much longer than the interaction.
The production of jet wiggling by the interaction
with a plane-parallel stratified environment was also
reported in previous numerical studies (GBB96; GB96).
The simulations also indicate that before a deflected
beam fades it may have time enough to
produce and deposit some knots into the
working surface at the jet head (see, e.g., Fig. 6)
which may contribute
to enrich and enlarge the knotty pattern
behind the bow shock.
Since this clumpy structure resembles the knotty pattern
commonly observed in HH jets, this result suggests that
transient jet/cloud interactions may also play an important role
in the formation of these HH structures.
Finally, we should also note that a jet undergoing many transient
interactions with compact clumps along its propagation and lifetime
may inject a considerable amount of
shocked jet material sideways
into the surrounding
ambient medium over a transverse extension which will depend on the deflection
angle of each interaction
(Figs. 6 and 7). This process
may be therefore, a powerful tool for momentum transfer
and
turbulent mixing with the ambient medium $-$ a process that
may help to feed the slower and wider molecular outflows often associated
with protostellar jets (see e.g., Raga et al. 1993, Chernin et al. 1994,
Cabrit et al. 1997).
\acknowledgements
This work was partially supported by the
Brazilian agencies FAPESP and CNPq.
The author is indebted to the referee Paul Wiita for
his fruitful comments.
The author also would like to acknowledge the kind
hospitality of the Star Formation Group of the
Astronomy Department of the
University of California at Berkeley where most of this
work was done and also relevant and clarifying discussions
with Frank Shu and Bo Reipurth. Technical support from A.H. Cerqueira
is also acknowledged.
\newpage
|
\section{Introduction}
NGC~6752 is a medium-rich globular cluster whose proximity, low
reddening and relatively high galactic latitude ($r \approx 3.8$~kpc,
$E(B-V)=0.04$, $b=-25.6$; Harris 1996) make it an excellent object for
detailed studies. The cluster was selected as one of the targets of an
ongoing survey for eclipsing binaries in globular clusters (Kaluzny,
Thompson \& Krzeminski 1997). The ultimate goal of the project is to
use observations of detached eclipsing binaries to determine the ages
and distances of globular clusters (Paczy\'nski 1997), and to study the
binary star fraction in these clusters.
Until recently only 3 variable stars were known in NGC 6752 (Hogg
1973; Clement 1996 and references therein). One of these is a
population II Cepheid and there is insufficient information on the
other two to define their types. The horizontal branch of NGC 6752 is
blue and the cluster contains no RR~Lyr stars. Three photometric
studies of variable and binary stars based on HST data have been
published during the last three years. Shara et al. (1995) reported
null results in a search for short period variables in the core region
of NGC~6752, while Bailyn et al. (1996) identified 3 candidate
cataclysmic variables, also in the core of the cluster. From an
analysis of the broadened, asymmetric main sequence in NGC 6752,
Rubenstein \& Bailyn (1997) determined that the binary fraction is
probably in the range 15\%--38\% within the core radius.
In this contribution we present an analysis of CCD photometry of NGC
6752. This data set is best suited for a search for variable stars in
the outer parts of the cluster. A separate paper will be devoted to
the analysis of photometry for the central part of the cluster based on
data obtained in 1998 with the 2.5-m du Pont telescope (Kaluzny et al.,
in preparation)
\section{Observations and Data Reduction}
Time-series photometry of NGC~6752 was obtained during the interval
1996 June 23 -- 1997 September 15 with the 1.0m Swope telescope at Las
Campanas Observatory. In 1996 a Loral CCD was used as the detector,
with a scale of 0.435 arcsec/pixel and a field of view of $14.8\times
14.8$ arcmin. Two cameras were used in 1997. The first (LCO camera
SITe1) has a field of view $23.8\times 23.8$ arcmin with a scale of
0.70 arcsec/pixel. The second (LCO camera SITe3) has a field of view
of $14.8\times 22.8$ arcmin with a scale of 0.435 arcsec/pixel. In all
cases the observations were approximately centered on the cluster. A
total of 539 $V$-band images, 42 $B$-band images, and 2 $U$-band images
were collected on 14 nights. Exposure times were sufficiency long --
ranging from 100 to 300 s for the $V$-band, depending on the seeing --
to ensure accurate photometry for stars located 2-3 mag below the
cluster turnoff. The cluster was monitored for 28h35m, 13h30m and
12h00m with the SITe3, Loral and SITe1 cameras, respectively.
Instrumental photometry was extracted using DoPHOT (Schechter, Mateo
\& Saha 1993). We used DoPHOT in the fixed-position mode, with the stellar
positions measured on "template" images (either the best images
obtained during a given run or a combination of the 2-3 best images).
For each one of the CCD cameras used in this survey a separate data
base was constructed using procedures described in detail in Kaluzny et
al. (1996). The total number of stars included in the $V$~filter data
bases for the SITe1, SITe3 and Loral cameras was 45049, 43882 and
22957, respectively. The quality of the derived photometry is
illustrated in Fig. 1 in which we have plotted the $rms$ deviation
versus average magnitude for stars measured with the SITe3 camera.
This plot includes 31442 stars with $12.1<V<20.8$ with at least 74
observations. Photometry for stars with $V<13.5$ is poor since these
stars were frequently over-exposed. To select potential variables we
employed three methods, as described in some detail in Kaluzny et al.
(1996). $V$-band light curves showing possible periodic signals or smooth
changes on time scales of weeks were selected for further examination.
Eleven certain variables were identified in this way. Note that the three
variables listed in Clement (1996) are all saturated on our CCD frames.
The instrumental photometry was transformed to the standard $BV$
system using observations of standard stars from Landolt (1992),
leading to relations of the form:
\begin{eqnarray}
v = a_{1} + V + a_{2}\times (B-V) \\
b = a_{3} + B + a_{4}\times (B-V)
\end{eqnarray}
The linear coefficients $a_{2}$ and $a_{4}$ were separately derived for
each of the CCD cameras. The additive constants $a_{1}$ and $a_{3}$ were
derived based on $BV$ photometry
of 12 secondary standards selected from Cannon \& Stobie (1973).
Average values of
the $B-V$ color were used
when transforming the observations of the variables from instrumental
$v$ magnitudes to standard $V$ magnitudes
This procedure leads to systematic errors not exceeding
0.003 mag.\footnote{The values of $a_{2}$ in Eq. 1 had values
from --0.02 to 0.03 depending on the CCD. None of the detected variables
has an observed variation of $B-V$ exceed 0.1 mag. Hence,
systematic errors due to the adoption of an average $B-V$ color in Eq. 1
do not exceed 0.003 mag}
On the night of Jun 26, 1997 we used the SITe1 camera to secure two
$600$-sec $U$-band frames centered on the cluster. These frames,
supplemented with a pair of short $BV$ exposures ($V$ 25-sec, $B$
35-sec) and a pair of long exposures $BV$ exposures ($V$ 120-sec, $B$
150-sec), were used to derive $UBV$ photometry for a large sample of
stars from the cluster field. The color-magnitude diagram based on
these data is discussed in Sec. 4. The transformation from the
instrumental to the standard $UBV$ system was determined using
observations of several Landolt (1992) fields obtained over the whole
observing sub-run:
\begin{eqnarray}
v = c_{1} + V + 0.04\times (B-V) \\
b-v = c_{2} + 0.912\times (B-V)\\
u-b = c_{3} + 0.957\times (U-B)
\end{eqnarray}
The zero points of our $UBV$ photometry of NGC 6752 were derived
from observations of secondary standards from Cannon \& Stobie (1973).
\section{Results for variables}
In Table 1 we list equatorial coordinates of the 11 newly identified
variables. Approximate angular distances
of the variables from the cluster center are given in the 4th column.
The last column gives a variability type for each object
(ECL -- eclipsing binary; EW -- contact binary; EA -- detached eclipsing
binary, SX -- SX Phe variable).
The limiting radius of the cluster is estimated to be 31 $arcmin$
(Webbink 1985)
and all variables from Table 1 are located within this radius.
Variables V4, V5 and V11, which are located at large radii from the
cluster center, were outside the field of view of the SITe3 and Loral
cameras and were observed only with the SITe1 camera. Variable V10 was
outside the field of view of the Loral camera. Variables V12 and V13,
which are located relatively close to the projected cluster center, are
absent in the data base for the SITe1 camera because of crowding. All
other variables are present in all data sets. Finding charts for the
11 variables are given in Figs. 2 and 3.
Variable V9 is the brighter component of an unresolved blend of two
images. However, this close visual pair could be resolved on images
obtained with the du Pont telescope. The fainter component has a
$V$-magnitude of $V=17.65 $ and color $B-V=0.45$, and we used these
values to decompose the observations of the combined image obtained
with the Swope telescope.
Figure 4 shows the location of all of the newly identified variables on
the cluster color-magnitude diagram (CMD). For the SX~Phe stars the
plotted positions correspond to the intensity-averaged magnitudes. For
the eclipsing variables the magnitudes at maximum light are plotted.
The phased light curves of variables V4-9 and V11-14 are shown in Fig.
5. The periods of variability were derived using an algorithm based on
an {\it analysis of variance} statistic introduced by
Schwarzenberg-Czerny (1989, 1991). Variable V10 showed flat light
curves on all but one night. On the night of June 7 1997 that variable
showed an eclipse event lasting more than 5 hours. Figure 6 shows
time-domain light curves of V10 obtained on the nights of June 2 1997
and June 7 1997.
\newpage
\setcounter{figure}{1}
\begin{figure}[h]
\vbox to9cm{\rule{0pt}{9cm}}
\special{psfile=thompson.fig2.ps angle=0 hoffset=60 voffset=0
vscale=100 hscale=100}
\caption{Finder charts for variables V7-10 and V12-13. Each chart is 44
arcsec on a side with north up and east to the left.}
\end{figure}
\begin{figure}[h]
\vbox to4cm{\rule{0pt}{4cm}}
\special{psfile=thompson.fig3.ps angle=0 hoffset=40 voffset=0
vscale=100 hscale=100}
\caption{Finder charts for variables V4-6, V11 and V14. Each chart is 44
arcsec on a side with north up and east to the left.}
\end{figure}
\newpage
\begin{figure}[h]
\vbox to14cm{\rule{0pt}{14cm}}
\special{psfile=thompson.fig4.ps angle=0 hoffset=-80 voffset=-150
vscale=120 hscale=120}
\caption{A $V - (B-V)$ CMD for NGC~6752 with the positions of the
variables marked: triangles -- SX~Phe stars; asterisks -- contact binaries;
open square -- probable detached binary.}
\end{figure}
\subsection{SX Phe stars}
The periods, average colors, intensity averaged $V$ brightnesses and
full amplitudes of the 3 identified SX~Phe stars are listed in Table 2.
All three variables are candidate blue straggler stars. The observed
luminosities of these SX~Phe stars are consistent with membership in
NGC~6752. This is demonstrated in Fig. 7 which shows the positions of
the variables in a period $vs.$ absolute magnitude diagram. The standard
relation for SX Phe stars from McNamara (1997) is also shown.
Absolute magnitudes for the SX~Phe stars were calculated assuming a
distance modulus of $(m-M)_{V}=13.02$ (Harris 1996).
While V17 is clearly a fundamental mode pulsator, based on its amplitude
and asymmetric light curve, a classification of V12 and V13 is more
problematical. Observational error and the low amplitudes of these stars combine to complicate a determination of the asymmetry of the light curves.
In addition, as McNamara (1997) points out, classification of a star
as a first overtone pulsator based only on low amplitude and
symmetrical light curves is itself questionable. We note here simply
that the properties of these three SX Phe stars are consistent with
pulsation in the fundamental mode and membership in NGC 6752.
\begin{figure}[h]
\vbox to15cm{\rule{0pt}{15cm}}
\special{psfile=thompson.fig5.ps angle=0 hoffset=0 voffset=-130
vscale=100 hscale=100}
\caption{Phased $V$-band light curves for the 10 newly identified variables in
the field of NGC~6752.}
\end{figure}
\newpage
\begin{figure}[h]
\vbox to4cm{\rule{0pt}{4cm}}
\special{psfile=thompson.fig6.ps angle=0 hoffset=0 voffset=-440
vscale=100 hscale=100}
\caption{The $V$ band light curves of variable V10 obtained on the
nights of 1997 June 2 and 7.}
\end{figure}
\begin{figure}[h]
\vbox to5cm{\rule{0pt}{5cm}}
\special{psfile=thompson.fig7.ps angle=0 hoffset=-60 voffset=-170
vscale=100 hscale=100}
\caption{Period vs. absolute magnitude diagram for SX~Phe stars from the field
of NGC~6752. The solid line represents the standard relation for
fundamental mode pulsators (McNamara 1997).}
\end{figure}
\subsection{Eclipsing binaries}
Our sample of newly identified variables includes 8 eclipsing
binaries. Table 3 lists some basic photometric characteristics of
their light curves. Seven objects can be classified as contact binaries
(EW type eclipsing binaries according to the GCVS scheme) and one --
variable V10 -- is most likely a detached eclipsing binary. We have
managed to catch just one eclipse-like event for V10. Our data indicate
that the orbital period of that binary is longer than 2 days. The
position of V10 on the cluster CMD (see Fig. 4) does not support its
membership in NGC~6752 unless the error in $B-V$ is unusually
large. In addition, the variable is located
relatively far from the cluster center. Determination of the radial
velocity and/or proper motion of V10 is necessary in order to clarify
the membership status of this potentially important binary.
Of the seven identified contact binaries only V8 is located among the
blue stragglers on the cluster CMD. Variable V6 occupies a position
slightly below the cluster main-sequence. V11 is located about 0.2 mag
to the red of the subgiant branch. Its position on the CMD indicates
that it does not belong to the cluster. The remaining 3 EW systems are
located to the red of the cluster main-sequence. We have applied the
absolute brightness calibration established by Rucinski (1995) to
estimate $M_{V}$ for the newly identified contact
binaries\footnote{Rucinski \& Duerbeck (1997) have derived a new
version of the calibration $M_{V}=M_{V}(log P, B-V)$ using a sample of
EW systems with $HIPPARCOS$ parallaxes. However, that new calibration
is based on stars from the solar neighbourhood and does not include a
metallicity term.}. That calibration gives $M_{V}$ as a function of
period, unreddened color and metallicity:
\begin{eqnarray}
M_{V} = -2.38 log P +4.26(B-V)_{0} +0.28 -0.3[{\rm Fe/H}]
\end{eqnarray}
\noindent
where $P$ is the period in days.
We adopt $[{\rm Fe/H}]= -1.61$ and $E(B-V)=0.04$ for NGC~6752
(Harris 1996). The formal errors of the estimated values of $M_{V}$ are
about 0.5 magnitude. Figure 8 shows a period versus apparent
distance modulus diagram for the 7 contact binaries from the cluster field.
The apparent distance modulus was calculated as the difference between
$V_{max}$ and $M_{V}^{cal}$ for each system. The data presented in
Fig. 8 indicate that only V14 can be considered a possible cluster
member. The remaining 6 EW systems are most likely
background/foreground objects.
\begin{figure}[h]
\vbox to5cm{\rule{0pt}{5cm}}
\special{psfile=thompson.fig8.ps angle=0 hoffset=-60 voffset=-170
vscale=100 hscale=100}
\caption{Period vs. apparent distance modulus diagram for contact
binaries from the field of NGC~6752. The horizontal line
corresponds to a cluster distance modulus of $(m-M)_{V}=13.02$. Vertical
bars represent formal errors of $M_{V}$
derived from the Rucinski (1995) calibration. }
\end{figure}
Although V9 is classified as a contact binary with $P=0.36$~d we cannot
exclude the possibility that the true period is $P=0.72$~d and that
the variable is a single spotted star, presumably of type RS CVn. Its
position on the cluster CMD suggests that in this case the star is a
subgiant belonging to the cluster (see Fig. 4).
\section{The color-magnitude diagrams}
Since the pioneering studies by Alcaino (1972) and Cannon \& Stobie
(1973) it has been known that the horizontal branch of NGC~6752 is
strongly dominated by stars located to the blue side of the instability
strip. The CMD of the cluster was studied in detail by Buonanno et al.
(1986) based on deep photographic photometry. They showed that the
horizontal branch of NGC~6752 spans about 4 magnitudes in $V$ reaching
$V\approx 18.0$ ($M_{V}\approx 5.0$) on its faint end. Detailed
studies of the properties of the hot subdwarfs forming extended
horizontal branch (EHB) of NGC 6752 have been published by Heber et
al. (1986) and Moehler, Heber \& Rupprecht (1997).
\begin{figure}[h]
\vbox to11cm{\rule{0pt}{11cm}}
\special{psfile=thompson.fig9.ps angle=0 hoffset=0 voffset=-130
vscale=100 hscale=100}
\caption{A $V - (B-V)$ CMD for NGC~6752 based on the
template images obtained with the SITe3 CCD camera.
Two candidate faint EHB stars are marked with triangles.
Note also the excess of stars on the red side of the EHB. }
\end{figure}
As a by-product of our variability program we have obtained medium-deep
CMD's for the surveyed fields. In Fig. 9 we present a $V/B-V$ CMD based
on the pair of "template" images obtained with the SITe3 camera. The
photometry was extracted using the Daophot/Allstar package (Stetson
1987). The CMD shows several features of the EHB discussed in some
detail by Buonanno et al. (1986). In particular, we confirm the
presence of stars in the EHB gap between $V\approx 16.0$ and $V\approx
17.0$, and an apparent faint limit to the EHB stars at $V\approx
18.0$. We comment briefly on two features of this CMD that bear on the
origin and evolution of EHB stars. The first is that there are several
stars located slightly to the red of the EHB in Fig. 9. Although we
cannot exclude the possibility that some of them are field objects, we
find that these stars are strongly concentrated toward the cluster
center. These stars are candidate composite systems, consisting of an
EHB star plus a red dwarf. We are planning a more detailed discussion
of these systems in a forthcoming paper based on data obtained with the
2.5m du Pont telescope (Kaluzny et al. in preparation). The CMD in
Fig. 9 also shows two blue stars with $B-V\approx -0.25$ which form an
apparent faint extension of the EHB of the cluster. These stars are
marked with triangles in Fig. 9 and their coordinates as well as $UBV$
photometry are listed in Table 4. In Fig. 10 we present a $V/U-B$ CMD
based on the data collected with the SITe1 camera. The positions of the
two faint blue stars in Fig. 9 are marked. Faint blue stars located
below the EHB have been observed in other stellar clusters. Kaluzny \&
Rucinski (1995) noted the presence of several $UV$-bright stars with
$M_{V}>8$ in the center of NGC~6791 and more recently Cool et al.
(1998) identified similar objects in the core of NGC~6397. They argue
that these stars are either low-mass helium white dwarfs or
very-low-mass core-He-burning stars. We note that both faint blue
objects discovered in the field of NGC~6752 are good targets for
spectroscopic follow up with large ground-based telescopes.
\begin{figure}[h]
\vbox to11cm{\rule{0pt}{11cm}}
\special{psfile=thompson.fig10.ps angle=0 hoffset=0 voffset=-120
vscale=100 hscale=100}
\caption{A $V - (U-B)$ CMD for NGC~6752 based on the
images obtained with the SITe1 CCD camera.
The two candidate faint EHB stars are marked with triangles.}
\end{figure}
Photometry and equatorial coordinates of all stars plotted in Figs. 9
and 10 are available on request from the second author of this paper.
\section{Conclusions}
We have used time series CCD observations to identify eleven new variables
in the direction of the globular cluster NGC~6752. Three of these
variables are SX Phe stars which are likely to be cluster members.
Six out of the seven identified contact binaries are probably field
objects. One candidate detached eclipsing binary has been discovered and
follow-up observations are planned
to get complete light curves for this potentially important variable.
As a side-result we obtained $UBV$ photometry for a large sample
of stars from the cluster field, and we note the presence of two faint blue
objects located below the apparent cut-off of the EHB of the cluster.
\acknowledgements
JK, WP and WK were supported by the Polish Committee of Scientific
Research through grant 2P03D-011-12 and by NSF grant AST-9528096 to
Bohdan Paczy\'nski.
We are indebted to Dr. B. Paczy\'nski for
his long-term support to this project. Thanks are due to Randy Phelps
for taking some
data which were used in this paper. Dr. Dona Dinescu
kindly provided us with positional data for stars from
the NGC~6752 field.
\newpage
|
\section{Introduction}
During the past years, systems which exhibit self--organized
criticality (SOC) have attracted much attention, since they might
explain part of the abundance of fractal structures in nature
\cite{bak87}. Their common features are slow driving or energy input
and rare dissipation events which are instantaneous on the time scale
of driving. In the stationary state, the size distribution of
dissipation events obeys a power law, irrespective of initial
conditions and without the need to fine-tune parameters. Examples for
such systems are the sandpile model \cite{bak87}, the self-organized
critical forest fire model \cite{dro92,cla94}, the earthquake model by
Olami, Feder, and Christensen \cite{ola92}, and the Bak-Sneppen
evolution model \cite{sne93}. Numerical as well as analytical studies
of those systems are usually based on the assumption that their
critical behaviour can be described in similar terms as that of
equilibrium critical systems. This assumption is given a basis in
\cite{sor95}, where it is suggested that SOC systems can be mapped on
conventional critical systems by interchanging control and order
parameters. Thus, the Bak-Sneppen model can be mapped on a depinning
problem \cite{sne93}. However, it has been shown in \cite{cla97} that
the mapping suggested in \cite{sor95} for the SOC forest-fire model
does not generate a system with a conventional critical
point. Instead, the phase transition shows hysteresis effects and is
discontinuous when approached from above. Other unconventional
features have also been seen in the SOC forest-fire model, like the
existence of more than one diverging length scale \cite{cla95,hon96},
the absence of a spanning cluster immediately beyond the critical
point \cite{cla95}, and the dependence of the large-scale behaviour on
details of the model rules \cite{hon96,dro96}. (For a review on the
SOC forest-fire model, see \cite{cla96}.) Other SOC systems show also
unconventional scaling behaviour. Thus, in the two-dimensional abelian
sandpile model finite-size scaling is violated \cite{teb99}, and the
critical exponents for the earthquake model by Olami, Feder, and
Christensen \cite{ola92} appear to depend continuously on the
parameters. There is substantial need to better understand the nature of
the scaling behaviour of those systems.
It is the purpose of this paper to shed some light on the
unconventional critical behaviour of the SOC forest-fire model by
studying its finite-size effects. We choose a version of the model
which is identical to the SOC forest-fire model for system sizes much
larger than the correlation length, and we discuss the changes that
occur in the model as the system size is decreased below the
correlation length. We find that instead of displaying finite-size
scaling, small systems undergo a rearrangement from a structure with
patches of different density to a more homogeneous structure with
large density fluctuations in time. We find also that, contrary to
conventional critical systems, small systems and small parts of large
systems differ in the probability distribution for the density and in
the fire-size distribution. We suggest that these results can be
explained by the fact that the system has two qualitatively different
types of fires.
The outline of this paper is as follows: In section II, we define the
model that we used for studying finite-size effects, and discuss
briefly known results. Section III shows computer simulation results
for the fire-size distribution and the tree density as the system size
changes from values larger than the correlation length to values much
smaller than it. In the conclusion, we summarize and discuss our
findings.
\section{The Model}
\label{model}
The version of the SOC forest-fire model studied in this paper is
defined on a square lattice with $L^2$ sites. Each site is either
occupied (``tree'') or empty (``no tree''). At each time step, the
system is updated according to the following rules: (i) ``Burning'': A
site in the system is chosen at random (``struck by lightning''). If
the site is occupied, the whole cluster of occupied sites connected to
this site (by nearest-neighbour coupling) is removed from the system
(``burnt''), i.e., the occupied sites of that cluster turn to empty
sites. If the chosen site is empty, nothing happens. (ii) ``Tree
growth'': We select randomly $s_0\equiv pL^2$ sites from the system
and occupy those that are empty (possibly also including sites which
have become empty due to the removal of the cluster). These sites are
selected one after another, allowing for the same site being selected
more than once during the same filling step. In principle, $s_0$ can
therefore be larger than $L^2$, however, in our simulations we chose
usually values smaller than $L^2$.
For fixed $s_0$ and very large system size $L$, these rules are
equivalent to having a lightning probability $f=1/L^2$ per site and
time step, and a tree growth probability $p$, and the model is
identical to the original SOC forest-fire model \cite{dro92}. Because
of this equivalence, which was first pointed out by Grassberger
\cite{gra93} most numerical studies of the SOC forest-fire model up to
now were performed using the above rules, which allow for fast and
efficient computer simulations. With the above rules, finite-size
effects can also be studied very efficiently, as was suggested in
\cite{hon96}. However, one has to keep in mind that the results are
somewhat different from those for the original model. While in the
original model lightning can strike the system between the growth of
any two trees, it can strike the system in the present model only
after growth step (ii) is finished. This leads to density peaks in
Figures~\ref{bild9} and \ref{bild10} below that are not present in the
original model. However, our main conclusions are not affected by the
particular choice of the dynamical rules, as will be discussed further
below.
Let us first summarize shortly the major numerical results for the case
$s_0 \ll L^2$, as reported in the literature on the SOC forest-fire
model\cite{cla94,gra93,hen93,chr93}. In this limit, only a small
number of trees grow at each time step (compared to the total number
of trees). After a transient time, a stationary state is reached where
the tree density has only small fluctuations around some average value
$\bar \rho(s_0)$ that does not depend on $L$. Throughout this paper,
we study only stationary states and do not evaluate the initial
transient behaviour. Since the mean number of trees $\bar s$ burnt
during a fire must be identical to the mean number of trees growing
between two fires, we have the relation
\begin{equation}
\bar s = s_0(1-\bar \rho)/\bar \rho.
\end{equation}
The leading finite-size corrections to this equation are of order
$s_0/L^2$ and can be neglected in the case $s_0 \ll L^2$ which we are
considering in this paragraph. As $s_0$ increases, the mean fire size
increases also, and we approach the critical point of the SOC
forest-fire model, where the mean tree density is given by $\bar
\rho_c \simeq 0.41$. The correlation length $\xi$ is a measure for the
radius of the largest tree cluster and is related to $s_0$ via $\xi
\sim s_0^{\nu}$, with $\nu \approx$ 0.58 in $d=2$ dimensions. The
size distribution of tree clusters near the critical point is well
described by the scaling form
\begin{equation}
n(s) \simeq s^{-\tau}{\cal C}(s/s_{max})\, , \label{scaling}
\end{equation}
with a cutoff function ${\cal C}$ that is constant for small arguments
and decays exponentially fast when the argument is considerably larger
than 1. The cutoff cluster size $s_{max}$ is related to the
correlation length $\xi$ via $s_{max} \sim \xi^\mu$, with $\mu$ being
the fractal dimension of tree clusters, which is found to be 1.95 \cite{hen93}
or 1.96 \cite{cla94,sch99}. The value of the exponent $\tau$ is
approximately 2.14. The relation between $s_{max}$ and $s_0$ is
$s_{max} \sim s_0^\lambda$, with $\lambda = \nu\mu \simeq 1.15$
\cite{cla94}.
All these numerical findings agree well with conventional scaling
assumptions based on a single diverging length scale. Analytical
studies of the model, such as mean-field theories
\cite{chr93,dro94c,ves98} and renormalization group calculations
\cite{pat94,lor95} are also based on conventional scaling assumptions.
Therefore, the violation of finite-size scaling described in the
following might appear surprising to many readers. However, one must
keep in mind that the simulation data do not cover much more than one
decade in the correlation length $\xi$. The observed scaling behaviour
eq.~(\ref{scaling}), together with the measured values of the critical
exponents, do not necessarily indicate an exact asymptotic scaling
form, but may simply be a good approximation to more complicated
scaling, which works well for the system sizes and parameter values
studied in simulations. A similar phenomenon is known for the sandpile
model, where good scaling collapses for the avalanche size
distribution could be achieved in \cite{lub97,chessa} and older
papers, although it has been recently shown \cite{teb99} that
finite-size scaling is violated and that the simple scaling ansatz
used for the data collapse is incorrect.
Figure~\ref{bild1} shows a snapshot of a system with a tree density
$\bar \rho$ just below $\bar \rho_c$. One can distinguish regions of
different densities with a rather homogeneous tree distribution within
a region. These regions are obviously created by a fire that burns
down a cluster of high tree density. After the fire, a burnt region is
almost empty and becomes slowly filled with trees according to the law
$\dot \rho = p(1-\rho)$. We call these regions of homogeneous tree
density ``patches'', as we did in \cite{cla97a}. If lightning strikes
a tree in a patch of low density, it usually burns down a small tree
cluster. If it strikes a patch of a density larger than the
percolation threshold, it burns down a tree cluster as large as the
patch itself. This observation indicates that there are two
qualitatively different types of fires in the system: those that span
an entire patch, and those that destroy a small percolation cluster
within a patch of a tree density below the percolation threshold. As
we will see below, this gives rise to the unusual finite-size
properties of the model.
If the correlation length $\xi$ is of the same order as or larger than
$L$, the behaviour sketched above is modified due to finite-size
effects. For not too small values of $s_0/L^2$, the tree density
increases by a noticeable amount between two fires, leading to large
density fluctuations and to fires that span the entire system. If the
SOC forest-fire model showed conventional critical behaviour, there
would be a single diverging length scale, namely the correlation
length $\xi$, which would be related to $f/p$ or, equivalently, to
$s_0$, via $\xi \sim (f/p)^{-\nu}$ or $\xi \sim s_0^\nu$. Finite-size
effects would then manifest themselves in a scaling form
\begin{equation}
n(s) \simeq s^{-\tau}{\cal C}(s/L^\mu)\, , \label{FSS}
\end{equation}
for the size distribution of tree clusters. Furthermore, on scales
smaller than $L$ and $\xi$, all measured quantities should be
indistinguishable from those measured in a small section of an
infinitely large critical system.
The following section presents simulation results that show that none
of these finite-size scaling assumptions is satisfied for the SOC
forest-fire model. In fact, the invalidity of the assumption of a
single diverging length scale has already been shown in
\cite{hon96}. The invalidity of the second assumption that
measurements in small systems and in small sections of large systems
should give identical results, can be understood by considering for
instance the mean time interval between two fires. In a small
subsystem of linear size $l$ of a large system with a correlation
length $\xi \gg l$, this is given by $(p(1-\bar
\rho_c))^{-1}$. Just before fire reaches the subsystem, its tree density is
far above the percolation threshold, and the spanning cluster of the
subsystem is part of a large tree cluster that extends far beyond the
limits of the subsystem. Lightning usually strikes this large cluster
outside the subsystem, the time interval between two lightning strokes
within the subsystem being $L^2/ l^2$, which diverges as $L$
diverges. In contrast, fire cannot enter a small system from outside,
but the tree density of a small system increases until lightning
strikes a tree within the system. According to our rules, time is
measured in units of the mean time interval between two lightning
strokes. On this time scale, the time between two fires within a small
system of linear size $l$ is finite. In contrast, the time interval
between two fires within a subsystem of size $l$ of a much larger
system is vanishingly small compared to the time interval between two
lightning strokes within the subsystem. All these arguments are backed
up and complemented by the numerical results reported in the
following.
\section{Results of Computer Simulations}
In this section we will present and explain data obtained from about
300 runs of the model for various values of $s_0$ and $L$. Since
many runs of the simulation were necessary, we chose a
cluster of workstations rather than a "supercomputer". The system size
$L$ varied between 10 and 2000 in these runs. We found that as
finite-size effects become more important, the system shows a
transition between two qualitatively different types of behaviour
which we call critical behaviour and percolation-like behaviour. The
critical behaviour is is characterized by a good scaling collapse of
the fire size distribution and by large spatial variations in the
local tree density. The percolation-like behaviour is characterized by
large temporal fluctuations in the global tree density, with a rather
homogeneous tree distribution within the system for any given
time. Snapshots of the system therefore resemble percolation systems
where each site is occupied by a tree with a probability $\rho$. (For an introduction to percolation theory, see e.g. \cite{sta92}.) The
following three subsections show how this transition manifests itself
in the mean tree density, the fire size distribution, and the
probability distribution for the tree density.
\subsection{Lines of constant tree density}
First, we measured the mean tree density
\begin{equation}
\bar \rho = (1/T) \sum_{t=1}^T \rho(t)\label{rho}
\end{equation}
in the system, averaged
over a large number of $T$ iterations, for various values of $L$
and $s_0$. The density $\rho(t)$ was always evaluated after the
refilling step (ii). Compared to a model where trees grow at a
rate $\dot \rho = p(1-\rho)$, the density values in our
model are somewhat larger when $s_0/L^2$ is not very small. If, for
instance, the density is increased from $\rho - \Delta \rho$ to $\rho$
during the refilling step (ii), the value $\rho$ enters the above
sum, while an evaluation based on a constant growth rate would give
$1-\Delta \rho/\ln(1+\Delta\rho/(1-\rho))$ instead of $\rho$.
It it not obvious what the relation between $s_0$, $L^2$ and $\bar
\rho$ should be if we want to deduce it from an analogy with
equilibrium critical systems. We have already mentioned that the
temporal fluctuations in $\rho(t)$ become larger as the ratio
$s_0/L^2$ increases. Similarly, temporal fluctuations increase in a
critical equilibrium system when the system size becomes smaller. One
might therefore expect that decreasing $L$ at fixed $s_0$ should drive
the system toward the critical point, where $\bar \rho = \bar \rho_c
\simeq 0.41$. However, we have argued in the previous section that a
given site burns down more often in a large system without finite-size
effects than in a smaller system with finite-size effects that has the
same value of $s_0$. From this, it follows that the mean tree density
increases with decreasing $L$, when $s_0$ is fixed. In the limit $s_0
\gg L^2$ it must go to one. From this point of view, a system with
sufficiently large finite-size effects should rather be compared to an
equilibrium system in the ordered phase, for instance to a percolation
system beyond the percolation threshold. The analogue of $s_0$ in a
percolation system is then the mean size of the cluster that a given
site belongs to, and it is proportional to $L^2$ beyond the
percolation threshold \cite{sta92}. Indeed, we find that the mean size
of fires becomes proportional to $L^2$ when finite-size effects are
strong (see below). However, there is nevertheless a fundamental
difference between a percolation system beyond the percolation
threshold and our system with a density above $\rho_c$: In our model,
a tree cluster that spans the system and has a size proportional to
$L^2$ occurs only rarely when the mean density is only slightly above
the critical density, while a percolation system above the percolation
threshold has always a system spanning tree cluster.
In Figure~\ref{bild2}, lines of constant mean tree density are plotted
on a double logarithmic scale in $s_0$ and $L^2$. This figure shows
that there are two qualitatively different regions in the $s_0$ vs. $L^2$
plane,
with a transition region between them. First, there is the region
where there are no finite-size effects, $s_0 \ll L^2$. In this region,
the size of the tree clusters is much smaller than the system size,
and the global density fluctuations are small. The mean tree density
of such a system is smaller than $\bar \rho_c \approx 0.41$. For
systems with small density fluctuations the probability that a given
empty site is filled with a tree during one time step, is given by
$s_0/L^2$, and the probability that a given tree is burnt by a fire is
$s_0 (1-\bar \rho)/(\bar \rho L^2)$. Since both probabilities decrease
as $1/L^2$ with increasing $L$, dynamics of larger systems are slower
than those of smaller systems. Apart from this change of the
characteristic time scales, the local dynamics is independent of $L$,
and consequently correlation functions and cluster size distributions
are the same for systems of different sizes (provided that $L^2 \gg
s_0$). This leads to the horizontal slope in the large $L$ regime of
the curves for $\bar \rho<\bar \rho_c$ in Figure~\ref{bild2}.
The curves show deviations from the horizontal behaviour when $L$
becomes as small as or smaller than the correlation length $\xi \sim
s_0^\nu$. These deviation occur in the transition region where
finite-size effects begin to become noticeable. The dynamics changes
from fires that are not affected by the finite system size to a fire
size distribution that includes sometimes events of the order of the
system size. Fires of such a large size destroy the patchy structure
of the forest described earlier, and cause a more random tree
distribution.
In the second region, lines of constant mean density are curves of
constant $s_0/L^2$. This feature can best be understood when
considering the parameter range where $s_0$ is of the order of $ L^2$
or above. In this range, the mean tree density is much larger than
$\bar \rho_c$, and the system contains a spanning cluster after each
filling. During the "burning" step, this cluster is removed with a
finite probability, leading to a large change in density in the
system. When a "finite" (i.e. not spanning) cluster is removed during
the "burning" step, the overall density hardly changes for large
$L$. The time series of the density is therefore determined almost
completely by the filling events and by the large burning
events. Since the filling events fill a large fraction of empty sites,
and since large burning events burn a large fraction of trees, the
tree distribution within the system is rather homogeneous. A snapshot
of the system at a given time looks therefore similar to a percolation
system with a density $\rho(t)$. From percolation theory we know that
for a given density the fraction of trees sitting in the spanning
cluster is independent of $L$, and consequently curves of constant
large density are curves of constant $s_0/L^2$ in Figure~\ref{bild2}.
Even curves for smaller $s_0/L^2$, which correspond to densities only
slightly above $\bar \rho_c$ show for sufficiently large $L$ an
asymptotic behaviour $s_0/L^2 =$ const, with the constant vanishing at
$\bar \rho_c$. The reason is again that finite fires do not reduce the
density of an infinitely large system, and that the system spanning
fires reduce the density by an amount that does not depend on $L$, but
only on the density itself.
Finally, the critical curve (for the density $\bar \rho= \bar \rho_c
\simeq 0.41$) is obtained from the condition
\begin{displaymath}
L \approx \xi \sim s_0^{\nu}, \; \; \mbox{with} \; \; \nu\simeq 0.58 \;.
\end{displaymath}
This curve is the separatrix between the two regions
described above and is indicated in Figure~\ref{bild2} by the bold
line.
As $L$ decreases, more and more curves merge
with the separatrix when their correlation length
becomes comparable to the system size.
\subsection{The fire size distribution}
Since lightning strikes each tree with the same probability, the size
distribution of fires is proportional to $sn(s)$, with $n(s)$ being
the size distribution of tree clusters. As mentioned above,
conventional scaling would imply a form $s n(s) \simeq s^{1-\tau}{\cal
C}(s/s_0^{\mu\nu})$ for the fire size distribution if the correlation
length $\xi \sim s_0^\nu$ is smaller than the system size, and a
finite-size scaling form $s n(s) \simeq s^{1-\tau}{\cal C}(s/L^\mu)$
in the opposite case. In both cases, one would obtain a scaling collapse of
the curves for different $s_0$ or $L$.
Figure ~\ref{bild4} shows the fire size distribution for parameters
such that $\xi < L$. While not perfect, the data collapse is good and
would not impose the conclusion that simple scaling is violated. The
bump near the end of the curves indicates that the cutoff function
${\cal C}$ increases first with increasing argument, before it shows
the exponential decay. This bump is believed to contain all the trees
that would sit in larger clusters if the system was exactly at the
critical point \cite{gra93}.
Figure ~\ref{bild5} shows the fire size distribution for parameter
values such that the mean density is ten percent above its critical
value. As discussed in the previous subsection, system spanning fires
occur, and their size scales as $L^2$. These fires are responsible
for the peaks in the fire size distribution. Similar peaks occur in
equilibrium critical systems in the ordered phase, for example in the
cluster size distribution of a percolation system above the
percolation threshold. In a percolation system, the occurrence of such
peaks implies that the system size is larger than the correlation
length, which is identical to the cutoff in the radius of the finite
(i.e., non system spanning) clusters. In our system, however, we do
not see such an exponential cutoff to the size distribution of the
finite clusters. Instead, the curve for $L=800$ in figure ~\ref{bild5}
appears to obey a power law from $s \simeq 100$ up to the point where
the peak begins. The explanation for this unusual behaviour must lie
in the large temporal fluctuations in the density. The density is only
sometimes so large that the large fires, which have a size of the
order $L^2/2$, occur. At other times, the density values are different
and allow for a broad range of other fire sizes.
The transition between critical scaling and $L^2$-scaling can be
observed when $L$ is varied for fixed $s_0$, as illustrated in
Figure~\ref{bild6}. One can see that the shape of the curves changes
continuously as $L$ is decreased. Clearly, because of this change in
shape, finite-size effects do not manifest themselves in a scaling
behaviour $s n(s) \simeq s^{1-\tau}{\cal C}(s/L^\mu)$. It is
impossible to generate a scaling collapse of different curves, even if
their density is close to the critical density. Furthermore, as
mentioned above, the cutoff introduced by the finite system size
always scales as $L^2$, due to the occurrence of system spanning
fires, and not as $L^\mu$, as expected for conventional critical
systems.
For small
system sizes, spanning clusters may already occur for densities below
$\bar \rho_c$, an effect which is clearly visible in the curve for
$L=63$. The formation of peaks due to finite
size effects, was also found in~\cite{mala98}.
As mentioned earlier, for conventional critical systems a
system of small size and a small section of a large system are
equivalent. We have argued that this is not true for the forest-fire
model since the mean tree density and the time interval between fires
are different in the two cases. The next two figures show that also
the fire size distributions are different. In Figure~\ref{bild6b} the
fire size distribution of a section of a large system and a
corresponding small system is shown. The scaling parts and the form of
the bumps near the cutoff are very different. The fire size
distribution of a small section of a large system is broader than that
of a small system. The reason is that a section of a large system can
contain a boundary between a patch of large tree density and a patch
of small tree density. This boundary can pass through the section in
different ways, and the number of trees in the dense part can take
different values. Since large fires only burn the dense part, the
size distribution of fires becomes broad.
Figure~\ref{bild6a} shows how the fire size distribution changes when
smaller and smaller sections of a large system are evaluated.
Comparing to Figure~\ref{bild6}, one sees again that the fire size
distribution of small sections of large systems is different from that
of small systems.
To summarize this subsection, the fire size distribution in the
presence of finite-size effects does not show the features of
finite-size effects in conventional critical systems. We find a
continuous change in the shape of the fire size distribution and
cutoffs that scale as $L^2$, rather than conventional finite-size
scaling. Furthermore, the fire size distribution in small sections of
large systems is different from small systems. Our results for the
fire-size distribution confirm the qualitative transition from a
parameter region unaffected by finite-size effects to a region
dominated by system spanning fires that we found in the previous
subsection.
\subsection{Probability distribution of the density}
Finally, we studied the temporal fluctuations in the values of tree
density $\rho(t)$ by measuring how often a given value of $\rho$
occurs within a sufficiently long time series. We denote by
$w(\rho)d\rho$ the probability that the tree density lies in the
interval between $\rho$ and $\rho+d\rho$. The quantity $w(\rho)$ is
therefore the probability density for the tree density $\rho$. We
measured $\rho$ always after the trees were refilled, i.e., after step
(ii). The results show again a qualitative change as the correlation
length becomes smaller than the system size, reflecting the transition
from critical to percolation-like behaviour.
In Figures~\ref{bild7} to~\ref{bild10} $w(\rho)$ is shown for
different values for $L$ and $s_0$. For large enough and fixed
$s_0/L^2$, the mean tree density increases with increasing system
size, until it reaches its asymptotic value above $\bar\rho_c$. For
fixed $L$, the mean tree density increases with increasing $s_0$.
Apart from this increase in mean tree density, the following other
trends are observed: (a) As the mean density approaches $\bar \rho_c$,
the curves for $w(\rho)$ become broader (Figure~\ref{bild7}). This is
because the patches of homogeneous density visible in
Figure~\ref{bild1} become larger with increasing $\bar \rho$, leading
to larger global density fluctuations. (b) As the mean tree density
increases above $\bar \rho_c$, the shape of the distribution becomes
asymmetric, with the maximum moving from $\bar \rho_c$ to the
percolation threshold $\rho_{perc} \simeq 0.59$
(Figure~\ref{bild8}). The reason is that for $\bar \rho > \bar \rho_c$
the patchy structure is replaced by a more homogeneous (percolation
like) structure, where the largest fires are system spanning and occur
for densities above the percolation threshold. Once the density lies
above the percolation threshold the probability that a system spanning
fire occurs is very high. This is why densities much higher than the
percolation occur seldom, explaining the rapid decrease of $w(\rho)$
above the percolation threshold. (c) As the system size increases for
fixed $p=s_0/L^2$, there occur peaks in the density distribution which
become sharper and more numerous for larger $L$ (Figures~\ref{bild9}
and \ref{bild10}). This can be explained by realizing that the
difference between finite and system spanning fires becomes more
pronounced as $L$ increases. In the limit $L\to \infty$, finite fires
do not affect the density at all, while system spanning fires reduce
it to a small value. Subsequent filling events then increase the
density to $1-\exp(-p)$, $1-\exp(-2p)$, $1-\exp(-3p)$, etc., until the
density is above the percolation threshold and another system spanning
fire can occur. These system spanning fires do not always occur at the
first instance where the density is above $\rho_{perc}$, since
lightning might strike and empty a site. Also, the density immediately
after a system spanning fire depends slightly on the density before
the fire. Therefore, the series of density values given above, becomes
slightly shifted, depending on the density just before the last system
spanning fire. These shifted series of peaks, in turn, give rise to
further possible density values above $\rho_{perc}$, leading to an
additional series of peaks, etc. This is the mechanism leading to the
fractal peak structure that emerges as $L$ is increased. For smaller
$L$, the effect of small fires leads to a larger width of the peaks,
which can therefore not be resolved when they are close together.
As mentioned in the introduction, the peaks in $w(\rho)$ are due to
the fact that lightning can strike the system only between two filling
steps. Had we instead performed our simulations using a small tree
growth probability $p$ and a small lightning probability $f$,
lightning could strike the system between the growth of any two
trees. However, such a simulation would be very slow. In oder to make
sure that our choice of the algorithm has no other effect on the
results apart from the peaks in $w(\rho)$, we performed a test
simulation where $s_0$ is not the same for each filling step. For each
filling step, we chose $s_0$ randomly from an exponential distribution
$P(s_0)=(L^2p)^{-1}\exp(-s_0/(L^2p))$. Such an exponential distribution
results when lightning can strike the system between the growth of any
two trees with the same probability. The mean number of trees growing
between two lightning strokes is now smaller than before. The reason
is that the majority of filling steps increase the tree number by a
value smaller than $L^2p$, and that during large filling events the
tree density becomes high and most of the sites chosen for filling are
already occupied. The mean tree densities evaluated according to
Eq.~(\ref{rho}) are consequently slightly smaller than before.
The probability density $w(\rho)$ for the tree
density resulting from this modified algorithm is shown in the insets
in Figures ~\ref{bild8} and~\ref{bild9}. As expected, the peaks have
vanished, while the change in shape from a curve with peak around 0.4
to a curve with peak near 0.6 due to finite-size effects is the same
as before.
\section{Conclusion}
In this paper, we have studied finite-size effects in the SOC
forest-fire model. As these effects become stronger, the system
rearranges from a structure with patches of different densities to a
more homogeneous structure with large density fluctuations in time.
This rearrangement is reflected in the structure of the fire size
distribution, in the mean tree density, and in the temporal density
fluctuations. Qualitatively similar (although quantitatively
different) rearrangements are observed when smaller and smaller
sections of a large SOC system are studied. Due to these qualitative
changes, conventional finite-size scaling does not hold. Our work thus
demonstrates that concepts from equilibrium critical phenomena cannot
be taken over to the study of SOC systems such as the forest-fire
model. Instead, these nonequilibrium critical systems show generically
new features unknown in equilibrium. As the scaling ansatz
eq.~(\ref{scaling}) which is based on a single length scale $\xi \sim
s_0^\nu$ can only be approximately correct, the true asymptotic
scaling behaviour of the model is still an open question.
We suggest that the reason for the unconventional behaviour of the SOC
forest-fire model is the fact that two qualitatively different types
of fires occur: those that burn down a patch of high tree density of
fractal dimension 2, and those that burn down a tree cluster of a
smaller fractal dimension within a region of a tree density below the
percolation threshold. As a consequence, the scaling behaviour of the
system cannot be characterized using only one length scale. While the
superposition of the two types of fires creates the impression of
simple scaling as long as finite-size effects are small, the
difference between them becomes clearly visible for smaller system
sizes, where system spanning fires receive a larger weight. We suggest
that the superposition of the two types of fires is also responsible
for the other unconventional features of the SOC forest-fire model
listed in the introduction.
Models related to the present one have been studied in
\cite{cla95,cla97a} and \cite{cla97}. In these models, the tree
density is globally conserved by filling exactly the same number of
trees into the system that have been burnt. As long as the density is
below the critical value, these models are equivalent and show the
critical behaviour of the SOC forest-fire model as the critical
density is approached from below. They were introduced for the
purpose of studying the SOC forest-fire model beyond the critical
point, i.e. for densities larger than the critical density $\bar
\rho_c$. As the density increases beyond the critical density, both
models undergo large-scale rearrangements. In \cite{cla95,cla97a},
where trees are refilled only after the end of a fire, the new
structure consists of a finite number of large domains of different
density. In \cite{cla97}, where each tree is refilled into the system
immediately after it is burnt, the new structure has a continuously
burning fire, and resembles the forest-fire model without lightning
introduced earlier by Bak, Chen, and Tang \cite{bak90}, which shows
spiral-shaped fire fronts \cite{gra91}. In both these models, the
dynamics in the restructured state are dominated by large fires
burning forests of a fractal dimension two, similarly to the
restructuring due to finite-size effects reported in this paper.
Let us conclude by noting that it is unclear whether the behaviour in
higher dimensions resembles that in two dimensions. Clearly, as long
as the ``patchy'' structure with two qualitatively different types of
fire occurs, mean-field theory which neglects all spatial structure
\cite{chr93,dro94c,ves98} cannot apply, and the system must be below
its upper critical dimension. The recent paper by Br\"oker and
Grassberger \cite{bro97} on the forest-fire model without lightning
indicates that unusual scaling behaviour can occur also in 3 and 4
dimensional forest-fire models. If 6 is the upper critical dimension
of the forest-fire model, as suggested in \cite{cla94,chr93,dro94c},
then the scaling behaviour of the forest-fire model should be
conventional above 6 dimensions.
\acknowledgements
This work was supported by EPSRC Grant
GR/K79307, and by the EU network project (TMR) ``fractal
structures and self organization'', EU-contract ERPFMRXCT980183.
|
\section{Introduction}
Perovskite manganites ($R_{1-x}, A_x$)$_{n+1}$Mn$_n$O$_{3n+1}$
(R=La, Pr, Nd, Sm ; A= Ca, Sr, Ba ; $n=1, 2, \infty$) have recently
attracted renewed interests from the viewpoint of the close connection
between magnetism and transport.
\cite{chaha,helmolt,tokura95,jin94,ram,moritomo95-96}
The stability of the perovskite structure enables the preparation
of high quality single crystals with systematically changing the
carrier density and the bandwidth by controlling $R$ and $A$ atoms.
In addition, the crystal structure changes from two-dimensional
($n=1$, single layer) to three-dimensional ($n=\infty$, cubic) as
$n$ increases.
The dimensionality of the electronic structure is controlled also by
the spin and the orbital orderings, for example, the transfer along $c$-axis
is forbidden by the anti-parallel spin configuration and/or the
$d_{x^2-y^2}$ orbital ordering.\cite{maezono981,maezono982}
Therefore even in the isotropic ($n=\infty$, cubic, 113-system)
crystal structure, the electronic dispersion can be quasi-two-dimensional
(layered antiferromagnetic (spin $A$) in Pr$_{1-x}$Sr$_{x}$MnO$_{3}$,
\cite{kawano97,tomioka}
Nd$_{1-x}$Sr$_{x}$MnO$_{3}$,\cite{kuwahara98,kajimoto99,kuwahara99}
La$_{1-x}$Sr$_{x}$MnO$_{3}$\cite{moritomo98-113})
or quasi-one-dimensional (rod type $AF$ (spin $C$) in
Nd$_{1-x}$Sr$_{x}$MnO$_{3}$ \cite{kuwahara98,kajimoto99}).
Another important related issue is the spin canting.
As discussed by de Gennes\cite{degenne}, $A$-type antiferromagnet
near the parent
insulator is unstable toward the spin canting when holes are doped.
This is because the kinetic energy gain along $c$-axis wins the energy
cost of the $AF$ exchange interaction for small canting angle.
In the metallic $A$-type $AF$ phase mentioned above, however, no spin canting
has been observed. \cite{kawano97}
\par
Double-layered manganites with $n=2$ (327-system) offer an interesting
opportunity
to study the interplay of the spin, the orbital and the crystal structure
in the above-mentioned issues of dimensionality and spin canting.
With increasing $x$, the spin ordering within one double-layer changes
from the ferromagnetic one (spin $F$, $x=0.3$)
to the spin $A$ ($x=0.5$).
\cite{moritomo95-96,battle96,kimura96-97,mitchell97,argyriou971,argyriou972,perring97,hirota98,moritomo97,moritomo98,kimura98,kubota98}
For $0.4<x<0.48$ the diffraction peak observed in the neutron-scattering
experiments indicates the coexistance of the spin $F$- and
$A$-components at low temperatures, which has been interpreted as
the spin canting.
\cite{argyriou971,argyriou972,perring97,hirota98,kimura98,kubota98}
However it could be interpreted also in terms of the phase separation,
and further theoretical studies are needed.
Lattice deformation depending upon $x$ is also observed,
\cite{mitchell97,hirota98,moritomo97,moritomo98,kimura98,kubota98}
implying active contribution to the spin transition, in comparison
with 113-system.
In 113-system, there is the one to one correspondence between the crystal
deformation and the orbital ordering, e.g., the $c$-axis contracts in the
metallic spin $A$ state \cite{kuwahara99} where the $d_{x^2-y^2}$ ordering
occurs. \cite{maezono982}
In 327-system, on the other hand, MnO$_6$ octahedra elongates along $c$-axis
during the spin structural change from spin $F$ to cant, and to spin $A$.
\cite{moritomo98}
This elongation decreases with increasing $x$ upto $x=0.5$ continuously,
but even in the spin $A$ phase a little elongation remains,\cite{moritomo98}
though this deformation prefers the $d_{3z^2-r^2}$.
\par
In this paper, we study the spin and the orbital ordering in the
double-layered compounds.
Possibility of the spin canting is studied for general
$x$, which includes the de Gennes's canting mechanism as
a special case.
In the metallic region, the spin $A$ is locally stable against the
spin canting.
Spin transition from the spin $A$ is therefore discontinous one
to the spin canting or to the spin $F$, depending on the ratio of
the transfer integral and the superexchange interaction between
$t_{2g}$ spins.
For the spin canting to occur in the metalic region, the small transfer
integral along $c$-axis, i.e., the planer orbital
ordering, turned out to be indispensable.
The canting angle is sensitive to the
bond-length along $c$-axis.
With this mechanism, calculated mean-field phase diagram
together with the observed lattice distortion
\cite{mitchell97,hirota98,moritomo97,moritomo98,kimura98}
can qualitatively explain the observed $x$-\cite{moritomo95-96,battle96,kimura96-97,mitchell97,argyriou971,argyriou972,perring97,hirota98,moritomo97,moritomo98,kimura98,kubota98} and $T$-dependence \cite{hirota98}
of the spin ordering
and also the spin-anisotropy \cite{argyriou972,hirota98,kubota98,perring98}
in 327-system.
\par
The plan of this paper follows.
In Sec. II, the model and the formulation are given.
Results and discussion are presented in Sec. III.
Notations used here are standard as in ref. 8.
\section{Model and formulation}
We start with the Hamiltonian
\begin{eqnarray}
H&=&
\sum\limits_{\sigma \gamma \gamma' \langle ij \rangle}
{t_{ij}^{\gamma \gamma '}d_{i\sigma \gamma }^{\dagger}d_{j\sigma \gamma '}}
\nonumber \\
&-&
J_H\sum\limits_i {\vec S_{t_{2g} i}\!\cdot\! \vec
S_{e_g i}}
\nonumber \\
&+& J_S\sum\limits_{\left\langle {ij} \right\rangle } {\vec S_{t_{2g} i}
\!\cdot\! \vec S_{t_{2g} j}} +H_{\rm on\ site}
\nonumber \\
&+&H_{\rm el-ph} \ ,
\label{eqn:eq1}
\end{eqnarray}
where $\gamma$ [$=a(d_{x^2-y^2}), b(d_{3z^2-r^2})$] specifies the
orbital and the other notations are standard.
The transfer integral $t_{ij}^{\gamma \gamma'}$ depends on the pair
of orbitals $(\gamma, \gamma')$ and the direction of the
bond $(i, j)$ \cite{ishihara97},
$J_H$ is the Hund coupling between $e_g$ and
$t_{2g}$ spins, and $J_S$ is
the $AF$ coupling between nearest neighboring $t_{2g}$ spins.
$H_{\rm on\ site}$ represents the on-site
Coulomb interactions between $e_g$ electrons, and is given by
\cite{maezono982,ishihara97}
\begin{equation}
H_{\rm on\ site}
= -\sum\limits_i {\left( {\tilde \beta \vec T_i^2+\tilde
\alpha \vec S_{e_{g} i}^2} \right)},
\label{eqn : eq2}
\end{equation}
where the spin operator for the $e_g$ electron is defined as
$\vec S_{e_g i}={1 \over 2}\sum\limits_{\gamma \alpha \beta}
{d_{i\gamma \alpha }^{\dagger}\vec \sigma _{\alpha \beta }
d_{i\gamma \beta }}$ with the Pauli matrices $\vec \sigma$,
while the orbital isospin operator is defined as
$
\vec T_i={1 \over 2}\sum\limits_{\gamma \gamma' \sigma} {d_{i\gamma \sigma }
^\dagger\vec \sigma _{\gamma \gamma '}d_{i\gamma '\sigma }} \ .
$
Coefficients of the spin and isospin operators, i.e.,
$\tilde \alpha $ and $\tilde \beta $, are given by
\cite{maezono982,ishihara97}
$
\tilde \alpha = U-{J \over 2}>0\ ,
$
and
$
\tilde \beta = U-{3J \over 2}>0 \ .
$
The minus sign in Eq. (\ref{eqn : eq2}) means
that the Coulomb interactions induce both spin and orbital
(isospin) moments.
The parameters $\tilde \alpha ,\tilde \beta ,
t_0$, used in the numerical calculation are chosen as
$t_0 = 0.72$ eV, $U=6.3$ eV, and $J=1.0$ eV, being
relevant to the actual manganese oxides. \cite{maezono981,maezono982}
The electron-lattice interaction is \cite{maezono982}
\begin{eqnarray}
H_{\rm el-ph}
= +\left|g\right|r\sum_{i}{\vec v_{i}\cdot\vec T_{i}}\ ,
\label{eqn : eq3.2.16}
\end{eqnarray}
where $g$ is the coupling constant and $r$ ($\vec v_{i}$) is
the magnitude (direction) of the lattice distortion of the MnO$_6$-octahedra.
Values of $r$ and $\vec v$ are taken from the observed elongation
along $c$-axis in (La$_{1-x}$, Sr$_x$)$_3$Mn$_2$O$_7$ ($0.3<x<0.4$)
\cite{moritomo98} leading to $r\sim0.01$, $\vec v /\!/ \hat z$.
We calculate the ground state energy by the meanfield approximation
with the two order parameters,\cite{maezono982}
\begin{eqnarray}
\vec \varphi_S
& = & \left\langle \vec S_{e_g} \right\rangle + {{J_H } \over
{2 \tilde \alpha }} \left\langle \vec S_{t_{2g}} \right\rangle \ ,\\
\vec \varphi_T
& = & \left\langle \vec T \right\rangle \ .
\label{eqn : eq3.2.16}
\end{eqnarray}
\par
As the exchange interaction between two double-layers
is reported to be less than 1/100 compared with the intra double-layer one,
\cite{kimura-p}
we consider an isolated double-layer, for which the
Brillouin zone contains only two $\vec k$-points along $c$-axis.
We consider four kinds of the spin alignment in the cubic cell: spin $F$,
$A$, $C$ and $G$ (NaCl-type).
For spin $A$, we also consider the possibility of the canting characterized
by an angle $\eta$ which is $0\ (\pi)$ for spin $F$ ($A$).
As for the orbital degrees of freedom, we consider two sublattices $I$,
and $I\!I$, on each of which the orbital is specified by the
angle $\theta_{I,I\!I}$ as \cite{maezono982}
\begin{eqnarray}
\left| {\theta _{I,I\!I}} \right\rangle =\cos {{\theta _{I,I\!I}}
\over 2}\left| {d_{x^2-y^2}} \right\rangle +\sin {{\theta _{I,I\!I}}
\over 2}\left| {d_{3z^2-r^2}} \right\rangle.
\label{eqn : eqN.3}
\end{eqnarray}
We also consider four types of orbital-sublattice ordering, i.e.,
$F$-, $A$-, $C$-, $G$-type in the cubic cell.
Henceforth, we often use a notation such as spin A, orbital $G$
($\theta_I,\theta_{I\!I}$) etc..
Denoting the wave vector of the spin (orbital) ordering as
$\vec q_{S}$ ($\vec q_{T}$), the ground state energy is
given as a function of the spin ordering ($\eta$, $\vec q_{S}$),
the orbital ordering ($\theta_{I,I\!I}$, $\vec q_{T}$), and
the lattice distortion ($g$, $r$, $\vec v$).
\section{Results and discussions}
\subsection{Phase diagram without canting and lattice distortion}
\begin{figure}[p]
\caption{}
\label{fig : F1}
\end{figure}
\noindent
Figure. \ref{fig : F1} shows the phase diagram at zero temperature
in the plane of the doping concentration ($x$) and the $AF$ interaction
between $t_{2g}$ spins ($J_S$), without the electron-lattice coupling
($g=0$).
Orbital shape is represented by $p$- (planer, orbital $F$ (0,0), i.e.,
$d_{x^2-y^2}$) and $n$- (non-planer, orbital G (100,-100) around $x=0.1$
and orbital A (-40,-40) around $x=0.7$) hereafter.
The possibility of the spin canting is not taken into account, i.e.,
$\eta=0$ or $\pi$.
For $J_S=0$, spin $F$ is the most stable for almost the whole region of $x$,
except around $x=0$.
This ferromagnetism comes from the superexchange interaction between
$e_g$ spins for small $x\ (x\sim0.1)$, and from the double-exchange
interaction for larger $x$ ($x\sim0.7$). \cite{maezono982,okamoto98}
Nonmonotonic behavior with double peaks ($x=0.1,\ 0.7$) of the phase
boundary between spin $F$ and $A$ ($J_S(F\!A)$) is then attributed
to the cross-over from the super- to the double-exchange interaction,
similarly to the case of 113-system. \cite{maezono982}
We identify the spin transition from spin $F$ to spin $A$
with increasing $x$ observed experimentally for $0.3<x_{exp.}<0.5$
\cite{moritomo95-96,battle96,kimura96-97,mitchell97,argyriou971,argyriou972,perring97,hirota98,moritomo97,moritomo98,kimura98,kubota98}
with that around $x\sim0.15$ when $J_S\lesssim0.006$ eV
(for example, arrow (a) or (b) in Fig. \ref{fig : F1}).
This value of $J_S$ is roughly of the same order in magnitude as the 113-case.
\cite{maezono982,maezono981}
Depicted orbital alignment is the one optimized at each point of the
phase diagram.
For most of the phase diagram, $d_{x^2-y^2}$ is the most stable one
reflecting the two-dimensional nature of the double-layered structure.
The spin $C$ phase completely disappears.
One might regard this self-evident because of the two-dimensionality.
This, however, is not so trivial because the valency along $c$-axis
can still lower the energy by the bonding/anti-bonding splitting
for the double-layered structure.
\par
\subsection{Spin canting}
We now study the spin canting between spin $F$ and $A$.
First we consider the possibility of the spin canting with fixed
$J_S=0.004$ eV, shown as the arrow (a) in Fig. \ref{fig : F1}.
\begin{figure}[p]
\caption{}
\label{fig : F2}
\end{figure}
\noindent
The $x$-dependence of the optimized canting angle $\eta$ is
shown in Fig. \ref{fig : F2}.
The electron-lattice coupling is not considered ($g=0$).
The canting angle changes continuously from $\pi$ (spin $A$, $x=0$)
to zero (spin $F$, $x\sim0.1$), corresponding to the spin canting
around the parent insulator.
In this region, the orbital alignment almost remains
that for $x=0$ (orbital $G$ (100,-100)), implying that the ordering
is still dominated by the superexchange interaction in this region.
This is consistent with the interpretation that the peak of the phase boundary
$J_S(F\!A)$ around $x=0.1$ is attributed to the superexchange interaction.
Coexisting double-exchange interaction due to the introduced carriers
competes with $J_S$, leading to the spin canting seen for $0<x<0.1$
in Fig. \ref{fig : F2} through the conventional mechanism
a la de Gennes \cite{degenne} for 113-system.
Battle $et\ al$. observed the coexistance of the spin $F$ and $A$
by the powder diffraction of the poly-crystal sample
La$_{2}$SrMn$_2$O$_7$ ($x=0$). \cite{battle97}
This can be explained by small amount of carriers introduced by the
oxygen deficiencies which brings about the spin canting by
this mechanism for small $x$ region, as reproduced in Fig.
\ref{fig : F2}.
When $x$ goes beyond $x\sim0.125$, $\eta$ discontinuously changes from zero
to $\pi$, corresponding to the first-order transition from spin $F$ to
spin $A$.
This behavior without canting is due to the discontinous orbital transition.
With increasing the double-exchange interaction for $x>0.1$,
the orbital discontinuously changes from orbital $G$ (100,-100) to
orbital $F$ (0,0), i.e., $d_{x^2-y^2}$, in order to maximize the kinetic
energy gain.
With fixed $J_S$ as in Fig. \ref{fig : F2}, the spin transition between
spin $F$ and $A$
is accompanied with this orbital transition ($n$-spin $F$ to $p$-spin $A$).
Therefore, the spin canting observed for $0.4<x_{exp.}<0.5$ in the
single-crystal
samples, \cite{argyriou971,argyriou972,perring97,hirota98,kimura98,kubota98}
therefore, cannot be reproduced in Fig. \ref{fig : F2}.
\par
In order to explain this experimental observation, we look for the possible
mechanisms for the continous change of spin/orbital structure starting
from the $p$-spin A.
One possibility is to change the orbital wave-function via the coupling
to the lattice.
In 327-system
\cite{mitchell97,hirota98,moritomo97,moritomo98,kimura98,kubota98}
the lattice constant $c$ is longer than $a$, $b$, which prefers
$d_{3z^2-r^2}$ compared with $d_{x^2-y^2}$, and $c$ changes as $x$.
This corresponds to the ``magnetic field'' to the orbital pseudo-spin,
and tends to enhance the transfer along $z$-direction and hence the spin
canting.
\begin{figure}[p]
\caption{}
\label{fig : F4}
\end{figure}
\noindent
In Fig. \ref{fig : F4} shown the phase diagram in the plane of $x$ and
$gr$ ($g$ : coupling constant, $r$ : elongation along $c$-axis).
This $gr$ corresponds to the ``orbital magnetic field'' to prefer
$d_{3z^2-r^2}$.
Here $J_S$ is fixed to be 0.004 eV which is comparable to the value
appropriate for 113-system. \cite{maezono981,maezono982}
The $p$-spin $A$ state (shaded in Fig. \ref{fig : F4}(a)) is surrounded
by the discontious ``orbital spin-flop'' transition.
This is because the ``orbital magnetic field'' $gr$ is applied not
perpendicularly but anti-parallel to the $d_{x^2-y^2}$-direction.
Therefore it seems unlikely that the spin canting occurs between the
$p$-spin $A$ and spin $F$ with {\it different} orbital structures, e.g.,
the arrow (a) in Fig. \ref{fig : F1}.
\par
We now look for the possibility of the spin canting with the orbital
structure fixed to be $d_{x^2-y^2}$, i.e., the arrow (b) in
Fig. \ref{fig : F1}.
The corresponding $J_S$ value might appear to be too small ($\sim$ 0.0002 eV),
but $J_S$ is very sensitive to the bond-length $l$ ( $J_S \sim l^{-14}$
as discussed later), and the lattice elongation along $c$-axis suggests
smaller value of $J_S$ in 327-system compared with 113.
Furthermore, the $x$-dependence of $l$ induces the $x$-dependence of $J_S$
and the arrow (b) in Fig. \ref{fig : F1} has finite vertical component.
Another remark is that even though the orbital pseudo-spin $\vec T_i$
is fully polarized along $d_{x^2-y^2}$-direction, the wave-fuction is the
hybridized one with $d_{x^2-y^2}$ and $d_{3z^2-r^2}$.
Namely the $d_{3z^2-r^2}$ has the weight $\sim \left(t_0/\tilde \beta
\right)^2$, which induces the effective transfer $t_z$ between layers.
However $t_z$ is much smaller than $t_0$, which is crutial to the spin
canting as discussed below.
\par
\begin{figure}[p]
\caption{}
\label{fig : F5}
\end{figure}
\noindent
In Fig. \ref{fig : F5} shown the energy as a function of the angle $\eta$
between the spins of two layers for several values of $J_S$.
Spin $F$ for $J_S=0$ is due to the above-mentioned double-exchange
interaction with the weight $\sim \left(t_0/\tilde \beta
\right)^2$.
For $J_S=$ 0.42-0.5 meV, the optimized angle $\eta$ is the intermediate between
$\eta=0$ and $\eta=\pi$, and the canting state is realized.
Therefore we conclude that the spin canting occurs between $p$-spin $A$ and
$p$-spin $F$ states when $J_S$ changes.
\begin{figure}[p]
\caption{}
\label{fig : F3}
\end{figure}
\noindent
In order to study further this canting state, we introduce a simplified model
shown in Fig. \ref{fig : F3}.
We assume that the spin polarization is perfect, i.e., all the electron
spins are aligned along the ordered direction.
Therefore we consider the spinless fermions.
The density of states is taken to be a constant $N_F$ for each layer.
We take the hole-picture, and $x$ holes are occupying this constant
density of states from the bottom.
These two bands are split into bonding and anti-bonding ones with the
splitting $\Delta=t_z\cos{\frac{\eta}{2}=t_z \xi}$.\cite{anderson}
The kinetic energy gain $\Delta E_{kin}(\xi)$ is given by
\begin{eqnarray}
\Delta E_{kin}\left(\xi\right)
\!\sim\!
\left\{
\begin{array}{ll}
\!\!-t_z^2\!\cdot\!N_F\cdot \xi^2
\ \ (\rm for\ \it \xi < \xi_c \equiv \frac{x}{N_Ft_z})
\\
-t_z\cdot x\cdot \xi \ \ (\rm for\ \it \xi > \xi_c)
\\
\end{array}
\right.
\ ,
\label{eqn : eq6.5.1}
\end{eqnarray}
while the energy cost of the exchange interaction is
$J_S\!\cos{\eta}=J_S\left(2\xi^2-1\right)$.
The lower line of Eq. (\ref{eqn : eq6.5.1}) is obtained by
de Gennes, and if this holds the canting always occurs.\cite{degenne}
The new aspect here is that $\Delta E_{kin}(\xi) \propto \xi^2$
when the splitting $\Delta=t_z \xi$ is smaller than the Fermi energy
$\epsilon_F = x/N_F$ and both the bonding and anti-bonding bands are occupied.
Therefore the spin $A$ structure ($\xi=0,\ \eta=\pi$) is at least locally
stable when $2J_S > t_z^2 N_F$.
This condition can be satisfied when the orbital is almost $d_{x^2-y^2}$
and $t_z$ is much reduced from $t_0$.
For general orbital configuration $t_z^2 N_F$ is order of magnetude
larger than $J_S$.
Now we look for the optimized $\xi$ to minimize the total energy
$\Delta E (\xi) = \Delta E_{kin}(\xi)+\Delta E_{ex}(\xi)$.
When $\xi_c > 1$ ($x > t_z N_F$), only the upper line of
Eq. (\ref{eqn : eq6.5.1}) is relevant and $\Delta E = \left(2J_S
-t_z^2 N_F\right)\cdot \xi^2$.
Therefore $\xi$ jumps from 1 (spin $F$) to 0 (spin $A$) as $J_S$
increases across $t_z^2 N_F/2$.
\begin{figure}[p]
\caption{}
\label{fig : F6}
\end{figure}
When $\xi_c<1$, the optimized $\xi$ as a function of $J_S$ is given in
Fig. \ref{fig : F6}.
As $J_S$ increases, the spin structure changes as spin $F$
($J_S < t_z x/4$) $\rightarrow$ spin canting ($t_z x/4 <J_S<t_z^2 N_F/4$)
$\rightarrow$ spin canting with fixed canting angle
($t_z^2 N_F/4<J_S<t_z^2 N_F/2$) $\rightarrow$ spin $A$ ($t_z^2 N_F/2<J_S$).
Note that the canting angle continuously evolves from spin $F$, but jumps
at the transition to the spin $A$.
Summarizing, $x<t_z N_F$ and $t_z x/4 <J_S<t_z^2 N_F/2$ are the condition
for the occurence of the spin canting.
Considering $N_F\sim 1/t_0$, and $J_S$ is about two orders of magnitude
smaller than $t_0$, this condition is satisfied only when $t_z$ is much
reduced from $t_0$ and $x$ is small.
Spin canting state does not appear in Fig. \ref{fig : F2} because of the large
$t_z$ for $n$-spin $F$.
\par
Spin canting observed in the metallic region $0.4<x_{exp.}<0.5$
\cite{argyriou971,argyriou972,perring97,hirota98,kimura98,kubota98}
implies therefore the planer orbital in this region.
This is consistent with the spin easy axis observed within the $ab$-plane
in this region \cite{argyriou972,hirota98,kubota98}
because the planer orbital leads to this anisotropy of the easy axis
according to the spin Hamiltonian derived by
Matsumoto \cite{matsumoto} taking into account the spin-orbit interaction.
The observed spin transition from $F$ to $A$ with increasing $x$ should
therefore corresponds to that from $p$-spin $F$
to $p$-spin $A$ in Fig. \ref{fig : F1}, for which
the increase in $J_S$ with increasing $x$ is needed
(arrow (b) in Fig. \ref{fig : F1}).
\par
This $x$-dependence of $J_S$ can be explained in terms of the
bond-length dependence of the inter-layer hopping integral.
According to the pseudo-potential theory \cite{harrison},
the hopping integral between the $d$ orbitals
depends on the bond length $l$ as $t \propto l^{-7}$,
leading to $\xi_0\propto t^z_{e_g}/\left(t^z_{t_{2g}}\right)^2
\propto l_c^{7}$,
where $l_c$ denotes the bond-length along $c$-axis.
The observed lattice contraction along $c$-axis with
increasing $x$, \cite{moritomo97} therefore, gives rise to
the decrease of $\xi_0$, leading to the spin structure changes
from spin $F$ to cant, and to spin $A$.
$J_S^z\propto \left(t_{t_{2g}}^z\right)^2\propto l_c^{-14}$ increases due to
this lattice contraction along $c$-axis, which can explain the
$x$-dependence of $J_S$ needed for the change
from $p$-spin $F$ to $p$-spin $A$, as shown by the arrow (b)
in Fig. \ref{fig : F1}.
\par
The neutron-diffraction experiments have revealed that the samples with
the spin canting at zero temperature ($0.4<x_{exp.}<0.48$) become
the spin $A$ at higher temperatures. \cite{hirota98}
This can also be explained in terms of the lattice deformation as follows.
The observed lattice contraction along $c$-axis
\cite{mitchell97,kimura98,hirota98} with increasing temperature
brings about the decrease of $\xi_0 \propto t_z/J_S\propto l^7_c$,
leading to the spin transition toward the spin $A$ discontinously.
Hirota $et\ al.$ also attributed this $T$-dependent spin transition
to the lattice deformation. \cite{hirota98}
\par
As seen above, the observed lattice deformation plays an important
role to explain the spin transition.
In our scenario, the bond-length dependence of the inter-layer
hopping is the most important mechanism which relates the lattice- and
the electron-systems.
One might, however, attribute it rather to the Jahn-Teller type
electron-lattice interaction through the electro-static coupling
(in other word, the ``orbital magnetic field'').
In this scenario, the lattice contraction along $c$-axis
reduces the hybridization of $d_{3z^2-r^2}$
orbital as $x$ increases, namely the stabilization of $d_{x^2-y^2}$
which prefers spin $A$. \cite{hirota98}
According to this scenario, the temperature dependence of the
spin canting mentioned above is also explained by the change in the
hybridization of $d_{3z^2-r^2}$ \cite{hirota98} which is brought about by the
temperature-depenedent lattice elongation.
This scenario, however, results in the decrease of the inter-layer
hopping as $x$ ($T$) increases, being opposite to our scenario, where
little change in the orbital ordering and the increase of the
inter-layer hopping $t_z\!\propto l_c^{-7}$
during the spin transition are predicted.
Main distinction between these two scenarios is whether such a large
change in the hybridization of $d_{3z^2-r^2}$ occurs during the spin transition
or not.
As seen in Fig. \ref{fig : F2} and Fig. \ref{fig : F4}, however, the spin
transition
should be of the first order without canting if it is accompanied with the
large change in the orbital.
Experimentally, the spin easy axis falls onto $ab$-plane
discontinously around $x_{exp}=0.32$ (spin $F$) \cite{moritomo97,kubota98}
and the spin transition from $F$ to cant, and to spin $A$ occurs
with this easy axis being unchanged.
This behavior implies that the hybridization of $d_{3z^2-r^2}$ decreases
discontinously before the spin transition, and during the transition
the wave-function is almost planer as $d_{x^2-y^2}$, namely the transition
is between $p$-spin F and $p$-spin A with canting.
Though, even in this case, the ``orbital magnetic field'' brings about
a little change in the hybridization of $d_{3z^2-r^2}$ via the
change of the weight mentioned before,
$\left(t_0/\tilde \beta\right)^2 \rightarrow
\left(t_0/\left(\tilde\beta-gr \right)\right)^2$,
it is a minor correction comparing with the sensitively changing
bond-length dependence, $t^z\propto l_c^{-7},\ J_S\propto l_c^{-14}$.
\par
For the spin canting, $\xi_0=t^z x/4J_S<1$, and hence
the planer orbital during the spin transition turned out to be
essential.
In 113-system, the spin $F$ phase neighboring the metallic spin $A$
has the quasi-three-dimensional orbital ordering
for the realistic parameters in the mean field theory.
\cite{maezono981,maezono982}
Therefore the canting does not occur because the orbital structures
are {\it different} between spin $A$ and $F$, and the transition between them
is always discontinous.
In the real 113-system no anisotropy has been reported in the spin $F$
state, and there is a theoretical suggestion of the orbital
liquid state for the spin $F$ metallic region.\cite{ishihara97}
In any case the orbital state is different from the $d_{x^2-y^2}$
in the metallic spin $A$ state.
In 327-case, the layered crystal structure brings
about the two-dimensional orbital anisotropy, reflecting the interplay
of the dimensionality of the crystal structure and that of
the orbital ordering.
\par
\begin{figure}[p]
\caption{}
\label{fig : F7}
\end{figure}
Figure \ref{fig : F7} summarizes our interpretation of the experiments
(notations $x_{th.}$ and $x_{exp.}$ are used to prevent
the confusion due to the quantitative discrepancy of $x$ between the
calculations and the experiments).
The hatched region corresponds to the spin canting phase due to the
planer orbital for the $p$-spin $F$.
$J_S\left(FA\right)$ is the phase boundary with the
first-order transition between spin $F$ (or canting) and spin $A$.
Spin $F$ phase at $x_{exp.}=0.3$ \cite{kimura96-97}
corresponds to the $n$-spin $F$ phase.
With increasing $x$, spin $F$ changes from $n$- to $p$- with
the orbital transition into $d_{x^2-y^2}$ around $x_{th.}\sim0.125$.
This transition is of the first-order, leading to the discontinous
reorientation of the spin easy axis from being along $c$-axis
into within the $ab$-plane by considering the spin-orbit interaction.
\cite{matsumoto}
This explains the observed spin anisotropy
\cite{argyriou972,hirota98,kubota98,perring98}
where the easy axis, pointing parallel to the $c$-axis for $x_{exp.}=0.3$,
\cite{perring98} discontinuously turns onto the
$ab$-plane at $x_{exp.}\sim 0.32$,\cite{kubota98}
for $T\lesssim100$ K.
The $p$-spin $F$ phase around $x_{th.}\gtrsim 0.125$ cants readily
within the $ab$-plane because of the decrease in $\xi_0 \propto l_c^7$
due to the contraction of the bond-length along $c$-axis with increasing $x$,
and then discontinously jumps to the $p$-spin $A$, corresponding to
the observation for $0.4<x_{exp.}<0.5$.
\cite{argyriou971,argyriou972,perring97,hirota98,kimura98,kubota98}
\par
In conclusions, we have studied the spin and the orbital ordering in
327-system, taking the orbital degrees of freedom and the
strong Coulombic repulsion into account.
Observed $x$-dependence of the spin ordering and its anisotropy
were qualitatively explained in terms of the orbital ordering and the
observed lattice deformation along $c$-axis.
The observed temperature depencence of the spin structure is also
explained by the lattice deformation.
The difference between 327-system with canting and 113-system
without canting is understood in terms of the difference in the dimensionality
of the crystal strucuture.
\par
The authors would like to thank K. Hirota, Y. Endoh,
T. Akimoto, Y. Moritomo, S. Ishihara, W. Koshibae,
S. Maekawa, H. Yoshizawa, T. Kimira, and Y. Tokura for their valuable
discussions.
This work was supported by Priority Areas Grants from the Ministry
of Education, Science and Culture of Japan.
\par
|
\section{introduction}
The spatial gradient expansion \cite{ge:a,ge:b,ge:c,ge:d} of the Einstein equation is non-linear approximation method which describes long-wavelength inhomogeneity in the universe. This approximation scheme expands the Einstein equation by the order of the spatial gradient. As the background solution, we solve the Einstein equation by neglecting all spatial gradient terms. The resultant solution has the same form as that for the spatially flat Friedman-Robertson-Walker(FRW) universe, but the three metric can have spatial dependence. It is possible to include the effect of spatial gradient terms by calculating the next order. This method can describes the long-wavelength non-linear perturbation without imposing any symmetry for a space, and is suitable for analyzing the global structure of an inhomogeneous universe. Furthermore, it is easy to include perfect fluid with pressure and scalar fields.
But this scheme is valid only for the perturbation whose wavelength is larger than the Hubble horizon scale. For the matter field which satisfies energy conditions, the perturbation terms induced by the spatial gradient terms grows in time and finally dominates the background solution. This occurs when the wavelength of the perturbation equals the Hubble horizon scale. After this time, the wavelength of the perturbation becomes shorter than the horizon and the result of the gradient expansion becomes unreliable.
The similar situation occurs in the field of non-linear dynamical systems. To obtain temporal evolution of the solution of a non-linear differential equation, we usually apply perturbative expansion. But naive perturbation often yields secular terms due to the resonance phenomena. The secular terms prevent us from getting approximate but global solutions. There are many techniques to circumvent the problem, for example, averaging method, multi-time scale method, WKB method and so on\cite{appro:a,appro:b}. Although these method yield globally valid solutions, they provide no systematic procedure for general dynamical systems because we must select a suitable assumption on the structure of perturbation series.
Renormalization group method\cite{rg:a,rg:b,rg:c,rg:d,rg:e} as a tool for global asymptotic analysis of the solution to differential equations unifies the techniques listed above, and can treat many systems irrespective of their features. Starting from a naive perturbative expansion, the secular divergence is absorbed to the constants of integration contained in the zeroth order solution by the renormalization procedure. The renormalized constants obey the renormalization group equation. This
method can be viewed as a tool of system reduction. The renormalization group equation corresponds to the amplitude equation which describes slow motion dynamics in the original system. We can describe complicated dynamics contained in the original equation by extracting a simpler representation using the renormalization group method.
In this paper, we apply the renormalization group method to the gradient expansion of the Einstein equation. Our purpose is to obtain the renormalized long-wavelength solution of the Einstein equation which is also valid for late time. Through the procedure of renormalization, we extract slow motion from the Einstein equation.
This paper is organized as follows. In section 2, we briefly review the gradient expansion. In section 3, we introduce the renormalization group method by using two examples. In section 4, the renormalization group method is applied to the solution of the gradient expansion. In section 5, we solve the renormalization group equation for several situations. Section 6 is devoted to summary and discussion. We use the unit in which $c=\hbar=8\pi G=1$ throughout the paper.
\section{long-wavelength expansion of Einstein+dust system}
We use synchronous reference frame:
\begin{equation}
ds^2=-dt^2+\gamma_{ij}(t,x)dx^idx^j.
\end{equation}
The Einstein equation with dust fluid is
\begin{eqnarray}
{}^{(3)}R&+&K^2-K^k_lK^l_k=2\rho, \label{eq:hc}\\
K_{,i}&-&K^j_{i;j}=\sqrt{1+u^ku_k}\,\rho\,u_i, \label{eq:mc}\\
{}^{(3)}R^j_i&+&\frac{1}{\sqrt{\gamma}}\frac{\partial}{\partial t}\left(\sqrt{\gamma}\,K^j_i\right)=\frac{\rho}{2}\left(\delta^j_i+2u^ju_i\right),\label{eq:evo}\\
K_{ij}&=&\frac{1}{2}\frac{\partial\gamma_{ij}}{\partial t},
\end{eqnarray}
where $K_{ij}$ is the extrinsic curvature, $\rho$ is the energy density of dust, $u_i$ is the three velocity of dust. If we neglect terms with spatial derivative, the solution can be written
\begin{equation}
\gamma^{(0)}_{ij}=a^2(t)h_{ij}(x),\quad \rho^{(0)}=\rho_0(t),\quad u_i^{(0)}=0,
\end{equation}
where $h_{ij}(x)$ is an arbitrary function of spatial coordinate (seed metric), and the scale factor $a(t)$ and the energy density $\rho_0(t)$ obey usual spatially flat FRW equation. Using this as the zeroth order solution, we solve this system by the following gradient expansion:
\begin{eqnarray}
\gamma_{ij}&=&a^2(t)h_{ij}(x)+\left(f_2(t)R_{ij}(h)+g_2(t)R(h)h_{ij}\right)+\cdots,\\
u_i&=&u_3(t)\nabla_iR(h)+\cdots,\\
\rho&=&\rho_0(t)+\rho_2(t)R(h)+\cdots,
\end{eqnarray}
where $R_{ij}(h)$ is the Ricci tensor of the seed metric $h_{ij}$.
The zeroth order solution is
\begin{equation}
\gamma^{(0)}_{ij}=t^{4/3}h_{ij}(x),\quad u^{(0)}_i=0,\quad \rho^{(0)}=\frac{4}{3t^2}.
\end{equation}
The solution to the next order is
\begin{eqnarray}
\gamma^{(2)}_{ij}&=&-\frac{9t^2}{5}\left(R_{ij}(h)-\frac{1}{4}R(h)h_{ij}\right), \\
\rho^{(2)}&=&\frac{3t^{-4/3}}{10}R(h),\quad u^{(3)}_i=0. \nonumber
\end{eqnarray}
The solution up to the second order of spatial gradient can be written as follows:
\begin{eqnarray}
\gamma_{ij}&=&t^{4/3}\left\{h_{ij}-\frac{9}{5}t^{2/3}\left(R_{ij}(h)-\frac{1}{4}R(h)h_{ij}\right)\right\},\\
\rho&=&\frac{4}{3t^2}\left(1-\frac{9}{20}t^{2/3}R(h)\right),\quad u_i=0.\nonumber
\end{eqnarray}
We can see that the perturbation term, which originated from the spatial gradient of the seed metric, grows as the universe expands and finally has the same amplitude as the background term at $t\sim H^{-1}$ when the wavelength of the perturbation equals Hubble horizon scale. After this time, the wavelength of perturbation becomes smaller than horizon scale and the long-wavelength expansion breaks down.
To make the gradient expansion applicable to the perturbation whose wavelength is smaller than the horizon scale, we use the renormalization group method.
\section{renormalization group method}
The renormalization group method\cite{rg:a,rg:b,rg:c,rg:d,rg:e} improve the long time behavior of naive perturbative expansion. We explain the basic concept of the renormalization group method using two examples. First one is a harmonic oscillator. The equation of motion is
\begin{equation}
\ddot x+x=-\epsilon\, x,
\end{equation}
where $\epsilon$ is a small parameter. We solve this equation perturbatively by expanding the solution with respect to $\epsilon$:
\begin{equation}
x=x_0+\epsilon\,x_1+\cdots.
\end{equation}
The solution up to $O(\epsilon)$ becomes
\begin{equation}
x=B_0\cos t+C_0\sin t+\frac{\epsilon}{2}\,(t-t_0)(C_0\cos t-B_0\sin t)+O(\epsilon^2),\label{eq:naive}
\end{equation}
where $B_0$ and $C_0$ are constants of integration determined by the initial condition at arbitrary time $t=t_0$. This naive perturbation breaks down when $\epsilon\,(t-t_0)>1$ because of the secular term. To regularize the perturbation series, we introduce an arbitrary time $\mu$, split $t-t_0$ as $t-\mu+\mu-t_0$, and absorb the divergent term containing $\mu-t_0$ into the renormalized counterparts $B$ and $C$ of $B_0$ and $C_0$, respectively.
We introduce renormalized constants as follows:
\begin{equation}
B_0=B(\mu)+\epsilon\,\delta B(\mu,t_0),\quad C_0=C(\mu)+\epsilon\,\delta C(\mu,t_0), \label{eq:counter}
\end{equation}
where $\delta B$ and $\delta C$ are counter terms that absorb the terms containing $\mu-t_0$ in naive solution. Substitute Eq.(\ref{eq:counter}) to Eq.(\ref{eq:naive}), we have
\begin{eqnarray}
x&=&B(\mu)\cos t+C(\mu)\sin t \nonumber \\
&&\quad +\epsilon \left\{\delta B\cos t+\delta C\sin t+\frac{1}{2}(t-\mu+\mu-t_0)(C(\mu)\cos t-B(\mu)\sin t)\right\}.
\end{eqnarray}
$\delta B$ and $\delta C$ are chosen as
\begin{equation}
\delta B(\mu,t_0)+\frac{1}{2}(\mu-t_0)C(\mu)=0,\quad \delta C(\mu,t_0)-\frac{1}{2}(\mu-t_0)B(\mu)=0. \label{eq:count}
\end{equation}
Using the relation $\epsilon\,\delta B=B_0-B(\mu),\, \epsilon\,\delta C=C_0-C(\mu)$,
\begin{equation}
B(t_0)=B(\mu)-\frac{\epsilon}{2}(\mu-t_0)C(\mu),\quad C(t_0)=C(\mu)+\frac{\epsilon}{2}(\mu-t_0)B(\mu). \label{eq:rgtransf}
\end{equation}
These equations define the transformation up to $O(\epsilon)$:
$${\cal R}_{\mu-t_0}:~(B(t_0), C(t_0))\rightarrow (B(\mu), C(\mu)).$$
Explicit form of the transformation is
\begin{equation}
\left(
\begin{array}{c}
B(\mu) \\ C(\mu)
\end{array}
\right)={\cal R}_{\mu-t_0}
\left(
\begin{array}{c}
B(t_0) \\ C(t_0)
\end{array}
\right)=
\left(
\begin{array}{cc}
1 & \frac{\epsilon}{2}(\mu-t_0) \\
-\frac{\epsilon}{2}(\mu-t_0) & 1
\end{array}
\right)
\left(
\begin{array}{c}
B(t_0) \\ C(t_0)
\end{array}
\right)+O(\epsilon^2).
\end{equation}
${\cal R}$ satisfies the composition law:
\begin{eqnarray}
{\cal R}_{\mu-\mu_1}\otimes{\cal R}_{\mu_1-t_0}&=&
\left(
\begin{array}{cc}
1 & \frac{\epsilon}{2}(\mu-\mu_1) \\
-\frac{\epsilon}{2}(\mu-\mu_1) & 1
\end{array}
\right)
\left(
\begin{array}{cc}
1 & \frac{\epsilon}{2}(\mu_1-t_0) \\
-\frac{\epsilon}{2}(\mu_1-t_0) & 1
\end{array}
\right)+O(\epsilon^2) \nonumber \\
&=&
\left(
\begin{array}{cc}
1 & \frac{\epsilon}{2}(\mu-t_0) \\
-\frac{\epsilon}{2}(\mu-t_0) & 1
\end{array}
\right)+O(\epsilon^2)\nonumber \\
&=&{\cal R}_{\mu-t_0},
\end{eqnarray}
and ${\cal R}_0=1$ and ${\cal R}_{-\mu}$ is the inverse transformation of ${\cal R}_\mu$. So we can conclude that the transformation forms a Lie group up to $O(\epsilon)$. Assuming the properties of Lie group, we can extend the locally valid expression (\ref{eq:rgtransf}) to a global one, which is valid for arbitrary large $\mu-t_0$. We apply this transformation to get $(B(\mu),C(\mu))$ at arbitrary large $\mu$. By differentiating Eq.\,(\ref{eq:rgtransf}) with respect to $\mu$ and setting $t_0=\mu$, we have renormalization group equations:
\begin{equation}
\frac{\partial B}{\partial\mu}=\frac{\epsilon}{2}\,C(\mu),\quad \frac{\partial C}{\partial \mu}=-\frac{\epsilon}{2}\,B(\mu). \label{eq:rge}
\end{equation}
The renormalized solution becomes
\begin{equation}
x(t)=B(\mu)\cos t+C(\mu)\sin t+\frac{\epsilon}{2}\,(t-\mu)(C(\mu)\cos t-B(\mu)\sin t). \label{eq:rsola}
\end{equation}
Solving the renormalization group equation (\ref{eq:rge}) and equating $\mu$ and $t$ in (\ref{eq:rsola}) eliminates the secular term and we get uniformly valid result:
\begin{equation}
x(t)=B(0)\cos\left(1+\frac{\epsilon}{2}\right)t+C(0)\sin\left(1+\frac{\epsilon}{2}\right)t.
\end{equation}
Second example is the Einstein equation for the FRW universe with dust. The spatial component of the Einstein equation is
\begin{equation}
\ddot\alpha+\frac{3}{2}\left(\dot\alpha\right)^2=-\frac{\epsilon}{2}e^{-2\alpha}, \label{eq:frw}
\end{equation}
where $\alpha(t)$ is the logarithm of the scale factor of the universe $a(t)$ and $\epsilon$ is the sign of the spatial curvature. The exact solution is given by
\begin{equation}
a(t)=e^{\alpha(t)}=\left\{
\begin{array}{lll}
a_0\,(1-\cos\eta), & t=a_0\,(\eta-\sin\eta) & \quad\mbox{for}~\epsilon=1 \\
a_0\,\frac{\eta^2}{2}, & t=a_0\,\frac{\eta^3}{6} & \quad\mbox{for}~\epsilon=0 \\
a_0\,(\cosh\eta-1), & t=a_0\,(\sinh\eta-\eta) & \quad\mbox{for}~\epsilon=-1
\end{array}
\right. \label{eq:esol}
\end{equation}
We solve Eq.\,(\ref{eq:frw}) perturbatively by assuming that the right hand side is small. This is expansion with respect to small spatial curvature around the flat universe. By substituting $\alpha=\alpha_0+\epsilon\,\alpha_1+\cdots$ to Eq.\,(\ref{eq:frw}), we have naive solution:
\begin{equation}
\alpha=\ln\tau+C_0-\epsilon\,\frac{9}{20}e^{-2C_0}(\tau-\tau_0)+O(\epsilon^2), \label{eq:nsol}
\end{equation}
where we introduced a new time variable $\tau=t^{2/3}$ and $C_0$ is a constant of integration determined by the initial condition at $\tau=\tau_0$. $O(\epsilon)$ term is secular and we regularize this term by introducing arbitrary time $\mu$ and renormalized constant $C_0=C(\mu)+\epsilon\,\delta C(\mu,\tau_0)$:
\begin{equation}
\alpha=\ln\tau+C(\mu)+\epsilon\,\delta C(\mu,\tau_0)-\epsilon\,\frac{9}{20}e^{-2C(\mu)}(\tau-\mu+\mu-\tau_0).
\end{equation}
The counter term $\delta C$ is determined by so as to absorb $\mu-\tau_0$ dependent term:
\begin{equation}
\delta C(\mu,\tau_0)-\frac{9}{20}e^{-2C(\mu)}(\mu-\tau_0)=0.
\end{equation}
This defines the renormalization group transformation ${\cal R}_{\mu-\tau_0}:~C(\tau_0)\ \rightarrow\ C(\mu)$
\begin{equation}
C(\mu)=C(\tau_0)-\epsilon\,\frac{9}{20}e^{-2C(\mu)}(\mu-\tau_0),
\end{equation}
and this transformation forms Lie group up to $O(\epsilon)$. So we can have $C(\mu)$ for arbitrary large value of $\mu-\tau_0$ by assuming the property of Lie group, and this makes possible to produce globally uniform approximated solution of the original equation. The renormalization group equation is
\begin{equation}
\frac{\partial C(\mu)}{\partial\mu}=-\epsilon\,\frac{9}{20}e^{-2C(\mu)},
\end{equation}
and the solution is
\begin{equation}
C(\mu)=\frac{1}{2}\ln\left(c-\frac{9\epsilon}{10}\mu\right),
\end{equation}
where $c$ is a constant of integration. The renormalized scale factor is given by
\begin{equation}
a(t)=e^{\alpha(t)}=\tau\,e^{C(\tau)}=t^{2/3}\left(c-\frac{9\epsilon}{10}t^{2/3}\right)^{1/2}. \label{eq:rsol}
\end{equation}
As the zeroth order solution, it is possible to include another integration constant $t_0$ which defines the origin of cosmic time $t$. By requiring that the renormalization group transformation forms Lie group, it can be shown that $t_0$ does not get renormalization. So it is sufficient to consider the solution with the boundary condition $a(t=0)=0$ which fixes the value of $t_0$ equals zero. This point is different from the example of harmonic oscillator, in which
case two integration constants $B$ and $C$ both get renormalization.
We compare the renormalized solution (\ref{eq:rsol}) with the exact solution (\ref{eq:esol}) and the naive solution (\ref{eq:nsol}) for the case of a closed universe ($\epsilon=1$). The scale factor of the exact solution has maximum at $t=\pi$ and goes to zero at $t=2\pi$. The naive solution does not show this behavior. The renormalized solution improves the naive solution and reproduces the expanding and contracting feature of the exact solution (Fig.1).
\section{renormalization of long-wavelength solution}
Now we proceed to the long-wavelength solution of Einstein equation. We renormalize the secular behavior of three metric $\gamma_{ij}$. By introducing a new time variable $\tau=t^{2/3}$, the longwavelength solution can be written as
\begin{equation}
\gamma_{ij}=\tau\,\left\{h_{ij}(x)-\frac{9\tau}{5}\left(R_{ij}(h)-\frac{1}{4}R(h)h_{ij}\right)\right\}.
\end{equation}
We renormalize the secular term $\frac{9\tau}{5}(R_{ij}-\frac{1}{4}R h_{ij})$. We introduce the initial time $\tau_0$ by re-defining the seed metric $h_{ij}$, and define renormalized metric and the counter term $h_{ij}(x)=h_{ij}(x,\mu)+\delta h_{ij}(x,\mu,\tau_0)$:
\begin{eqnarray}
h_{ij}(x)&-&\frac{9}{5}(\tau-\tau_0)\left(R_{ij}-\frac{1}{4}R(h)h_{ij}\right) \nonumber \\
&=&h_{ij}(x,\mu)+\delta h_{ij}-\frac{9}{5}(\tau-\mu+\mu-\tau_0)\left(R_{ij}-\frac{1}{4}R(h)h_{ij}\right).
\end{eqnarray}
The counter term is determined so that it absorbs terms containing $\mu-\tau_0$:
\begin{equation}
\delta h_{ij}(\mu,\tau)-\frac{9}{5}(\mu-\tau_0)\left(R_{ij}(h)-\frac{1}{4}R(h)h_{ij}\right)=0.
\end{equation}
Using the definition of $\delta h_{ij}$,
\begin{equation}
h_{ij}(x,\mu)=h_{ij}(x,\tau_0)-\frac{9}{5}(\mu-\tau_0)\left(R_{ij}(h(\mu))-\frac{1}{4}R(h(\mu))h_{ij}(\mu)\right). \label{eq:rgtrnsf2}
\end{equation}
This equation defines the renormalization group transformation ${\cal R}_{\mu-\tau_0}:h_{ij}(\tau_0)\rightarrow h_{ij}(\mu)$ and this transformation is Lie group up to $O(\epsilon)$. We therefore can get the value of $h_{ij}(\mu)$ for arbitrary $\mu$ using the relation (\ref{eq:rgtrnsf2}) by assuming the property of Lie group. The renormalization group equation is obtained by differentiating Eq.\,(\ref{eq:rgtrnsf2}) with respect to $\mu$ and setting $\tau_0=\mu$:
\begin{equation}
\frac{\partial}{\partial\tau}h_{ij}(x,\mu)=-\frac{9}{5}\left[R_{ij}(h(\mu))-\frac{1}{4}R(h(\mu))h_{ij}(\mu)\right], \label{eq:rge2}
\end{equation}
and renormalized solution is
\begin{eqnarray}
\gamma_{ij}&=&t^{4/3}h_{ij}(x,t), \\
\rho &=&\frac{4}{3t^2}+\frac{3t^{-4/3}}{10}R(h(x,t)).
\end{eqnarray}
\section{solution of renormalization group equation}
We solve the renormalization group equation (\ref{eq:rge2}) for some special cases and see how the renormalization group method improves the behavior of the long-wavelength solution.
\subsection{FRW case}
The metric is
$$
h_{ij}(x,t)=\Omega^2(t)\,\sigma_{ij}(x),\quad R_{ij}(\sigma)=\frac{1}{3}\sigma_{ij}\,R(\sigma),
$$
where $\sigma_{ij}(x)$ is the metric of three dimensional maximally symmetric space. In this case, the renormalization group equation reduces to
$$
\frac{\partial}{\partial\tau}\Omega^2(\tau)=-\frac{9}{10}k,\quad (k=\pm1, 0),
$$
and the solution is
$$
\Omega(t)=\sqrt{c-\frac{9k}{10}t^{2/3}},
$$
where $c$ is a constant of integration.
The renormalized solution is
\begin{eqnarray}
\gamma_{ij}&=&t^{4/3}\left(c-\frac{9k}{10}\,t^{2/3}\right)\sigma_{ij}(x), \nonumber \\
\rho&=&\frac{4}{3t^2}\left(1+\frac{27}{20}\frac{k\,t^{2/3}}{c-\frac{9k}{10}\,t^{2/3}}\right), \label{sol:frw}
\end{eqnarray}
and the scale factor of the universe is given by
\begin{equation}
a(t)=t^{2/3}\sqrt{c-\frac{9k}{10}t^{2/3}}.
\end{equation}
This is the same as the solution (\ref{eq:rsol}).
\subsection{spherically symmetric case}
In spherical coordinate $(r, \theta, \phi)$, the metric is
\begin{equation}
h_{ij}=\mbox{diag}\left(A^2(\tau, r), B^2(\tau, r), B^2(\tau, r)\sin^2\phi\right).
\end{equation}
The renormalization group equation (\ref{eq:rge2}) becomes
\begin{eqnarray}
2A\frac{\partial A}{\partial\tau}&=&\frac{9}{5}\left(\frac{A^2}{2B^2}-\frac{A_{,r}B_{,r}}{AB}-\frac{\left(B_{,r}\right)^2}{2B^2}+\frac{B_{,rr}}{B}\right), \quad (rr~ \mbox{component})\\
2B\frac{\partial B}{\partial\tau}&=&\frac{9}{5}\left(-\frac{1}{2}+\frac{\left(B_{,r}\right)^2}{2A^2}\right).\quad (\theta\theta~ \mbox{component})
\end{eqnarray}
The solution of this equation is given by
\begin{eqnarray}
B&=&\left(1-\frac{9\alpha(r)}{10}\tau\right)^{1/2}\beta(r), \\
A&=&\frac{B_{,r}}{\sqrt{1-\alpha(r)\,\beta^2(r)}},
\end{eqnarray}
where $\alpha(r)$ and $\beta(r)$ are arbitrary function of $r$. The renormalized metric and density are
\begin{eqnarray}
ds^2&=&-dt^2+\frac{\left(t^{2/3}B_{,r}\right)^2}{1-\alpha\,\beta^2}\,dr^2+\left(t^{2/3}B\right)^2d\Omega^2_2,\label{eq:rgstb} \\
\rho&=& \frac{4}{3t^2}+\frac{3}{5t^{4/3}}\frac{3\alpha\,\beta_{,r}\left(1-\frac{9\alpha}{10}\tau\right)+\alpha_{,r}\,\beta\,\left(1-\frac{27\alpha}{20}\tau\right)}{\left(1-\frac{9\alpha}{10}\tau\right)\left[\left(1-\frac{9\alpha}{10}\tau\right)\beta_{,r}-\frac{9}{20}\alpha_{,r}\,\beta\,\tau\right]}.
\end{eqnarray}
By choosing $\beta=r, \alpha=0,\pm 1$, we can recover the solution of FRW case (\ref{sol:frw}). This solution corresponds to Toleman-Bondi solution\cite{TB}:
\begin{equation}
ds^2=-dt^2+\frac{R_{,r}(t,r)^2}{1+2E(r)}dr^2+R(t,r)^2 d\Omega_2^2
\end{equation}
where
\begin{equation}
R(t,r)=\left\{
\begin{array}{lll}
-\frac{M(r)}{2E(r)}(1-\cos\eta),& t=\frac{M(r)}{(-2E(r))^{3/2}}(\eta-\sin\eta) & \quad\mbox{for}~E(r)<0 \\
\left(\frac{9M(r)}{2}\right)^{1/3}t^{2/3}, & &\quad\mbox{for}~E(r)=0 \\
\frac{M(r)}{2E(r)}(\cosh\eta-1), & t=\frac{M(r)}{(2E(r))^{3/2}}(\sinh\eta-\eta) &\quad\mbox {for}~E(r)>0
\end{array}
\right.
\end{equation}
and
\begin{equation}
\rho(t,r)=\frac{2M_{,r}(r)}{R^2\,R_{,r}}.
\end{equation}
$E(r), M(r)$ are arbitrary function of radial coordinate $r$ and are connected to the initial distribution of the spatial curvature and the initial distribution of mass density of dust, respectively. They have the following correspondence with the functions $\alpha, \beta$ contained in the renormalization group solution (\ref{eq:rgstb}):
\begin{equation}
2E(r)\leftrightarrow -\alpha(r)\beta^2(r),\qquad \frac{9}{2}M(r)\leftrightarrow \beta^3(r).
\end{equation}
The renormalized solution well reproduces the feature of the metric of spherically symmetric gravitational collapse of dust.
\subsection{Szekeres solution}
In Cartesian coordinate $(x,y,z)$, the metric is assumed to be
\begin{equation}
h_{ij}=\mbox{diag}\left(1, 1, A^2(\tau,x,y,z)\right).
\end{equation}
The renormalization group equation (\ref{eq:rge2}) reduces to the following three equations:
\begin{eqnarray}
0&=&-\frac{A_{,xy}}{A}, \quad (xy~\mbox{component})\nonumber \\
0&=&\frac{A_{,yy}-A_{,xx}}{2A}, \quad (xx~\mbox{component})\\
\frac{\partial A^2}{\partial\tau}&=&-\frac{1}{2}A(A_{,xx}+A_{,yy}), \quad(zz~\mbox{component})\nonumber
\end{eqnarray}
The solution of this equation is
\begin{equation}
A=g(z)(x^2+y^2)+\frac{9}{5}g(z)\,\tau+c(z),
\end{equation}
where $c, g$ are arbitrary function of $z$. The renormalized metric is
\begin{equation}
ds^2=-dt^2+t^{4/3}\left[dx^2+dy^2+\left(g(z)\,(x^2+y^2)+\frac{9g(z)}{5}t^{2/3}+c(z)\right)^2\,dz^2\right].
\end{equation}
This is the Szekeres' exact solution\cite{seke}, which represents one dimensional gravitational collapse. It is known that the ``naive" gradient expansion reproduces this solution by including the fourth order spatial gradient\cite{ge:c}. We obtained the solution using the second order spatial gradient with renormalization. In this case, renormalization procedure much improves the naive solution.
\section{summary and discussion}
We have investigated renormalization of the long-wavelength solution of the Einstein equation which has secular behavior. We applied the renormalization group method to improve the long time behavior of the solution. We introduced renormalization point $\mu\sim L/H^{-1}$ where $L$ is physical wavelength of perturbation. For $\mu\ll 1(L\gg H^{-1})$, the naive expansion(gradient expansion) well approximate the exact solution. But as the universe expands, $\mu\sim H^{-1}/L\propto t^{1/3}$ becomes large and at $\mu\sim 1(L\sim H^{-1})$, perturbative expansion breaks down. After this time, the wavelength of the perturbation is smaller than the horizon scale. Using renormalization transformation which has the property of Lie group, we can renormalize the secular terms caused by the spatial gradient of the background seed metric. The renormalized metric can be extended to large value of $\mu$, which corresponds to the small scale fluctuation.
We obtained the solution of the renormalization group equation for FRW, spherically symmetric and Szekeres cases. The behavior of the renormalized solution indicates that they describe the collapsing phase of the system qualitatively well. The renormalization group method is regarded as the procedure of system reduction. This means the renormalization group Eq.\,(\ref{eq:rge2}) is reduced version of the original Einstein equation and describes slow motion dynamics of the original equation. We expect the renormalization group equation (\ref{eq:rge2}) has physically interesting properties and solutions which contained in the Einstein equation.
We can look the renormalization of the long-wavelength solution from the view point of the backreaction problem in cosmology. The naive solution represents the evolution of the perturbation with the fixed background metric. By renormalizing the naive solution, the constants $h_{ij}(x)$ contained in the background solution becomes to have the time dependence due to the spatial inhomogeneity. So we can investigate how the spatial inhomogeneity affects ``background'' metric $h_{ij}$ by solving the renormalization group equation. The remarkable feature of the renormalization group approach is that it does not need any spatial averaging which is necessary in conventional approach to the backreaction problem in cosmology\cite{back}.
In this paper, we considered only dust fluid as the matter field. But the gradient expansion can treat the perfect fluid with pressure and scalar fields. For the scalar field system, although we cannot write down exact solutions of the naive expansion in general, it is possible to define the renormalization group transformation and obtain the renormalized solution formally. It is interesting to apply the renormalization group method to the inflationary model and investigate the effect of the stochastic quantum noise on the background geometry. We can construct the inhomogeneous model of stochastic inflation\cite{inf:a,inf:b,inf:c,inf:d} and this is our next subject to be explored.
\vspace{2cm}
\begin{center}
{\bf ACKNOWLEDGEMENT}
\end{center}
We would like to thank K. Nozaki for indicating us the possibility of higher order renormalizability of the long-wavelength solution.
|
\section{Introduction}
Quantum computation is a most challenging project involving research both
by physicists and computer scientists. The principles of quantum
computation differ from the principles of classical computation very much.
The classical computation is based on classical mechanics while
quantum computation attempts to exploit phenomena specific to
quantum physics.
One of features of quantum mechanics is that a quantum process can be
in a combination (called {\em superposition}) of several states and these
several states can interact one with another.
A computer scientist would call this {\em a massive
parallelism}.
This possibility of massive parallelism is very important for
Computer Science. In 1982, Nobel prize winner physicist Richard Feynman
(1918-1988) asked what effects the principles of quantum
mechanics can have on computation\cite{Fe 82}. An exact simulation of quantum
processes demands exponential running time.
Therefore, there may be other computations
which are performed nowadays by classical computers but might be
simulated by quantum processes in much less time.
R.Feynman's influence was (and is) so high that rather soon this
possibility was explored both theoretically and practically.
David Deutsch\cite{De 89} introduced quantum Turing machines,
quantum physical counterparts of probabilistic Turing machines.
He conjectured that they may be more efficient that classical
Turing machines. He also showed the existence of a universal quantum
Turing machine. This construction was subsequently improved
by Bernstein and Vazirani \cite{BV 97} and Yao \cite{Ya 93}.
Quantum Turing machines might have remained relatively unknown
but two events caused a drastical change.
First, Peter Shor \cite{Sh 94} invented surprising polynomial-time
quantum algorithms for computation of discrete logarithms and for
factorization of integers.
Second,
unusual quantum circuits having no
classical counterparts (such as quantum bit teleportation) have
been physically implemented. Hence, there is a chance that
universal quantum computers may be built.
Moreover, since the modern public-key
cryptography is based on intractability of discrete logarithms and
factorization of integers, building a quantum computer implies
building a code-breaking machine.
In this paper, we consider quantum finite automata (QFAs),
a different model of quantum computation.
This is a simpler model than quantum Turing machines and
and it may be simpler to implement.
Quantum finite automata have been studied in \cite{AF 98,BP 99,KW 97,MC 97}.
Surprisingly, QFAs do not generalize deterministic finite automata.
Their capabilities are incomparable.
QFAs can be exponentially more space-efficient\cite{AF 98}.
However, there are regular languages that cannot be recognized by quantum
finite automata\cite{KW 97}.
This weakness is caused by reversibility.
Any quantum computation is performed by means of unitary
operators. One of the simplest properties of these operators
shows that such a computation is reversible. The result always determines
the input uniquely. It may seem to be a very strong limitation.
Luckily, for unrestricted quantum algorithms (for instance,
for quantum Turing machines) this is not so.
It is possible to embed any irreversible computation in an appropriate
environment which makes it reversible\cite{Be 89}.
For instance, the computing agent
could keep the inputs of previous calculations in successive order.
Quantum finite automata are more sensitive to the reversibility requirement.
If the probability with which a QFA is required to be correct
decreases, the set of languages that can be recognized increases.
In particular\cite{AF 98}, there are languages that can be recognized
with probability 0.68 but not with probability 7/9.
In this paper, we extend this result by constructing a hierarchy of
languages in which each next language can be recognized with
a smaller probability than the previous one.
\section{Preliminaries}
\subsection{Basics of quantum computation}
\label{basics}
To explain the difference between classical and quantum mechanical
world, we first consider one-bit systems.
A classical bit
is in one of two
classical states $true$ and $false$.
A {\em probabilistic} counterpart of the
classical bit can be $true$ with a
probability $\alpha $ and $false$ with probability $\beta $, where
$\alpha + \beta = 1$.
A {\em quantum bit (qubit)} is very much like to it
with the following distinction.
For a {\em qubit} $\alpha $ and $\beta $ can be
arbitrary complex numbers with the property
$\|\alpha \|^2 + \|\beta \|^2 = 1$.
If we observe a qubit, we get $true$ with probability
$\|\alpha\|^2$ and $false$ with probability $\|\beta\|^2$, just like
in probabilistic case.
However, if we modify a quantum system without observing it
(we will explain what this means), the set of transformations
that one can perform is larger than in the probabilistic case.
This is where the power of quantum computation comes from.
More generally, we consider quantum systems with $m$ basis states.
We denote the basis states $\ket{q_1}$, $\ket{q_2}$, $\ldots$,
$\ket{q_m}$. Let $\psi$ be a linear combination of them
with complex coefficients
\[ \psi=\alpha_1\ket{q_1}+\alpha_2\ket{q_2}+\ldots+\alpha_m\ket{q_m} .\]
The $l_2$ norm of $\psi$ is
\[ \|\psi\|=\sqrt{|\alpha_1|^2+|\alpha_2|^2+\ldots+|\alpha_m|^2}. \]
The state of a quantum system can be any $\psi$ with $\|\psi\|=1$.
$\psi$ is called a {\em superposition} of $\ket{q_1}$, $\ldots$, $\ket{q_m}$.
$\alpha_1$, $\ldots$, $\alpha_m$ are called {\em amplitudes} of
$\ket{q_1}$, $\ldots$, $\ket{q_m}$.
We use $l_2(Q)$ to denote the vector space consisting
of all linear combinations of $\ket{q_1}$, $\ldots$, $\ket{q_m}$.
Allowing arbitrary complex amplitudes is essential for physics.
However, it is not important for quantum computation.
Anything that can be computed with complex amplitudes can be done
with only real amplitudes as well.
This was shown for quantum Turing machines in \cite{BV 93}\footnote{For
unknown reason, this proof does not appear in \cite{BV 97}.}
and the same proof works for QFAs.
However, it is important that {\em negative} amplitudes are allowed.
For this reason, we assume that
all amplitudes are (possibly negative) reals.
There are two types of transformations that can be performed
on a quantum system. The first type are unitary transformations.
A unitary transformation is a linear transformation $U$ on $l_2(Q)$
that preserves $l_2$ norm. (This means that any $\psi$ with $\|\psi\|=1$
is mapped to $\psi'$ with $\|\psi'\|=1$.)
Second, there are measurements.
The simplest measurement is observing
$\psi=\alpha_1\ket{q_1}+\alpha_2\ket{q_2}+\ldots+\alpha_m\ket{q_m}$
in the basis $\ket{q_1}, \ldots, \ket{q_m}$.
It gives $\ket{q_i}$ with probability $\alpha_i^2$.
($\|\psi\|=1$ guarantees that probabilities of different outcomes sum to 1.)
After the measurement, the state of the system changes to $\ket{q_i}$
and repeating the measurement gives the same state $\ket{q_i}$.
In this paper, we also use {\em partial measurements.}
Let $Q_1, \ldots, Q_k$ be pairwise disjoint subsets of $Q$ such that
$Q_1\cup Q_2 \cup \ldots \cup Q_k=Q$.
Let $E_j$, for $j\in\{1, \ldots, k\}$, denote the
subspace of $l_2(Q)$ spanned by $\ket{q_j}$, $j\in Q_i$.
Then, a {\em partial measurement} w.r.t. $E_1, \ldots, E_k$
gives the answer $\psi\in E_j$ with probability $\sum_{i\in Q_j} \alpha_i^2$.
After that, the state of the system collapses to the projection
of $\psi$ to $E_j$.
This projection is $\psi_j= \sum_{i\in Q_j} \alpha_i \ket{q_i}$.
\subsection{Quantum finite automata}
Quantum finite automata were introduced twice. First this was done by C.
Moore and J.P.Crutchfield \cite{MC 97}. Later in a different and
non-equivalent way these automata were introduced by A. Kondacs and J.
Watrous \cite{KW 97}.
The first definition just mimics the definition of
1-way probabilistic finite automata
only substituting {\em stochastic} matrices by {\em unitary} ones.
We use a more elaborated definition \cite{KW 97}.
A QFA is a tuple
$M=(Q;\Sigma ;V ;q_{0};Q_{acc};Q_{rej})$ where $Q$ is a finite set
of states, $\Sigma $ is an input alphabet, $V$ is a transition function,
$q_{0}\in Q$ is a starting state, and $Q_{acc}\subset Q$
and $Q_{rej}\subset Q$ are sets of accepting and rejecting states.
The states in $Q_{acc}$ and $Q_{rej}$ are called {\em halting states} and
the states in $Q_{non}=Q-(Q_{acc}\cup Q_{rej})$ are called
{\em non halting states}.
$\kappa$ and $\$$ are symbols that do not belong to $\Sigma$.
We use $\kappa$ and $\$$ as the left and the right endmarker,
respectively. The {\em working alphabet} of
$M$ is $\Gamma = \Sigma \cup \{\kappa ;\$\}$.
The transition function $V$ is a mapping from $\Gamma\times l_2(Q)$
to $l_2(Q)$ such that, for every $a\in\Gamma$, the function
$V_a:l_2(Q)\rightarrowl_2(Q)$ defined by $V_a(x)=V(a, x)$ is a
unitary transformation.
The computation of a QFA starts in the superposition $|q_{0}\rangle$.
Then transformations corresponding to the left endmarker $\kappa$,
the letters of the input word $x$ and the right endmarker $\$$ are
applied. The transformation corresponding to $a\in \Gamma$ consists
of two steps.
1. First, $V_{a}$ is applied. The new superposition $\psi^{\prime}$
is $V_{a}(\psi)$ where $\psi$ is the superposition before this step.
2. Then, $\psi^{\prime}$ is observed with respect to
$E_{acc}, E_{rej}, E_{non}$ where
$E_{acc}=span\{|q\rangle:q\in Q_{acc}\}$,
$E_{rej}=span\{|q\rangle :q\in Q_{rej}\}$,
$E_{non}=span\{|q\rangle :q\in Q_{non}\}$
(see section \ref{basics}).
If we get $\psi^{\prime} \in E_{acc}$, the input is accepted.
If we get $\psi^{\prime} \in E_{rej}$, the input is rejected.
If we get $\psi^{\prime} \in E_{non}$, the next transformation is applied.
We regard these two transformations as reading a letter $a$.
We use $V'_a$ to denote the transformation consisting of
$V_a$ followed by projection to $E_{non}$.
This is the transformation mapping $\psi$ to the non-halting part
of $V_a(\psi)$.
We use $\psi_y$ to denote the non-halting part of
QFA's state after reading the left endmarker $\kappa$ and the
word $y\in\Sigma^*$.
\comment{For probabilistic computation, the probability
of correct answer can be easily increased
by executing several copies of an automaton (or Turing machine)
in parallel. Hence, if a language can be recognized by a probabilistic
Hence, it is not surprising that \cite{KW 97} wrote
"with error probability bounded away from $1/2$", thinking
that all such probabilities are equivalent.
However, mixing reversible (quantum computation) and non-reversible
(measurements after each step) components in one model
makes it impossible.}
We compare QFAs with different probabilities of correct answer.
This problem was first considered by A. Ambainis
and R. Freivalds\cite{AF 98}. The following theorems were proved there:
\begin{theorem}
\label{T1}
Let $L$ be a language and $M$ be its minimal automaton.
Assume that there is a word $x$ such that $M$
contains states $q_1$, $q_2$ satisfying:
\begin{enumerate}
\item
$q_1\neq q_2$,
\item
If $M$ starts in the state $q_1$ and reads $x$,
it passes to $q_2$,
\item
If $M$ starts in the state $q_2$ and reads $x$,
it passes to $q_2$, and
\item
$q_2$ is neither "all-accepting" state, nor "all-rejecting" state.
\end{enumerate}
Then $L$ cannot be recognized by a 1-way quantum finite automaton with
probability $7/9+\epsilon$ for any fixed $\epsilon>0$.
\end{theorem}
\begin{theorem}
\label{T2}
Let $L$ be a language and $M$ be its minimal automaton.
If there is no $q_1, q_2, x$ satisfying conditions of Theorem \ref{T1}
then $L$ can be recognized by a
1-way reversible finite automaton (i.e. $L$ can be recognized by a 1-way
quantum finite automaton with probability 1).
\end{theorem}
\begin{theorem}
\label{T4}
The language $a^{*}b^{*}$ can be recognized by a 1-way QFA
with the probability of correct answer $p=0.68...$ where
$p$ is the root of $p^3+p=1$.
\end{theorem}
\begin{corollary}
\label{C1}
There is a language that can be recognized by a 1-QFA with probability
$0.68...$ but not with probability $7/9+\epsilon$.
\end{corollary}
For probabilistic automata, the probability of correct answer can be
increased arbitrarily and this property of probabilistic computation
is considered as evident. Theorems above show thatits counterpart
is not true in the quantum world! The reason for that is that the model
of QFAs mixes reversible (quantum computation) components with
nonreversible (measurements after every step).
In this paper, we consider the best probabilities of acceptance by 1-way
quantum finite automata the languages $a^{*}b^{*}\dots z^{*}$. Since the
reason why the language $a^{*}b^{*}$ cannot be accepted by 1-way
quantum finite automata is the property described in the Theorems
\ref{T1} and \ref{T2}, this new result provides an insight on what the
hierarchy
of languages with respect to the probabilities of their acceptance by
1-way quantum finite automata may be.
We also show a generalization of Theorem \ref{T4} in a style similar
to Theorem \ref{T2}.
\section{Main results}
\begin{lemma}
\label{lemma}
For arbitrary real $x_1>0$, $x_2>0$, ..., $x_n>0$, there exists
a unitary $n\times n$ matrix $M_n(x_1,x_2,...,x_n)$ with elements $m_{ij}$
such that
$$
m_{11}=\frac{x_1}{\sqrt{x_1^2+...+x_n^2}},\
m_{21}=\frac{x_2}{\sqrt{x_1^2+...+x_n^2}},\ ...,
m_{n1}=\frac{x_n}{\sqrt{x_1^2+...+x_n^2}}.
$$
\end{lemma}
\hspace*{118mm} $\Box$
Let $L_n$ be the language $a_1^*a_2^*...a_n^*$.
\begin{theorem}
\label{part1}
The language $L_n$ ($n>1$) can be recognized by a 1-way QFA
with the probability of correct answer $p$ where $p$ is the root of
$p^\frac{n+1}{n-1}+p=1$ in the interval $[1/2, 1]$.
\end{theorem}
{\bf Proof:}
Let $m_{ij}$ be the elements of the matrix $M_k(x_1,x_2,...,x_k)$
from Lemma \ref{lemma}.
We construct a $k\times (k-1)$ matrix $T_k(x_1,x_2,...,x_k)$
with elements $t_{ij}=m_{i,j+1}$.
Let $R_k(x_1,x_2,...,x_k)$ be a $k\times k$ matrix
with elements $r_{ij}=\frac{x_i\cdot x_j}{x_1^2+...+x_k^2}$
and $I_k$ be the $k \times k$ identity matrix.
For fixed $n$,
let $p_n\in[1/2, 1]$ satisfy $p_n^\frac{n+1}{n-1}+p_n=1$
and $p_k$ ($1\leq k<n$) = $p_n^\frac{k-1}{n-1}-p_n^\frac{k}{n-1}.$
It is easy to see that $p_1+p_2+...+p_n=1$ and
\begin{equation}
1-\frac{p_n(p_k+...+p_n)^2}{(p_{k-1}+...+p_n)^2}=
1-\frac{p_np_n^\frac{2(k-1)}{n-1}}{p_n^\frac{2(k-2)}{n-1}}=
1-p_n^\frac{n+1}{n-1}=p_n.
\end{equation}
Now we describe a 1-way QFA accepting the language $L_n$.
The automaton has $2n$ states:
$q_1$, $q_2$, ... $q_{n}$ are non halting states,
$q_{n+1}$, $q_{n+2}$, ... $q_{2n-1}$ are rejecting states
and $q_{2n}$ is an accepting state.
The transition function is defined by unitary block matrices
$$V_\kappa=
\left (
\begin{array}{cc}
M_n(\sqrt{p_1},\sqrt{p_2},...,\sqrt{p_n})&{\bf 0}\\
{\bf 0}&I_n
\end{array}
\right ),
$$
$$V_{a_1}=
\left (
\begin{array}{ccc}
R_n(\sqrt{p_1},\sqrt{p_2},...,\sqrt{p_n})&
T_n(\sqrt{p_1},\sqrt{p_2},...,\sqrt{p_n})&{\bf 0}\\
T_n^T(\sqrt{p_1},\sqrt{p_2},...,\sqrt{p_n})&{\bf 0}&{\bf 0}\\
{\bf 0}&{\bf 0}&1
\end{array}
\right ),
$$
$$
V_{a_2}=
\left (
\begin{array}{ccccc}
0&{\bf 0}&1&{\bf 0}&0\\
{\bf 0}&R_{n-1}(\sqrt{p_2},...,\sqrt{p_n})&
{\bf 0}&T_{n-1}(\sqrt{p_2},...,\sqrt{p_n})&{\bf 0}\\
1&{\bf 0}&0&{\bf 0}&0\\
{\bf 0}&T_{n-1}^T(\sqrt{p_2},...,\sqrt{p_n})&
{\bf 0}&{\bf 0}&{\bf 0}\\
0&{\bf 0}&0&{\bf 0}&1
\end{array}
\right ),
$$
$$...,$$
$$
V_{a_k}=
\left (
\begin{array}{ccccc}
{\bf 0}&{\bf 0}&I_{k-1}&{\bf 0}&{\bf 0}\\
{\bf 0}&R_{n+1-k}(\sqrt{p_k},...,\sqrt{p_n})&
{\bf 0}&T_{n+1-k}(\sqrt{p_k},...,\sqrt{p_n})&{\bf 0}\\
I_{k-1}&{\bf 0}&{\bf 0}&{\bf 0}&{\bf 0}\\
{\bf 0}&T_{n+1-k}^T(\sqrt{p_k},...,\sqrt{p_n})&
{\bf 0}&{\bf 0}&{\bf 0}\\
{\bf 0}&{\bf 0}&{\bf 0}&{\bf 0}&1
\end{array}
\right ),
$$
$$...,$$
$$
V_{a_n}=
\left (
\begin{array}{cccc}
{\bf 0}&{\bf 0}&I_{n-1}&{\bf 0}\\
{\bf 0}&1&{\bf 0}&{\bf 0}\\
I_{n-1}&{\bf 0}&{\bf 0}&{\bf 0}\\
{\bf 0}&{\bf 0}&{\bf 0}&1
\end{array}
\right ),
$$
$$
V_\$=
\left (
\begin{array}{cc}
{\bf 0}&I_n\\
I_n&{\bf 0}
\end{array}
\right ).
$$
{\it Case 1.} The input is $\kappa a_1^*a_2^*...a_n^*\$$.
The starting superposition is $\ket{q_1}$.
After reading the left endmarker
the superposition becomes
$\sqrt{p_1}\ket{q_1}+\sqrt{p_2}\ket{q_2}+\ldots+\sqrt{p_n}\ket{q_n}$
and after reading $a_1^*$ the superposition remains the same.
If the input contains $a_k$ then
reading the first $a_k$ changes the non-halting part of the superposition to
$\sqrt{p_k}\ket{q_k}+\ldots+\sqrt{p_{n}}\ket{q_{n}}$
and after reading all the rest of $a_k$
the non-halting part of the superposition remains the same.
Reading the right endmarker maps $\ket{q_{n}}$ to $\ket{q_{2n}}$.
Therefore, the superposition after reading it contains
$\sqrt{p_n}\ket{q_{2n}}$.
This means that the automaton accepts with probability $p_n$
because $q_{2n}$ is an accepting state.\\
{\it Case 2.} The input is $\kappa a_1^*a_2^*...a_k^*a_ka_m...\ (k>m)$.
After reading the last $a_k$ the non-halting part of the superposition is
$\sqrt{p_k}\ket{q_k}$\\ $+\ldots+\sqrt{p_{n}}\ket{q_{n}}$.
Then reading $a_m$
changes the non-halting part to\\
$\frac{\sqrt{p_m}(p_k+...+p_n)}{(p_m+...+p_n)}\ket{q_m}+\ldots+
\frac{\sqrt{p_n}(p_k+...+p_n)}{(p_m+...+p_n)}\ket{q_n}.$
This means that the automaton accepts with probability
$\leq \frac{p_n(p_k+...+p_n)^2}{(p_m+...+p_n)^2}$ and rejects
with probability at least
$$1-\frac{p_n(p_k+...+p_n)^2}{(p_m+...+p_n)^2}
\geq 1-\frac{p_n(p_k+...+p_n)^2}{(p_{k-1}+...+p_n)^2}=p_n$$
that follows from (1).
$\Box$
\begin{corollary}
\label{cor1}
The language $L_n$ can be recognized by a 1-way QFA with
the probability of correct answer at least $\frac{1}{2}+\frac{c}{n}$,
for a constant $c$.
\end{corollary}
\noindent
{\bf Proof:}
By resolving the equation $p^{\frac{n+1}{n-1}}+p=1$, we get
$p=\frac{1}{2}+\Theta(\frac{1}{n})$.
$\Box$
\begin{theorem}
\label{part2}
The language $L_n$ cannot be recognized by a 1-way QFA with probability
greater than $p$ where $p$ is the root of
\begin{equation}
\label{e1}
(2p-1)=\frac{2(1-p)}{n-1}+4\sqrt{\frac{2(1-p)}{n-1}}
\end{equation}
in the interval $[1/2, 1]$.
\end{theorem}
\noindent
{\bf Proof:}
Assume we are given a 1-way QFA $M$.
We show that, for any $\epsilon>0$, there is a word
such that the probability of correct answer is less than $p+\epsilon$.
\begin{lemma}
\cite{AF 98}
\label{LT1}
Let $x\in \Sigma^{+}$.
There are subspaces $E_1$, $E_2$ such that $E_{non}=E_1\oplus E_2$ and
\begin{enumerate}
\item[(i)]
If $\psi\in E_1$, then $V_x(\psi)\in E_1$,
\item[(ii)]
If $\psi\in E_2$, then $\| V'_{x^k}(\psi)\|\rightarrow 0$ when
$k\rightarrow\infty$.
\end{enumerate}
\end{lemma}
We use $n-1$ such decompositions: for $x=a_2$,
$x=a_3$, $\ldots$, $x=a_n$.
The subspaces $E_1$, $E_2$ corresponding to $x=a_m$ are
denoted $E_{m, 1}$ and $E_{m, 2}$.
Let $m\in\{2, \ldots, n\}$, $y\in a_1^* a_2^* \ldots a_{m-1}^*$.
Remember that $\psi_{y}$ denotes the superposition after reading $y$
(with observations w.r.t. $E_{non}\oplus E_{acc}\oplus E_{rej}$
after every step).
We express $\psi_y$ as $\psi^1_{y}+\psi^2_{y}$, $\psi^1_{y}\in E_{m, 1}$,
$\psi^2_{y}\in E_{m, 2}$.
{\em Case 1.}
$\|\psi^2_{y}\|\leq \sqrt{\frac{2(1-p)}{n-1}}$
for some $m\in\{2, \ldots, n\}$ and $y\in a_1^* \ldots a_{m-1}^*$.
Let $i>0$.
Then, $y a_{m-1}\in L_n$ but $y a_m^i a_{m-1}\notin L_n$.
Consider the distributions of probabilities on $M$'s answers ``accept"
and ``reject" on $y a_{m-1}$ and $y a_m^i a_{m-1}$.
If $M$ recognizes $L_n$ with probability $p+\epsilon$,
it must accept $y a_{m-1}$ with probability at least $p+\epsilon$
and reject it with probability at most $1-p-\epsilon$.
Also, $y a_m^i a_{m-1}$ must be rejected with
probability at least $p+\epsilon$
and accepted with probability at most $1-p-\epsilon$.
Therefore, both the probabilities of accepting and the probabilities of
rejecting must differ by at least
\[ (p+\epsilon)-(1-p-\epsilon)=2p-1+2\epsilon .\]
This means that the {\em variational distance} between two probability
distributions (the sum of these two distances) must be
at least $2(2p-1)+4\epsilon$.
We show that it cannot be so large.
First, we select an appropriate $i$.
Let $k$ be so large that $\|V'_{a_m^k}(\psi^2_{y})\|\leq \delta$ for
$\delta=\epsilon/4$.
$\psi^1_{y}, V'_{a_m}(\psi^1_{y}), V'_{a_m^2}(\psi^1_{y})$, $\ldots$
is a bounded sequence in a finite-dimensional space. Therefore, it
has a limit point and there are $i, j$ such that
\[ \|V'_{a_m^j}(\psi^1_{y})-V'_{a_m^{i+j}}(\psi^1_{y})\|<\delta.\]
We choose $i, j$ so that $i>k$.
The difference between the two probability distributions comes from two sources.
The first source is the difference between $\psi_{y}$ and
$\psi_{y a_m^i}$ (the states of $M$ before reading $a_{m-1}$).
The second source is the possibility of $M$ accepting while
reading $a_m^i$ (the only part that is different in the two words).
We bound each of them.
The difference $\psi_{y}-\psi_{y a_{m}^i}$
can be partitioned into three parts.
\begin{equation}
\label{eq5}
\psi_{y}-\psi_{y a_{m}^i}=(\psi_{y}-\psi^1_{y})+
(\psi^1_{y}-V'_{a_m^i}(\psi^1_{y}))+
(V'_{a_m^i}(\psi^1_{y})-\psi_{y a_{m}^i}).
\end{equation}
The first part is
$\psi_{y}-\psi^1_{y}=\psi^2_{y}$ and
$\|\psi^2_{y}\|\leq\sqrt{\frac{2(1-p)}{n-1}}$.
The second and the third parts are both small.
For the second part, notice that $V'_{a_m}$ is unitary
on $E_{m, 1}$ (because $V_{a_m}$ is unitary and $V_{a_m}(\psi)$ does not
contain halting components for $\psi\in E_{m, 1}$).
Hence, $V'_{a_m}$ preserves distances on $E_{m, 1}$ and
\[ \|\psi^1_{y}-V'_{a_m^i}(\psi^1_{y})\|=
\|V'_{a_m^j}(\psi^1_{y})-V'_{a_m^{i+j}}(\psi^1_{y}) \| <\delta \]
For the third part of (\ref{eq5}), remember that
$\psi_{y a_m^i}=V'_{a_m^i}(\psi_y)$.
Therefore,
\[ \psi_{y a_{m}^i}-V'_{a_m^i}(\psi^1_y) = V'_{a_m^i}(\psi_y) -
V'_{a_m^i}(\psi^1_y) = V'_{a_m^i}(\psi_y-\psi^1_y) = V'_{a_m^i}(\psi^2_y) \]
and $\|\psi^2_{y a_{m}^i}\|\leq \delta$ because $i>k$.
Putting all three parts together, we get
\[ \|\psi_{y}-\psi_{y a_{m}^i}\|\leq \|\psi_{y}-\psi^1_{y}\|+
\|\psi^1_{y}-\psi^1_{y a_{m}^i}\|+
\|\psi^1_{y a_{m}^i}-\psi_{y a_{m}^i}\|\leq
\sqrt{\frac{2(1-p)}{n-1}}+ 2\delta.\]
\begin{lemma}
\cite{BV 97}
\label{BVTheorem}
Let $\psi$ and $\phi$ be such that $\|\psi\|\leq 1$, $\|\phi\|\leq 1$ and
$\|\psi-\phi\|\leq\epsilon$.
Then the total variational distance resulting from measurements
of $\phi$ and $\psi$ is at most $4\epsilon$.
\end{lemma}
This means that the difference between any probability distributions
generated by $\psi_{y}$ and $\psi_{y a^i_{m}}$ is at most
\[ 4\sqrt{\frac{2(1-p)}{n-1}}+ 8\delta.\]
In particular, this is true for the probability distributions obtained
by applying $V_{a_{m-1}}$, $V_{\$}$ and the corresponding measurements
to $\psi_{y}$ and $\psi_{y a_m^i}$.
The probability of $M$ halting while reading $a_m^i$ is at most
$\|\psi^2_{\kappa}\|^2=\frac{2(1-p)}{n-1}$. Adding it increases the
variational distance by at most $\frac{2(1-p)}{n-1}$.
Hence, the total variational distance is at most
\[ \frac{2(1-p)}{n-1}+4\sqrt{\frac{2(1-p)}{n-1}}+ 8\delta=
\frac{2(1-p)}{n-1}+4\sqrt{\frac{2(1-p)}{n-1}}+ 2\epsilon .\]
By definition of $p$, this is the same as $(2p-1)+2\epsilon$.
However, if $M$ distinguishes $y$ and $y a_m^i$ correctly,
the variational distance must be at least $(2p-1)+4\epsilon$.
Hence, $M$ does not recognize one of these words correctly.
{\em Case 2.}
$\|\psi^2_{y}\|> \sqrt{\frac{2(1-p)}{n-1}}$
for every $m\in\{2, \ldots, n\}$ and $y\in a_1^* \ldots a_{m-1}^*$.
We define a sequence of words $y_1, y_2, \ldots, y_m\in a_1^* \ldots a_n^*$.
Let $y_1=a_1$ and $y_{k}=y_{k-1} a_k^{i_k}$ for $k\in\{2, \ldots, n\}$ where
$i_k$ is such that
\[ \|V'_{a_k^{i_k}}(\psi^2_{y_{k-1}})\|
\leq \sqrt{\frac{\epsilon}{n-1}}.\]
The existence of $i_k$ is guaranteed by (ii) of Lemma \ref{LT1}.
We consider the probability that $M$ halts on
$y_n=a_1 a_2^{i_2} a_3^{i_3} \ldots a_n^{i_n}$
before seeing the right endmarker.
Let $k\in\{2, \ldots, n\}$.
The probability of $M$ halting while reading the $a_k^{i_k}$ part
of $y_n$ is at least
\[ \| \psi^2_{y_{k-1}}\|^2 -
\| V'_{a_k^{i_k}}(\psi^2_{y_{k-1}}) \|^2
> \frac{2(1-p)}{n-1}-\frac{\epsilon}{n-1} .\]
By summing over all $k\in\{2, \ldots, n\}$, the probability
that $M$ halts on $y_n$ is at least
\[ (n-1)\left(\frac{2(1-p)}{n-1}- \frac{\epsilon}{n-1}\right)=
2(1-p)-\epsilon .\]
This is the sum of the probability of accepting and the probability
of rejecting. Hence, one of these two probabilities must be at least
$(1-p)-\epsilon/2$. Then, the probability of the opposite
answer on any extension of $y_n$ is at most $1-(1-p-\epsilon/2)=p+\epsilon/2$.
However, $y_n$ has both extensions that are in $L_n$ and extensions that are not.
Hence, one of them is not recognized with probability $p+\epsilon$.
$\Box$
By solving the equation (\ref{e1}), we get
\begin{corollary}
\label{cor2}
$L_n$ cannot be recognized with probability greater than
$\frac{1}{2}+\frac{3}{\sqrt{n-1}}$.
\end{corollary}
\noindent
{\bf Proof:}
The right-hand side of (\ref{e1}) is at most
$\frac{1}{n-1}+4\sqrt{\frac{1}{n-1}}$ because $p\geq 1/2$ and,
hence, $1-p\leq 1/2$. This implies
\[ 2p-1 \leq \frac{1}{n-1}+4\sqrt{\frac{1}{n-1}}, \]
\[ p\leq \frac{1}{2}+2\sqrt{\frac{1}{n-1}}+\frac{1}{2(n-1)}\leq
\frac{1}{2}+3\sqrt{\frac{1}{n-1}} \]
and $L_n$ cannot be recognized with probability greater than $p$
by Theorem \ref{part2}.
$\Box$
Let $n_1=2$ and $n_k=\frac{9n_{k-1}^2}{c^2}+1$ for $k>1$
(where $c$ is the constant from Theorem \ref{part1}).
Also, define $p_k=\frac{1}{2}+\frac{c}{n_k}$.
Then, Corollaries \ref{cor1} and \ref{cor2} imply
\begin{theorem}
For every $k>1$,
$L_{n_k}$ can be recognized with by a 1-way QFA with the probability
of correct answer $p_{k}$ but cannot be recognized with the
probability of correct answer $p_{k-1}$.
\end{theorem}
\noindent
{\bf Proof:}
By Corollary \ref{cor1}, $L_{n_k}$ can be recognized with
probability $\frac{1}{2}+\frac{c}{n_k}=p_k$.
On the other hand, by Corollary \ref{cor2}, $L_{n_k}$ cannot be
recognized with probability $\frac{1}{2}+\frac{3}{\sqrt{n_k-1}}$.
The definition of $n_k$ implies
$n_k-1=\frac{9n_{k-1}^2}{c^2}$, $\sqrt{n_k-1}=\frac{3 n_{k-1}}{c}$,
\[ \frac{1}{2}+\frac{3}{\sqrt{n_k-1}}=\frac{1}{2}+\frac{c}{n_{k-1}}=p_{k-1} .\]
$\Box$
Thus, we have constructed a sequence of languages $L_{n_1}$, $L_{n_2}$, $\ldots$
such that, for each $L_{n_k}$, the probability with which $L_{n_k}$
can be recognized by a 1-way QFA is smaller than for $L_{n_{k-1}}$.
Our final theorem is a counterpart of Theorem
\ref{T2}. It generalizes Theorem \ref{T4}.
\begin{theorem}
\label{T9}
Let $L$ be a language and $M$ be its minimal automaton.
If there is no $q_1, q_2, q_3, x, y$ such that
\begin{enumerate}
\item~
the states $q_1 , q_2 , q_3$ are pairwise different,
\item
If $M$ starts in the state $q_1$ and reads $x$,
it passes to $q_2$,
\item
If $M$ starts in the state $q_2$ and reads $x$,
it passes to $q_2$, and
\item
If $M$ starts in the state $q_2$ and reads $y$,
it passes to $q_3$,
\item
If $M$ starts in the state $q_3$ and reads $y$,
it passes to $q_3$,
\item
both $q_2$ and $q_3$ are neither "all-accepting" state, nor
"all-rejecting" state,
\end{enumerate}
then $L$ can be recognized by
a 1-way quantum finite automaton with probability $p=0.68... $.
\end{theorem}
|
\section{Introduction}
Recently a fascinating conjecture by Maldacena \cite{Malda} has been
proposed. According to \cite{Malda} the
supergravity theory in d+1 dimensional Anti-de Sitter space (AdS) with a
compact extra space is related by holographic correspondence principle
\cite{Sus} to a certain conformal field theory (CFT) living on the boundary
of AdS space. The underlying principle behind this AdS/CFT
correspondence was elaborated in explicit form by Gubser, Klebanov and
Polyakov \cite{gkp} and Witten \cite{w}. According to \cite{gkp} and
\cite{w}, the action for the supergravity theory on AdS considered as a
functional of the asymptotic values of the fields on the boundary is
interpreted as a generating functional for the correlation functions in the
conformal field theory living on the boundary. The explicit form of this
interpretation is:
\begin{equation}
\int\limits_{\Phi}{\mathcal D}\Phi exp\{-S\[\Phi\]\}=
\langle exp\int\limits_{\partial AdS}d^d{\mathbf x}{\mathcal O}\Phi_0\rangle
\end{equation}
where $\Phi_0$ is the boundary data for AdS theory which couples to a
certain conformal operator $\mathcal O$ on the boundary. This interpretation
has already a large number of examinations by computing various correlation
functions of a local operators in CFT induced by AdS scalar fields
\cite{w,4,5,6,7}, spinor fields \cite{HS,8}, vector fields \cite{w,6,8,9},
antisymmetric fields \cite{14,15,16} and Rarita-Schwinger fields
\cite{C,A,kr}. All the examples confirmed the validity of the AdS/CFT
correspondence principle. The essence of these examinations is in
studying of the field behavior near the boundary of AdS space and
calculation of the boundary terms which produce the corresponding
correlation functions in CFT. While in the case of an action of AdS theory
with derivatives of order higher than one the considerations are, more or less,
transparent, in the case of Dirac-like actions (Dirac and Rarita-Schwinger)
the situation is more subtle. That is because the naive limit to the boundary
lead to vanishing of the action (it is zero on shell) which obviously
spoil the correspondence principle.
Recently two different but equivalent ways of treatment of the boundary term
problem in the case of spinor field were proposed \cite{ar,H}. In \cite{ar}
the considerations are based on the Hamiltonian approach treating $x^0$ as
an evolution parameter. A half of the components of the boundary spinor
turn out to be cannonically conjugated to the other half and the
boundary term naturally appears to be $p\dot q $.
In the paper \cite{H} another approach is used, namely the stationary phase
method. It is based on the fact that one can expand the path integral of a
given theory with action $S[\Phi]$:
$$
{\mathbf Z}=\int{\mathcal D}\Phi e^{\frac{i}{\hbar} S[\Phi]}
$$
near the stationary point given by the solutions of the classical theory. In
the classical limit ($h\to 0$) the leading order is simply:
$$
{\mathbf Z}\sim exp\{\frac{i}{\hbar} S_{class}\}
$$
where $S_{class}$ is the action functional evaluated on the classical
field configurations. The stationary point is determined by the requirement:
\begin{equation}
\frac{\delta S}{\delta\Phi}=0
\label{a}
\end{equation}
which is nothing but the classical equations of motion with
given boundary data. The main immediate but important observation is that while
the action $S[\Phi]$ may satisfy (\ref{a}) on the clasical field
configurations, it is not true in general if there is a surface term
$\mathcal{B}_\infty$, i.e. there may be $\delta\mathcal{B}_\infty\neq 0$ and therefore
$\delta(S+\mathcal{B}_\infty)\neq 0$. Such a situation appears for instance in
the gauge field theory where $\mathcal{B}_\infty$ gives the conserved charges of the
symmetry currents. It follows that in order the classical configurations to
be a true stationary point it is necessary a boundary term to be added
ensuring the wanted requirement. This scheme was used in \cite{H} in order
to obtain the boundary term in the case of spinor field $\Psi$.
In this letter we would like to study the boundary term in the case of
massless and massive Rarita-Schwunger field. In our analysis we will use
the receipt described in \cite{H} for derivation of the boundary term in
the spinor case.
The paper is organized as follows. In Section 2 the derivation of the
boundary term in case of massless Rarita-Schwinger field is given. It is
shown that the result reproduce exactly the two-point correlation functions
derived in \cite{A}. In Section 3 the same analysis is performed in
case of massive Rrita-Schwinger field. The result for the correlation
functions is in complete agreement with that given in \cite{kr}. In the
Conclusions we give some remarks and brief comments.
\section{Boundary term for massless Rarita-Schwinger field}
In this Section we will consider a massles Rarita-Schwinger field on
$AdS_{d+1}$ given by the following action \cite{N,A}:
\begin{equation}
S_{R-S}=\int\limits_{AdS}d^{d+1}x\sqrt{G}\bar\Psi_\mu\(\Gamma^{\mu\nu\lambda}
D_\nu-m\Gamma^{\mu\nu}\)\Psi_\lambda
\label{1}
\end{equation}
We used the notations:
\begin{align}
&\Gamma^{\mu\nu\lambda}=\Gamma^{\[\mu\right.}\Gamma^\nu\Gamma^{\left.\lambda\]},\quad
\Gamma^{\mu\nu}=\Gamma^{\[\mu\right.}\Gamma^{\left.\nu\]}\\ \notag
\label{1'}&\Psi_{\mu}=\Psi_{\mu \alpha}
\end{align}
where $\Gamma^\mu$ are the gamma matrices in AdS connected to the flat gamma
matrices by the relation $\Gamma^\mu=e^\mu_a\gamma^a$. The vielbein is given below
and the flat gamma matrices satisfy the usual anticommutation relations
$\{\gamma^a,\gamma^b\}=2\delta^{ab}$. We choose to work in coordinates
$x^a=(x^0,x^i)=(x^0,{\mathbf x}); i=1,\dots d$ defining $d+1$-dimensional
Euclidean Anti-de Sitter space as Lobachevski upper half plane $x^0>0$ with a
metric of the form:
\begin{equation}
ds^2=\frac{1}{x_0^2}\(dx^0+d{\mathbf x}^2\)
\label{2}
\end{equation}
With this choice the vielbein and the corresponding non-zero components
of the spin connection are given by the expressions:
\begin{equation}
e^\mu_a=\frac{\delta^a_\mu}{x^0};\quad \omega_i^{0j}=-\omega_i^{j0}=
\frac{\delta^j_i}{x^0};\quad a=0,\dots d
\label{3}
\end{equation}
The boundary of the AdS space consists in a hypersurface $x^0=0$ and a
single point $x^0=\infty$.
In this frame the covariant derivative and the Dirac operator reads off:
\begin{equation}
D_\nu=\partial_\nu+\frac{1}{2x^0}\gamma_{0\nu};\quad \Gamma^\mu D_\mu=\,\,{\raise.15ex\hbox{/}\mkern-12mu D}=x_0\gamma^0\partial_0+
x_0{\mathbf\gamma}.{\mathbf\nabla}-\frac d2\gamma^0
\label{4}
\end{equation}
where ${\mathbf\gamma}=(\gamma^i);\,\, {\mathbf\nabla}=(\partial_i);\,\, i=1\dots d$.
The equation of motion for Rarita Schwinger field following from the
action (\ref{1}):
\begin{equation}
\Gamma^{\mu\nu\lambda}D_\nu\Psi_\lambda=m\Gamma^{\mu\nu}\Psi_\lambda
\label{5}
\end{equation}
can be rewitten in the form \cite{A}:
\begin{align}
&x_0\gamma^\nu\partial_\nu\psi_0-\(\frac d2+1\)\gamma_0\psi_0 +m\psi_0=0\\ \notag
\label{6}&x_0\gamma^\nu\partial_\nu\psi_i-\frac d2\gamma_0\psi_i+m\psi_i=\gamma_a\psi_0
\end{align}
where $\psi_a=e^\mu_a\Psi_\mu$.
In order to find the boundary contributions we are interested in studying
of the behavior of the solutions near the boundary $x^0=0$. For this purpose
one can use the Frobenius procedure looking for solutions of the form \cite{H}:
$$
\(x^0\)^\rho\sum\limits_{n=0}^\infty c_n^a\(\vec{\mathbf x}\)\(x^0\)^n
$$
Subsitution of these series into the equations of motion determines the
values of the parameter $\rho$:
$$
\rho=\frac d2\pm m+\delta_{a0}
$$
(the values of $\rho$ turn out to be the same as in the spinor case
\cite{H} for $a\neq0$). Therefore, we are dealing with two types of
solutions:
\begin{equation}
\psi_a^-\(x_0,{\mathbf x}\)=\(x_0\)^{\frac d2-m+\delta_{a0}}\varphi_a\({\mathbf x}\)+
o\(\(x_0\)^{\frac d2-m+\delta_{a0}}\)
\label{7}
\end{equation}
\begin{equation}
\psi_a^+\(x_0,{\mathbf x}\)=\(x_0\)^{\frac d2+m+\delta_{a0}}\chi_a\({\mathbf x}\)+
o\(\(x_0\)^{\frac d2+m+\delta_{a0}}\)
\label{8}
\end{equation}
where:
\begin{align}
&\gamma^0\varphi_a\({{\mathbf x}}\)=-\varphi\({{\mathbf x}}\)\notag\\
\label{9}&\gamma^0\chi_a\({{\mathbf x}}\)=\chi\({{\mathbf x}}\)
\end{align}
It follows that $\psi^{\pm}_a$ are eigenvectors of $1/2(I\pm\gamma_0)$ with
eigenvalues $\pm 1$ respectively. The conjugated Rarita-Schwinger field
can be treated analogously giving the following result:
\begin{align}
\label{10}&\bar\psi^+_a\({\mathbf x}\)=\bar\varphi_a\({\mathbf x}\)\(x_0\)^{\frac d2+m+\delta_{a0}}+
o\(\(x_0\)^{\frac d2+m+\delta_{a0}}\)\\
\label{11}&\bar\psi^-_a\({\mathbf x}\)=\bar\chi_a\({\mathbf x}\)\(x_0\)^{\frac d2-m+\delta_{a0}}+
o\(\(x_0\)^{\frac d2-m+\delta_{a0}}\)
\end{align}
where the fields $\bar\varphi_a$ and $\bar\chi_a$ are subject to the
constraints:
$$
\bar\varphi_a\gamma^0=\bar\varphi_a;\quad \bar\chi_a\gamma^0=-\bar\chi_a
$$
Note that using the equations of motion one can find the subleading
terms recursively but since they doesnt contribute to the boundary we
will skip their explicit form.
Let us consider the general solutions of the equations of motion:
$$
\psi_a\(x_0,]bx\)=\int d^d{\mathbf p} e^{i{\mathbf p}.{\mathbf x}}\[F_+\(x_0,{\mathbf p}\)\tilde\varphi_a^+\({\mathbf p}\)
+F_-\(x_0,{\mathbf p}\)\tilde\varphi_a^-\({\mathbf p}\)\]
$$
Since the solutions must be regular in the bulk (up to
$x_0=\infty$) the boundary spinor fields $\tilde\varphi^\pm_a$ are not independent
and the solutions can be expressed in terms of $\tilde\varphi^-_a$ only
\footnote{In what follows we will always suppose that the Fourier transform
is well defined, i.e. $\varphi_a\({\mathbf x}\)$ and $\bar\varphi_a\({\mathbf x}\)$ are of
compact support and vanish at ${\mathbf x}\to\infty$ and therefore can be
Fourier transformed.}\cite{A}:
\begin{equation}
\psi_0\(x_0,{\mathbf x}\)=\int d^dpe^{i{\mathbf p}
.{\mathbf x}}\(x_0p\)^{\frac{d+3}{2}}\[i\frac{\hat p}{p}{\mathcal K}_{m+\frac 12}\(x_0p\)+
{\mathcal K}_{m-\frac 12}\(x_0p\)\]\tilde\varphi^-_0\({\mathbf p}\)
\label{12}
\end{equation}
\begin{multline}
\psi_i\(x_0,{\mathbf x}\)=\int d^dp\(x_0p\)^{\frac{d+1}{2}}
\{\[i\frac{\hat p}{p}{\mathcal K}_{m+\frac 12}\(x_0p\)+
{\mathcal K}_{m-\frac 12}\(x_0p\)\]\tilde\varphi^-_i\({\mathbf p}\)\\
+\[\(\(2m+1\)\frac{p_i\hat p}{p^2}-ip_ix_0+\gamma_i\){\mathcal K}_{m+\frac 12}
\(x_0p\)-\frac{p_i\hat p}{p}x_0{\mathcal K}_{m-\frac 12}\(x_0p\)\]\tilde\varphi^-_0
\({\mathbf p}\)\}
\label{13}
\end{multline}
($\hat p=\gamma^ip_i$).In order to see how the fields $\tilde\varphi^-_a$ are
related to the fields $\varphi^-_a$ (which gives the asymptotic of $\psi_a$
at $x_0\to 0$) we use the small argument expansion of the modified
Bessel function ${\mathcal K}_\nu$:
\begin{equation}
{\mathcal K}_\nu\(z\)=\frac 12\[\(\frac z2\)^{-\nu}\Gamma(\nu)\[1+\dots\]+\(\frac
z2\)^\nu\Gamma(-\nu)\[1+\dots\]\]
\label{14}
\end{equation}
where dots stands for positive integer powers of $z^2$. Subsitution of
(\ref{14}) into (\ref{12},\ref{13}) gives the following behavior:
\begin{align}
&\psi_0^+\(x_0,{\mathbf x}\)=x_0^{\frac d2-m+1}\int d^dpe^{i{\mathbf p} .{\mathbf x}}\[i\hat p
p^{\frac d2-m}2^{m-\frac 12}\Gamma\(m+\frac 12\)\]\tilde\varphi^-_0\({\mathbf p}\)+
o\(\(x_0\)^{\frac d2-m+1}\)\notag \\
\label{15}&\psi_0^-\(x_0,{\mathbf x}\)=x_0^{\frac d2+m+1}\int
d^dpe^{i{\mathbf p} .{\mathbf x}}\[\frac{
p^{\frac d2+m+1}\Gamma\(\frac 12-m\)}{2^{m+\frac 12}}\]\tilde\varphi^-_0\({\mathbf p}\)+
o\(\(x_0\)^{\frac d2+m+1}\)
\end{align}
\begin{multline}
\psi^-_i\(x_0,{\mathbf x}\)=
x_0^{\frac d2-m}\int d^dpe^{i{\mathbf p} .{\mathbf x}}\{\[i\hat p
p^{\frac d2-m-1}2^{m-\frac 12}\Gamma\(m+\frac 12\)\]\tilde\varphi^-_i\({\mathbf p}\)
\\
+\[p^{\frac d2-m}2^{m-\frac 12}\Gamma\(m+\frac 12\)\(\(2m+1\)\frac{p_i\hat
p}{p^2}+\gamma_i\)\]\}\tilde\varphi^-_0\({\mathbf p}\)+o\(\(x_0\)^{\frac d2-m}\)
\label{16}
\end{multline}
\begin{equation}
\psi^+_i\(x_0,{\mathbf x}\)=x_0^{\frac d2+m}\int d^dpe^{i{\mathbf p} .{\mathbf x}}\frac{p^{\frac
d2+m}\Gamma\(\frac 12-m\)}{2^{m+\frac 12}}\tilde\varphi^-_i\({\mathbf p}\)+
o\(\(x_0\)^{\frac d2+m}\)
\label{16'}
\end{equation}
The components of the Rarita-Schwinger field are subject to one more
constraint:
$$
\gamma^0\psi_0+\gamma^i\psi_i=0
$$
which relate the $\tilde\varphi_0^-$ and $\tilde\varphi_i^-$ components as
follows \cite{A}:
\begin{equation}
\tilde\varphi_0^-\({\mathbf p}\)=-\frac{2ip_j\tilde\varphi_j^-\({\mathbf p}\)}{\(2m+d-1\)p};
\quad \gamma^i\tilde\varphi_i^-\({\mathbf p}\)=0
\label{19}
\end{equation}
Using the asymptotic expressions for $\psi_a$ (\ref{7},\ref{8}) it is
easy to find the relation between $\varphi_a\({\mathbf p}\)$ and $\chi_a\({\mathbf p}\)$ and
$\tilde\varphi^-_a\({\mathbf p}\)$:
\begin{align}
&\varphi_0\({\mathbf p}\)=i\hat pp^{\frac d2-m}2^{m-\frac 12}\Gamma\(\frac 12+m\)
\tilde\varphi_0^-\({\mathbf p}\)
\notag\\
\label{17}&\chi_0\({\mathbf p}\)=\frac{p^{\frac d2+m+1}}{2^{m+\frac 12}}\Gamma\(\frac 12-m\)
\tilde\varphi^-_0\({\mathbf p}\)
\end{align}
and:
\begin{align}
&\varphi_i\({\mathbf p}\)=p^{\frac d2-m}2^{m-\frac 12}\Gamma\(m+\frac 12\)\[i\frac{\hat
p}{p}\tilde\varphi_i^-+\(\(2m+1\)\frac{p_i\hat p}{p^2}+\gamma_i\)\tilde\varphi^-_0\]
\notag\\
\label{18}& \chi_i\({\mathbf p}\)=\frac{p^{\frac d2+m}\Gamma\(\frac 12-m\)}{2^{m+\frac 12}}
\tilde\varphi^-_i\({\mathbf p}\)
\end{align}
From (\ref{15},\ref{16}, \ref{19}) it follows that $\psi^\pm_a$ can be
expressed in terms of $\varphi^+_a\({\mathbf p}\)$ only and that $\chi_a$ and $\varphi_a$
are related "on-shell".
The final lesson from the above considerations is that a half of the
boundary data is expressible in terms of the other half but the
relations are valid "on-shell". The main conclusion is that the general
solutions of the equations of motion in the whole AdS space are
determined by the fields annihilated by $(I+\gamma_0)$, a quite similar
result as in case of spinor field \cite{H}.
Now we are going to apply the variational principle to the
Rarira-Schwinger action (\ref{1}). Since the Rarira-Schwinger equations
of motion are first order differential equations we cannot fix all the
components of $\psi_a$ at the boundary but only a half of them, $\varphi_a$ or
$\chi_a$. The basic idea is to use AdS correspondence principle which
tells us that the fields $\varphi_a$ serve as a sources for the
bulk-boundary Green functions \cite{HS,C,A}. Thus, it is appropriate to fix
$\varphi_a$ at the boundary and to leave $\chi_a$ to vary. The variational
principle will be applied to all configurations of the form:
\begin{align}
&\psi_a=\psi^-_a+\psi^+_a\notag\\
\label{20}&\bar\psi_a=\bar\psi^-_a+\bar\psi^+_a
\end{align}
The fields $\psi^-_a$ and $\bar\psi^+_a$ have the asymptotic (\ref{7},
\ref{10}) while $\varphi_a$ and $\bar\varphi_a$ are fixed on the boundary. The
other part of $\psi^+_a$ and $\bar\psi^-_a$ behaves near the boundary as
it is described in (\ref{8},\ref{11}), but in this case the values of
$\chi_a$ and $\bar\chi_a$ on the boundary are free to vary. The relations between
$\varphi_a$ and $\chi_a$ (\ref{17}, \ref{18}) are only on-shell and will not
affect the variarional principle (the same is true for $\bar\varphi_a$ and
$\bar\chi_a$).
After the above preparations we are ready to vary the action (\ref{1})
with respect to $\psi_a$ and $\bar\psi_a$. As in \cite{H}, the variation
will be in the class of fields defined in (\ref{20}) but varying (\ref{1})
we will take into account the surface terms:
\begin{equation}
\delta S_{R-S}=B_\infty+\[\,\,0\,\,\]_{on-shell}
\label{21}
\end{equation}
where:
\begin{multline}
B_\infty=-\frac 12\int d^d{\mathbf x}\[\bar\varphi_i\({\mathbf x}\)g^{ij}\delta\chi_j\({\mathbf x}\)+
\delta\bar\chi_i\({\mathbf x}\)g^{ij}\varphi_j\({\mathbf x}\)+
\bar\varphi_i\({\mathbf x}\)\gamma^i\gamma^j\delta\chi_j\({\mathbf x}\)\right.\\
\left.+\delta\bar\chi_i\({\mathbf x}\)\gamma^i\gamma^j\varphi_j\({\mathbf x}\)\]
\label{22}
\end{multline}
(we recall that $\sqrt{G}=\(x_0\)^{-d-1}$ and $g^{ij}$ is the induced
metric on the boundary).
The term $B_\infty$ is nothing but the variation of the surface term at
infinity \cite{H}:
\begin{equation}
B_\infty=-\delta C_\infty
\label{23}
\end{equation}
where:
\begin{equation}
C_\infty=\frac 12\int d^d{\mathbf x}\[\bar\varphi_ig^{ij}\chi_j+\bar\chi_ig^{ij}\varphi_j+
\bar\varphi_i\gamma^i\gamma^j\chi_j+\bar\chi_i\gamma^i\gamma^j\varphi_j\]
\label{24}
\end{equation}
Note that since $\gamma^i\chi_i=0$ the last two terms don't contribute. The
requirement for the action $S_{R-S}$ to be stationary on the solutions of
the equations of motion imposes to consider a new, improved action of
the form:
\begin{equation}
S=S_{R-S}+C_\infty
\label{25}
\end{equation}
It is obvious that $\delta S=0$ on-shell and reproduce the correct solutions
of the equations of motion.
Of course, the above boundary term is not unique. This can be achieved
by imposing three natural conditions, namely:
a) $C_\infty$ is local
b) $B_\infty$ is without derivatives
c) $C_\infty$ preserves the AdS symmetry.
Under the above requirements $C_\infty$ is unique.
Having the explicit solutions for $\psi_a$ and its asymptotics, one can
rewrite the boundary term (up to irrelevant for our considerations
contact terms) as:
\begin{equation}
C_\infty=\lim\limits_{\varepsilon\to 0}\frac 12\int\limits_{M_\varepsilon}d^d
\sqrt{g_\varepsilon}\bar\psi_ig^{ij}\psi_j
\label{26}
\end{equation}
where $M_\varepsilon$ is a d-dimensional surface approaching the
boundary $\varepsilon\to 0$ and $g_\varepsilon$ in the induced metric on
$M_\varepsilon$. The boundary term (\ref{26}) is in complete agreement
with that of \cite{C,A}.
Using the explicit expressions for ($\varphi_i,\chi_i$) and
($\bar\varphi_i,\bar\chi_i$) it is straightforward to calculate the correlation
functions produced on the boundary. Since the Rarita-Schwinger action is
zero on-shell the contributions will come only from the boundary terms
(\ref{24}). According to the AdS/CFT correspondence principle one must replace
$\chi_i$ and $\bar\chi_i$ in (\ref{24}) with their on-shell values
(\ref{18}). The substitution gives:
\begin{multline}
S_{class}=\int\frac{d^d{\mathbf p}}{(2\pi)^d}\[\bar\varphi_i\(-{\mathbf p}\)\chi_i\({\mathbf p}\)+
\bar\chi_i\({\mathbf p}\)\varphi_i\(-{\mathbf p}\)\]\\
=i\frac{\Gamma\(\frac 12-m\)}{2^{2m}\Gamma\(\frac
12+m\)}\int\frac{d^d{\mathbf p}}{(2\pi)^d}\bar\varphi_i\({\mathbf p}\)\(\delta_{ij}\frac{\hat
p}{p}-\frac{2\(2m+1\)}{d+2m-1}\frac{p_ip_j\hat p}{p^3}\)\varphi_i\({\mathbf p}\)\\
=-\frac{\Gamma\(\frac{2m+d+1}{2}\)}{\pi^{\frac d2}\Gamma\(\frac 12+m\)}\int
d^d{\mathbf x} d^d{\mathbf y}\bar\varphi_i\({\mathbf x}\)\frac{\(x-y\)_i\gamma^i}{|{\mathbf x}-{\mathbf
y}|^{2m+d+1}}\[\delta_{ij}-2\frac{\(x-y\)_i\(x-y\)_j}{|{\mathbf x}-{\mathbf y}|^2}\]
\varphi_j\({\mathbf y}\)
\label{27}
\end{multline}
The above expression coincides with the two point correlation functions
found in \cite{C,A}:
\begin{equation}
\Omega\({\mathbf x},{\mathbf y}\)\sim\frac{\(x-y\)_i\gamma^i}{|{\mathbf x}-{\mathbf
y}|^{2m+d+1}}\[\delta_{ij}-2\frac{\(x-y\)_i\(x-y\)_j}{|{\mathbf x}-{\mathbf y}|^2}\]
\label{28}
\end{equation}
corresponding to conformal operator of dimension $\Delta=\frac d2+m$.
\section{Massive case}
We proceed with the analysis of the boundary term in the case of massive
Rarita-Schwinger field. The most general action is given by \cite{N}:
\begin{equation}
S_{mR-S}=\int d^d{\mathbf x}\sqrt{G}\[\bar\Psi_\mu\Gamma^{\nu\n\rho}D_\nu\Psi_\rho-
-m_1\bar\Psi_\mu\Psi^\nu-m_2\bar\Psi_\mu\Gamma^{\mu\nu}\Psi_\nu\]
\label{30}
\end{equation}
The equations of motion following from (\ref{30})
$$
\Gamma^{\mu\nu\rho}D_\nu\Psi_\rho-m_1\Psi^\mu-m_2\Gamma^{\mu\nu}\Psi_\nu=0
$$
can be rewritten in more convenient form \cite{C,kr}:
\begin{equation}
\Gamma^\nu\(D_\nu\Psi_\rho-D_\rho\Psi_\nu\)-m_-\Psi_\rho+\frac{m_+}{d-1}\Gamma_\rho
\Gamma^\nu\Psi_\nu=0
\label{31}
\end{equation}
where we use the notations of \cite{kr}: $m_\pm=m_1\pm m_2$. Applying
the standard procedure of passing to flat space equations ($\psi_a=
e^\mu_a\Psi_\mu$) it is straightforward to obtain the equations for
$\psi_0$ and $\psi_i$ \cite{C,kr}:
\begin{align}
\label{32}&x_0\gamma^a\partial_a\psi_0-\(\frac d2+1\)\gamma_0\psi_0-m_-\psi_0=
x_0\partial_0\eta-
\frac{m_+}{d-1}\gamma_0\eta-\eta\\
\label{33}&x_0\gamma^a\partial_a\psi_i-\frac d2\psi_i-m_-\psi_i=
x_0\partial_i\eta+\frac 12\gamma_0
\gamma_i\eta-\frac{m_+}{d-1}\gamma_i\eta+\gamma_i\psi_0
\end{align}
where $\eta=\gamma^a\psi_a$.
One can try to apply the standard Frobenius procedure in solving the
system (\ref{32},\ref{33}) by looking for solutions in form:
\begin{equation}
\psi^a\sim\(x_0\)^\rho\sum\limits_{n=0}^\infty c_n^a\({\mathbf x}\)x_0^n
\label{35}
\end{equation}
where $c_n^a({\mathbf x})$ are $x_0$-independent Rarita-Schwinger fields. If one
try to solve the system for fields depending on $x_0$ only the leading
terms will have rather different values for $\rho$ \footnote{As it was
noted in \cite{kr} simple
algebraic operations leads to the following relation between $\psi_0$ and
$\eta$ when we consider the dependence on $x_0$ {\it only}:
$$
\(d-1-2m_-\gamma_0\)\psi_0=\(d-1+2m_2\gamma_0\)\eta.
$$
Note that $b^\pm_a$ are constant spinors.} \cite{kr}:
\begin{align}
\label{36}&\psi_0\sim x_0^{\frac d2+C}b_0^++x_0^{\frac d2-C}b_0^-\\
\label{37}&\psi_i\sim x_0^{\frac d2+C}\gamma_ib_0^++x_0^{\frac d2-C}\gamma_ib_0^-
+x_0^{\frac d2+m_-}b_i^++x_0^{\frac d2-m_-}b_i^-
\end{align}
where the constant $C$ is given by:
\begin{equation}
C=\frac{d\(d-1\)}{4m_1}+\frac{m_-\(m_1+dm_2\)}{m_1\(d-1\)}
\label{38}
\end{equation}
Similar expressions for the conjugated fields hold. The splitting of
$\varphi_a$ into $\pm$ parts is subject to the conditions:
\begin{equation}
\gamma_0b_a^\pm=\pm b^\pm_a;\quad \bar b_a^\pm=\mp\bar b^\pm_a
\label{41}
\end{equation}
Let us discuss the general solutions of the system (\ref{32}, \ref{33}).
For the $\psi_0$ component it reads off\footnote{In order to introduce ${\mathbf x}$
dependence, in \cite{kr} O(d+1,1) transformations are used. Such
transformations are accompanied by rotation for the spinor fields. One can
show that our expressions are equivalent to those of \cite{kr}.} \cite{kr}:
\begin{multline}
\psi_0\(x_0,{\mathbf p}\)=
\left\{\frac{x_0^{\frac{d+1}{2}}p^{\frac 12-C\gamma_0}}{2^{\frac 32-C\gamma_0}
\Gamma\(\frac d2+\frac 32-C\gamma_0\)}\[\(d-3+2C\gamma_0+2i\hat px_0\){\mathcal K}_{\frac 12
-C\gamma_0}\(x_0p\)\right.\right.\\
+\(2x_0\gamma_0p-i\(d+3-2C\gamma_0\)\frac{\hat p}{p}\){\mathcal K}_{\frac
12+C\gamma_0}\(x_0p\)-\\
\left.\frac{4m_1x_0p}{d\(d-1-2m_2\)}\(\(\frac{d}{x_0}-i\gamma_0\hat p\){\mathcal K}_{\frac
12-C\gamma_0}\(x_0p\)-p{\mathcal K}_{\frac 12+C\gamma_0}\(x_0p\)\)\]\tilde\varphi_0\({\mathbf p}\)\\
\left. -\frac{2x_0^{\frac{d+1}{2}}p^{\frac 12-m_-\gamma_0}}{2^{\frac
32-m_-\gamma_0}\Gamma\(\frac d2+\frac 32-m_-\gamma_0\)}\(-i\gamma_0p_i{\mathcal K}_{\frac
12-m_-\gamma_0}\(x_0p\)-\frac{p_i\hat p}{p}{\mathcal K}_{\frac 12+m_-\gamma_0}\(x_0p\)\)
\tilde\varphi_i\right\}
\label{42}
\end{multline}
\begin{multline}
\psi_i\(x_0,{\mathbf p}\)=
\left\{\frac{x_0^{\frac{d+1}{2}}p^{\frac 12-C\gamma_0}}{2^{\frac 32-C\gamma_0}
\Gamma\(\frac d2+\frac 32-C\gamma_0\)}\[2\frac{p_i\hat p}{p}x_0{\mathcal K}_{\frac 12+
C\gamma_0}\(x_0p\)\right.\right.\\
-2\(\gamma_i-ip_i\gamma_0x_0\){\mathcal K}_{\frac 12-C\gamma_0}\(x_0p\)
+\frac{2m_1}{d\(d-1+2m_2\)}\times\\
\times\(\(\(d-1-2C\gamma_0\)\gamma_0\gamma_i+2ip_ix_0\)
{\mathcal K}_{\frac 12-C\gamma_0}\(x_0p\)
-\(\(d-1-2C\gamma_0\)i\frac{\hat p}{p}\gamma_i\right.\right.\\
\left.\left.\left.+2i\(1+2C\gamma_0\)\frac{p_i}{p}-
-2i\frac{dp_i+\gamma_i\hat p}{p}
-2\frac{p_i\hat p}{p}\gamma_0x_0\){\mathcal K}_{\frac
12+C\gamma_0}\(x_0p\)\)\]\tilde\varphi_0\({\mathbf p}\)\\
+\frac{x_0^{\frac{d+1}{2}}p^{\frac 12-m_-\gamma_0}}
{2^{\frac 32-m_-\gamma_0}\Gamma\(
\frac d2+\frac 32-m_-\gamma_0\)}
\[\(\(d-1-2m_-\gamma_0\)\gamma_0\delta_{ij}+2i\frac
{p_ip_j\hat p}{p^2}x_0\){\mathcal K}_{\frac 12-m_-\gamma_0}\(x_0p\) \right.\\
-\(\(d-1-2m_-\gamma_0\)i\frac{\hat p}{p}\delta_{ij}+2i\(1+2m_-\gamma_0\)\frac
{p_ip_j\hat p}{p^3}-2i\frac{p_j\gamma_i}{p}\right.\\
-\left.\left.\left. \frac{2p_ip_j}{p}\gamma_0 x_0\)
{\mathcal K}_{\frac 12+m_-\gamma_0}\]\tilde\varphi_j\({\mathbf p}\)\right\}
\label{43}
\end{multline}
where $\tilde\varphi_a$ satisfy the relations:
$$
\gamma_0\tilde\varphi_a=-\tilde\varphi_a;\quad
\gamma^i\tilde\varphi_i=0\notag
$$
and analogous expression for conjugated fields. Since we have two rather
different leading terms (of powers $d/2\pm C$ and $d/2\pm m_-$),
some of the solutions have to be fixed to zero, i.e. we must fix
$\tilde\varphi_0$ or $\tilde\varphi_i$.
A natural criterion for this is the requirement that in
the limit $m_2\to 0$ to reproduce the massless case. This uniquelly
determine $\tilde\varphi_0=0$. The same arguments as in the massless case
for regularity of the solution in
the interior of the AdS relate $\tilde\varphi_i^+$ and $\tilde\varphi_i^-$
which reminiscent again
the principle that only a half of the components can be fixed on the boundary.
\footnote{This requirement lead to the condition $p_j\tilde\varphi_j^+=
-i\frac{\(d-1+2m_-\)}{\(d-1-2m_-\)}p^{2m_-}\frac{\hat p}{p}p_j\tilde\varphi_j^-$.
The relation holds only on-shell. Note that in our expressions the spinors
are rotated compared to \cite{kr} and will be denodet by $\hat{}$.}
In order to extract the contribution to the boundary we will use the small
argument expansion of the modified Bessel function (\ref{14}). The result
for $\psi_0$ is:
\begin{align}
&\psi_0\(x_0,{\mathbf p}\)=x_0^{\frac d2+m_-+1}\chi_0\({\mathbf p}\)+
x_0^{\frac d2-m_-+1}\varphi_0\({\mathbf p}\)\label{45}\\
&\chi_0=\frac{\Gamma\(\frac 12-m_-\)}{d-1+2m_-}
\[
\frac{p_i\hat p}{p}\(\frac p2\)^{2m_-}\hat\varphi_i^- \]\label{46}\\
&\varphi_0=-\frac{\Gamma\(\frac 12+m_-\)}{d-1+2m_-}
\[ip_i\hat\varphi^-_i\]\label{47}
\end{align}
One can express $\chi_0$ in terms of $\varphi_0$, but since $\psi_0\sim
x_0^{d/2\pm m_-+1}$ there will be no contribution to the boundary term.
Let us apply the same analysis to the $\psi_i$ components. We split $\psi_i$
into "chiral" and "anti-chiral" parts:
\begin{align}
\label{48}&\psi_i\(x_0,{\mathbf p}\)=
x_0^{\frac d2+m_-}\chi_i\(x_0,{\mathbf p}\)+x_0^{\frac d2-m_-}\varphi_i\(x_0,{\mathbf p}\)\\
\label{49}&\gamma_0\chi_i=\chi_i;\quad \gamma_0\psi_i=-\psi_i
\end{align}
The analysis of the $x_0\to0$ behavior of the corresponding component gives
the following expressions:
\begin{align}
&\varphi_i\({\mathbf p}\)=\frac{\Gamma\(\frac 12-m_-\)}{2^{m_-+\frac 12}}\(
\(d-1+2m_-\)i\frac{\hat p}{p}\delta_{ij}+2i\(1-2m_-\)\frac{p_ip_j\hat p}
{p^3}-2i\frac{p_j\gamma_i}{p}\)p^{2m_-}\hat\varphi_j^-\label{50}\\
&\chi_i\({\mathbf p}\)=-\frac{\Gamma\(\frac 12+m_-\)}{2^{-m_-+\frac 12}}\(d-1+2m_-\)
\hat\varphi_i^-\label{51}
\end{align}
From the above formulae follows that the two expressions (\ref{50},
\ref{51}) are not independent and a half of the boundary data can be
expressed (on-shell) in terms of the other half.
We now turn to the variational principle applied to the action for the
massive Rarita-Schwinger field (\ref{30}). Repeating the same considerations
as in the massless case we have found the boundary terms in the same form
\footnote{We note that since $\gamma^i\chi_i=0$ the $3^{th}$ and $4^{th}$ terms
in (\ref{55}) and the second term in (\ref{57}) below
will not contribute.}
:
\begin{equation}
C_\infty=\frac 12\int d^d\[\bar\varphi_ig^{ij}\chi_j+\bar\chi_ig^{ij}\varphi_j+\bar\varphi_i
\gamma^i\gamma^j\chi_j+\bar\chi_i\gamma^i\gamma^j\varphi_j\]
\label{55}
\end{equation}
which again can be written as in the massless case:
\begin{equation}
C_\infty=\lim\limits_{\varepsilon\to 0}\frac 12\int\limits_{M_\varepsilon}d^d
\sqrt{g_\varepsilon}\(\bar\psi_ig^{ij}\psi_j+\bar\psi_i\gamma^i\gamma^j\psi_j\)
\label{57}
\end{equation}
The action ensuring that the classical solutions are true stationary point
of the action (\ref{30}) gets the modification:
\begin{equation}
S=S_{mR-S}+C_\infty
\label{56}
\end{equation}
which is unique under the requirement of locality, absence of derivatives
and presevation of the AdS symmetry.
Now, using (\ref{50}, \ref{51}) it is straightfoward to reproduce the
correlation functions in the CFT living on the boundary found in \cite{kr}:
\begin{equation}
\Omega\({\mathbf x},{\mathbf y}\)\sim\frac{\(x-y\)_i\gamma^i}{|{\mathbf x}-{\mathbf
y}|^{2m_-+d+1}}\[\delta_{ij}-2\frac{\(x-y\)_i\(x-y\)_j}{|{\mathbf x}-{\mathbf y}|^2}\]
\label{58}
\end{equation}
Note that in the massless limit $m_2\to 0$ the above result coincide with
(\ref{28}).
\ \\
{\bf{\large Conclusions}}
\ \\
In this letter we have analysed the boundary term for Rarita-Schwinger
field in the AdS/CFT correspondence. It is shown that, as in the spinor case,
one cannot fix simultaneously all the components of the Rarita-Schwinger
field on the boundary but only a half of them. Following \cite{H} we used the
stationary phase method to determine the surface term.
We apply variation of the action over
an appropriate class of field configurations. Since we are dealing with
a theory with a boundary, it turns out that one half of the components of the
spinor field must be kept fixed but the other half are free to vary on the
boundary (at $x_0=0$). We choose to impose the boundary conditions on the
"chiral" part of the Rarita-Schwinger field (annihilated by $1/2(I-\gamma_0)$)
which have a minimal value of the leading term in $x_0$. The choice is quite
natural thinking of this part as of a source of the bulk-boundary Green
function. The "anti-chiral" starts at higher power of $x_0$ but it is shown
that these components play an important role since they contribute to the
boundary term. Moreover, these components are not off-shell subject to
boundary conditions.
It would be interesting to study interacting Rarita-Scwinger--scalars in
AdS/CFT correspondence as in \cite{Ge} and to proceed with an
investigation of the $S$-matrix along the lines of \cite{Sus1,P,Bala,Ar}.
Work on the subject is in progress.
\ \\
{\bf Note added}
After this paper was completed I was informed about paper \cite{af}
where, in the case of $AdS_5\times S^5$ geometry of type IIB
supergravity, analogous boundary terms for massless gravitino field were found.
I thank Dr. S.Frolov for information and comments. In \cite{vl} was
described a procedure of obtaining the boundary terms based on the
intertwining operators considerations. I am obligated to Prof. V.K.
Dobrev for information.
\ \\
{\bf Acknowledgments}
I am grateful to N.I.Karchev and M.Stanishkov for comments and critical
reading the manuscript.
\ \\
|
\section{Introduction}
The formation of stellar clusters is yet an unsolved problem of
theoretical astrophysics. Clusters form in giant molecular clouds
complexes on time scales of order $10^6$ yrs (Carpenter et al. 1997,
Hillenbrand, 1997). This time scale is similar to the dynamical time
scale of the clouds, it is however one order of magnitude shorter than
the inferred cloud lifetime (Blitz \& Shu 1980). The mechanisms that
stabilize molecular clouds for such a long time are not well
understood. Turbulence has been proposed to stir clouds, supporting
them against gravitational collapse (Arons \& Max 1975). Indeed,
molecular emission lines tend to be an order of magnitude broader than
the thermal linewidth (Blitz 1993), indicating supersonic velocities.
Magnetic fields have been detected by direct measurements of field
strength in some star forming regions (Myers \& Goodman 1988, Crutcher
et al. 1993 \& 1999, Crutcher 1999, see however Verschuur 1995a,b) and may
affect the cloud dynamics and stability.
Recent simulations suggest that freely decaying turbulence dissipates its kinetic energy
on time scales
shorter than a dynamical time scale (Gammie \& Ostriker 1996, Mac Low et al. 1998, Stone, Ostriker \& Gammie 1998), regardless of the adopted equation of state or the presence of magnetic fields.
In order to stabilize clouds for a longer time, energy must be supplied, either
by internal processes like stellar winds and outflows (Franco \& Cox 1983, McKee 1989) or externally,
e.g. by the shear motions induced through Galactic differential rotation (Fleck 1981).
The spread in stellar ages for a given cluster is however similar to the
dissipation time scale indicating that clusters form in regions where
turbulence could decay leading to gravitational collapse. How turbulence is driven
and why at some point in the evolution of the cloud the driving mechanism fails
remains an outstanding issue.
Previous numerical models of isolated gaseous spheres have shown that
stellar clusters could form as a result of gravitational collapse and
fragmentation (see e.g. Larson 1978, Keto, Lattanzio \& Monaghan
1991). These models were however strongly constrained by numerical
resolution. More recently, Whitworth et al (1995), Turner et
al. (1995) and Bhattal et al. (1998) investigated in detail the
fragmentation of shocked interfaces of colliding molecular clumps into
small stellar systems. The effects of gas accretion on the evolution
of a newly formed, young stellar clusters has been investigated by
Bonnell et al. (1997). They found that gas accretion is highly
non-uniform with a few stars accreting significantly more than the
rest and concluded that competitive accretion processes play an
important role in shaping the initial stellar mass function. Finally,
the individual collapse of perturbed protostellar cores and their
break-up into binaries and multiple systems has recently been studied
with high resolution (see e.g. Burkert \& Bodenheimer 1993, 1996;
Burkert, Bate \& Bodenheimer 1997; Boss 1997, Truelove et al. 1997,
1998, Bate 1998).
In this paper we extend previous studies of the collapse of isolated
gas clumps to the molecular cloud regime and study the formation of
stellar clusters localized within globally stable molecular
clouds. Our simulations combine self-consistently previous studies of
the dynamics of turbulent molecular clouds with studies of their
collapse and fragmentation stage till the formation of a stellar
cluster including its competitive accretion phase. In a previous
short Letter (Klessen et al. 1998) we discussed a high-resolution
simulation which lead to a stellar cluster with a log-normal mass
distribution in excellent agreement with observations of multiple
stellar systems. The current paper will explore the formation and early
evolution of a star cluster in greater detail, starting from different
initial realizations of clumpy molecular clouds with initial
power-spectrum $P(k) \propto k^{-2}$. Section 2 presents the numerical
method. Model properties and initial conditions are presented in
section \ref{sec:scaling}. The global evolution of the models is
discussed in section \ref{sec:time-evolution}. Section
\ref{sec:discussion} analyzes the dynamics and structure of the gas
cloud and the properties the newly formed stellar system and compares
it with observations. Our results are summarized in section \ref{sec:summary}.
\section{The numerical model}
\label{sec:numerics}
\subsection{SPH in combination with GRAPE}
SPH ({\em smoothed particle hydrodynamics}) is a particle-based scheme
to solve the equations of hydrodynamics. As the fluid is represented
by an ensemble of particles, the technique can be regarded as an
extension of the pure gravitational $N$-body system. Besides being
characterized by its mass $m_i$, velocity ${\vec v}_i$ and location
${\vec r}_i$, each particle $i$ is associated with a density $\rho_i$,
an internal energy $\epsilon_i$ (equivalent to a temperature $T_i$),
and a pressure $p_i$. The time evolution of the fluid is represented
by the dynamical evolution of the SPH particles. Their behavior is
governed by the equation of motion supplemented by additional
equations to modify the hydrodynamic properties. Thermodynamic
observables are obtained by averaging over an appropriate subset of
the SPH particles. Excellent overviews over the method, its numerical
implementation, and some of its applications have been written by Benz
(1990) and Monaghan (1992). We use SPH because it is intrinsically
Lagrangian: As opposed to mesh-based methods, it does not require a
fixed grid to represent fluid properties and calculate spatial
derivatives (see e.g.~Hockney \& Eastwood 1988). The fluid particles
are free to move and -- in analogy -- constitute their own grid. The
method is therefore able to resolve very high density contrasts
because the particle concentration increases where needed.
Our code is based on a version originally developed by Benz (1990).
It uses a standard description of a von~Neumann-type artificial
viscosity (Monaghan \& Gingold 1983) with adopted parameters
$\alpha_{\rm v} = 1$ and $\beta_{\rm v} = 2$ for the linear and
quadratic terms. The system is integrated in time using a second-order
Runge-Kutta-Fehlberg scheme, allowing individual time steps for each
particle. In any self-gravitating fluid, regions with masses exceeding
the Jeans limit become unstable and collapse. In the current code,
once a highly-condensed object forms in the center of a collapsing gas
clump and has passed a certain density threshold, the dense core is
substituted by a `sink' particle (Bate, Bonnell \& Price 1995). This
particle has a fixed accretion radius, which is of order of the Jeans
length at the density at which the sink particle is created. It
inherits the combined masses, linear and `spin' angular momenta of the
particles it replaces. It has the ability to accrete further SPH
particles from its infalling gaseous surrounding. Again, the mass of
the accreted particles, their linear and angular momenta will be added
to the sink particle in order to guarantee mass and momentum
conservation. The accreted particles are then removed from the
calculation. By adequately replacing high-density cores by sinks and
keeping track of their further evolution in a consistent way the code
time stepping is prevented from becoming prohibitively small and we
are able to follow the dynamical evolution of the system over many
free-fall times. However, this procedure implies that information
about the evolution of gas inside the sink particle is lost. In our
case a sink particle corresponds to a gravitationally collapsing
protostellar core. For a detailed description of the physical
processes inside a protostellar core, a new simulation just
concentrating on this single object would be required with the
appropriate initial and boundary conditions taken from the
larger-scale simulation (Burkert, Klessen \& Bodenheimer 1998).
To achieve high computational speed, we use SPH in combination with
the special-purpose hardware device GRAPE (Sugimoto et al.~1990,
Ebisuzaki et al.~1993). This device calculates the forces and the
potential in the gravitational $N$-body problem by direct summation on
a specifically designed chip with very high efficiency. This allows
calculations at supercomputer level on a normal
workstation. Additionally, GRAPE returns the list of nearest neighbors
for each particle. This feature makes it attractive for use in
smoothed particle hydrodynamics (Umemura et al.~1993, Steinmetz
1996). For particles near the surface of the integration volume,
`ghost' particles are created to correctly extend the neighbor search
beyond the borders of the computational volume; no forces are computed
for these particles.
Since we wish to describe the dynamical evolution of a gravitationally
unstable region in the interior of a considerably larger globally
stable molecular cloud, we adopt periodic boundary conditions to
prevent overall collapse. As {\sc Grape} cannot treat periodic
particle distributions directly due to its restricted force we have to
introduce the Ewald (1921) method to prevent global collapse. The
basic idea is to compute a periodic correction force for each particle
on the host computer, applying a particle-mesh like scheme: We first
compute the forces in the isolated system using direct summation on
{\sc Grape}, then we assign the particle distribution to a mesh and
compute the periodic correction force for each grid point, by
convolution with the adequate Green's function in Fourier
space. Finally, we add this correction to each particle in the
simulation. The corrective Green's function can be constructed as the
offset between the periodic solution (calculated via the Ewald
approximation) and the isolated solution on the grid. This method has
proven to be numerically stable and inexpensive in terms of the
computational effort (Klessen 1997).
\subsection{Some cautionary remarks on the limitations of SPH}
\label{subsec:resolution-limit}
To make full use of the Lagrangian nature and resolving power of SPH
the smoothing volume over which hydrodynamic quantities are averaged
in the code is freely adjustable in space and time such that the
number of neighbors considered is always kept in the range 30 and 70,
with the optimum value being 50. This sets a natural limit to the
spatial resolution of the code. It is furthermore constrained by the
Courant-Friedrichs-Lewy (1928) criterion. It demands that the minimum
time stepping in the SPH code is always less than the time required
for a sound wave to cross the minimum smoothing volume. In order to
prevent the time stepping required to resolve very high density peaks
to become prohibitively small, one has to introduce a minimum
smoothing length which defines the smallest resolvable length scale.
In our simulations of self-gravitating gas the spatial resolution is
subject to an additional constraint. In order to correctly treat the
dynamical evolution of high-density peaks, the mass contained within
the smoothing volumes of two interacting particles must be less than
the local Jeans mass. Otherwise, the stability of the clump against
gravitational collapse depends on the detailed implementation of the
gravitational force law and on the kernel function used for the
simulation, rather than on physical processes. The minimum Jeans mass
that is reached during the calculation must always be greater than
approximately twice the mass of particles in the SPH kernels (Bate \&
Burkert 1997). If one bears this caveat in mind, the SPH method
calculates the time evolution of gaseous systems very reliably and
accurately. Additionally it offers spatial and dynamical flexibility
that has yet to be achieved by grid-based methods.
\section{The model}
\label{sec:scaling}
The numerical models discussed in this paper describe
self-gravitating, isothermal gas. This is justified by the typical
densities observed in molecular clouds and determines the equation of
state as well as the physical processes considered in the model. The
dynamical evolution of the gas is scale-free and depends only on the
ratio of the internal to gravitational energy. We start with an
initially clumpy cloud region with density fluctuations following a
Gaussian random distribution. The velocities are computed
self-consistently from the Poisson equation. Assuming that stellar
clusters form through the gravitational collapse of clumpy cloud
regions that have lost their turbulent support, we study the detailed
behavior of this process and the properties of the newly formed
cluster.
\subsection{Scaling properties of isothermal, self-gravitating gas}
\label{subsec:scale-free}
For isothermal gas, the energy density is a function of temperature
only and the equation of state reduces to $p = c^2_{\rm s}\;\! \rho$,
with $c_{\rm s}$ being the thermal sound speed. This approximation is
valid for physical regimes where the cooling time scales are much less
than the dynamical ones. In molecular clouds, this is the case for gas
densities $1 \lesssim n({\rm H}_2) \lesssim 10^{10}\,$cm$^{-3}$, where
the gas is optically thin for the dominant cooling processes and
energy is radiated away very efficiently (e.g.~\cite{toh82}). Average
densities in star forming regions typical are in the range $10\,{\rm
cm}^{-3} < n({\rm H}_2) < 10^5\,{\rm cm}^{-3}$ which is the density
range of our models. Even when resolving a density contrast of
$10^4$, the isothermal equation of state is appropriate
throughout the entire simulation.
The self-gravitating, isothermal model studies the interplay between
gravity and gas pressure, it is thus scale free. Besides the dependence on the
initial density and velocity distribution, the dynamical evolution of
the system depends only on one additional free parameter, the ratio between
the internal energy $\epsilon_{\rm int}$ and the potential energy
$\epsilon_{\rm pot}$. This ratio can be interpreted as a dimensionless
{\em temperature},
\begin{equation}
\alpha \equiv \epsilon_{\rm int}/|\epsilon_{\rm pot}|\:.
\end{equation}
Line widths in molecular clouds are super-thermal, implying the
presence of supersonic turbulent motions (e.g.~Blitz 1993). In case of
isotropic turbulence, these non-thermal (turbulent) contributions can
be accounted for by introducing an {\em effective} energy
$\epsilon_{\rm int} = \epsilon_{\rm therm} + \epsilon_{\rm turb} =
\gamma\cdot(c^2_{\rm s} + \sigma^2_{\rm turb})/2$ adding up the
thermal and non-thermal contributions to the kinetic energy. This is
equivalent to defining an {\em effective} temperature $\alpha_{\rm
eff}$. The turbulent velocity dispersion is denoted by $\sigma_{\rm
turb}$ and the factor $\gamma$ depends on the degree of freedom. In
case of anisotropic turbulent motions, the system has (locally)
preferred axes and the concept of one single effective temperature is
not valid.
\subsection{Normalization and relation to observed star-formating regions}
\label{subsec:conversion}
In analogy to the temperature, we adopt dimensionless and normalized
units for all physical quantities. All masses are scaled relative to
the total mass of the simulated molecular cloud region, and the length
scale is its size $L$. The numerically simulated area is then a cube
$[-L,+L]^3=[-1,+1]^3$ with periodic boundary conditions. The density
of the homogeneous cube is $\rho = 1/8$. In order to trigger
gravitational collapse and star formation we consider systems that are
highly unstable against gravitational collapse and set $\alpha =
0.01$. The volume will then contain $N_J=222$ Jeans masses. We assume
the gravitational constant $G\equiv1$ and the gas constant ${\cal R} =
1/\gamma$ with $\gamma=3/2$ for an ideal gas having three degrees of
freedom. In these units the sound speed is $c_{\rm s} = ({\cal R}
\alpha)^{1/2}$ and the Jeans mass follows as
\begin{equation}
M_{\rm J} = 1.6\cdot\rho^{-1/2}\;\alpha^{3/2}\;.
\end{equation}
In order to scale to physical units let us consider a dark cloud like
Taurus with $n({\rm H}_2) \approx 10^2\,{\rm cm}^{-3}$ and
$T\approx10\,$K. Assuming a mean molecular weight of $\mu = 2.36$,
the units of mass and length correspond to $M = 6\;\!300\,$M$_{\odot}$
and $L = 5.2\,$pc, respectively. The time unit is equivalent to
$t=2.2\times10^6\,$years and the average Jeans mass transforms to $
M_{\rm J} = 28\,{\rm M}_{\odot}$. Applied to a dense, massively
star-forming cloud with $n({\rm H}_2) \approx 10^5\,{\rm cm}^{-3}$ and
$T\approx10\,$K, similar to the BN region in Orion, the simulated cube
translates into a mass of $M = 200\,$M$_{\odot}$ and a size scale $L =
0.16\,$pc. The time unit now converts to $t=7.0\times10^4\,$years and
the mean Jeans mass for the homogeneous distribution is $M_{\rm J} =
0.9\,{\rm M}_{\odot}$.
Molecular clouds are stabilized against global collapse by the
presence of supersonic turbulence as indicated by the large observed
line width (e.g.~Blitz 1993). However, molecular clouds do form
stars. Star formation occurs in their interior in regions which {\em
locally} lose turbulent support and which begin to contract to form
stars or clusters of stars. The time scale for freely decaying
turbulence is of the order of the free-fall time or shorter (e.g.~Mac
Low et al.~1998). As the processes that lead to the local loss of
turbulent support are not well understood, we do not attempt to model
that stages of the evolution and start our simulations when turbulence
has already decayed. This is equivalent to the assumption of
instantaneous loss of support. The sizes of local collapse regions
are typically much smaller than the overall extent of the cloud. To a
good approximation we therefore neglect the influence of cloud
boundaries and place the considered volume inside an infinite
cloud. Hence, to describe a gravitationally unstable volume inside an
overall stable cloud of much larger extent we adopt periodic boundary
conditions. This is a commonly used scheme when describing turbulent
cloud dynamics (e.g. Mac Low et al. 1998, Stone et al. 1998, Klessen
2000).
With the adopted values of $\alpha = 0.01$ collapse progresses quite
rapidly and the influence of the boundary conditions on the overall
dynamics is weak. The situation changes when the scaling parameter
$\alpha$ is increased. The total number of Jeans masses contained in
the computed cube then decreases, which in our isothermal models is
equivalent to `zooming' in onto smaller spatial volumes. Hence,
the adopted boundary conditions become more important and may begin to
influence the collapse behavior of individual protostellar cores. As
cores are never isolated, but instead interact with each other or
accrete gas from their environment, in the case of $\alpha
{\:\lower0.6ex\hbox{$\stackrel{\textstyle >}{\sim}$}\:} 0.1$
boundaries allowing for mass inflow would become more appropriate. Indeed,
this study may be used for finding the optimum boundary conditions for
high-resolution simulations which focus on the evolution of individual
protostellar cores. We defer a detail discussion of this issue to
paper II where we address the influence of different initial and
boundary conditions. Decreasing $\alpha$ on the other hand means
increasing the corresponding physical extend of the computed
volume. The simulation treats a larger fraction of molecular cloud
material and results in a larger number of protostellar cores that
form during the dynamical evolution and fragmentation of the gas.
This, however, also increases the computational demands (if a fixed mass
and spatial resolution for individual cores is retained). The adopted
value $\alpha = 0.01$ in the present paper corresponds to the maximum
number of Jeans masses which we can handle with sufficient numerical
resolution.
\subsection{Initial conditions -- Gaussian random fields}
\label{subsec:initial}
The dynamical evolution of gas in a molecular cloud region depends on its
initial density and velocity distribution. In a self-consistent model,
both are related via Poisson's and the hydrodynamic equations. We
adopt random Gaussian fluctuations for the density as starting
condition for the SPH simulations. We use this approach, because these
distributions have well determined statistical properties and can
easily be generated by the Zel'dovich approach (see appendix
\ref{app:zel}). Furthermore, their properties resemble the end stages
of decaying turbulence and are suited to mimic observed features of
molecular clouds once advanced into the non-linear regime (see
e.g.~Stutzki \& G{\"u}sten 1990, who deploy a Gaussian decomposition
technique to describe the clumpy structure of molecular clouds).
Gaussian random fields $\rho(\vec{r})$ are completely characterized by
their first two moments, the mean value $\rho_0 \equiv \langle
\rho(\vec{r})\rangle_{\scriptsize \vec{r}}$ and the 2-point
correlation function $\xi(\vec{r}) \equiv \langle
\rho(\vec{r}')\rho^*(\vec{r} +\vec{r}')\rangle_{\scriptsize
{\vec{r}'}}$, which is equivalent to the power spectrum $P(\vec{k})$
in Fourier space. All higher moments can be expressed in terms of
$\xi(\vec{r})$. For an isotropic fluctuation spectrum, $P(k) =
P(|\vec{k}|)$, the 2-point correlation degrades to a function of
the distance between two points, $\xi(r)$. By defining a
normalization $\rho_0$ and the power spectrum $P(k)$ in Fourier space,
all statistical properties of the field $\rho(\vec{r})$ are
determined.
The function $P(k)$ identifies the contribution of waves with wave
number $k$ to the statistical fluctuation spectrum. In Gaussian random
fields, the phases are arbitrarily chosen from a {\em uniform}
distribution in the interval $[0,2\pi[$, and the amplitudes for each
mode $k$ are randomly drawn from a {\em Gaussian} distribution with
width $P(k)$ centered on zero. Since waves are generated from random
processes, only the properties of an {\em ensemble} of fluctuation
fields are determined in a statistical sense. Individual realizations
(from different sets of random numbers) may deviate considerably from
this mean value, especially at small wave numbers $k$, i.e.~at long
wave lengths, where only a few modes $(k_x,k_y,k_z)$ contribute to the
wave number $k=|\vec{k}|$. This means that different density fields
generated from the same power spectrum $P(k)$ may {\em look} quite
differently, especially on large scales (see
Fig.~\ref{fig:nine-models}), despite having identical statistical
properties. This variance effect can be reduced by considering and
averaging over a large enough ensemble of realizations. We generate
the initial density fluctuation fields by applying the Zel'dovich
(1970) approach. The method and its applicability to gaseous systems
is discussed in detail in the appendix \ref{app:zel}.
In the current paper, we concentrate our analysis on a simple
power-law functional form for the fluctuation spectrum, $P(k) \propto
k^{\nu}$, with ${\nu} = -2$. As ${\nu}$ is a negative number, most
power is in large-scale modes. This choice gives a good representation
of the observed patchiness and inhomogeneity of molecular clouds. The
dependence of the dynamical evolution of the model cloud on variations
of the initial density distribution with different exponents ${\nu}$
of the power law will be discussed in paper II.
\subsection{Model realization}
\label{subsec:model-properties}
We present results from a
detailed analysis of nine different initial realizations of $P(k) \propto k^{-2}$
and use different particle numbers to address the
issue of numerical resolution: six models with $50\!\;000$ SPH
particles, two with $200\!\;000$ and one with $500\!\;000$ particles,
respectively. All models are generated using the Zel'dovich (1970)
approximation (see appendix \ref{app:zel}) and their properties are summarized in
Tab.~\ref{tab:models-N=2}. The projection into the $xy$-plane of the
initial particle distribution in each model at the start of the
dynamical evolution with SPH is presented in
Fig.~\ref{fig:nine-models}.
All but one model assume a random uniform particle distribution before
applying the Zel'dovich shift. Placing particles randomly into a given
volume produces an overall distribution that is homogeneous on large
scales, but is subject to statistical fluctuations on small scales
which introduce white noise into the correlation function. The scale
at which this effect becomes important is of the order of the mean
inter-particle distance. These undesired small-scale fluctuations may
influence the fragmentation behavior of the gas. However, with the
adopted temperature, the mass of small-scale regions of enhanced
density is much less than the local Jeans mass. Therefore, these
fluctuations are quickly damped and dynamically unimportant. As
alternative to the random distribution, in model $\cal F$ particles
are placed on a regular grid before applying the Zel'dovich
approximation. This distribution is force-free and has exactly uniform
density. On the other hand, the grid has preferred axes which
introduce anisotropies in the neighbor list of each particle. This
fact may again influence the small-scale behavior. On larger scales,
the system is increasingly isotropic and this effect becomes
negligible. In the hydrodynamic evolution phase after the Zel'dovich
shift traces of the grid are quickly erased. Comparing the dynamical
evolution, the model with the particles placed on a grid is
statistically indistinguishable from the models generated from a
random distribution, which we use as standard.
As discussed in appendix~\ref{app:zel}, the Zel'dovich shift interval
$\delta t$ determines the density contrast in the particle
distribution. Typically, larger $\delta t$ leads to higher initial
peak densities. On the other hand, smaller $\delta t$ implies that a
larger fraction of the total time evolution of the gas has to be
computed with the SPH method and pressure forces, which are not
included in the Zel'dovich approximation, have more time to act on the
gas. In general, $\delta t$ should be chosen small enough so that the
subsequent evolution is not dependent on the choice of $\delta t$. In
our models this is indeed the case for $\delta t \lesssim 2.0$ as is
shown in appendix \ref{subsec:validity}. To address this issue further and
to examine how a variation of the Zel'dovich shift interval influences
the properties of the protostellar cluster that forms during the
dynamical evolution and collapse of the gas, three models are
generated with smaller shift intervals $\delta t$: model $\cal E$ with
$\delta t = 1.0$, and model $\cal H$ together with the high-resolution
model $\cal I$ having $\delta t = 1.5$. Within the statistical
variance between different models the results are not dependent on the
initial shift interval.
\section{Global time evolution}
\label{sec:time-evolution}
Due to the small value of $\alpha$ the gravitational energy outweighs
the internal energy by far and the system is highly unstable to
gravitational collapse and fragmentation. As a result, typically
about sixty collapsed cores form during the dynamical evolution. A
complete time evolution is illustrated in
Fig.~\ref{fig:3D-cube-A}. Representative for all nine models it shows
snapshots of the model $\cal A$ at twelve different stages of its
dynamical evolution. Note that the cube has to be seen periodically
replicated in all directions. With the start of the SPH simulation,
pressure forces begin to act on the gas and smear out fluctuations
which are below the local Jeans limit. On the other hand, large modes
are unstable against gravitational collapse and start to contract. At
$t\approx 0.4$ the first highly-condensed cores form in the centers of
the most massive and densest Jeans-unstable gas clumps and are
replaced by sink particles (see Sec.~\ref{sec:numerics}). Soon,
clumps of smaller initial mass and density follow. The density
threshold for the formation of sink particles is chosen to be
$\rho_{\rm c} = 5000$ and the diameter of the sink particles is 1/100
of the linear size of the simulated cube. It is visible in
Fig.~\ref{fig:3D-cube-A} that the system evolves into a network of
intersecting sheets and filaments. The gas density is highest along
filaments and at their intersections. These are the locations where
dense cores form predominantly and soon a hierarchically-structured
cluster of accreting protostellar cores, represented by sink particles
is built up. Whereas the overall dynamical evolution of the system is
initially dominated by hydrodynamic effects (all the mass is in the
gas phase), the later evolution is increasingly determined by the
gravitational $N$-body interaction between protostellar cores and
between cores and their clumpy gaseous environment because more and
more gas is accreted onto dense cores. The final result is a bound
dense cluster of protostars. After a few cluster crossing times, core
motions have been randomized by close encounters and the protostars
have lost knowledge of their initial conditions.
At this stage it is necessary to draw attention to one caveat: The gas
in the models is treated isothermal. This implies that there exists no
feedback mechanism which may prevent the complete accretion of all the
gas into cores and at late phases the global core-formation
efficiency\footnote{We use the word core-formation efficiency to
distinguish from the commonly quoted star-formation efficiency. We
identify the sink particles with unresolved, collapsing protostellar
cores. Stars and cores are connected via the ability and
effectiveness of {\em individual} cores to form stars.}, defined by
the percentage of gas that is converted into dense cores, approaches
unity. This appears unphysical and is not observed in star-forming
regions. At some stage during the protostellar evolution phase,
feedback processes from nascending stars inside protostellar cores or
from already existing massive stars in their vicinity will become
relevant and terminate the accretion and collapse phase. These effects
cannot be treated in an isothermal model. However, the early
(isothermal) phases of the collapse of molecular cloud regions and the
formation of a stellar cluster are well described. The time at which
these assumptions become less appropriate is difficult to estimate. We are
therefore hesitant to interpret the models beyond the phase at which
more than $\sim60\;\!$\% of the gas has been accreted onto
protostellar cores.
In Fig.~\ref{fig:3D-cube-I}, four stages of the evolution of the
high-resolution model $\cal I$ are
presented. Figure~\ref{fig:3D-cube-I} plots the system
initially, at time $t=1.4$, when 10\% of the gas is condensed into
protostellar cores and t=2.0 and t=2.8 when 30\% and 60\%
of the total mass has been accreted, respectively.
Instead of placing individual SPH particles,
Fig.~\ref{fig:3D-cube-I} plots the distribution of the gas
density. The density field is scaled logarithmically with darker areas
denoting higher densities. Hence, dark dots identify the location of
dense collapsed cores. When comparing the time scales for core
formation and subsequent accretion with the previous model, the
different Zel'dovich shift intervals have to be taken into account. In
the high-resolution calculation $\cal I$ it is chosen to be $\delta t=
1.5$ instead of $\delta t=2.0$ for model $\cal A$. Furthermore, the
adopted accretion radius of sink particles in the code was reduced by
a factor of two, it is now $1/200$ of the linear size of the simulation
box. This delays the formation and accretion onto sink particles by
$\Delta t\approx 0.3$. Altogether, the state of the system where the mass
fraction accumulated in collapsed cores exceeds a certain percentage
is reached somewhat later by $\Delta t \approx 0.8$ in simulation
$\cal I$. Besides
this delay, the dynamical evolution of both systems is very
similar, leading to the formation of a cluster of protostellar
cores which grow in
mass by accretion from its surrounding gas reservoir.
We conclude that our basic results are not resolution dependent.
\section{Discussion}
\label{sec:discussion}
Supplementing the general outline of the dynamical evolution in the
last section, we now present a detailed analysis of the large-scale
collapse and fragmentation behavior of the system. First, we discuss
the one-point probability density functions of the density and of the
distribution of line-of-sight velocity centroids. Then we describe the
clumping properties of the gas at various stages of its dynamical
evolution, the kinematic and spatial properties of the protostellar
cluster that forms as the system advances in time and the mass
spectrum of protostellar cores. Furthermore, we discuss the
boundedness of the cluster and the rotational properties of
protostellar cores. Finally, we speculate about the implications of
our results for understanding the initial stellar mass function (IMF)
by comparing the numerically calculated core mass spectrum with the
IMF for multiple stellar systems.
\subsection{Evolution of the one-point probability density functions for
density and line-of-sight velocity centroids}
\label{subsec:pdfs}
There has been a series of recent attempts to use one-point
probability density functions (pdf) as statistical descriptors of
properties of the interstellar medium, in particular as diagnostic to
distinguish between different physical processes influencing its
dynamics and time evolution. For a discussion of the one-point
probability density function of the density field and its relevance
for the star formation process see e.g.~Scalo et al.~(1998a) and
references therein. Detailed analyses of the pdf of line centroid
velocity fluctuations are presented by Lis et al.~(1996, 1998),
Miesch, Scalo \& Bally (1998), and Klessen (2000). Here we derive
both sets of functions for our model of larger-scale cloud collapse.
\subsubsection{The density pdf}
We define the pdf of gas densities, $f(\rho)$, such that
$f(\rho)d\rho$ measures the fraction of the total mass in the system
that falls into the density range $[\rho, \rho+d\rho]$. The function
$f(\rho)$ has of inverse density. Basically it is the histogram of the
density summed over all SPH particles in the simulation. This defines
$f(\rho)$ as {\em mass} weighted average. Note that some authors
define the function by measuring the fractional {\em volume} occupied
by gas within the density range $[\rho,\rho+d\rho]$. The later
definition is convenient when analyzing numerical simulations
performed with a grid-based method. SPH on the other hand is a
particle based method, therefore it is more appropriate to obtain
statistical measures by summing over the contributions of individual
particles and hence obtain mass weighted averages.
Pure hydrodynamic simulations of turbulent isothermal gas {\em
without} self-gravity or additional physical processes result in
density pdf's of log-normal functional shape, i.e.~the logarithm of
the density follows a normal (Gaussian) distribution. However, there
is increasing evidence that the inclusion of additional physical effects
(like self-gravity or turbulent driving mechanisms) will lead to
density pdf's which significantly deviate from being log-normal (Scalo
et al.~1998, Passot \& V{\'a}zquez-Semadeni 1998, \cite{nordlund98},
Klessen 2000, Mac~Low \& Ossenkopf 2000).
This is also found in our simulations. Using data from the
high-resolution simulation $\cal I$, Fig.~\ref{fig:rho-pdfs} clearly
shows that the evolution of Jeans unstable isothermal self-gravitating
gas leads to {\em non}-log-normal density pdf's
which can be described as power laws during most evolutionary
stages. The initial state of the SPH evolution ($t=0$,
Fig.~\ref{fig:rho-pdfs}a) still exhibits a well established log-normal
density pdf. This is expected after applying the Zel'dovich shift to
transfer a Gaussian fluctuation spectrum -- which has {\em per
definitionem} a normal pdf in the linear regime -- into the non-linear
regime (\cite{kofman94}). Later on self-gravity and gas pressure begin
to shape the mass distribution into a network of intersecting sheets
and filaments and ultimately into a cluster of collapsed dense cores
(see Fig.~\ref{fig:3D-cube-A} and \ref{fig:3D-cube-I}). More and more
gas converges into regions of high density and consequently the
density pdf develops a high-density `tail' of a power-law functional
form. The power-law distribution is best established at that stage of
the evolution, when the first and most massive Jeans unstable gas
clumps have collapsed to high enough central densities that their
innermost regions can be identified as compact cores (and subsequently
are substituted by sink particles in our code). This happens at
$t \approx 1.2$ and leads to a slope of the distribution function of
$d\log_{10}N/d\log_{10}\rho = -0.8$ as can be seen in
Fig.~\ref{fig:rho-pdfs}b. Densities of $\log_{10} \rho \approx 4$ in
this figure arise from the contributions of the collapsed cores. For
those, the values of $\log_{10} \rho$ are lower limits since they are
computed by dividing the masses of the cores by their (fixed) volume;
isothermal gas would continue to collapse to ever increasing central
densities (until the gas becomes optically thick and begins to heat
up). Subsequently, with the number and masses of collapsed cores
increasing, the density distribution is best
described by multiple power laws. A steep
part describes collapsing gas that forms a new core or that is
accreted onto dense cores, and a shallower part that describes matter
that does not (yet) participate in collapse and fills the
low-density regions between filaments and
cores (see Fig.'s \ref{fig:rho-pdfs}c -- e). In the final stage of the
evolution (Fig.~\ref{fig:rho-pdfs}f) almost all of the mass is contained in a
cluster of dense protostellar cores (with densities of $\log_{10}
\rho \approx 4$) immersed into gas of low density exhibiting an almost
flat density pdf.
In general, the spread and shape of pdf's can be characterized by using their
statistical moments. For each of the six epochs of simulation $\cal I$
that are shown in Fig.~\ref{fig:rho-pdfs} we compute
the first four moments which are defined in the following way:
\begin{eqnarray}
{\rm mean} & = \,\mu\, = & \frac{1}{N}\sum_{i=1}^N \rho_i\;,\label{eqn:mom1}\\
{\rm standard \;\;deviation} & =\, \sigma \,= & \sqrt{\frac{1}{N}\sum_{i=1}^N (\rho_i-\mu)^2}\;,\label{eqn:mom2}\\
{\rm skewness} & =\, \theta \,= & \frac{1}{N}\sum_{i=1}^N \frac{(\rho_i-\mu)^3}{\sigma^3}\;,\label{eqn:mom3}\\
{\rm kurtosis} & =\, \kappa \,= & \frac{1}{N}\sum_{i=1}^N \frac{(\rho_i-\mu)^4}{\sigma^4}\;.\label{eqn:mom4}
\end{eqnarray}
The summations extend over all $N=500\,000$ SPH particles in the
system. Mean $\mu$ and standard deviation $\sigma$ quantify the
location and the width of the pdf and are given in density units. The
third and fourth moments are dimensionless quantities characterizing
the shape of the distribution. The skewness $\theta$ describes the
degree of asymmetry of the distribution around its mean. A negative
value indicates an asymmetric tail extending towards the negative
side, a positive value indicates a positive tail in the distribution.
The kurtosis $\kappa$ measures the relative peakedness or flatness of
the distribution. Gaussian distributions are characterized by
$\kappa=3$. Smaller values indicate distributions that have flatter
peaks, and larger ones point towards strong peaks or equivalently the
existence of prominent tails. A pure exponential distribution leads to
$\kappa=6$. The values for model $\cal I$ are listed in
Tab.~\ref{tab:moments}. Since we analyze a mass weighted density
distribution, the first moment $\mu$ is not identical to the global
average density, $\rho = 1/8$. Most of the mass is in gas of larger
densities. Using only the first two moments (i.e.~a Gaussian fit) to
reconstruct the logarithmic density pdf never leads to a good
representation of the data except for the initial configuration
(Fig.~\ref{fig:rho-pdfs}a).
In summary, the filamentary morphology of the gas distribution and the
power-law appearance of the density pdf in our models is clearly a
result of the presence of self-gravity which leads to collapse and
compression and thus to the occurance of high density contrasts. The
hydrodynamic evolution of isothermal gas naturally produces
log-normal density pdf's (see also \cite{passot98}). The inclusion of
self-gravity leads to deviations from this simple analytical form
towards more complex pdf's which during most stages of the dynamical
evolution of the system can be described by power laws at the large
density part of the distribution.
\subsubsection{The line-of-sight velocity centroid pdf}
One way to infer the three-dimensional structure of molecular clouds
from the observed column density distribution is to include the
velocity information along the line-of-sight. Using velocity coherence
as indication for spatial correlation one is able to deconvolve the
cloud into smaller subunits, i.~e.~into clumps on different
scales. This is commonly applied to a variety of molecular clouds
(\cite{lor89}, \cite{stu90}, \cite{willi94}, \cite{oni96},
\cite{kram98}, \cite{heithausen98}) and we will discuss the clump
structure of our models in the next section. However, other
combinations of spatial and velocity information are possible and
useful as well. There are attempts to characterize the structure of
molecular clouds (and the turbulent interstellar medium in general) by
using the pdf's of molecular line centroid velocity fluctuations
(\cite{miesch95}, \cite{lis98}, \cite{miesch98}). The idea is to
determine the centroids of molecular lines for a large number of
different positions across the face of a molecular cloud and from that
compute their distribution function in velocity space. Like in the
case of the density pdf, by doing so direct spatial information is
lost, but the velocity centroid pdf still contains statistical
information about the cloud and may be used to differentiate between
different theoretical models. For instance, it has been shown
experimentally and in numerical simulations that the pdf of the
velocity field for incompressible turbulence is very nearly Gaussian
although non-zero skewness should exist to some degree (e.g.~Vincent
\& Meneguzzi 1991, Cao et al.~1996, Vainshtein 1997, Lamballais et
al.~1997). This situation changes in the regime of highly compressible
supersonic turbulence, the velocity pdf can deviate quite
significantly from the Gaussian form (Lis et al.~1998, Klessen 2000,
Mac~Low \& Ossenkopf 2000). This behavior is also found in the
simulations of clustered star formation discussed in this paper. Here
it is due to the presence of self-gravity which introduces strong
collapse motions into the velocity field.
Again using data from the high resolution model $\cal I$,
Fig.~\ref{fig:vel-pdfs} shows centroid pdf's for the line-of-sight
velocities along the $x$-, $y$-, and $z$-axis at four different stages
of the dynamical evolution of the system, (a) at $t=1.2$ just when the
first dense cores have formed and contain roughly 1\% of the total
mass, (b) at $t=2.0$ when altogether 33 cores have formed and accreted
30\% of the total mass, (c) at $t=2.8$ when 60\% of the available gas
mass has been accreted, and finally (d) at $t=3.9$ when 90\% of the
mass is accounted for by a cluster of 56 cores. We divide the face of
the simulation cube into $50^2$ cells of size $0.04^2$. For each cell,
we compute the line profile by binning the normal (line-of-sight)
velocity components of all gas particles that are projected into that
cell; the width of each velocity bin is $\Delta v = 0.05$ which is
60\% of the sound speed $c_{\rm s} = 0.08$. This procedure
corresponds to the formation of optically thin lines in molecular
clouds where all molecules within a certain column through the clouds
contribute to the shape and intensity of the line. We obtain the
centroid of each line and the desired pdf by plotting the histogram
of all centroid velocities. To reduce the sampling uncertainties, we
repeat the same procedure, but with the location of each cell shifted
by half a cell size. Altogether 5000 lines contribute to the pdf,
each line being determined by the velocities of on average 200 SPH
particles. The left column depicts the pdf for the line-of-sight along
the $x$-axis (i.e.~for particle projections onto the $yz$-plane), and
the middle and right column for the $y$- and $z$-component,
respectively. Note that in the linear-log plots the Gaussian
distribution has a parabolic shape. The small inlay in the upper
left corner of each histogram gives the {\em overall} line profile
resulting from observing the entire cube, i.e.~including the
line-of-sight velocities of {\em all} gas particles in the
system. Analog to plotting the pdf, the horizontal axis of the inlay gives the
velocity in linear scale (each tick mark denoting a velocity increment
$\delta v = 1.0$) and the vertical axis the logarithm of the number of
particles that contribute to a velocity bin (with the separation of
each tick mark being $\log_{10} N = 1.0$).
Since the thermal sound speed is $c_{\rm s} = 0.08$ all line widths and
the widths of the centroid pdf's are highly supersonic. Furthermore,
none of the pdf's resembles a Gaussian. For comparison we have plotted
for each of the pdf's a Gaussian curve constructed from using the
first two moments of the distribution presented in
Tab.~\ref{tab:moments} (the dashed parabolas in each plot). This curve
never fits any of the pdf's and is typically too small in width. The
distributions show extended tails and can appear quite asymmetric as
indicated by the large values for the third and fourth moments, also
listed in Tab.~\ref{tab:moments}. Furthermore the pdf's show
considerable substructure, for instance notice the double peaked
distribution in Fig.~\ref{fig:vel-pdfs}b ($z$-component) or
Fig.~\ref{fig:vel-pdfs}c ($y$-component). This is a sign of the
streaming motion of gas onto a local center of gravity along filaments
which by chance are aligned with the line-of-sight. Some pdf's show
nicely developed exponential tails, like Fig.~\ref{fig:vel-pdfs}c
($z$-component) or Fig.~\ref{fig:vel-pdfs}d ($y$-component). As
collapse and star formation progresses in time the width of the pdf's
grows. However, pdf's at equal times but obtained from different
projection angles also exhibit considerable deviations from each
other. Depending on whether one looks along a major axis of
contraction or perpendicular to it the width of the pdf appears larger
or smaller. In addition, massive gas clumps moving through the cloud
will lead to individual features in the velocity pdf. Altogether, the
clumpy structure of molecular clouds causes considerable substructure
in the observed velocity pdf, it will never appear as a smooth function.
One of the implications from our sample is that the velocity centroid
pdf's in star forming regions are expected to drastically deviate from
the Gaussian shape predicted from hydrodynamic turbulence models. This
is indeed observed (\cite{lis98}, \cite{miesch98}). In our models this
is due to the presence of self-gravity inducing strong contracting
motions and clumpiness. However, other effects may lead to quite
similar distortions as well (see Klessen 2000 for more details).
\subsection{Evolution of clumps and cores}
\label{subsec:evolution}
\subsubsection{Clump mass spectrum}
\label{subsubsec:clump-mass-spectrum}
The structure of molecular clouds is extremely complex and
observations reveal a network of intersecting filaments and clumps on
all scales (e.g.~\cite{bal87,falg92,wise96}). Molecular clouds may be fractal and
various studies of the mass spectrum $N(m)$ of molecular clouds and of the
gas clumps inside of clouds indicate that their distribution may be
approximated by a power law of the form $dN/dm \propto m^{\alpha}$
with exponent $\alpha \approx -1.5$ (see e.g.~\cite{lor89,stu90,willi94,oni96,kram98,heithausen98}). This universal law is an important
constraint and test for models of molecular cloud
evolution. Numerical simulations {\em must} be able to reproduce this
structural feature of observed clouds. Indeed, our isothermal gas models
naturally lead to a power-law clump mass spectrum which is a
result of the interaction between gravity and gas pressure.
Representative for the entire set of isothermal larger-scale collapse
calculations, Fig.~\ref{fig:clump-spectrum-Z} shows the mass
distribution of clumps and condensed cores at four different stages of
the dynamical evolution of the high-resolution model $\cal I$, namely
initially and when 10\%, 30\% and 60\% of the available gas has been
accreted onto protostellar cores. For each of these times the upper
panels compare the mass distribution of detected gas clumps (thin
line) with the observed clump mass distribution $dN/dm \propto
m^{-1.5}$, which translates into a slope of $-0.5$ when using
logarithmic mass bins (dashed line). To identify
individual clumps, we have developed a clump-finding algorithm similar
to that of Williams et al.~(1994), but based on the framework of
SPH. For details see appendix \ref{app:clumps}. The thick line depicts the
mass distribution of condensed protostellar cores that have formed
within the more massive gas clumps (a detailed analysis of this
process is given in Sec.~\ref{subsubsec:growth-protostellar-cores}).
The lower panels show the position of each gas clump in a
mass--density diagram. Clumps without cores are shown as crosses.
Clumps containing one single core are denoted by open circles, those
with multiple cores by filled circles. Unresolved, very condensed
cores are plotted as stars with the density of cores being defined as
the core mass divided by the volume within the fixed accretion
radius. Therefore, they all fall along a straight line with slope
$1/3$. The isothermal Jeans mass as function of density is indicated
by the diagonal line which separates the diagram into two regions.
Clumps that lie to the right exceed their Jeans mass and are due to
collapse, whereas clumps to the left are stabilized by gas pressure.
The vertical line indicates the SPH resolution limit
(Sec.~\ref{subsec:resolution-limit}). Clumps to the right of this line
are well resolved, objects to the left are ill-defined and may be
spurious results of the clump find algorithm.
The clumping properties in all simulations, from model $\cal A$ with
$50\,000$ SPH particles to model $\cal I$ with ten times more
particles, are remarkably similar. This suggests that the dominant
dynamical processes are well treated and resolved. In all simulations,
the clump spectrum of the initial gas distribution {\em cannot} be
described by a simple power-law, it reflects the structural properties
of the Gaussian random field from which the initial conditions are
generated. The initially linear growth of density perturbations is
not able to generate a hierarchical structure on all scales. This
requires considerable non-linear gravitational evolution to take
place. In the subsequent self-consistent dynamical evolution, the
clump distribution quickly achieves a universal mass distribution with
a power-law slope which is in excellent agreement with the observed
exponent $\alpha \approx -1.5$ (dashed line). The core distribution,
on the other hand, deviates significantly from a power-law
distribution for smaller masses. Note, that the relative
underabundance of low-mass cores with respect to low-mass clumps is
not a resolution effect (see the vertical thin lines). All cores are
well resolved in our models. Considerable deviations from a simple
power-law spectrum of clump masses occur again in the very late phases
of the evolution when most of the material is accreted onto dense
cores and the gas reservoir becomes significantly depleted. Then, huge
voids of very low density open up and the entire system more strongly
resembles a (proto)star cluster than a molecular cloud. Within these
two extreme phases of the isothermal gas evolution, from the linear
initial gas distribution to the final star cluster, the complex
interplay between gravity and gas pressure naturally creates a
hierarchical structure with a distribution of clumps that is best
described by a simple power-law, in agreement with the
observations. As a reminder, when scaled to low density molecular
cloud regions with densities $n({\rm H_2}) \approx 10^2\,{\rm
cm}^{-3}$ (e.g.~Taurus) the total mass, $M=1$, of the system corresponds to
$6\;\!300\,$M$_{\odot}$. When considering high density regions with
$n({\rm H}_2) \approx 10^5\,{\rm cm}^{-3}$ where clusters of stars can
form (e.g.~the BNKL-region in Orion) then the physical mass of our
simulation is $200\,$M$_{\odot}$ (see Sec.~\ref{subsec:conversion}).
The lower panels in Fig.~\ref{fig:clump-spectrum-Z} indicate that
initially only very few high-mass clumps exist which exceed the local
Jeans mass. These clumps collapse rapidly and form the first
condensed cores in their central region (see also
Sec.~\ref{subsubsec:ind-clumps}). During the subsequent evolution the
number of Jeans-unstable clumps and subsequently of protostellar cores
increases, populating a larger region to the right of the diagonal
line. If clumps merge which already contain protostellar cores, the
new more massive clump will contain multiple cores in its interior.
Competitive accretion (Price \& Podsiadlowski 1995; Bonnell et
al.~1997) and the gravitational interaction between the cores has
important consequences for the further evolution. It influences their
gas accretion and thus the final mass spectrum (more details are
discussed in Sec.~\ref{subsubsec:dyn-WW}). These often unpredictable,
probabilistic processes, resulting from chaotic multiple $N$-body
interactions of cores, are responsible for the difference in the
distribution of clump masses and of core masses. Whereas gas clumps
evolve according to the laws of hydrodynamics, cores behave like
accreting and gravitationally interacting $N$-body system. As a
result, the mass spectrum of protostellar cores is well approximated
by a log-normal functional form whereas the clump spectrum shows a
power-law distribution.
\subsubsection{Formation and mass growth of cores}
\label{subsubsec:growth-protostellar-cores}
The location and the time at which protostellar cores form, is
determined by the dynamical evolution of their gas cloud. Besides
collapsing individually, clumps stream towards a common center of
attraction, where they may merge with each other or undergo further
fragmentation. As can be seen in Fig.'s \ref{fig:3D-cube-A} and
\ref{fig:3D-cube-I}, isothermal models evolve into a network of
intersecting sheets and filaments. The gas density is highest along
filaments and at their intersections. These are the locations where
dense cores form predominantly. The time of their formation in the
centers of unstable clumps depends strongly on the relation between
the time scales for individual collapse and fragmentation and
clump-clump merging. When clumps merge that already contain cores
the dynamical interaction between the cores becomes important.
As illustration, we show in Fig.~\ref{fig:accretion-history} the
accretion history for a number of selected protostellar cores (a) from
model $\cal A$ and (b) from model $\cal I$. The objects are numbered
chronologically according to their time of formation. Model $\cal A$
forms altogether 60 cores and model $\cal I$ 55. The figure shows the
following trends:
(a) The cores which form first tend to have the largest final
masses. They emerge from clumps that were present from the beginning
with the largest masses and highest densities and which are identified
with the most significant peaks in the fluctuation field. Jeans
unstable clumps with these qualities have very short collapse time
scales. Hence, these clumps quickly form a single cores which
are likely to accrete a large fraction of the parental gas before
the dynamical evolution of their environment changes the clump
properties too much, i.e.~before the clumps get dispersed or merge
with other clumps. If a clump which already contains a core merges
with a non-collapsed gas clump, for example when streaming along a
filament towards a common center of attraction, then also the new
clump is likely to become completely accreted onto these cores. Since
the first cores form in the highest-density regions more or less
independent of each other, their mass growth rate is dominated by the
matter originating from their vicinity.
(b) On the other hand, matter that contracts into dense cores at later
times ($t\gtrsim2$) has already undergone considerable dynamical
evolution. Clumps that were initially not massive enough to collapse
onto themselves merge until enough mass is accumulated for them to
collapse and build up a new protostellar core. This needs time. For
those cores that form predominantly in the late stages of the
dynamical evolution the available gas reservoir is already
considerably depleted. Therefore, their average mass is rather
small. The final mass of the protostellar cores (i.e.~at the time when
90\% of the available gas mass has been accreted) for the standard
model $\cal A$ and the high-resolution model $\cal I$ in order of
their formation is plotted in Fig.~\ref{fig:sink-masses}.
Another aspect of the accretion process onto individual cores is
illustrated in Fig.~\ref{fig:mass-region}. For the first nine cores
(Fig.~\ref{fig:mass-region}a) and for the last nine cores
(Fig.~\ref{fig:mass-region}b) that form in the high-resolution model
$\cal I$, it plots the average initial distances at $t=0.0$ between
the accreted SPH particles and the particle that gets converted into
the sink particle during the course of the simulation. This shows the
volume from which particles accrete onto protostellar cores. As
indicated above, the cores in Fig.~\ref{fig:mass-region}a accrete most
particles and the bulk of their final mass from their vicinity,
i.e.~from a distances less than $\sim 0.5$. With the total size of the
simulated cube being $2^3$, this corresponds to roughly $1/64$ of the
total volume. If the material would be randomly sampled from a
homogeneous cube of size $2^3$, the average distance would be $\sim
1.3$. On the other hand, the smaller cores in
Fig.~\ref{fig:mass-region}b consist of matter that originates from a
larger volume, much closer to the value for random sampling. This
indicates that these clumps accrete from matter that already has
undergone considerable dynamical evolution and is well mixed.
Further information about the processes determining the formation of
and accretion onto dense cores is given in
Fig.~\ref{fig:clump-accretion}. It plots the contributions from
individual clumps to the final mass of selected cores. As a general
trend, cores that form very early can accrete a large fraction of
their parental and neighboring clumps before clump interaction and
merging becomes important. On the other hand, matter that builds up
protostellar cores at the late stages of the dynamical evolution has
participated in large-scale motions and successive clump merging. It
is well mixed and the initial clump assignment is no longer
significant. Furthermore, the competitive accretion amongst groups of
cores in the interior of multiple merged clumps becomes more important
with the progression of time. This phenomenon also contributes to the
distribution of available gas onto a larger number of
cores. Therefore, the fraction $\left \langle f_j \right \rangle$ of
material of individual clumps that gets accreted onto the cores
decreases with time.
\subsubsection{The Importance of dynamical interaction processes}
\label{subsubsec:dyn-WW}
Besides the effects discussed in the previous section, the growth rate
of protostellar masses is strongly affected by the dynamical
interaction between the cores themselves. This phenomenon is closely
related to the competitive accretion of multiple cores within one
common gas envelope (\cite{price95}, \cite{bonnell97}) and becomes
important when clumps merge that already contain dense cores. In the
center of the larger merged clump protostellar cores interact
gravitationally with each other. Like in dense stellar clusters,
close encounters may lead to the formation of unstable triple or
higher-order systems and alter the orbital parameters of cluster
members. During the decay of unstable subsystems, protostellar cores
can get accelerated to very high velocities and may leave the
cluster. During this process, the less massive cores are more likely
to become ejected. Protostellar cores that get expelled from their
parental clump are suddenly bereft of the massive gas inflow from
their collapsing surrounding. They effectively stop accreting
and their final mass is determined. The dynamical interaction between
cores is an important agent in shaping the mass distribution.
As an example, we discuss the history of protostellar cores in
the standard model. From Fig.~\ref{fig:accretion-history}a it is evident
that core \#5 stops accreting at $t\approx 0.8$, long before the
overall gas reservoir is depleted. The same applies to core \#7 at
$t\approx 0.9$. In both cases, the objects were involved in a 3-body
encounter that resulted in the expulsion from their gas rich parental
clumps. Fig.~\ref{fig:trajectory} depicts the trajectories of the cores
\#3, \#5, \#11, \#31, and \#46. At $t \approx 0.3$ core \#3 forms
within an overdense region and slightly later cores \#5 and \#11 form
in its vicinity from other Jeans-unstable density fluctuations. Their
parental gas clumps merge and the whole systems flows towards a local
minimum of the gravitational potential. The three cores soon build an
unstable triple system, continuing to accrete from the converging gas flow they
are embedded in (the detailed trajectories are shown in
Fig.~\ref{fig:trajectory}b, note the larger scale). At $t\approx 0.8$,
core \#5 is expelled and the remaining two cores form a wide binary,
which at $t \approx 1.6$ suddenly hardens due to the gravitational
encounter with core \#46 (see Fig.~\ref{fig:trajectory}c). Also this
hard binary is transient, the interaction with core \#31 pumps energy
into its orbit, and in the subsequent evolution, the orbital period
increases further during the encounter with a dense gas filament whose
tidal influence finally disrupts the binary at $t\approx 2.4$. At
that time, accretion stops.
As another example from the high-resolution simulation,
Fig.~\ref{fig:accretion-history}b shows that at $t\approx 1.8$, core
\#19 stops accreting. It is thrown out of a dense clump at the
intersection of two massive filaments by a triple interaction with
cores \#1 and \#17. However, it remains bound to the gas clump
which grows in mass due to continuous infall and falls back into the
clump and resumes accreting at $t\approx2.0$. Cores \#9 and \#41 are
also expelled from their parental clumps but, unlike core \#17, their
accretion is terminated completely. These dynamical interactions
between protostellar cores influence their mass distribution
significantly. In reality, ejected protostars may travel quite far and
could explain the extended distribution of weak-line T Tauri stars
found via X-ray observations in the vicinities of star-forming
molecular clouds (e.g.~\cite{neu95}, \cite{sterzik95}, \cite{kraut97},
\cite{wich97}).
As in our numerical scheme cores (i.e.~sink particles) only interact
gravitationally with each other, possible hydrodynamic processes are
suppressed. Dense cores that come close to each other exert mutual
tidal torques, they may merge or on the contrary strip off matter and
loose mass during this process. However, we estimate these effects to
be small. Matter accreted onto core particles would continue to
collapse inwards to form a young stellar object surrounded by a disk
on a very short time scale. The boundary introduced by the sink
particle is determined by the resolution limit of the code. Scaled to
physical units the diameter of the core particle is roughly 600 AU in
case of high densities of $n({\rm H}_2) \approx 10^5\,{\rm cm}^{-3}$
(as in the Orion star forming region) and about $10\,000$ AU in low
density regions like Taurus (using $n({\rm H}_2) \approx 10^2\,{\rm
cm}^{-3}$ and $T\approx10\,$K). Matter inside the core is always
expected to be strongly concentrated in the protostar in the
center. The necessary cross section for merging or significant tidal
perturbation (e.g.~Hall, Clarke \& Pringle 1996) is therefore
typically much smaller than the size of the sink particles. To
estimate how possible core mergers could influence the conclusions
derived in this paper we took the initial conditions of simulation
$\cal A$ and repeated the calculation allowing for core particles to
join together. We studied three different cases. The most stringent
one was that we merged sink particles if they overlap by at least
90\%, then if they overlap by 10\% or more, and finally if two sink
particles just touch each other. The resulting shape and width of the
final core mass distribution remained essentially unchanged in all
three cases compared to simulation $\cal A$, only the number of cores
decreased.
\subsection{Properties of (individual) gravitationally unstable clumps}
\subsubsection{Properties of individual clumps}
\label{subsubsec:ind-clumps}
Whereas the previous section discussed global features of the clump
and core distribution, this section investigates in detail the
properties of individual clumps. This is important for studies of
the collapse of single clumps and their fragmentation into binaries
and multiple stellar systems
(\cite{burkert93}, \cite{burkert96}, \cite{boss97}, \cite{burkert98},
\cite{burkert98b}). The
simulations presented here are no longer able to resolve the collapse
and sub-fragmentation of individual protostellar cores once they have
been substituted by sink particles. To study this subsequent
evolution, a new simulation just concentrating on the collapse of one
single core would be required. To connect the final stages of the
larger-scale simulations discussed here and the initial conditions for
detailed collapse calculations, knowledge about the properties of
individual cores is necessary. Important parameters are their density
structure and their rotational properties. We note, however, that gas
clumps are never isolated. Interactions with the environment are very
important for the mass spectrum, the spatial distribution and the
geometrical shape of clumps and cores.
The clumps generally are highly distorted and triaxial. Depending on
the projection angle, they often appear extremely elongated, being
part of a filamentary structure which may appear as a chain of
connecting, elongated individual clumps. More complicated irregular
shapes are also common which result from recent clump mergers at the
intersections of filaments. A selection of clump shapes is presented
in Fig.~\ref{fig:ind-clumps} which plots the contour lines for eight
high-density clumps in simulation $\cal I$ at time $t=1.2$ projected into the
$xy$-, $xz$- and $yz$-plane. At this state of the dynamical
evolution, four collapsed cores have formed which accreted 1\% of the
total gas mass. This time is most appropriate to determine clump
properties, because the system has already undergone substantial
evolution (the power-law clump spectrum is well established) but most
of the gas has not yet been accreted onto protostellar cores. The
clumps that already formed dense cores in their interior are plotted
in the first part of the figure, the second part shows clumps
whose central densities are not yet high enough to form a
condensed core. The clumps are numbered according to their peak
density, i.e.~clump \#1 has the highest central density. The (surface)
density contours are spaced logarithmically with two contour levels
spanning one decade, $\log_{10}\Delta \rho = 1/2$. The lowest contour
is a factor of $10^{1/2}$ above the mean density $\left \langle \rho
\right \rangle = 1/8$. The black dots indicate the positions of the
dense protostellar cores. Note the similarity to the appearance of
observed dense (pre-stellar) clumps (for instance Fig.~1 in
\cite{myers91}). It is clearly visible that the clumps are very
elongated. The ratios between the semi-major and the semi-minor axis
measured at the second contour level are typically between 2:1 and
4:1. However, there may be significant deviations from simple triaxial
shapes, see e.g.~clump \#4 which is located at the intersection of two
filaments. This clump is distorted by infalling material along the
filaments and appears `y'-shaped when projected into the
$xz$-plane. As a general trend, high density contour levels typically
are regular and smooth, because there the gas is mostly influenced by
pressure and gravitational forces.
On the other hand, the lowest level samples gas that is
strongly influenced by environmental effects. Hence, it appears patchy
and irregular. The location of the condensed core within a clump is
not necessarily identical with the center-of-mass of the clump,
especially in irregularly shaped clumps.
Typically, the overall density distribution of identified clumps in
our simulations follows a power law and the density increases from the
outer regions inwards to the central part as $\rho(r) \propto
1/r^2$. For clumps that contain collapsed cores, this distribution
continues all the way to the central protostellar object. However, for
clumps that have not yet formed a collapsed core in their center, the
central density distribution flattens out. As an example,
Fig.~\ref{fig:radial-density-profile} plots the averaged radial
(surface) density profile of the $xz$-projection of clump \#4 and of
clump \#12. This is in accordance with analytical models of
isothermal collapse (see \cite{lars69}, \cite{penston69}; and also
\cite{whit85}) and with the observational data for dense cores in
dark molecular clouds (see e.g.~\cite{myers91}, \cite{ward94},
\cite{chini97}, \cite{motte98}).
\subsection{Properties of the evolving protostellar cluster}
\subsubsection{Binarity and clustering properties of protostellar cores}
\label{subsubsec:clustering}
An important statistical quantity derived from observational data is
the 2-point correlation function or equivalently the mean surface
density of companion stars $\xi$ as function of separation $r$.
Larson (1995) found that in the Taurus star-forming region the mean
surface density of companions follows a power law as function of
separation with a break in the slope at the transition from the binary
to the large-scale clustering regime. It occurs at separations for
which the binary system blend into the background distribution of
cluster stars. This analysis has also been applied to other
star-forming regions (\cite{simon97}, \cite{bate98},
\cite{gladwin98}) . Larson identified the transition separation with
the typical Jeans length in the molecular cloud. For Taurus this may
be true. In Orion this conclusion fails. In general, it can be shown
that the transition separation depends on the volume density of stars,
the extent of the star-forming regions along the line-of-sight, the
volume-filling nature of the stellar distribution and on the details
of the binary distribution (\cite{bate98}). Additionally, the
transition separation evolves in time due to dynamical interactions
amongst the cluster members. Note that the 2-point correlation
function is never unique. It cannot differentiate between hierarchical
(fractal) and non-hierarchical structure, and different stellar
distributions may lead to the same 2-point correlation
function. Therefore, the additional analysis of the spatial
distribution by eye is essential for a meaningful interpretation of
$\xi$. Alternatively, to obtain quantitative results, the calculation
of higher-order correlation functions may be useful, which is common
in studying the large-scale structure of the universe
(\cite{peebles93}). However, for star-forming regions this has never
gone beyond the second order terms. Hence, we restrict our current
analysis to a discussion of the 2-point correlation function $\xi$ of
the spatial distribution of protostellar cores (sink particles).
Figure~\ref{fig:larson-X} plots the mean surface density of companions
$\xi$ as function of separation $r$ for the protostellar cluster that
forms in the standard model $\cal A$ at four different times of its
evolution. At each point in time, the left panel gives the spatial
distribution of the protostellar cluster projected in the $xy$-, $xz$-
and $yz$-plane. The resulting functions $\xi$ for each projection are
shown in the right panel. We use data from model $\cal A$ because it is the
model which we advanced furthest in time. However, the conclusions
apply to all models discussed in this paper: The initial binary
fraction is very large. Roughly 75\% of the cores are part of a
(bound) binary or higher-order system which dominate the function
$\xi$ at small separations $r$. At larger $r$ the function $\xi$
flattens out which reflects the fact that the binary systems are
relatively homogeneously distributed throughout the entire volume
(Fig.~\ref{fig:larson-X}a). The value of
the plateau of $\xi$ indicates the projected density of this
homogeneous distribution.
At the intersections of filaments clumps which already contain one or
more cores merge, forming larger clumps which subsequently contain a
cluster of cores. As a result of this process, in
Fig.~\ref{fig:larson-X}b one sees several small aggregates distributed
throughout the simulated volume. Within these small dense clusters
complex dynamical interaction between cores takes place (cf.~with
Sec.~\ref{subsubsec:dyn-WW}) leading to the destruction of soft
(i.e.~weakly bound) binaries. At the same time hard binaries become
even harder (i.e.~more strongly bound) and the function $\xi$ extends
to smaller spatial separations. However, due to this secular evolution
the overall binary fraction decreases. During the progression of the
evolution, the small protostellar aggregates follow the streaming
motion of their surrounding gas envelopes and merge, thereby forming a
single bound cluster (see Fig.~\ref{fig:larson-X}c). The protostellar
system has reached a state of minimum spatial extent. The overall size
of the cluster is reflected by the steep decline of the function $\xi$
at large separations.
At that stage about 90\% of the available gas has been
accreted and the subsequent evolution of the cluster is almost
entirely determined by gravitational $N$-body interactions. The cluster
re-expands and rapidly develops a core/halo structure which is typical
for collision-dominated self-gravitating $N$-body systems (see
e.g.~\cite{BT87}). This process is expedited by the fact that cores
which have been accelerated to high velocities by close encounters
with hard binaries cannot leave the simulation box due to the periodic
boundary conditions. Once those `bullets' have trespassed the
boundaries of the cube, they are reinserted at the opposite side with
the same velocity and may again interact with other cluster members to
transfer energy to more slowly moving cores. In reality they would be
ejected from the cluster and leave the star-forming cloud. At these
late stages of the evolution, a clear change in the slope of the
function $\xi$ can be seen at the break between the binary regime and
the large-scale clustering (see Fig.~\ref{fig:larson-X}d). The
fraction of fast moving ejected cores has grown to roughly 50\%. At
the same time, the binary fraction has decreased to 15\%.
Duquennoy \& Mayor (1991) analyzed a sample of nearby G-dwarfs and
found that the distribution of orbital periods of binary stars in
their sample follows approximately a broad Gaussian with a peak at
$\sim 10^{4.8}$ days which corresponds to a separation of $\sim
32\;$AU (using Kepler's third law and assuming the typical mass of a
G-dwarf is $\sim 1\,$M$_{\odot}$). From the broad distribution of
orbital periods it follows immediately that the function $\xi$ of mean
surface density of binary companions decreases with separation as
$1/r^2.$\footnote{A broad distribution of orbital periods (in
$\log_{10} P$) is equivalent to a broad distribution of binary
separations $\log_{10} r$. To the lowest order it can be approximated
as being flat, i.e.~$dN/d\log_{10} r = K$. Then, $dN = K/r\;dr$ is the
number of binary companions with separations in the range $r$ to $r +
dr$ and the mean surface density of companions in the 2-dimensional
projection follows as $\xi(r) = N^{-1}dN/(2\pi r dr) = N^{-1}K/ (2\pi
r^2) \propto 1/r^2$.} The same behavior is found in our
systems. Figure \ref{fig:binary-separations} illustrates the
distribution of semi-major axes for the identified {\em bound} pairs
of cores in simulation $\cal A$ at selected times. For each core its
closest neighbor is found and if the pair is bound (i.e.~if the sum of
relative kinetic and potential energy is negative) the kinematic
data at the present time are used to calculate the orbital
parameters. In case of isolated pairs this works well, however, most
cores are part of higher-order systems. In this case the given value
is only a rough estimate of the true orbital characteristics as in
unstable and highly chaotic few-body system well defined orbital
parameters do not exist. We find that the logarithmic distribution of
binary separations is broad and relatively flat and correspondingly
that the function $\xi$ falls off as $1/r^2$ in the binary regime
(Fig.~\ref{fig:larson-X} at small $r$). When compared with the
observations it should be noted furthermore that the systems of
protostellar cores in our simulations correspond to wide
binaries. Close binary system would form from {\em sub}-fragmentation
of individual protostellar cores (e.g.~Burkert \& Bodenheimer 1996).
This is a process which we cannot treat in the present numerical
scheme. However, as indicated in Fig.~\ref{fig:ang-momentum-2}, also
the angular momenta of individual protostellar cores in our models
follow a relatively flat and broad logarithmic distribution spanning
at least three orders of magnitude. If we assume that a rotating
protostellar core typically breaks up into a binary system with
similar angular momentum the period distribution of close binaries
should also be broad, as observed. This result follows from the
stochastic processes which determine the formation and evolution of
protostellar cores. They influence the properties of binary systems
and at the same time also affect the angular momentum distribution of
individual cores.
\subsubsection{Rotational properties of protostellar cores}
\label{subsubsec:rotational-properties}
A very important parameter for core collapse and possible
sub-fragmentation is their specific angular momentum.
Conservation of angular momentum prevents material from the envelope
of a star forming core
to directly accrete onto the central protostar. Only the inner part
is able to immediately fall onto the star, the bulk of the
infalling envelope accumulates in a rotationally supported disk around
the central object. Angular momentum is transferred outward on
viscous time scales $\tau_{\rm vis}$, by this moving matter inwards
towards the central star. Typically, $\tau_{\rm vis}$ is by a factor of
$10\;$--$\;100$ larger than the rotational time scale which is
comparable to the free-fall time scale $\tau_{\rm ff}$ (see
e.g.~\cite{pringle81}).
These processes take place deep inside the protostellar cores, hence they
cannot be resolved by our numerical scheme. However, we can keep track of the
the total angular momentum accreted onto each condensed core in our
simulation. Figure~\ref{fig:ang-momentum-1} plots the time evolution of the
angular momentum vector $\vec{L}$ of all protostellar cores in the
high-resolution model $\cal I$. The left panel describes the evolution of each
individual component of the vector. The right panel plots the distribution of
$L_x$, $L_y$ and $L_z$ at the end of the simulation at $t=3.9$. For
comparison, we overlay the final distribution (at $t=5.6$) of the angular
momenta of the cores in the standard model $\cal A$ with dashed lines. Both
distributions are statistically indistinguishable. From the plots, we see that
the evolution of the angular momentum of individual cores can be very complex
and intimately reflects the rotational properties of the environment they are
embedded in: the angular momentum of the protostellar core is determined by
the angular momentum of the clump it forms in. Clumps that merge at the
intersection of two filaments can accumulate considerable angular momentum
which is transfered onto the core by accretion. The angular
momentum vector of cores may even change its sign if material is accreted
which rotates counterclockwise with respect to the core.
The distribution of the absolute values of angular momenta
$|\vec{L}\;\!|$ of the protostellar cores in simulation $\cal I$ is
plotted in Fig.~\ref{fig:ang-momentum-2}a. Values are shown for
$t=2.8$ (solid line), i.e.~when 60\% of the gas is converted into
dense cores, and at the final stage, $t=3.9$ (dashed line). As already
indicated in Fig.~\ref{fig:ang-momentum-1}, the angular momenta are
very broadly distributed and peak at $\sim 10^{-5}$. Note again that
all values are given in dimensionless units with length and mass
scales set to one. Figure~\ref{fig:ang-momentum-2}b plots the
distribution of the specific angular momenta $|\vec{L}\;\!|/m$ of the
cores, again at $t=2.8$ (solid line) and at $t=3.9$ (dashed line).
The mass dependence of $|\vec{L}\;\!|/m$ is shown in
Fig.~\ref{fig:ang-momentum-2}c. There are no massive cores with low
specific angular momentum. However, there is no clear correlation
between both quantities.
Using the scaling properties discussed in
Sec.~\ref{subsec:conversion}, the dimensionless $|\vec{L}\;\!|/m$
values can be converted into physical units. When we apply our model
to the high-density regime (e.g.~to the BN region in the Orion
molecular cloud with $n({\rm H}_2) \approx 10^5\,{\rm cm}^{-3}$) then
$|\vec{L}\;\!|/m=1$ corresponds to $10^{23}\,{\rm cm}^2{\rm
s}^{-1}$. In the case of low densities of $n({\rm H}_2) \approx
10^2\,{\rm cm}^{-3}$ (the average value in Taurus) this converts into
$3.7\times 10^{24}\,{\rm cm}^2{\rm s}^{-1}$. The specific angular
momenta of protostellar cores in our simulations are therefore of the
order of $10^{20}$ to $10^{21}\,{\rm cm}^2{\rm s}^{-1}$ and agree
remarkably well with the observed values for wide binaries and
protostellar objects (e.g.~Bodenheimer 1995).
When looking at the spatial distribution of the angular momentum
vectors of protostellar cores, there is a correlation between the
location and the orientation. As is visible in
Fig.~\ref{fig:ang-momentum-3}, the angular momentum vectors of cores
that form in the same region tend to be aligned. These cores form
from gas that has similar global flow patterns. In clumps which
contain multiple cores as the result of merging, all cores accrete
from a common environment with a certain well defined angular momentum
vector. As a result, the rotational properties of these cores tend to
be comparable. In some star-forming regions, molecular outflows from
young stellar objects indeed appear to be correlated and aligned
(e.g.~in the northern part of the L1641 cloud, see Fig.~14 in the
review article by \cite{reip89}). Assuming that protostellar outflows
are associated with the rotational properties of the protostellar
object they emerge from, this implies a correlation between the
angular momenta of the protostellar cores similar to the one found in
our numerical models. On the other hand, in other regions no
correlation is found at all (see Eisl{\"o}ffel \& Mundt 1997, Stanke,
McCaughrean \& Zinnecker 1998, Reipurth, Devine \& Bally 1998 for
recent surveys). Taking all together, the observational data are not
conclusive.
\subsubsection{Boundedness of protostellar clusters}
\label{subsubsec:bound}
In the simulations described in this paper, the dynamical evolution of
the gas results in the formation of dense clusters of collapsing cores
that form protostars (see Fig.'s~\ref{fig:3D-cube-A} and
\ref{fig:3D-cube-I}, and the left panels of
Fig.~\ref{fig:larson-X}). These clusters are bound throughout their
entire evolution. To specify this in more detail,
Fig.~\ref{fig:boundedness}a plots for the protostellar clusters
forming in simulation $\cal A$ (dashed lines) and $\cal I$ (solid
lines) the time evolution of the total kinetic energy subdivided into
the contribution from the internal velocity dispersion (thick line)
and from the center-of-mass motion (thin line). The kinetic energy is
almost entirely dominated by the internal random motions of the
cluster members. In Fig.~\ref{fig:boundedness}b, the evolution of the
potential energy is given. Only the gravitational interaction between
cores themselves is taken into account. The potential of the gas in
which the cluster is embedded is excluded in order to estimate whether
the cluster would dissolve in case of a sudden removal of the gas.
The cluster in the high-resolution model $\cal I$ (solid lines) is
still contracting when the simulation stops at $t=3.9$. The cluster
in model $\cal A$ is allowed to evolve further and develops the
typical core/halo structure of collisional $N$-body systems. The
global virial coefficient $\eta_{\rm vir} \equiv 2E_{\rm int}/|E_{\rm
pot}|$ is plotted in Fig.~\ref{fig:boundedness}c. Once the clusters
have formed they remain marginally bound, i.e.~$\eta_{\rm vir}
\lesssim 1$, even if all gas is removed. These conclusions do not
change when taking into account only the cores that lie within the
half-mass radius of the clusters. The cumulative mass of the clusters
is given in Fig.~\ref{fig:boundedness}d. For $t \geq 2.0$ more than
50\% of the total mass is condensed into protostellar cores (sink
particles). It is, however, interesting that already much earlier,
during the formation of the cluster $\eta_{\rm vir}$ becomes constant
with $\eta_{\rm vir} \approx 0.7$. Dense protostellar cores tend to
form such that the increase in the absolute values of the potential
energy of the stellar cluster neglecting the potential of the gas
component is always more or less balanced by an increase of its
kinetic energy, even at very early times when the gravitational
potential is strongly dominated by the gas. The 3-dimensional
velocity dispersion and the line-of-sight velocity dispersion of the
two clusters as function of time is described in
Fig.~\ref{fig:velocity-dispersion}. Again the dashed line denotes
model $\cal A$ and the solid line model $\cal I$. The fact that in
our models the line-of-sight measurements along the different axes are
almost identical to each other demonstrates that the cluster is
kinematically well mixed and isotropic. When scaled to low-density
star-forming regions like Taurus, a dimensionless value of $\sigma=1$
corresponds to $2.2\,$km$\,$s$^{-1}$. In the case of a high-density
region, this corresponds to $3.0\,$km$\,$s$^{-1}$ (see
Sec.~\ref{sec:scaling}). The values calculated from our simulations
when roughly 50\% to 60\% of the gas has been accumulated in cores ($t
\approx 2$) are in agreement with the measurements in Taurus and in
the Trapezium cluster in Orion which both have comparable velocity
dispersions of $\sigma \approx 2.5\,$km$\,$s$^{-1}$ (for Taurus see
\cite{frink97}, and for Orion see \cite{jones88}). However, for very
high core-formation efficiencies, when more than 60\% of the gas has
been converted into condensed cores, the derived velocity dispersions
appear to be too high. In order to reproduce the observed velocity
dispersions we predict core formation efficiencies of order 50\% to
60\%. This is reasonable as one would expect that energetic feedback
processes between the newly formed stars and their gaseous environment
disperse gas clouds before they have completely condensed into stars.
Note that the total star formation efficiency in the considered
molecular cloud region is the product of the core formation efficiency
and the efficiency with which the material accumulated in dense cores
accretes onto their central stellar objects. The latter one is
expected to be close to unity (for detailed simulations see Wuchterl
\& Tscharnuter 2000; for observational evidence see Motte, Andr\'{e}
\& Neri 1998)
\subsection{Implication for the IMF}
\label{subsec:implications-IMF}
In this section, the time evolution and the properties of the mass
distribution of protostellar cores in our simulations and their
relation to the initial stellar mass function are discussed. As
analyzed in the previous sections, protostellar cores form in the
centers of Jeans-unstable massive gas clumps and grow in mass via
competitive accretion. This process is strongly influenced by the
presence of unpredictable dynamical events which determine the shape
of the mass spectrum. In Fig.~\ref{fig:clump-spectrum-Z} the mass
distribution of identified gas clumps and protostellar cores in the
high-resolution model $\cal I$ has been introduced at four different
stages of the dynamical evolution of the system, i.e. initially and
when 10\%, 30\% and 60\% of the available gas has been converted into
condensed cores. Whereas the mass spectrum of gas clumps is best
described by a power-law function, the distribution of core masses
follows a Gaussian. Again for the high-resolution model,
Fig.~\ref{fig:mass-spectrum-1} plots the mass distribution of
protostellar cores at times when the cluster of cores has accreted a
total mass fraction of (a) $M_*=5$\%, (b) 15\%, (c) 30\%, (d) 45\%,
(e) 60\%, and (f) 85\%. Spanning two orders of magnitude, the mass
distribution of protostellar cores is very broad and peaks
approximately at the overall Jeans mass of the system, $\left \langle
M_{\rm J} \right \rangle = 1/222 = 10^{-2.3}$ (see
Sec.~\ref{sec:scaling}). This is somewhat surprising, given the
complexity of the overall dynamical evolution. The Jeans mass is a
function of density and may vary strongly for different clumps. In a
statistical sense, the system retains `knowledge' of its (initial)
average properties and the `typical' core mass is closely related to
the `typical' Jeans mass. In the initial conditions the density
contrast is limited. Hence, the gravitationally unstable clumps which
in Fig.~\ref{fig:clump-spectrum-Z} are located to the right of the
tilted line indicating the Jeans limit as function of density, all
have densities comparable to the mean density of the system,
i.e.~their local Jeans mass is $\sim \left \langle M_{\rm J} \right
\rangle$. These clumps form the first generation of cores which will
become very massive and populate the upper end of the mass
spectrum. At later stages, initially smaller clumps have merged and
grown dense enough to become Jeans unstable too. They begin to
collapse and form cores preferably at the low-mass end of the
spectrum. As long as there is enough gaseous material, the growth rate
of already existing cores to larger masses and the formation rate of
new low-mass cores are comparable and the mass distribution evolves
symmetrically (Fig.~\ref{fig:mass-spectrum-1}a -- e). However, at very
late stages of the evolution, the distribution gets skewed towards
higher masses which is an effect of the depletion of the gas
reservoir. There is still mass available for accretion, but it is not
sufficient to form new cores (see Fig.~\ref{fig:mass-spectrum-1}a --
f). The fact that the total gas reservoir is limited not only modifies
the shape of the distribution, it also influences its peak value. At
early stages, the median core mass is slightly below the average Jeans
mass, at late stages it lies above (see also
Tab.~\ref{tab:fitting}). Competitive accretion and the dynamical
interaction of protostellar cores as members of dense clusters (see
Sec.~\ref{subsubsec:dyn-WW}) do not alter the shape of the mass
distribution, but may widen it further.
Figure~\ref{fig:mass-spectrum-2} presents the mass spectra of
protostellar cores for {\em all} simulations discussed in this paper
(see Tab.~\ref{tab:models-N=2}) at the time when we would expect gas
to he heated and dispersed by newly formed stars (see section 6.4.3),
that is when roughly 60\% of the gas is accumulated in protostellar
cores. The distributions are very similar and the variations in widths
and centroid are small. Since the data sets of observed protostellar
cores are not yet large and accurate enough to derive well established
mass distributions\footnote{For $\sim$ 60 pre-stellar cores (i.e.\ for
self-gravitating gas clumps without central collapsed object) in the
$\rho$-Ophiuchus cloud, a mass spectrum is presented by Motte et al.\
(1998). See their Fig.~5.}, we have to go one step further and compare
the core mass spectrum with the initial mass function of stars (IMF).
The present simulations cannot resolve the conversion of individual
protostellar cores into stars. Since detailed collapse simulations
show that perturbed cores tend to break up into multiple systems we
can only make predictions about the system mass function. We adopt the
IMF for multiple stellar systems (i.e.~without corrections for binary
stars and higher-order systems) from Kroupa et al.~(1990) which we
compare with our numerically obtained core mass distribution. For
this comparison we need to know the efficiency of individual dense
cores to form stars inside. As their mass loss due to radiation, winds
and outflows is expected to be small, most of the material accumulated
in dense cores will be accreted onto the central protostars (Wuchterl
\& Tscharnuter 2000). Hence, we take the efficiency to convert dense
cores into stars to be of order unity. The overall star-formation
efficiency then depends only on the rate at which molecular gas forms
dense cores. It general, this rate evolves with time and is determined
by the properties of the underlying turbulent velocity field. This is
particularly important in the case of `isolated' star formation
(Klessen, Heitsch \& Mac\ Low 2000). Here, we concentrate on the
`clustered' mode of star formation and investigate molecular cloud
regions where turbulence has decayed completely. The overall
efficiency may also be influenced by the presence of massive O and B
stars in the vicinity or within of the considered molecular cloud
region. Once formed, their radiation would ionize the molecular gas
and could prevent subsequent core formation and growth. As this effect is
not included in our present investigation, the overall core formation
efficiency therefore in principle remains a free parameter.
To estimate the appropriate density scaling in our models
necessary for the computed and the observed mass distributions to
agree, we require that both distributions peak at the same mass.
Recall that the simulated core mass distribution reaches its maximum
roughly at the overall Jeans mass of the system. To allow for a
statistically significant comparison with observational data, we merge
together the mass distribution of all nine models with $\alpha=0.01$
and $P(k) \propto 1/k^2$. The resulting combined mass spectrum is
shown in Fig.~\ref{fig:mass-spectrum-3} at three stages of the
evolution, when the fraction of mass in protostellar cores is $M_*
=30$\%, $M_* = 60$\%, and $M_* = 90$\%, respectively. The
best-fitting Gaussian representation of each distribution is plotted
using open circles. The location of the centers and the widths are
specified in Tab.~\ref{tab:fitting}. The log-normal description of the
observed IMF for multiple stellar systems (the MS model in Kroupa et
al.\ 1990) is indicated by the dashed lines. We use the following
functional form,
\begin{equation}
\label{eqn:log-normal-function}
\xi(\log_{10}m) = \xi_0
\exp\left[-\frac{(\log_{10}m-\log_{10}\mu)^2}{2(\log_{10} \sigma)^2}\right]\,,
\end{equation}
where $N=\xi(\log_{10}m) d\log_{10}m$ is the number of objects per
logarithmic mass interval. The quantities $m$, $\mu$ and $\sigma$ are
given in units of the solar mass, $\log_{10}\mu$ and $\log_{10}\sigma$
determine peak and width of the distribution, and $\xi_0$ is the
normalization. With this functional form the observed IMF peaks at
$\log_{10} \mu = -0.31$ and has a width of $\log_{10} \sigma =
-0.38$. For the efficiency $M_* = 30$\%, the core mass distribution
has its maximum slightly below the average Jeans mass, $m_{\rm peak} =
0.86 \:\!\langle M_{\rm J}\rangle$. For agreement with the IMF the
system needs to be scaled such that the average Jeans mass is $\langle
M_{\rm J} \rangle = 0.58\,{\rm M}_{\odot}$. For a gas temperature
$T=10\,$K, this corresponds to a mean density $n({\rm H}_2) = 2.3
\times 10^5\,$cm$^{-3}$. For $M_* = 60$\% and $M_* = 90$\%, where
$m_{\rm peak} = 1.4\:\!\langle M_{\rm J}\rangle$ and $m_{\rm peak} =
2.0 \:\!\langle M_{\rm J}\rangle$, these values are $n({\rm H}_2) =
6.2 \times 10^5\,$cm$^{-3}$ and $n({\rm H}_2) = 1.3 \times
10^6\,$cm$^{-3}$, respectively (see Tab.~\ref{tab:fitting}). These
densities agree well with those observed in cluster forming regions
(see e.g.\ Motte et al.\ 1998 for the $\rho$-Ophiuchus cloud). Note,
that the width of the core mass distribution is slightly larger than
the one of the IMF of multiple stellar system by Kroupa et al.\
(1990), however, it is somewhat smaller compared to the original
Miller \& Scalo (1979) mass function (their model of constant
birthrate over $12\times 10^9$ years has $\log_{10} \sigma = -0.68$;
see the dotted lines in Fig.\ \ref{fig:mass-spectrum-3}). Given the
uncertainties involved in the observational determination of the IMF
(e.g.~Scalo 1986, 1998a) the agreement is remarkable and the
significance of these deviations is low.
Our dynamical model of clustered star formation predicts a universal
initial mass function with a log-normal functional form similar to the
observational data for multiple stellar systems. The overall nature
of the star formation process can only be understood in the frame work
of a statistical theory, where a sequence of probabilistic events may
naturally lead to a log-normal IMF (e.g.~Zinnecker 1990, Larson 1992,
1995, Richtler 1994, Price \& Podsiadlowski 1995, Murray \& Lin 1996,
Elmegreen 1997; also Adams \& Fatuzzo 1996). Since the final mass
distribution of protostellar cores in our self-gravitating, isothermal
models is a consequence of the chaotic dynamical evolution during
the accretion phase, our simulations support this hypothesis.
\section{Summary}
\label{sec:summary}
In this paper we numerically investigated the dynamical evolution and
fragmentation of molecular clouds and discussed the interplay between
gravity and gas pressure. We identified the processes that dominate
the formation and evolution of (proto)stellar clusters and determine
their properties. From the results we can conclude that even simple,
isothermal models of self-gravitating clumpy clouds are able to
explain many of the observed features of star-forming regions. This is
rather surprising, given the fact that magnetic fields and energetic
heating processes of newly formed stars could in principle be
important, both of which have been neglected. At this point the
reader should be cautioned again that our isothermal models do not
include any feedback effects from the star formation process
itself. At late stages of the evolution energy and momentum input from
young stars is likely to modify the dynamical properties of the
accreting gas and the simple isothermal approach which we follow here
needs to be extended by more elaborate schemes.
Our simulations show that, in general, the formation of a cluster of
condensed cores and protostars through gravitational collapse and
fragmentation of a molecular cloud region is extremely complex. The
dynamical evolution of molecular gas is determined by the interplay
between self-gravity and gas pressure. This creates an intricate
network of filaments, sheets and dense clumps. Some clumps will become
gravitationally unstable and undergo rapid collapse. While contracting
individually to form protostellar cores in their interior, gas clumps
stream towards a common center of attraction: the dynamical evolution
of molecular clouds involves processes acting simultaneously
on different length scales and time scales.
While following the large-scale flow pattern, gas clumps
can undergo further fragmentation or merge at the intersections of
filaments. At that stage, the central regions of some clumps will
have already collapsed to sufficiently high densities to be identified
as protostellar cores. These cores rapidly grow in mass via accretion
from their parental gas envelope. When clumps merge, the
newly formed clump may contain a multiple system of
protostellar cores which subsequently compete with each other for
accretion from the same limited and rapidly changing reservoir of
contracting gas in which they are embedded. Since the cores are
dragged along with the global gas flow, a dense cluster of
accreting protostellar cores builds up quickly. Analog to dense stellar
clusters, its dynamical evolution is subject to the complex
gravitational interaction between the cluster members: close
encounters occur frequently and will drastically alter the orbital
parameters of cores. This leads to the formation of unstable triple or
higher-order systems, and consequently a considerable fraction of
protostellar cores becomes expelled from the cloud. These cores
effectively stop accretion and their final mass is determined.
The presence of unpredictable dynamical events in the overall gas flow
and the evolution of the nascending protostellar cluster very
efficiently erases the memory of the initial configuration. For this
reason, we cannot predict the detailed evolution of individual objects
just from the initial state of the system. Only the properties of an
{\em ensemble} of protostellar cores, for example their kinematics and
mass distribution, can be determined in a probabilistic sense. A
comprehensive theory of star formation needs to be a {\em statistical
theory}. Some first attempts to formulate a statistical model of the
star-formation process appear very promising (e.g.~Zinnecker 1990,
Larson 1992, 1995, Richtler 1994, Price \& Podsiadlowski 1995, Murray
\& Lin 1996, Adams \& Fatuzzo 1996, Elmegreen 1997; for an overview
see Scalo 1998b) and are supported by the results of our numerical
study. Taken together, our simulations strongly suggest that {\em
gravitational} processes and accretion dominate the early phases of
star formation.
We extend the above overview by giving a detailed list of the features
and results of our calculations derived from the comparison of our
numerical models with specific observational properties of molecular
clouds and young stellar clusters:
\begin{itemize}
\item {\bf One-point probability distribution function of density and
line-of-sight velocity centroids:} During the dynamical evolution of
the system gravitational contraction on large scales and the collapse
of individual gas clumps result in considerable distortions of the
pdf's away from the initial Gaussian behavior. The pdf of the
logarithm of the density is best described by a (multiple) power-law
distribution and the pdf of the line-of-sight velocity centroids
develop extended high-velocity wings. This also holds for the
line profiles themselves (Sec.~\ref{subsec:pdfs}).
\item {\bf Clump mass spectrum:} During the early phases of
star formation our simulations of self-gravitating isothermal
gas are able to reproduce the observed power-law clump-mass spectrum
of molecular clouds. This is due to the progression of non-linear
gravitational attraction and the disintegration of small clumps by gas
pressure. The observed spectrum is best fit at times between the
formation of the first condensed objects up to the time when depletion of
the gas reservoir becomes considerable. Neither the initial Gaussian
fluctuation spectrum, nor the final stages of the evolution when most
of the gas is condensed into protostellar cores, give a clump-mass
distribution which is in agreement with the observations. This is discussed in
Sec.~\ref{subsubsec:clump-mass-spectrum}.
\item {\bf The Shapes of Individual Clumps:} Dense Jeans-unstable gas
clumps are the precursors of protostellar cores. As being part of a
complex network of filaments, individual clumps are typically very
elongated objects with ratios between semi-major and semi-minor axis
of 2:1 to 4:1. However, in many cases (especially at the intersection
of two filaments) they can be quite irregularly shaped. These features
are in agreement with observed dense pre-stellar cores in dark clouds
(Sec.~\ref{subsubsec:ind-clumps}).
\item {\bf Formation and Growth of Protostellar Cores:} The formation
and growth of protostellar cores is subject to a progression of
statistical events. However, we can identify the following trends in
our models (Sec.~\ref{subsubsec:growth-protostellar-cores}): (a) The
protostellar cores that form first are generally formed in the clumps
with the highest initial densities, and tend to have the highest final
masses. They accrete the bulk of their final mass from their close
vicinity. (b) On the other hand, matter that forms cores at later
times has already undergone considerable dynamical evolution; these
cores form from gas that was initially in widely distributed
low-density clumps. Along filaments, they stream towards a common
center of gravity and may merge at the intersections. Once enough mass
is accumulated, these clumps undergo rapid collapse and build up new
protostellar cores. The cores which form a late stages tend to have
very low final masses.
\item {\bf Competitive Accretion and the Importance of Dynamical
Interaction:} Once a gas clump becomes Jeans unstable it collapses and
forms a condensed core. This core grows in mass via accretion from the
infalling envelope. Merging may lead to clumps that contain multiple
cores. These compete with each other for the material of a common gas
reservoir. The succession of clump mergers leads to the formation of
an embedded, dense protostellar cluster, whose dynamical behavior is
dominated by close encounters between cluster members. Competitive
accretion and collisional dynamics determine the kinematical and
spatial properties of the cluster and the mass distribution of
protostellar cores (Sec.'s \ref{subsubsec:growth-protostellar-cores}
\& \ref{subsubsec:dyn-WW}).
\item {\bf Rotational Properties of Protostellar Cores:} The rotation
of a protostellar core is an important parameter for its collapse. It
determines the properties and stability of accretion disks and their
tendency for sub-fragmentation. Within the complex network of
intersecting filaments, gas clumps gain angular momentum from tidal
torque and shear. In the accretion process it is transfered onto the
embedded cores. Since the angular momentum is gained from large-scale
motion, the orientation of the spin vectors of individual protostellar
cores is correlated with their location
(Sec.~\ref{subsubsec:rotational-properties}). A similar correlation is
often found in observed star-forming regions between the orientation
of the molecular outflows from young stellar objects and their
location. However, the observational data are not conclusive.
\item {\bf Clustering Properties of Protostellar Cores:} The time
evolution of a highly Jeans-unstable region within a molecular cloud
leads to the formation of a dense cluster of protostellar cores. The
final dynamical state of the system closely resembles the properties
of observed stellar clusters: it exhibits the typical core/halo
structure of collision-dominated $N$-body systems, and when
calculating its 2-point correlation function, or equivalently the mean
surface density of companions as function of separation, one can
clearly distinguish between the binary regime and the large-scale
clustering regime (Sec.~\ref{subsubsec:clustering}).
\item {\bf Boundedness of Protostellar Clusters:} The clusters of
protostellar cores in our simulations form as bound entities: the
conversion of gas into condensed cores is such that the decrease of
potential energy is always balanced by the increase of kinetic energy
(Sec.~\ref{subsubsec:bound}). Whereas the protostellar cluster is
bound, this may not be true for the resulting stellar cluster. Its
kinematical properties depend strongly on the details of the
conversion of individual cores into stars: on the speed and the
overall efficiency of the process. However, this cannot be treated in
our simulations and needs to be addressed in detailed calculations of
individual core collapse.
\item {\bf Mass Spectrum of Protostellar Cores --- The Star Formation
Efficiency and Implications for the IMF:} The distribution of stellar
masses is one of the most important properties of the star-formation
process. Any comprehensive model of star formation must be able to
derive this quantity or at least address this issue. In our isothermal
models the masses of protostellar cores are the result of a sequence
of unpredictable stochastic events. In a natural way this leads to a
{\em log-normal} mass spectrum which peaks roughly at the {\em average
Jeans mass} of the system (Sec.~\ref{subsec:implications-IMF}).
Detailed collapse calculations show that perturbed rotating cores tend
to break up into multiple stellar systems, which cannot be resolved in
the larger-scale simulations presented here. Therefore, we compare the
numerical mass function of protostellar cores with the IMF derived for
multiple stellar systems. Our simulations require densities in the
range $n({\rm H}_2) = 10^5\,{\rm cm}^{-3}$ to $10^6\,{\rm cm}^{-3}$
for both distributions to agree.
\end{itemize}
\acknowledgements We thank Pavel Kroupa for valuable discussions on
the topic of binary stars and the IMF, Philippe Andr\'{e} for helpful
comments and remarks on the mass spectrum of prestellar condensations,
and Matthew Bate for his help with the SPH-code. We thank the referee
for many useful comments and for the prompt reply.
\begin{appendix}
\section[Clump Finding]{The Clump Finding Method}
\label{app:clumps}
This appendix describes the method used to identify the clumping
properties of the numerical calculations. It applies a scheme similar
to the one introduced by Williams et al.~(1994), but which is fully
integrated into the SPH formalism.
In SPH, the densities $\rho_i$ of individual particles $i$ are
obtained in a local averaging process over a list of neighbors within
distances less than twice the smoothing length $h_i$ (e.g.~Benz 1990,
Monaghan 1992). To identify clumps, the resulting 3-dimensional
density field is subdivided into ten bins equally spaced in the
logarithm of the density. Starting with the highest density level, the
particles are sorted by decreasing density, i.e.~the first particle in
the list is the one with the highest density. Going through this list,
for each particle $i$ it is checked whether the particles in its
neighbor list are already assigned to a clump. If this is not the case,
then particle $i$ is assigned to a new gas clump, together with all the
particles in its neighbor list. All additional particles at that
level which are connected to the particles of the new clump by means
of overlapping smoothing volumes are identified and assigned to the
same clump. The assigned particles are finally removed from the
density list. On the other hand, if the neighbor list of the tested
particle $i$ contains contributions from already identified clumps the
particle will be associated to the clump that contributes the largest
fraction of particles in the neighbor list, only particle $i$ will be
removed from the sorted list. This scheme is repeated until all
particles at that density level are assigned to clumps. The procedure is
repeated at the next lower level and so on up to the lowest level.
The method is
illustrated in Fig.~\ref{fig:clumps}. In the end, all SPH particles in
the system are assigned to individual clumps.
This scheme is free of assumptions about the geometrical shape of the
clumps\footnote{An alternative approach to determine clump properties
in (observed) molecular clouds was introduced by \cite{stu90}. These
authors explicitly assumed that the gas clumps have Gaussian density profiles.
They decomposed their intensity maps into clumps with such
a profile by
minimizing the residual.} which is of great advantage when dealing with
highly irregular and filamentary structure as in our numerical
simulations of gravitational fragmentation and collapse.
Note that two clumps that may at a high level of
density be well separated entities, may at lower levels have common
contour lines. In our scheme, they are still separated even at these
lower levels by introducing an artificial `interface' between the two
clouds. Individual particles are {\em always} assigned to the clump
that contributes most neighbors. This sets a clear division line
between two competing clumps and enables their separation. Besides
this basic separation criterion, the method is free of independent
parameters and uses only intrinsic properties of the SPH scheme.
\section{The Zel'dovich approximation and its validity to gaseous
systems}
\label{app:zel}
\subsection{The Zel'dovich Approximation}
In 1970, Zel'dovich proposed for cosmological simulations a method to
extrapolate the linear theory into the non-linear regime. The
dynamical evolution of an idealized self-gravitating and pressureless
continuous medium can be expressed in terms of a function ${\vec
f}({\vec r}_0,\delta t)$. Its value is the position $\vec r$ of a fluid
element after a time interval $\delta t$, whose original position was
${\vec r}_0$, i.e.~${\vec r} = {\vec f}({\vec r}_0,\delta t)$. If we
denote the original density field as $\rho({\vec r}_0)$ then the
density field after $\delta t$ is given by
\begin{equation}
\rho({\vec r},\delta t) = \frac{\rho({\vec r}_0)}{|{\rm det}
f_{i,j}({\vec r}_0,\delta t)|}\;.
\end{equation}
Here, $f_{i,j}$ denotes the partial derivative of the $i$-th component
of $\vec f$ in ${\vec r}_j$. The time evolution is given by Poisson's
equation and to first approximation it follows
\begin{equation}
\label{eqn:zel-shift}
{\vec r}(\delta t) = {\vec f}({\vec r}_0,\delta t) = {\vec r}_0 + {\vec v}({\vec r}_0)\,\delta t\;,
\end{equation}
with $\vec{\nabla}_{{\vec {\scriptsize r}}_{\scriptscriptstyle 0}}
\cdot {\vec v}({\vec r}_0) \propto \rho({\vec r}_0)$. This assumes
that the velocity field is rotation-free, because then the existence
of a potential $\phi\:\!({\vec r}_0)$ with ${\vec v}\;\!({\vec r}_0)
\equiv \vec{\nabla}_{{\scriptsize {\vec r}}_{\scriptscriptstyle 0}}
\phi({\vec r}_0)$ is guaranteed and is connected to the density field
via $\Delta_{{\scriptsize {\vec r}}_{\scriptscriptstyle 0}} \phi\:\!({\vec
r}_0) \propto \rho({\vec r}_0)$.
Using this method for the extrapolation of initially small density
fluctuations into the non-linear
regime (i.e.~into regions where the density contrast exceeds the value
of one) has certain limitations. For example, if $|{\rm det} f_{i,j}|$
becomes very small or zero, pressure forces would become
important, preventing infinite densities. Furthermore, if one follows
trajectories along which $|{\rm det} f_{i,j}|$ vanishes, the density
decreases again after the singularity and at the same time the
solution is no longer unique. For practical purposes by
appropriately choosing the shift time $\delta t$ one can minimize these
problems and the Zel'dovich (1970) approximation is valid. We will
show that this also applies for gaseous systems. This procedure considerably
speeds up the computation of the early evolutionary phase since the
system is advanced in one single large step as opposed to solving the
complete set of hydrodynamic equations and integrating over many small time
intervals.
We use the following method to generate random Gaussian fields:
After the desired power spectrum $P(k)$ is defined
we generate a {\em hypothetical} density field from all contributing
modes in Fourier space (according to Sec.~\ref{subsec:initial}). This
field is transformed back into real space and Poisson's equation is
solved in order to obtain the velocities which would generate the
density field self-consistently. Starting from a homogeneous initial density
distribution, this velocity field is used to advance the particles in
the system in one single large time step. As illustration,
Fig.~\ref{fig:zel-shift} plots the 2-dimensional projection of the
homogeneous starting condition (Fig.~\ref{fig:zel-shift}a) and of the system after the
Zel'dovich shift has been applied for various shift intervals $\delta t$
(Fig.~\ref{fig:zel-shift}b -- h).
Note that the method implicitly assumes periodic boundary
conditions. Therefore, the images have to be seen periodically
replicated in all directions.
\subsection[Validity of the Zel'dovich Approach]{Validity of the Zel'dovich Approach}
\label{subsec:validity}
To test the validity of the Zel'dovich approximation and its parameter
dependence we compare systems which have been generated with different
sets of shift intervals $\delta t$ and slopes ${\nu}$ of the power
spectrum. Furthermore, we compare the results with systems that have
been advanced in time using the full SPH formalism.
If particles move along trajectories on which $|{\rm det} f_{i,j}|$
vanishes the result is no longer unique, i.e.~different paths may lead
to the same location. For converging flows, the density increases
before reaching the singularity in $|{\rm det} f_{i,j}|$ and decreases
afterwards. This is unphysical for collision dominated
systems. Depending on the size and the strength of individual
fluctuations, the time interval $\delta t$ to reach this singular
point may differ. Assuming comparable amplitudes, for small
perturbations $\delta t$ will be short and for large-scale modes
$\delta t$ will be longer. This has to be taken into account when
determining the optimum choice of the shift interval. The effect of
varying $\delta t$ is addressed in Fig.~\ref{fig:zel-shift} which
shows a sequence of particle distributions generated by the Zel'dovich
method from a power spectrum $P(k) \propto k^{-2}$ with $0 \le \delta
t \le 30$. The homogeneous starting field is identical to a shift of
$\delta t=0$. Increasing $\delta t$, the density contrast starts to
grow and the distribution becomes more structured. However, at $\delta
t \approx 3$ small-scale fluctuations begin to disperse again and the
density contrast starts to diminish. At $\delta t\approx 30$ the
system appears homogeneous again. The same is true for different
slopes ${\nu}$ of the power spectrum. Figure~\ref{fig:zel-shift}i
illustrates the dependence of the average (open circles) and maximum
(solid circles) particle density on the time shift $\delta t$. The
dashed line denotes the density of the homogeneous cube as reference.
Figure~\ref{fig:corr-pow-dt} describes the influence of different
shift intervals $\delta t$ on (a) the 2-point correlation function
$\xi(r)$ and on (b) the measured power spectrum $P(k)$. In analogy to
the above, the correlation strength and length increase with shift
interval $\delta t$ for systems generated with $t\lesssim3$ and
decrease again for larger shifts. Similar conclusions apply to the
power spectrum. For $\delta t\approx2$ the initial slope of ${\nu}=-2$
is best reproduced. Shorter shifts $\delta t$ do not establish the
mode spectrum sufficiently, the spectrum is too flat. Longer intervals
produce overshooting on small scales, i.e.~perturbations with short
wave lengths are smeared out, whereas large modes are still
amplified. Therefore the spectrum gets steeper. For $\delta t>10$,
overshooting occurs for the largest modes as well and the entire power
spectrum flattens again.
Finally, we compare systems which have been generated by the
Zel'dovich method with systems that have been advanced in time solely
with SPH. The Zel'dovich shift generates fluctuations on all scales,
since only gravitational forces are considered. Large perturbations
are Jeans unstable and start to collapse. Most small ones are Jeans
stable and pressure forces smear them out in the subsequent evolution
with the SPH method. However, also some low-mass fluctuations may have
been generated with sufficiently high density to be gravitationally
unstable, since the Jeans mass is inversely proportional to the square
root of the density. These clumps will collapse as well. Furthermore,
while flowing towards a common center of gravity small clumps may
merge to form more massive clumps, which again may exceed the Jeans
limit. The probability for that to happen depends on the time scale
for the collapse and dispersal of individual clumps relative to the
time scale for clumps to merge or fragment. This can be
determined only in a statistical sense.
If we advance the system from the homogeneous state with the full SPH
formalism, we compute the self-consistent initial velocity field of
the hypothetical fluctuation field. Pressure forces are included from
the beginning and small perturbations have no possibility to grow
unless their mass exceeds the Jeans limit of the homogeneous system,
which is determined by the mean density. Therefore, only high mass
clumps can form. This is different from the Zel'dovich case, where
perturbations are created on all scales and also some low-mass clumps
may have large enough densities such that their local Jeans mass is
smaller than the clump mass. Therefore, slightly smaller and more
fragments are expected in systems that have been generated using the
Zel'dovich method. The detailed discrimination of two systems subtly
depends on the desired inhomogeneity and density contrast in the
system, i.e.~on the choice of ${\nu}$ and $\delta t$. These
small-scale differences decrease, the more homogeneous the
distribution and the smaller the Zel'dovich shift. On large scales,
the statistical properties are much less affected.
This is exemplified, using a distribution of $50\,000$ particles with
a power spectrum $P(k) \propto k^{-2}$. Again the simulation cube
contains roughly 222 Jeans masses. Figure~\ref{fig:comp-zel} shows
snapshots of the time evolution of the system initially generated
applying a Zel'dovich shift with $\delta t=2$ and subsequently
advanced with SPH. On the other hand, Fig.~\ref{fig:comp-sph}
describes the evolution of the system that was evolved from the
homogeneous state using SPH without applying the Zel'dovich method.
The large-scale behavior of the two systems is very similar, however
differences on small scales occur. The distribution initially evolved
with the Zel'dovich approximation is patchier at comparable
times. Note, that time is measured from the begin of the evolution
with SPH. To compare both systems at equal times, the Zel'dovich shift
interval $\delta t=2$ has to be added in
Fig.~\ref{fig:comp-zel}. However, with the progression of the SPH
calculation the higher degree of irregularity is reduced. Small
perturbations are smoothed by pressure.
In summary, the Zel'dovich approximation is very well suited to
generate fluctuation fields with well defined statistical properties
determined by the power spectrum $P(k)$. With the appropriate choice
of the shift interval $\delta t$, every spectrum can be
generated. When applying the Zel'dovich method to gaseous systems, one
has to take the effect of neglecting pressure forces into account. On
small scales, small deviations from the fully self-consistent time
evolution may occur. Our test calculations do however
show that applying the Zel'dovich method is fully appropriate
for the purpose of the current investigation.
\end{appendix}
\clearpage
\newpage
|
\section{Introduction to NLL BFKL}
Early last year, after many years of hard work involving many
participants, Fadin and Lipatov \cite{nll} presented the NLL
corrections to the BFKL equation. Since then, there has been much
lively discussion of the interpretation of these corrections. In
this talk I will discuss some of the results of this activity.
I will not address phenomenology here, but some applications of the
BFKL resummation are
inclusive dijet production at large rapidity separation $
y=\ln\hat s/(p_{1\perp}p_{2\perp})$, forward jet production in
deep-inelastic scattering (DIS), DIS structure functions at small-$x$,
and others.
In order to discuss the NLL corrections it is useful
to consider first the solution to the BFKL equation at LL \cite{FKL}.
The BFKL equation is used to resum all powers of
$\alpha_{s}\log(\hat s)$ in the cross section. This leads to the
familiar prediction that at very high energies the cross section
scales as a power of the energy:
\begin{equation}
\hat\sigma\ \approx e^{Ay} \approx\
{\hat s}^{A}\ .
\label{saddle}
\end{equation}
The quantity $(1+A)$ is often referred to as the BFKL Pomeron
intercept.
This scaling behavior is obtained from the solution to
the BFKL equation, which is given at LL by the integral
\begin{equation}
f(y,p_{1\perp},p_{2\perp})\ =\ {1\over 2\pi i}
\int_{1/2-i\infty}^{1/2+i\infty} d\gamma\,
(p_{1\perp}^2)^{\gamma-1}\,(p_{2\perp}^{2})^{-\gamma}
e^{\bar\alpha_{s}\chi^{(0)}(\gamma)y}\ ,\label{sold}
\end{equation}
where $\bar \alpha_{s}=\alpha_{s}N_{c}/\pi$, and we have performed an
azimuthal average over the transverse momenta for convenience. The
function
\begin{equation}
\chi^{(0)}(\gamma)\ =\ 2\psi(1)-\psi(\gamma)-\psi(1-\gamma)
\label{eig}
\end{equation}
is the eigenvalue of the LL BFKL kernel, where $\psi$ is the
logarithmic derivative of the gamma function. Performing the
integral in the saddle-point approximation leads to the exponential
rise in the cross section (\ref{saddle})
with $A=\bar\alpha_{s}\chi^{(0)}({1\over2})=4\bar\alpha_{s}\ln2$.
At the heart of the BFKL equation, which is used to derive (\ref{sold}),
is the kernel
$K(p_{1\perp},p_{2\perp})$, which is an
integral operator in transverse momentum space that is used to build
up the BFKL ladder. The contributions to the kernel are shown,
schematically, at LL and NLL in Fig.~1.
At LL each application of the kernel adds one more factor of
${\cal O}(\alpha_{s}\Delta y)$ to the resummation. It is composed of
two types of contributions. The first type corresponds to an emission
of a real gluon, in the approximation that it is widely separated in
rapidity from any other emissions (known as multi-Regge kinematics).
The factor of $\Delta y$ just comes
from the integration over the rapidity of this gluon. The second type
corresponds to the virtual contributions which are enhanced by the
logarithmic factor $\Delta y$. The simplest contribution of this
second type can be found by considering the one-loop
corrections to $gg\rightarrow gg$ scattering.
At NLL one also includes terms of ${\cal O}(\alpha_{s}^{2}\Delta y)$
with each application of the kernel.
They consist of three types, corresponding to: the emission of
two gluons nearby in rapidity, the virtual correction to the emission
of one gluon in the multi-Regge kinematics, and the subleading
purely-virtual corrections. This last contribution can be found by
considering
the two-loop corrections to $gg\rightarrow gg$ scattering.
It is the calculation of these three contributions which took many
years and many papers to sort out the technical details\footnote{A
list of references can be found in ref.~\cite{schmidt}, but with no
guarantee of completeness.} Although the full kernel has not been
checked in a completely independent manner, many of the pieces of the
calculation have received independent confirmation. Two particularly
significant checks are the
calculation of the virtual correction to the gluon emission in
multi-Regge kinematics \cite{ddsii}, and the compilation of the three
NLL terms into a single kernel with the cancellation of all collinear and
soft singularities \cite{cc}.
\begin{figure}[t]
\centerline{\epsfxsize 4.0 truein \epsfbox{kernel.eps}}
\vskip 0.0 cm
\caption[]{
\label{kernel}
\small Contributions to the BFKL kernel at LL and NLL. Gluons that
cross the dashed line correspond to real emission.}
\end{figure}
The final result of this calculation is usually presented by applying
the kernel to the LL eigenfunctions, with azimuthal averaging, yielding
\begin{eqnarray}
\int d^{2}p_{2\perp}K(p_{1\perp},p_{2\perp})
(p_{2\perp}^{2}/p_{1\perp}^{2})^{\gamma-1}\ &=& \ \
\bar\alpha_{s}(\mu)\,\chi(\gamma)\nonumber\\
&=&\ \ \
\bar\alpha_{s}(\mu)\,
\chi^{(0)}(\gamma)\left[
1-\bar\alpha_{s}(\mu)b_{0}
\ln(p_{1\perp}^{2}/\mu^{2})\right]\nonumber\\
&&
+\ \bar\alpha_{s}^{2}(\mu)\chi^{(1)}(\gamma)\ ,
\label{nll}
\end{eqnarray}
where $b_{0}= {11/12}-{n_{f}/(6N_{c})}$ and $\mu$ is the
$\overline{\rm MS}$ renormalization scale.
The NLL correction has been separated into two terms. The first term
depends on the scale $p_{1\perp}$ and is associated with the running
of the coupling in the LL kernel:
$\alpha_{s}(\mu)\rightarrow \alpha_{s}(p_{1\perp})$.
The second term, $\bar\alpha_{s}^{2}\chi^{(1)}(\gamma)$, is independent
of scale and contains the remainder of the NLL corrections \cite{nll}.
\section{Problems at NLL.}
After completion of the NLL corrections to the BFKL kernel, several
issues with the NLL solution quickly became apparent. Depending on
one's point of view, these may even signify critical problems
for the entire BFKL resummation program.
Roughly speaking, they can be separated into
issues associated with the running coupling term and issues associated
with the scale-invariant term. Although my main focus will be
on the scale-invariant term, I briefly
touch on the topic of the running coupling term in NLL BFKL.
\subsection{Running coupling problems.}
The issues that arise with the running of the coupling in BFKL were
not entirely unanticipated. They are related to the fact
that the emission of each real gluon causes the momentum carried
down the BFKL ladder to
diffuse as one moves away from the starting rapidity. It
can diffuse to larger values or to smaller values; however, if it
diffuses below values around $\Lambda_{QCD}$ then nonperturbative
effects become important, and one can no longer make unambiguous
predictions from the perturbative BFKL resummation. This issue was
known even before the NLL corrections were completed, although it can
be ignored in LL BFKL. This is because logarithms
involving two transverse scales do not arise until NLL, and so
LL BFKL is calculated at fixed coupling. At NLL, however,
this issue can no longer be swept under the rug.
The effects of the running coupling term in the NLL BFKL solution
have been considered in several papers. Armesto, Bartels, and Braun
\cite{bartels} considered the modification of the eigenvalues and the
eigenfunctions due to the NLL terms in the kernel. They found that, due
to the running coupling term in the kernel, the NLL eigenvalues can take
on any value along the real axis. This is unlike at LL, where the
eigenvalues (\ref{eig}) have a maximum and the equation
(\ref{sold}) is well-defined. Thus, the interpretation of the NLL
corrections is not entirely well-defined in this approach.
Correspondingly, the NLL eigenfunctions
contain pieces displaying non-perturbative behavior.
The effects of the running coupling term in the NLL BFKL solution
were also included through a different approach by Kovchegov and
Mueller \cite{mueller}. They obtained a NLL solution by explicitly
iterating the NLL corrections to the kernel, starting from the LL solution
evaluated in the saddle point approximation. They found that the
running coupling term leads to a non-Regge behavior in the energy
dependence of the cross section. (This was also shown in
ref.~\cite{levin}). This
non-Regge behavior is exhibited as a term of the form $\alpha_{s}^{5}y^{3}$
in the exponential of eq.~(\ref{saddle}) at high energies. In
addition, they showed that the nonperturbative effects from resumming
the logarithms in the running coupling should be small
as long as $14\zeta(3)b_{0}^{2}\bar \alpha_{s}(\mu)^{3}y\lstsim1$.
From these results, it appears that for some region of kinematics the
nonperturbative effects should be small; however, the proper way to
deal with them at NLL is not yet completely clear.
\subsection{Scale-invariant problems.}
Whereas the problems at NLL due to the running coupling were
anticipated to some degree, the problems due to the scale-invariant
term were a big surprise. The first indication of this problem was
seen immediately by Fadin and Lipatov. The corrections to the
leading eigenvalue are large and negative! If we ignore the running
coupling term, we obtain
\begin{equation}
\bar\alpha_{s}\chi(\mbox{$1\over2$})\ =\
2.77\bar\alpha_{s}-18.34\bar\alpha_{s}^{2}\ ,
\label{nllchi}
\end{equation}
for three active flavors. At the not-unreasonable value
of $\alpha_{s}=0.16$ the NLL corrections exactly cancel the LL term,
while for larger values of $\alpha_{s}$ the eigenvalue becomes negative.
Naively, this would indicate that the BFKL Pomeron intercept also
becomes negative, leading to a cross section that decreases, rather
than increases, as a power of the energy.
Of course, this interpretation relies on the saddle-point evaluation
of the NLL generalization of the BFKL solution (\ref{sold}). Upon
closer analysis Ross \cite{ross} showed that the NLL eigenvalue
function $\chi(\gamma)$ no longer has a maximum at $\gamma={1\over2}$, but
has a minimum with two maxima occuring symmetrically on either side
of this point\footnote{The standard procedure in these analyses is to
modify the LL eigenfunctions used in eq.~(\ref{nll}) in order to make the
eigenvalues manifestly symmetric under $\gamma\rightarrow1-\gamma$,
following
ref.~\cite{nll}.}. Performing a higher-order expansion of $\chi(\gamma)$,
Ross found a smaller correction to the BFKL Pomeron intercept.
However, the solution he obtained was not positive definite. It
contained oscillations as one varied $p_{1\perp}$ and $p_{2\perp}$.
This led Levin \cite{levin} to declare that NLL BFKL has a serious
pathology.
One might wonder whether the approximate evaluation of the integral
performed by Ross is adequate at this stage. Perhaps an exact evaluation
is necessary. However, negative cross sections have also arisen when
the resummed small-$x$ anomalous dimensions, obtained from the NLL BFKL
solution, were used to study DIS scattering at
small-$x$ \cite{ball}. In any event the NLL corrections to the
BFKL solution are large, leading one to question the stability and
applicability of the BFKL resummation procedure in general.
\section{Attempts to Fix/Understand the large NLL Corrections.}
In this section I will discuss several attempts to understand the origin
of the large NLL corrections and to control them. The first two
proposals, although very different in implementation, both can be
traced to correlations that arise when two neighboring gluons are
emitted close to each other in rapidity \cite{bo}. Essentially, the
LL BFKL equation greatly overestimates the contribution of this
collinear
configuration because of the lack of ordering in transverse momentum.
The first proposal by Salam \cite{salam} was to resum the double transverse
logarithms of the form
$\bar\alpha_{s}\ln^{2}(p_{1\perp}^{2}/p_{2\perp}^{2})$. This idea
is based on the studies of Camici and
Ciafaloni \cite{cc} on the energy-scale dependence of NLL BFKL.
Instead of choosing the symmetric rapidity
$y=y_{2}-y_{1}=\ln\hat s/(p_{1\perp}p_{2\perp})$ as the large
logarithm to resum, one could equally well have chosen $y^{+}=
\ln x^{+}_{2}/x^{+}_{1}=\ln \hat s/p_{1}^{2}$
or $y^{-}=\ln x^{-}_{1}/x^{-}_{2}=
\ln \hat s/p_{2}^{2}$, where $x^{\pm}_{i}$ is the momentum fraction along
the positive or negative light-cone for the emitted gluon $i$.
Although these choices are all
equivalent at LL, at NLL a change in the logarithm produces a change
in the NLL kernel and can introduce the double transverse logarithms.
Motivated by DGLAP-type resummation \cite{dglap} one finds that the
appropriate choice is to resum $y^{+}$ when $p_{1\perp}^{2}\gg
p_{2\perp}^{2}$ and $y^{-}$ when $p_{2\perp}^{2}\gg
p_{1\perp}^{2}$. In refs.~\cite{cc} and \cite{nll} it was shown
that changing
from $y^{+}$ to $y$ shifts the NLL eigenvalue by terms with
$1/\gamma^{3}$ singularities. Similarly, changing
from $y^{-}$ to $y$ shifts the NLL eigenvalue by terms with
$1/(1-\gamma)^{3}$ singularities. Both the $1/\gamma^{3}$ and the
$1/(1-\gamma)^{3}$ singularities can be identified in
$\chi^{(1)}(\gamma)$, and methods for resumming these singularities
were given in ref.~\cite{salam}.
Results of this resummation of double transverse logarithms are shown in
Fig.~2, where the leading eigenvalue $\bar\alpha_{s}\chi({1\over2})$ and
its second derivative are plotted as a function of $\bar\alpha_{s}$.
The different schemes 1--4 give some measure of the ambiguity in this
resummation procedure. In general the eigenvalue is found to be
positive after resummation, although less than at LL. In addition the
point $\gamma={1\over2}$ remains a maximum over a wider range of
$\alpha_{s}$, especially in schemes 3 and 4.
The physical implications of this proposed solution
can be seen by further investigating the relation between
resummation in $y^{\pm}$ and $y$. When $p_{1\perp}^{2}\gg
p_{2\perp}^{2}$, the resummation in $y^{+}$ requires the ordering
$x^{+}_{2}>x^{+}_{1}$.
Translating back into the symmetric variable
$y$, this implies $y_{2}-y_{1}>\ln(p_{1\perp}/p_{2\perp})$. Similarly,
when $p_{2\perp}^{2}\gg
p_{1\perp}^{2}$, the resummation in $y^{-}$ requires the ordering
$x^{-}_{1}>x^{-}_{2}$, implying
$y_{2}-y_{1}>\ln(p_{2\perp}/p_{1\perp})$. These constraints hold for
any two successively emitted gluons. Therefore, the resummation of
the double transverse logarithms corresponds to imposing a
$p_{\perp}$-dependent cut,
$y_{i+1}-y_{i}>|\ln(p_{i\perp}/p_{i+1\perp})|$,
on the separation in rapidity between the
neighboring gluons.
This leads to the second proposal \cite{schmidt} (first suggested in
\cite{bo} and \cite{liptalk}) for dealing with the large corrections
to BFKL at NLL. It is to introduce explicitly a rapidity separation
parameter $\Delta$ into the BFKL equation, enforcing the condition
$y_{i+1}-y_{i}>\Delta$, where $\Delta$ is assumed to be much less than
the total rapidity interval $y$. This parameter can be included
systematically
at any order in the resummation, and it plays a role for the
rapidity resummation similar to the role played by the $\overline{\rm MS}$
renormalization scale $\mu$ for the resummation of logarithms in the
running coupling. A change in $\Delta$ shifts pieces of the
calculation between LL and NLL, such that any differences are always
next-to-next-to-leading logarithm
(NNLL). Thus, the dependence on $\Delta$ can be regarded as an estimate
of the uncertainty due to NNLL corrections.
\begin{figure}[t]
\vskip-1.0in
\centerline{\epsfxsize 5.5 truein
\epsfbox{ltnl.eps}\hskip-2.2in
\epsfxsize 5.5 truein
\epsfbox{ltnl.deriv.eps}}
\vskip-4.2in
\caption[]{
\label{salamFigure}
\small The result of resumming double transverse logarithms on $\chi$
and its second derivative, from
Ref.~\cite{salam}.}
\end{figure}
\vskip-0.2in
\begin{figure}[t]
\centerline{\epsfxsize 4.0 truein \epsfbox{BFKLA.eps}\hskip-0.8in
\epsfxsize 4.0 truein \epsfbox{BFKLB.eps}}
\vskip -.2 cm
\caption[]{
\label{schmidtFigure}
\small Dependence of $\tilde A=\bar\alpha_{s}\chi({1\over2})$ and
$\tilde B=-{1\over2}\bar\alpha_{s}\chi''({1\over2})$ on $\Delta$
for $\alpha_{s}=0.15$, from Ref.~\cite{schmidt}.}
\end{figure}
Figure 3 shows the dependence on $\Delta$ of the leading eigenvalue
and its second derivative at LL and NLL for $\alpha_{s}=0.15$.
We note that the corrections to $\bar\alpha_{s}\chi({1\over2})$ are not
large for $\Delta\gtrsim2$ and have weak dependence on $\Delta$ for
large $\Delta$. Also, the point $\gamma={1\over2}$ is a maximum for this
coupling as long as
$\Delta\gtrsim2.2$. Thus, the BFKL resummation is
stable for large enough $\Delta$. We also note that this procedure
gives predictions of
$\bar\alpha_{s}\chi({1\over2})$ and $\bar\alpha_{s}\chi''({1\over2})$ for large
$\Delta$ which are similar to the previous proposal. However,
the implications of a large value of $\Delta$ for the
phenomenological use of BFKL is open to interpretation.
Recently, Forshaw, Ross, and Sabio Vera \cite{frv} have studied the inclusion
of both the double transverse logarithm resummation and the rapidity veto
simultaneously. Whereas ref.~\cite{schmidt} emphasized the weak
dependence on $\Delta$ at large $\Delta$, they were more concerned by
the strong dependence at small $\Delta$. They showed
that after including the resummation of the double transverse
logarithms, the dependence on the rapidity veto parameter $\Delta$
was significantly reduced. Given the discussion above, this is
reasonable since both the double transverse logarithm resummation and
the rapidity veto incorporate the same physical effect: a suppression
of gluon emissions close by in rapidity.
A third proposal to deal with the large NLL corrections was presented
by Brodsky {\it et al.} \cite{blm}. They re-evaluated
the NLL corrections in a suitable physical renormalization scheme, and
then used the BLM procedure \cite{blmone} to find the optimal scale
setting for the QCD coupling. The physical connection of this
proposal to the other two is less obvious; however, as in the previous
proposals it works by reducing the LL prediction (in this case by the
choice of the large scale dictated by BLM) combined with a subsequent
reduction in the NLL corrections. In addition it yields a very weak
dependence on the gluon virtuality $p_{1\perp}^{2}$ and leads to an
approximate conformal invariance.
\section{Conclusions.}
In this talk I have given a brief overview of the BFKL resummation
program, and have discussed some of the issues that have arisen
from the incorporation of the NLL corrections. The surprisingly large
size of the corrections at NLL, as well as the subtle issues related to the
running of the coupling, have spurred investigations which will lead
to a better understanding of the physics of QCD at high energies.
Clearly, this is a
challenging and lively field of theoretical research which will
significantly impact our understanding of QCD.
|
\section{Introduction}
Recently there was proved the exact integrability
of one discrete $3d$ model, classical as well as quantum
\cite{s-3dsympl,s-qem}. This model was formulated originally
in terms of the pure time evolution as the map of the dynamical
variables from time $t$ to time $t+1$. Namely, for $t$ fixed
the system of the dynamical variables are the system of
\begin{equation}
\displaystyle [\;\mbox{\sf u}_{\alpha,a,b}\;,\;\mbox{\sf w}_{\alpha,a,b}\;]\;,\;\;\;
\alpha=1,2,3,\;\;\;a,b\in Z_M\;,
\end{equation}
where $M$ is the finite spatial size of the system,
and the Poisson brackets in the classical case are
\begin{equation}\label{Poisson}
\displaystyle
\{\mbox{\sf u}_{\alpha,a,b}\;,\;\mbox{\sf w}_{\alpha',a',b'}\}\;=\;
\delta_{\alpha,\alpha'}\;\delta_{a,a'}\;\delta_{b,b'}\;
\mbox{\sf u}_{\alpha,a,b}\,\mbox{\sf w}_{\alpha,a,b}\;.
\end{equation}
The evolution was formulated as the explicit form of functions $f,g$,
\begin{equation}
\displaystyle
[\mbox{\sf u},\mbox{\sf w}]\;=\;[\mbox{\sf u}(t),\mbox{\sf w}(t)]\;\mapsto\;
[\mbox{\sf u}(t+1),\mbox{\sf w}(t+1)]\;=\;[f(\mbox{\sf u},\mbox{\sf w}),g(\mbox{\sf u},\mbox{\sf w})]\;,
\end{equation}
such that the Poisson brackets are conserved by this map.
The existence of the complete set of involutive integrals of motion
was proved in \cite{s-3dsympl,s-qem}.
The map $t\;\mapsto\;t+1$ may be considered as a sort of
the Hamiltonian
equations of motion for this classical discrete model.
In this
paper we will derive their Lagrangian counterpart. Following
\cite{rmk-lybe}, one may guess \`a priori that the Lagrangian
variable is Hirota's tau function and it might obey Hirota's discrete
bilinear equations \cite{Hirota}.
But we will show that this is not true in
general, although equations we will derive resemble Hirota's ones.
In this note we recall the reader the Hamiltonian form
of the equations of motion first, then introduce the analogous
of the tau function and derive the Lagrangian form of the
equations of motion, and finally give the multi-soliton
solution of them.
\section{Hamiltonian equations of motion}
In this paper we investigate the equations of motion,
so we need no the evolution operator and may choose more
appropriate coordinate system. Actually, the coordinates
we will use here are the light cone frame with respect to previous
$t,a,b$. Let $\mbox{\sf e}_1,\mbox{\sf e}_2,\mbox{\sf e}_3$ be three translations making
\begin{equation}
\displaystyle\begin{array}{ccl}
\displaystyle\mbox{\sf e}_1 & : & \displaystyle (t,a,b)\mapsto(t+1,a,b)\;,\\
&&\\
\displaystyle\mbox{\sf e}_2 & : & \displaystyle (t,a,b)\mapsto(t+1,a+1,b)\;,\\
&&\\
\displaystyle\mbox{\sf e}_3 & : & \displaystyle (t,a,b)\mapsto(t+1,a,b+1)\;.
\end{array}\end{equation}
$\mbox{\sf e}_1,\mbox{\sf e}_2,\mbox{\sf e}_3$ are the elementary orthonormal
shifts of the three dimensional
cubic lattice. So for any site of the cubic lattice $\mbox{\sf p}$ given,
the following eight sites form the elementary cube of the lattice:
\begin{equation}\label{cube}
\displaystyle\left(\begin{array}{cccc}
\displaystyle \mbox{\sf p}\;,&\displaystyle\mbox{\sf p}+\mbox{\sf e}_2+\mbox{\sf e}_3\;,&\displaystyle\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_3\;,&\displaystyle\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_2\;,\\
&&&\\
\displaystyle\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_2+\mbox{\sf e}_3\;,&\displaystyle\mbox{\sf p}+\mbox{\sf e}_1\;,&\displaystyle\mbox{\sf p}+\mbox{\sf e}_2\;,&\displaystyle\mbox{\sf p}+\mbox{\sf e}_3\;.
\end{array}\right)\end{equation}
Equations of motion for $\mbox{\sf u}_{\alpha,\mbox{\sf p}}$, $\mbox{\sf w}_{\alpha,\mbox{\sf p}}$ may
be extracted from the form of $f,g$ in the definition of the
evolution in \cite{s-3dsympl,s-qem}, and in our coordinates
look like
\noindent
(i)
\begin{equation}\label{ev-1}
\displaystyle\left\{\begin{array}{ccl}
\displaystyle\mbox{\sf w}_{1,\mbox{\sf p}+\mbox{\sf e}_1} & = & \displaystyle
{\mbox{\sf w}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,+\,\mbox{\sf u}_{3,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,+\,
\kappa_3\,\mbox{\sf u}_{3,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}\over\mbox{\sf w}_{3,\mbox{\sf p}}}\;,\\
&&\\
\displaystyle\mbox{\sf u}_{1,\mbox{\sf p}+\mbox{\sf e}_1} & = & \displaystyle
{\kappa_2\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\over
\kappa_1\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,+\,
\kappa_3\,\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}\,+\,
\kappa_1\,\kappa_3\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}}\;,
\end{array}\right.
\end{equation}
(ii)
\begin{equation}\label{ev-2}
\displaystyle\left\{\begin{array}{ccl}
\displaystyle\mbox{\sf w}_{2,\mbox{\sf p}+\mbox{\sf e}_2} & = & \displaystyle
{\mbox{\sf w}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}\over
\mbox{\sf w}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,+\,
\mbox{\sf u}_{3,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,+\,
\kappa_3\,\mbox{\sf u}_{3,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}}\;,\\
&&\\
\displaystyle\mbox{\sf u}_{2,\mbox{\sf p}+\mbox{\sf e}_2} & = & \displaystyle
{\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf u}_{3,\mbox{\sf p}}\over
\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf u}_{3,\mbox{\sf p}}\,+\,\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{1,\mbox{\sf p}}\,+\,
\kappa_1\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{1,\mbox{\sf p}}}\;,
\end{array}\right.\\
\end{equation}
(iii)
\begin{equation}\label{ev-3}
\displaystyle\left\{\begin{array}{ccl}
\displaystyle\mbox{\sf w}_{3,\mbox{\sf p}+\mbox{\sf e}_3} & = & \displaystyle
{\kappa_2\,\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}\over
\kappa_1\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}}\,+\,
\kappa_3\,\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}\,+\,
\kappa_1\,\kappa_3\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{3,\mbox{\sf p}}}\;,\\
&&\\
\displaystyle\mbox{\sf u}_{3,\mbox{\sf p}+\mbox{\sf e}_3} & = & \displaystyle
{\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf u}_{3,\mbox{\sf p}}\,+\,
\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf w}_{1,\mbox{\sf p}}\,+\,
\kappa_1\,\mbox{\sf u}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{1,\mbox{\sf p}}\over \mbox{\sf u}_{1,\mbox{\sf p}}}\;.
\end{array}\right.
\end{equation}
Describe in a couple of words the idea of the derivation of
equations (\ref{ev-1}-\ref{ev-3}).
These relations may be obtained as the zero curvature condition
of the following system. Let $\varphi_{\mbox{\sf p}}$ be an auxiliary
variable assigned to the sites of the cubic lattice.
Consider three orthogonal plaquettes of the cube endowed by the
following
relations
\begin{equation}\label{linear}
\displaystyle\begin{array}{ccc}
\displaystyle 0\;=\;f_{1,\mbox{\sf p}} &\stackrel{def}{=}& \displaystyle
\varphi_{\mbox{\sf p}}\;\mbox{\sf w}_{1,\mbox{\sf p}}\;-\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_2+\mbox{\sf e}_3}\;\mbox{\sf u}_{1,\mbox{\sf p}}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_2}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_3}\;\kappa_1\;\mbox{\sf u}_{1,\mbox{\sf p}}\;\mbox{\sf w}_{1,\mbox{\sf p}}\;,\\
&&\\
\displaystyle 0\;=\;f_{2,\mbox{\sf p}} &\stackrel{def}{=}& \displaystyle
\varphi_{\mbox{\sf p}}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_3}\;\kappa_2\;\mbox{\sf u}_{2,\mbox{\sf p}}\;\mbox{\sf w}_{2,\mbox{\sf p}}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_1}\;\mbox{\sf w}_{2,\mbox{\sf p}}\;-\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_3}\;\mbox{\sf u}_{2,\mbox{\sf p}}\;,\\
&&\\
\displaystyle 0\;=\;f_{3,\mbox{\sf p}} &\stackrel{def}{=}& \displaystyle
-\;\varphi_{\mbox{\sf p}}\;\mbox{\sf u}_{3,\mbox{\sf p}}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_2}\;\mbox{\sf w}_{3,\mbox{\sf p}}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_2}\;+\;
\varphi_{\mbox{\sf p}+\mbox{\sf e}_1}\;\kappa_3\;\mbox{\sf u}_{3,\mbox{\sf p}}\;\mbox{\sf w}_{3,\mbox{\sf p}}\;.
\end{array}\end{equation}
Easy to see that the number of linear equations is
three times greater then the number of $\varphi_{\mbox{\sf p}}$.
This means that the coefficients of the system of linear
equations must obey extra conditions.
As an example consider
a cube. Six plaquettes give six linear relations
for eight corner $\varphi_{\mbox{\sf p}}$:
\begin{equation}
\displaystyle f_{\alpha,\mbox{\sf p}}\;=\;f_{\alpha,\mbox{\sf p}+\mbox{\sf e}_\alpha}\;=\;0\;,\;\;\;
\alpha\;=\;1,2,3\;.
\end{equation}
Demand that between these
six relations there are only four linearly independent,
so that the cube relations fix only four of eight corner
linear variables $\varphi_{\mbox{\sf p}}$
(e.g. we just may express
$\varphi_{\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_2}$, $\varphi_{\mbox{\sf p}+\mbox{\sf e}_2+\mbox{\sf e}_3}$,
$\varphi_{\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_3}$ and $\varphi_{\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_2+\mbox{\sf e}_3}$
via independent $\varphi_{\mbox{\sf p}}$, $\varphi_{\mbox{\sf p}+\mbox{\sf e}_1}$,
$\varphi_{\mbox{\sf p}+\mbox{\sf e}_2}$ and $\varphi_{\mbox{\sf p}+\mbox{\sf e}_3}$ and nothing more).
This demand gives equations of motion (\ref{ev-1}-\ref{ev-3})
immediately.
\section{Lagrangian equations of motion}
Three simple relations between the equations of motion are to be
mentioned:
\begin{equation}\label{simple}
\displaystyle\left\{\begin{array}{ccc}
\displaystyle\mbox{\sf w}_{1,\mbox{\sf p}}\,\mbox{\sf w}_{2,\mbox{\sf p}} & = & \displaystyle \mbox{\sf w}_{1,\mbox{\sf p}+\mbox{\sf e}_1}\,\mbox{\sf w}_{2,\mbox{\sf p}+\mbox{\sf e}_2}\;,\\
&&\\
\displaystyle\mbox{\sf u}_{2,\mbox{\sf p}}\,\mbox{\sf u}_{3,\mbox{\sf p}} & = & \displaystyle \mbox{\sf u}_{2,\mbox{\sf p}+\mbox{\sf e}_2}\,\mbox{\sf u}_{3,\mbox{\sf p}+\mbox{\sf e}_3}\;,\\
&&\\
\displaystyle\mbox{\sf u}_{1,\mbox{\sf p}}/\mbox{\sf w}_{3,\mbox{\sf p}} & = & \displaystyle \mbox{\sf u}_{1,\mbox{\sf p}+\mbox{\sf e}_1}/\mbox{\sf w}_{3,\mbox{\sf p}+\mbox{\sf e}_3}\;.
\end{array}\right.\end{equation}
Following \cite{rmk-lybe},
a parametrization of $\mbox{\sf u},\mbox{\sf w}$ in terms of tau functions
must turn relations (\ref{simple}) into tautologies.
Thus without lost of generality (we deal with the infinite system,
avoiding hence all the boundary problems)
\begin{equation}\label{tau-param}
\displaystyle\left\{\begin{array}{ll}
\displaystyle\mbox{\sf w}_{1,\mbox{\sf p}}^{}\;=\;
{{\cal T}_{3,\mbox{\sf p}+\mbox{\sf e}_2}\over{\cal T}_{3,\mbox{\sf p}}}\;,\;\;\;\;&
\displaystyle\mbox{\sf u}_{1,\mbox{\sf p}}^{}\;=\;
{{\cal T}_{2,\mbox{\sf p}}\over{\cal T}_{2,\mbox{\sf p}+\mbox{\sf e}_3}}\;,\\
&\\
\displaystyle\mbox{\sf w}_{2,\mbox{\sf p}}^{}\;=\;
{{\cal T}_{3,\mbox{\sf p}}\over{\cal T}_{3,\mbox{\sf p}+\mbox{\sf e}_1}}\;,&
\displaystyle\mbox{\sf u}_{2,\mbox{\sf p}}^{}\;=\;
{{\cal T}_{1,\mbox{\sf p}}\over{\cal T}_{1,\mbox{\sf p}+\mbox{\sf e}_3}}\;,\\
&\\
\displaystyle\mbox{\sf w}_{3,\mbox{\sf p}}^{}\;=\;
{{\cal T}_{2,\mbox{\sf p}}\over{\cal T}_{2,\mbox{\sf p}+\mbox{\sf e}_1}}\;,&
\displaystyle\mbox{\sf u}_{3,\mbox{\sf p}}^{}\;=\;
{{\cal T}_{1,\mbox{\sf p}+\mbox{\sf e}_2}\over{\cal T}_{1,\mbox{\sf p}}}\;.
\end{array}\right.\end{equation}
As usual, ${\cal T}_{\alpha,\mbox{\sf p}}$ may contain a quadratic and linear
pre-exponent,
\begin{equation}\label{tau-pure}
\displaystyle{\cal T}_{\alpha,\mbox{\sf p}}\;=\;
\mbox{{\large\bf e}}^{{1\over 2}\;(\mbox{\sf p} ,Q_\alpha,\mbox{\sf p})\;+\;
(\mbox{\sf q}_\alpha,\mbox{\sf p})}\,\cdot\,\tau_{\alpha,\mbox{\sf p}}\;,
\end{equation}
where $\mbox{\sf q}_\alpha$ and diagonals of
$Q_\alpha$ are arbitrary, but all non-diagonal
elements of $Q_\alpha$ must coincide (in the natural basis
$\mbox{\sf e}_\alpha$).
Substituting now the parametrization
(\ref{tau-param}) to (\ref{ev-1}-\ref{ev-3})
and taking (\ref{tau-pure}) into account,
we obtain Three Trilinear relations for Three $\tau$
functions:
\noindent
\begin{equation}\label{tril-1}
\displaystyle\left\{\begin{array}{ccc}
\displaystyle r_1^{}\;\tau_{1,\mbox{\sf p}+\mbox{\sf e}_2+\mbox{\sf e}_3}\;\tau_{2,\mbox{\sf p}}\;\tau_{3,\mbox{\sf p}} & = &
\displaystyle \tau_{1,\mbox{\sf p}}\;\tau_{2,\mbox{\sf p}+\mbox{\sf e}_3}\;\tau_{3,\mbox{\sf p}+\mbox{\sf e}_2}\\
&&\\&+&
\displaystyle s_2^{}\;\tau_{1,\mbox{\sf p}+\mbox{\sf e}_2}\;\tau_{2,\mbox{\sf p}+\mbox{\sf e}_3}\;\tau_{3,\mbox{\sf p}}\\
&&\\&+&
\displaystyle s_3^{-1}\;\tau_{1,\mbox{\sf p}+\mbox{\sf e}_3}\;\tau_{2,\mbox{\sf p}}\;\tau_{3,\mbox{\sf p}+\mbox{\sf e}_2}\;,
\end{array}\right.\end{equation}
\begin{equation}\label{tril-2}
\displaystyle\left\{\begin{array}{ccl}
\displaystyle r_2^{}\;\tau_{1,\mbox{\sf p}}\,\tau_{2,\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_3}\,\tau_{3,\mbox{\sf p}} & = &
\displaystyle \tau_{1,\mbox{\sf p}+\mbox{\sf e}_3}\,\tau_{2,\mbox{\sf p}}\,\tau_{3,\mbox{\sf p}+\mbox{\sf e}_1}\\
&&\\&+&
\displaystyle s_3^{}\;\tau_{1,\mbox{\sf p}}\,\tau_{2,\mbox{\sf p}+\mbox{\sf e}_3}\,\tau_{3,\mbox{\sf p}+\mbox{\sf e}_1}\\
&&\\&+&
\displaystyle s_1^{-1}\;\tau_{1,\mbox{\sf p}+\mbox{\sf e}_3}\,\tau_{2,\mbox{\sf p}+\mbox{\sf e}_1}\,\tau_{3,\mbox{\sf p}}\;,
\end{array}\right.\end{equation}
\begin{equation}\label{tril-3}
\displaystyle\left\{\begin{array}{ccl}
\displaystyle r_3^{}\;\tau_{1,\mbox{\sf p}}\,\tau_{2,\mbox{\sf p}}\,\tau_{3,\mbox{\sf p}+\mbox{\sf e}_1+\mbox{\sf e}_2} & = &
\displaystyle \tau_{1,\mbox{\sf p}+\mbox{\sf e}_2}\,\tau_{2,\mbox{\sf p}+\mbox{\sf e}_1}\,\tau_{3,\mbox{\sf p}}\\
&&\\&+&\displaystyle
s_1^{}\;\tau_{1,\mbox{\sf p}+\mbox{\sf e}_2}\,\tau_{2,\mbox{\sf p}}\,\tau_{3,\mbox{\sf p}+\mbox{\sf e}_1}\\
&&\\&+&
\displaystyle s_2^{-1}\;\tau_{1,\mbox{\sf p}}\,\tau_{2,\mbox{\sf p}+\mbox{\sf e}_1}\,\tau_{3,\mbox{\sf p}+\mbox{\sf e}_2}\;.
\end{array}\right.\end{equation}
Here we gather all $\kappa$-s and all the pre-exponents
into arbitrary $s_\alpha,r_\alpha$, $\alpha=1,2,3$.
One may change them in any appropriate way.
Because of the possibility to introduce the Poisson brackets
(\ref{Poisson}) in the system of $\mbox{\sf u}_{\alpha,\mbox{\sf p}},\mbox{\sf w}_{\alpha,\mbox{\sf p}}$,
the original equations of motion (\ref{ev-1}-\ref{ev-3})
may be regarded as a kind of the Hamiltonian equations of motion.
Contrary to that, the tautological parametrization of (\ref{simple})
resembles the introduction of a velocity instead of a momentum,
so that the number of variables decreases twice, so the trilinear
relations (\ref{tril-1}-\ref{tril-3}) are nothing but the
Lagrangian form
of the equations of motion.
(\ref{tril-1}-\ref{tril-3}) may be reduced to
Hirota's bilinear discrete equation in the limit
\begin{equation}
\displaystyle\kappa_1\;<<\;\kappa_2\;=\;\kappa_3\;<<\;1\;.
\end{equation}
\section{Solitons}
Now describe the simple solitons of the trilinear
relations (\ref{tril-1}-\ref{tril-3}).
Let $(\alpha,\beta,\gamma)$ be any cyclic permutation of $(1,2,3)$.
Fix $r_\alpha$ via
\begin{equation}
\displaystyle
r_\alpha\;=\;1\;+\;s_\beta\;+\;{1\over s_\gamma}\;.
\end{equation}
Then let the exponent
\begin{equation}
\displaystyle W\;=\;\mbox{{\large\bf e}}^{i\,(\mbox{\sf k},\mbox{\sf p})}\;=\;\mbox{{\large\bf e}}^{\displaystyle
i\,k_1\,p_1+i\,k_2\,p_2+i\,k_3\,p_3}
\end{equation}
for $s_\alpha$ given, is defined by
\begin{equation}\label{spectr}
\displaystyle
\mbox{{\large\bf e}}^{\displaystyle i\;k_\alpha}\;=\;
{\lambda_\alpha\;((\lambda_\alpha-\lambda_\gamma)\;+\;
s_\alpha\,(\lambda_\alpha-\lambda_\beta))\over
\lambda_\beta\,(\lambda_\alpha-\lambda_\gamma)\;+\;
s_\alpha\,\lambda_\gamma\,(\lambda_\alpha-\lambda_\beta)}\;.
\end{equation}
This parametrization
of the dispersion relation we will exhibit as $W=W(\lambda)$.
For the dispersion curve $W=W(\lambda)$ given, let for the shortness
\begin{equation}
\displaystyle
W^{(k)}_\alpha\;=\;\lambda_\alpha^{(k)}\;W(\lambda^{(k)})\;.
\end{equation}
The phase shift $D$ is defined by
\begin{equation}
\displaystyle
D(\lambda,\lambda^\prime)\;=\;
{d(\lambda,\lambda^\prime)\,\cdot\,d(\lambda^{-1},\lambda^{\prime-1})
\over
d(\lambda^{-1},\lambda^\prime)\,\cdot\,d(\lambda,\lambda^{\prime-1})}
\;,
\end{equation}
where
\begin{equation}
\displaystyle
d(\lambda,\lambda^\prime)\;=\;\det\;\left(
\begin{array}{ccc}
\displaystyle 1 &\displaystyle 1 &\displaystyle 1 \\
&&\\
\displaystyle \lambda_1^{} & \displaystyle \lambda_2^{} & \displaystyle \lambda_3^{}\\
&&\\
\displaystyle \lambda_1' & \displaystyle \lambda_2' & \displaystyle \lambda_3'
\end{array}\right)\;,
\end{equation}
and $\lambda^{-1}$ stands for
$\{\lambda_1^{-1},\lambda_2^{-1},\lambda_3^{-1}\}$.
For the shortness let
\begin{equation}
\displaystyle D^{(k,l)}\;=\;D(\lambda^{(k)},\lambda^{(l)})\;.
\end{equation}
One soliton solution is
$\displaystyle\tau_{\alpha,\mbox{\sf p}}^{(1)}\;=\;1\;-\;W_\alpha$.
Two solitons are
\begin{equation}
\displaystyle\tau_{\alpha,\mbox{\sf p}}^{(2)}\;=\;
1\;-\;W_\alpha^{(1)}\;-\;W_\alpha^{(2)}\;+\;
D^{(1,2)}\,W_\alpha^{(1)}\,W_\alpha^{(2)}\;.
\end{equation}
Three solitons are
\begin{equation}
\displaystyle\begin{array}{ccl}
\displaystyle\tau_{\alpha,\mbox{\sf p}}^{(3)} & = & \displaystyle
1\;-\;W_\alpha^{(1)}\;-\;W_\alpha^{(2)}\;-\;W_\alpha^{(3)}\\
&&\\
&&\displaystyle
+\;D^{(1,2)}\,W_\alpha^{(1)}\,W_\alpha^{(2)}
+D^{(1,3)}\,W_\alpha^{(1)}\,W_\alpha^{(3)}
+D^{(2,3)}\,W_\alpha^{(2)}\,W_\alpha^{(3)}\\
&&\\
&&-\;
D^{(1,2)}\,D^{(1,3)}\,D^{(2,3)}\,
W_\alpha^{(1)}\,W_\alpha^{(2)}\,W_\alpha^{(3)}\;.
\end{array}\end{equation}
And so on, common expression is usual.
Remarkable is that the dispersion relation
for the exponents $\mbox{{\large\bf e}}^{i\,k_\alpha}$ may be parametrized by
their origin $\lambda_\alpha$, (\ref{spectr}),
and parameters $s_\alpha$ are defined by an arbitrary linear
pre-exponent.
\section{Discussion}
System (\ref{tril-1}-\ref{tril-3}) contains a set of bilinear
relations
as its necessary conditions. Looking for a solution of trilinear
relations in a form of holomorphic functions, we get the following
set of necessary relations
\begin{equation}\label{holomorphic}
\Lambda_{\alpha,\mbox{\sf p}+\mbox{\sf e}_\alpha}\,\tau_{\alpha,\mbox{\sf p}}\;=\;
\tau_{\beta,\mbox{\sf p}+\mbox{\sf e}_\alpha}\,\tau_{\gamma,\mbox{\sf p}}\;+\;
s_\alpha\;\tau_{\beta,\mbox{\sf p}}\,\tau_{\gamma,\mbox{\sf p}+\mbox{\sf e}_\alpha}\;,
\end{equation}
where as before $(\alpha,\beta,\gamma)$ are any cyclic permutation
of $(1,2,3)$, and auxiliary functions
$\Lambda_{\alpha,\mbox{\sf p}}$ are supposed to be holomorphic. Thus
(\ref{tril-1}-\ref{tril-2}) have a host of solutions similar
to that of Hirota's equation. Namely, identifying (\ref{holomorphic})
with a set of Fay's or Pl\"ucker's relations, we may obtain
elliptic or determinant solution of (\ref{tril-1}-\ref{tril-3}).
Probably, the exception is a Bethe -- ansatz -- type solution of
(\ref{holomorphic}), because in this case (\ref{tril-1}-\ref{tril-3})
do not follow from (\ref{holomorphic}) tautologically.
Because of all these do not contain anything so surprising as
the solitons, we do not give any explicit formula here.
Mention only two aspects.
First one is the symmetry group of (\ref{tril-1}-\ref{tril-3}).
For Hirota's equation all the symmetries with respect to
permutations and reflections of the space directions are obvious.
In the case of (\ref{tril-1}-\ref{tril-3}) the Cube group is less
trivial because of it acts on the parameters $s_\alpha,r_\alpha$.
As an example give the realisation of the complete reflection $P$.
For a solution $\tau_{\alpha,\mbox{\sf p}}$ of the trilinear equations with
some $s_\alpha,r_\alpha$ given let
\begin{equation}
\overline{\tau}_{\alpha,\mbox{\sf p}}\;=\;\tau_{\alpha,\mbox{\sf p}-\mbox{\sf e}_\alpha}\;,\;\;
\overline{s}_\alpha\;=\;
{r_\beta\over r_\gamma}\;{s_\alpha\over s_\beta\,s_\gamma}\;,\;\;
\overline{r}_\alpha\;=\;
{s_\beta\over s_\gamma}\;{r_\beta\,r_\gamma\over r_\alpha}\;.
\end{equation}
Then the set of ``overlined'' objects obey the
same set of the trilinear relations (\ref{tril-1}-\ref{tril-3})
but with reflected $\mbox{\sf e}_\alpha\;\mapsto\;-\;\mbox{\sf e}_\alpha$ for all
$\alpha$.
Mention also the possibility to construct the complete solution
of original finite evolution model in terms of the elliptic functions
on a finite genus curve.
In such case one may consider a non-homogeneous and non-equidistant
version of the trilinear relations with a boundary conditions
taken into account. The number of the parameters
(moduli, initial divisor and homogeneousless data) will coincide
with the number of variables of the initial state.
This will be the subject of a separate paper.
\noindent
{\bf Acknowledgments.} I would like to thank Rinat Kashaev
for wery useful explanaions and comments.
The work was supported by the RFBR grant No. 98-01-00070.
|
\section{Introduction}
In 1905 Albert Einstein described Geomagnetism as one of the five unsolved
problems of physics\cite{r1}. After nearly a century, inspite of a tremendous
amount of work which has culminated in the dynamo model of Geomagnetism,
\cite{r2}, it cannot be said with confidence that the problem has been
solved. From time to time, simulations are developed which improve upon
earlier models\cite{r3,r4,r5}, but several unexplained features persist.
These include the problems of Geomagnetic reversals\cite{r6,r7,r8,r9}. In
particular it may be mentioned (cf.ref.\cite{r9} and \cite{r1})that Muller and
Morris have attributed the reversals to asteroid impacts, which in turn
have been related to mass extinctions.\\
We will point out in what follows that in the light of recent work on
semionic and anomalous behaviour of Fermions under special conditions, the
solid core of the earth which has hitherto not been considered could
contribute significantly to Geomagnetism, and could even facilitate an
explanation for the magnetic reversals.
\section{Magnetism of the Solid Core}
It is well known\cite{r10,r11} that the earth has a solid core with a radius
of about 1200 kilometers, composed mostly of Iron $(90\%)$ and Nickel $(10\%)$,
at a temperature of about 6000 degrees centigrade and with a relative
density around 10. This in turn is surrounded by the liquid core which
it is believed gives rise to the dynamo model of Geomagnetism.\\
Given the above data, using the atomic weight of iron, we can easily
calculate that the number of atoms in the solid core, $N$ is given by,
\begin{equation}
N \approx 10^{48}\label{e1}
\end{equation}
where the symbol $\approx$ denotes, "of the order of".\\
We next calculate the Fermi temperature of the conduction electrons in the
solid core. This is given by\cite{r12},
\begin{equation}
kT_F = \left\{6\pi^2\left(\frac{N}{V}\right)\right\}^{2/3} \frac{\hbar^2}{2m}\label{e2}
\end{equation}
where $V$ is the volume of the solid core, $k$ is the Boltzmann constant
$\hbar$ is the reduced Planck constant and $m$ the electron mass.\\
It follows from (\ref{e2}) that
\begin{equation}
T_F \approx 10^{5^o}C\label{e3}
\end{equation}
Thus one can see that the temperature of the solid core is below the
Fermi temperature of the conduction electrons.\\
In recent years it has been realized that under special conditions like low
dimensionality or temperatures, conduction electrons do not strictly obey
Fermi-Dirac statistics, but rather they are semionic, that is
they obey statistics between the Fermi-Dirac and Bose-Einstein statistics
\cite{r13,r14}. In particular, this is true below the Fermi temperature
\cite{r15}. The implications are interesting:\\
Given Fermi Dirac statistics, at temperatures below the Fermi temperature
we would have (cf.ref.\cite{r12}) for the magnetisation $M$ per unit volume
the formula
\begin{equation}
M = \frac{\mu (2\bar N_+ - N)}{V}\label{e4}
\end{equation}
where $\mu$ is the electron magnetic moment and $\bar N_+$ is the
average number of electrons with spin up, say, where $\bar N_+ \approx
\frac{N}{2}$, so that $M$ in (\ref{e4}) is very small.\\
However if the behaviour is not Fermionic, but rather Bosonic,
then
$$\bar N_+ \approx N$$
In our case,
$$\frac{N}{2} < \bar N_+ < N$$
With this input (\ref{e4}) becomes,
\begin{equation}
M \leq \frac{\mu N}{V}\label{e5}
\end{equation}
From (\ref{e5}) we can easily deduce that the terrestrial magnetic field
$H$ is given by,
$$H \leq \frac{MV}{r^3} \approx 1 G$$
The above order of magnetic calculation thus gives the correct order of the
terrestrial magnetism. Moreover, this would have the added advantage that
it could explain geomagnetic reversals: The semionic behaviour of the electrons
in the solid core is sensitive to external magnetic influences and could
thus flip or reverse polarity. On the other hand, the explanation in the
case of the convective dynamo model would be contrived in comparison.
|
\section{Introduction}
\label{sec:Introdution}
Carbon and oxygen are two of the most abundant elements in the universe
and lines from these elements provide valuable plasma diagnostics for
almost all classes of cosmic sources. Essential for many of these
diagnostics are accurate electron-ion recombination rate coefficients,
particularly of dielectronic recombination (DR), which for most ions in
electron-ionized plasmas is the dominant means of electron capture
(Arnaud \& Rothenflug 1985). Producing accurate theoretical DR rate
coefficients is, however, theoretically and computationally
challenging.
In the past, semi-empirical expressions such as the Burgess (1965)
formula along with modified versions by Burgess \& Tworkowski (1976)
and Merts et al.\ (1976) were developed to calculate DR rates. More
recently, a number of more sophisticated theoretical approaches have
been used to calculate DR, among them single-configuration
$LS$-coupling (Bellantone \& Hahn 1989), multiconfiguration
intermediate-coupling (Pindzola, Badnell, \& Griffin 1990), and
multiconfiguration fully-relativistic (Chen 1988) techniques, as well
as undamped and damped, unified radiative recombination (RR) and DR
calculations in $LS$-coupling (Nahar \& Pradhan 1997; Nahar 1999).
Approximations, though, need to be made to make any of these techniques
computationally tractable (Hahn 1993). Currently, sophisticated DR
calculations are non-existent for many ions, and in the absence of
anything better, semi-empirical formulae are often still used for
plasma modeling.
Laboratory measurements can be used to test the different theoretical
and computational techniques for calculating DR. Recently, Savin et
al.\ (1997, 1999) developed a technique for obtaining rate coefficients
from laboratory measurements of DR resonance strengths and energies.
They successfully used this technique to derive rates for $\Delta n=0$
DR of Fe XVIII and Fe XIX and to benchmark existing theoretical
calculations. Here, we describe this technique in detail for the first
time and apply it to recent DR measurements in C~V and O~VIII.
Kilgus et al.\ (1990, 1993) and Mannervik et al.\ (1997) have measured
the resonance strengths and energies for DR of C~V to C~IV and O~VIII
to O~VII. We use their results to produce DR rate coefficients to
benchmark existing C~V and O~VIII DR calculations and to provide rates
for use in plasma modeling.
In electron-ionized plasmas, lines from heliumlike C~V and hydrogenic
O~VIII trace gas at $T_e \sim 10^{5.1-5.9}$ K and $\sim 10^{6.4}$ K,
respectively (Arnaud \& Rothenflug 1985; Mazzotta et al.\ 1998). C~V
and O~VIII lines have been observed in solar spectra (Doschek \& Cowan
1984) and O~VIII lines in supernova remnants (Winkler et al.\ 1981).
And with the upcoming launches of {\it Chandra} and {\it XMM} and the
high-resolution spectrometers aboard, C~V and O~VIII lines are expected
to be seen in may other electron-ionized, cosmic sources.
Using different heavy-ion storage rings, Kilgus et al.\ (1993) and
Mannervik et al.\ (1997) have measured DR for C~V via the capture
channels
\begin{equation}
{\rm C}^{4+}(1s^2) + e^- \rightarrow {\rm C}^{3+}(1s2lnl^\prime)
\ (n=2,\ldots,n_{max}).
\end{equation}
where $n_{max} \sim 28$ for the results of Kilgus et al. and $\sim 16$
for the results of Mannervik et al. Kilgus et al.\ (1990) have also
measured DR for O~VIII via the capture channels
\begin{equation}
{\rm O}^{7+}(1s) + e^- \rightarrow {\rm O}^{6+}(2lnl^\prime)
\ (n=2,\ldots,n_{max})
\end{equation}
where $n_{max}\sim 69$. The radiative stabilization of these
autoionizing C~V and O~VII states to bound configurations results in
DR. Details of the experimental techniques used are given in the
references cited.
The paper is organized as follows: We describe in Section
\ref{sec:MethodofCalculation} how one produces a DR rate coefficient
using measured DR resonance strengths and energies. In Section
\ref{sec:ResultsandDiscussion} we present the resulting rate
coefficients and compare the derived DR rates with published theoretical
rates. We also give a simple fitting formula for use in plasma
modeling.
\section{Method of Calculation}
\label{sec:MethodofCalculation}
DR is a resonance process consisting, in the zero-density limit,
of an infinite number of
resonances. The DR rate coefficient $\alpha$ for a plasma with a
Maxwell-Boltzmann electron distribution is given by
\begin{equation}
\alpha(T_e)=\int \sum_d \sigma_d(E) v_e(E) P(T_e,E) dE
\label{eq:rate1}
\end{equation}
where $T_e$ is the electron temperature; $\sigma_d(E)$ is the
energy-dependent DR cross section for resonance $d$; $v_e(E)$ is the
relative electron-ion velocity at energy $E$, which is taken to be the
electron energy as the ions are nearly to stationary in the
center-of-mass frame; and the sum is over all DR resonances. The
Maxwell-Boltzmann distribution $P(T_e,E)$ is given by
\begin{equation}
P(E,T_e)dE ={2 E^{1/2} \over \pi^{1/2} (k_BT_e)^{3/2}} \exp\Biggl( {-E\over
k_BT_e}\Biggr)dE
\label{eq:pofe}
\end{equation}
where $k_B$ is the Boltzmann constant.
Kilgus et al.\ (1990, 1993) and Mannervik et al.\ (1997) published
measured DR resonance strengths $\hat{\sigma}_d$ and energies
$E_d$. The DR resonance strength is defined
\begin{equation}
\hat{\sigma}_d(E)=
\int_{E-\delta E/2}^{E+\delta E/2} \sigma_d(E^\prime) dE^\prime
\label{eq:strength1}
\end{equation}
where $\sigma_d$ is the cross section for a resonance or group of
resonances labeled $d$ and \{$E-\delta E$,$E+\delta E$\} is a region in
energy chosen such that it contains only those resonances comprising
$d$.
Here we are interested in calculating rate coefficients. This involves
convolving the DR resonances with the slowly varying function
$P(T_e,E)$. Because the energy widths of the measured resonances are
smaller than the scale over which $P(T_e,E)$ changes, for our purposes
we can accurately approximate $\sigma_d(E)$ as
\begin{equation}
\sigma_d(E)=\hat{\sigma}_d(E_d)\delta(E-E_d)
\label{eq:cross1}
\end{equation}
where $E_d$ is the energy of resonance $d$ and $\delta(E-E_d)$ is the Dirac
delta function. The DR rate coefficient for Maxwellian plasmas is
found by substituting Equation \ref{eq:cross1} into Equation
\ref{eq:rate1} which yields
\begin{equation}
\alpha(T_e)=\sum_d \hat{\sigma_d}(E_d) v_e(E_d) P(T_e,E_d).
\label{eq:rate2}
\end{equation}
Kilgus et al.\ (1993) and Mannervik et al.\ (1997) do not report
measured resonance energies for capture by C~V into levels where $n \ge 4$.
To calculate these resonance energies $E_n$ we use the Rydberg formula
\begin{equation}
E_n=\Delta E - q^2R_\infty/n^2
\label{eq:rescond}
\end{equation}
where $q=4$ is the charge of the ion before recombination,
$\Delta E=307.8$ eV is the energy of the $1s^2 (^1S_0) - 1s2p
(^1P_1)$ core excitation (Kelly 1987), and $R_\infty$ is the Rydberg
constant. For O~VIII, $q=7$ and $\Delta E=653.6$ eV is the
energy of the $1s (^1S_{1/2}) - 2p (^1P_{3/2,1/2})$ core excitation
(Kelly 1987).
Mannervik et al.\ (1997) estimate that they measured DR for capture into
levels $n_{max}\lesssim16$. Kilgus et al.\ (1993) do not report a value of
$n_{max}$, so we derive an estimate here. Using their ion energy, the
bending radius of the dipole magnets in their experiment (Linkemann
1995), and the semiclassical formula for field ionization (Brouillard
1983), we estimate electrons captured into levels where $n = n_{cut}
\gtrsim 19$ will be field ionized and thus not detected. However, the
captured electrons can radiatively decay below $n_{cut}$ during the
$\sim 5.1$ m distance they travel between the electron-ion interaction
region in the experiment and the dipole magnet (cf., Kilgus et al.\ 1992;
Savin et
al.\ 1997). Using the ion velocity, the hydrogenic formula for
radiative lifetimes of Marxer \& Spruch (1991), and calculations which
show that DR for heliumlike ions essentially populates only angular
momentum levels where $l\le4$ (Chen 1986), we estimate that electrons
captured into levels $n_{max} \lesssim 28$ will radiatively decay below
$n_{cut}$ and thus were detected by Kilgus et al. For OVIII, Kilgus et
al.\ (1990) estimate $n_{max}\sim 69$.
For a given DR series, as $n$ increases the energy spacing between DR
resonances decreases. Due to the energy resolutions of the experiments,
above some $n$ level it is not possible to resolve the DR resonances and
determine individual values of $\sigma_n$. For these high $n$ levels,
Kilgus et al.\ (1990; 1993) presents total resonance strengths summed
from $n\ge9$ to $n_{max}$ for C~V and from $n\ge8$ to $n_{max}$ for
O~VIII. We divide and spread out these summed resonance strengths into
$\sim 1$ eV wide bins when using Equation \ref{eq:rate2}. The
``resonance'' energies of these bins are chosen to lie between $E_9$
and $E_{n_{max}}$ for C~V and between $E_8$ and $E_{n_{max}}$ for
O~VIII. The bin widths are smaller than the scale over which $P(T_e,E)$
changes and the small errors in the exact resonance energies of these high
$n$ levels has an insignificant affect on the derived DR rates.
\section{Results and Discussion}
\label{sec:ResultsandDiscussion}
In Figures \ref{fig:CV} and \ref{fig:OVIII} we present the C~V and O
VIII DR rate coefficients, respectively, derived using the measured
resonance strengths and energies of Kilgus et al.\ (1990, 1993) and
Mannervik (1997). The unmeasured contribution to the DR rate due to capture
into levels where $n \gtrsim n_{max}$ is predicted to be insignificant
(Chen 1986; Pindzola et al.\ 1990). The absolute uncertainties in the
derived DR rates are estimated to be $\lesssim \pm25\%$ which corresponds
to the reported absolute experimental uncertainties.
Existing theoretical DR rates are also shown in Figures \ref{fig:CV}
and \ref{fig:OVIII} as well as the RR rates of Verner \& Ferland
(1996). For C~V, the single-configuration $LS$-coupling calculations
of Bellantone \& Hahn (1986) and Romanik (1988), multiconfiguration
intermediate-coupling calculations of Badnell, Pindzola, \& Griffin
(1990), and multiconfiguration fully-relativistic calculations of Chen
(1988) are all in good agreement with our experimentally inferred rate.
The $LS$-coupling calculations of Younger (1983) are $\sim 30\%$ larger
than the experimental rate. The recommended and commonly used DR rate
of Shull \& van Steenberg (1982) peaks at a value $\sim 45\%$ larger
than ours and has a steeper low $T_e$ behavior.
Pindzola et al.\ (1990) calculated resonance strengths for O VIII using
intermediate-coupling, multiconfiguration Hartree-Fock (MCHF) and
intermediate-coupling, multiconfiguration Thomas-Fermi (MCTF)
techniques. Using their MCHF results for capture into levels $n\le6$
and their MCTF results for $n\ge7$, as well as the experimental
resonance energies, we calculate the corresponding DR rate using
Equation \ref{eq:rate2}. The resulting rate is in good agreement with
our derived rate. The single-configuration $LS$-coupling rate of
Bellantone \& Hahn (1986) is also in good agreement with the
experimental rate, though with decreasing $T_e$ their rate falls off
sooner than ours. The recommended and commonly used DR rate of Shull
\& van Steenberg (1982) is $\sim 31\%$ larger than our derived rate.
Not shown in Figure \ref{fig:OVIII}, for reasons of clarity, is the
rate of Zhdanov (1978) whose peak rate is $\sim 7$ times larger than
the experimental rate.
Nahar \& Pradhan (1997) and Nahar (1999) present unified RR+DR rates.
To compare with their results we add the RR rates of Verner \& Ferland
(1996) to the experimentally derived DR rates. Thus we treat RR and DR
as independent processes and do not allow for the possibility of
interference between the two recombination channels. The validity of
this approach is supported by recent theoretical work (for DR on systems
ranging in complexity from C II and Mg II to U XCII) which has shown the
effect of interferences to be small (\cite{Pind92a}).
Figures~\ref{fig:CVNah} and \ref{fig:OVIIINah} show for C V and O VIII,
respectively, the unified RR+DR rate of Nahar \& Pradhan and the total
RR+DR rate using the derived DR results and the RR results of Verner \&
Ferland. For C V, at peak value the undamped, unified rate of Nahar \&
Pradhan is $\sim 43\%$ larger than our resulting total RR+DR rate.
This is consistent with the estimate by Pradhan \& Zhang (1997) that
allowance for radiation damping would reduce the undamped rate by $\sim
20-30\%$. For O VIII, at peak value the damped, unified rate of Nahar
is $\sim 20\%$ larger than our resulting total RR+DR rate.
DR rate coefficients are sometimes calculated using the Burgess (1965)
general formula (GF) or versions of the GF as modified by Merts et
al.\ (1976) and by Burgess \& Tworkowski (1976). We have calculated
the Burgess and Merts et al.\ rates using the formulae as given by
Cowan (1981). We use oscillator strengths and excitation energies from
Wiese, Smith, \& Glennon (1966). Shown in Figure \ref{fig:CVBurg} for
C~V is the Merts et al.\ rate which is in fair agreement with our
derived rate, though with decreasing $T_e$ it goes to zero sooner than
the experimental rate. Also shown in Figure \ref{fig:CVBurg} is the
Burgess GF rate which is a factor of $\sim 2.2$ times larger than the
experimental rate. The Burgess \& Tworkowski (1976) modification of
the GF yields a rate $\sim 50\%$ larger than our derived rate. This is
surprisingly good considering that their modification is meant for DR
forming heliumlike, not lithiumlike, ions. Figure \ref{fig:OVIIIBurg}
shows for O VIII the GF rate which is $\sim 38\%$ larger than the
experimental rate and, considering the expected accuracy of the Burgess
formula, in fair agreement. The Burgess \& Tworkowski rate is in good
agreement. Also shown is the Merts et al.\ formula rate which is a
factor of $\sim 2.2$ smaller than the experimental rate. Our C V and O
VIII results strongly suggest that for $\Delta N=1$ DR it is not
possible {\it a priori} to know which formula will yield a result
closer to the true rate.
For use in plasma modeling, we have fit the experimentally derived C~V
and O~VIII DR rates using the simple formula (Arnaud \& Raymond 1992)
\begin{equation}
\alpha(T_e)=T_e^{-3/2}\sum_i c_i e^{-E_i/k_BT_e}.
\label{eq:drratefit}
\end{equation}
Here $c_i$ and $E_i$ are, respectively, the strength and energy
parameters for the $i$th fitting component. Best fit values are listed
in Table \ref{tab:fitparameters}. For C~V, the fit is good to better
1\% for $2.6\times10^5 \le T_e \le 10^8$ K. Below $2.6\times10^5$ K,
with decreasing $T_e$ the fit goes to zero faster than the derived
rate. But this error is insignificant for plasma modeling as the RR
rate is $\sim 3$ orders of magnitude larger than the DR rate at these
temperatures. For O~VIII, the fit is good to better 1\% for
$7.1\times10^5 \le T_e \le 10^8$ K. Below $7.1\times10^5$ K, with
decreasing $T_e$ the fit goes to zero faster than the derived rate.
This error is insignificant for plasma modeling as the RR rate is
$\sim 2$ orders of magnitude larger than the DR rate at these
temperatures.
\section{Summary}
\label{sec:Summary}
We have presented a simple technique for obtaining DR rate coefficients
from laboratory measurements of DR resonance strengths and energies.
With this technique, we have derived DR rates for C~V to C~IV and
O~VIII to O~VII using published resonance strengths and energies. Our
derived rates are in good agreement with multiconfiguration,
intermediate-coupling and multiconfiguration, fully-relativistic
calculations as well as with most $LS$-coupling calculations. Our
rates are not in agreement with the recommended DR rates commonly used
for astrophysical plasma modeling. We have used theoretical radiative
recombination (RR) rates in conjunction with our derived DR rates to
produce a total recombination rate for comparison with unified RR+DR
calculations in $LS$ coupling. Our results are not in agreement with
undamped, unified calculations for C V but are in reasonable agreement
with damped, unified calculations for O VIII. Also, neither the
Burgess general formula, the Merts et al.\ formula, nor the Burgess \&
Tworkowski formula consistently yield a rate which is in agreement with
our derived rates. This suggests that for $\Delta n=1$ DR it is not
possible to know {\it a priori} which formula will yield a rate closer
to the true DR rate. We have also presented simple fitting formula of
the experimentally derived DR rates for use in plasma modeling.
\acknowledgements
The author would like to thank N. R. Badnell, M. H. Chen, T. W.
Gorczyca, S. M. Kahn, D. A. Liedahl, S. N. Nahar, and A. Wolf for
stimulating conversation and for critically reading the manuscript.
This work was supported in part by NASA High Energy Astrophysics X-Ray
Astronomy Research and Analysis grant NAG5-5123.
\vfill
\clearpage
\eject
\begin{deluxetable}{cccccc}
\footnotesize
\tablecaption{Fit parameters for the experimentally
derived C~V to C~IV and O~VIII to O~VII DR rate coefficient.
The units for $c_i$ are cm$^3$~s$^{-1}$~K$^{1.5}$ and for
$E_i$ are eV.
\label{tab:fitparameters}}
\tablewidth{0pt}
\tablehead{
\colhead{ } & \multicolumn{2}{c}{C V} & \colhead{ } &
\multicolumn{2}{c}{O VIII} \\
\cline{2-3} \cline{5-6} \\
\colhead{$i$} & \colhead{$c_i$} & \colhead{$E_i$} & & \colhead{$c_i$}
& \colhead{$E_i$}
}
\startdata
1 & 3.02e-3 & 246 & & 8.56e-3 & 497 \nl
2 & 1.75e-2 & 296 & & 5.43e-2 & 632 \nl
\enddata
\end{deluxetable}
\vfill
\clearpage
\eject
|
\section{Introduction}
Complete or, at least, statistically well defined samples of optical
identifications of faint X-ray sources, are important for a number
of scientific goals. For example, combined optical and X-ray data allow to
obtain information about the luminosity functions of various types of
X-ray sources as well as their evolution with redshift. In turn, these
informations can be used to further constrain models for the production of the
X-ray background, discovered more than thirty years ago (Giacconi et al.
1962). There is a general consensus that the majority of the optically
identified X-ray sources, at least for fluxes
$ S_{0.5 - 2 keV} \ge 5 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ , are active galactic nuclei
(AGNs), i.e. quasars and Seyfert galaxies with broad emission lines
(Shanks et al. 1991, Georgantopoulos et al. 1996, McHardy et al. 1998,
Hasinger et al. 1998 (Paper I) and Schmidt et al. 1998 (Paper II)).
In the first two papers of this series we have presented the
X-ray and optical data
for a complete catalogue of 50 X-ray sources with PSPC fluxes (0.5-2 keV)
above $ 5.5 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ in the Lockman Field. Optical and X-ray
results at a much fainter flux limit in this field, which has been observed
with the deepest ROSAT PSPC and HRI observations, will be presented
elsewhere.
In this paper we present and discuss the X-ray data (Section 2)
and the optical data (Section 3) for 50 sources detected with the PSPC at a
flux limit $ S_x \ge 3.7 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ in the Marano Field.
A discussion of the main results (percentages of identifications with
different classes of objects, hardness ratio as a function of X-ray flux,
comparison between X-ray and optically selected AGNs) is given in Section 4.
Throughout the paper we use $H_0$ = 50 km s$^{-1}$ Mpc$^{-1}$ and $q_0$ = 0.5.
\section{The X-ray Data}
\subsection{The sample}
The ROSAT--PSPC pointed observations of the Marano field, centered at
\hbox{RA = $ 03^h 15^m 09^s $}, \hbox{DEC = -55$ ^\circ $ 13$^{\prime}\,$ 57$^{\prime\prime}\,$}
(epoch 2000.0), have been carried out in the time interval
December~1992 -- July~1993, for a total of 56 ksec of
observing time. The observations were performed in the ``wobble mode''.
The X-ray analysis (e.g. source detection, position and flux determination
of the detected sources, determination of sensitivity as a function of distance
from the center, etc.) has been done applying a maximum likelihood
method. For details about the various steps in this analysis we refer the
reader to Hasinger et al. (1993 and 1998). Although detection of the
sources has been run in various ROSAT bands, we report here results
only for the hard band (0.5--2 keV), which has been shown (Hasinger
et al. 1993) to be the most efficient for point source detection.
\begin{table*}
\caption[]{ The X-ray Data}
\begin{tabular}{ccccrrccr}
\hline\noalign{\smallskip}
{\#} & RA (2000) & DEC (2000) & Err & Off-axis & ML & Net Counts & S$_x$ & HR~~~~~~\\
& & &[arcsec]&[arcmin]& & &[$\times 10^{-14}$ cgs]& \\
\noalign{\smallskip}
\hline\noalign{\smallskip}
~~X013--01 & 3 13 47.1 & $-$55 11 48 & 1.6 & 12.1~~ & 448.2 & 210.1 & 5.00 $\pm$ 0.42 & $-$0.03 $\pm$ 0.06 \\
~~X012--02 & 3 13 29.2 & $-$55 10 20 & 2.7 & 14.8~~ & 203.3 & 155.7 & 3.82 $\pm$ 0.37 & 0.41 $\pm$ 0.12 \\
~~X046--03 & 3 14 32.4 & $-$55 14 43 & 1.6 & 5.5~~ & 303.1 & 130.1 & 2.99 $\pm$ 0.32 & 0.20 $\pm$ 0.10 \\
~~X036--04 & 3 16 50.3 & $-$55 11 10 & 2.4 & 14.5~~ & 175.4 & 118.9 & 2.91 $\pm$ 0.30 & $-$0.10 $\pm$ 0.08 \\
~~X021--05 & 3 14 56.4 & $-$55 20 08 & 1.8 & 6.7~~ & 190.4 & 102.3 & 2.36 $\pm$ 0.25 & 0.88 $\pm$ 0.20 \\
~~X025--06 & 3 15 49.5 & $-$55 18 11 & 1.8 & 7.1~~ & 184.7 & 99.8 & 2.31 $\pm$ 0.30 & 0.28 $\pm$ 0.13 \\
~~X027--07 & 3 16 05.5 & $-$55 15 44 & 2.2 & 8.1~~ & 124.6 & 71.8 & 1.66 $\pm$ 0.28 & $-$0.21 $\pm$ 0.10 \\
~~X041--08 & 3 15 28.7 & $-$55 10 32 & 2.6 & 4.1~~ & 99.7 & 62.0 & 1.42 $\pm$ 0.25 & 0.58 $\pm$ 0.15 \\
~~X240--09 & 3 16 38.2 & $-$55 06 41 & 3.9 & 14.4~~ & 56.1 & 50.9 & 1.25 $\pm$ 0.24 & 0.45 $\pm$ 0.25 \\
~~X033--10 & 3 15 58.3 & $-$55 26 40 & 4.2 & 14.6~~ & 46.7 & 50.9 & 1.25 $\pm$ 0.23 & $-$0.48 $\pm$ 0.11 \\
~~X042--11 & 3 15 40.4 & $-$55 12 25 & 2.3 & 4.5~~ & 77.1 & 52.4 & 1.20 $\pm$ 0.18 & 0.32 $\pm$ 0.32 \\
~~X043--12 & 3 15 09.8 & $-$55 13 18 & 2.3 & 0.5~~ & 88.2 & 48.5 & 1.10 $\pm$ 0.25 & 0.37 $\pm$ 0.32 \\
~~X023--13 & 3 15 25.2 & $-$55 18 28 & 2.7 & 5.2~~ & 59.1 & 43.5 & 1.00 $\pm$ 0.17 & $-$0.27 $\pm$ 0.19 \\
~~X108--14 & 3 15 37.9 & $-$55 01 42 & 4.4 & 12.7~~ & 30.8 & 41.0 & 0.98 $\pm$ 0.20 & 0.63 $\pm$ 0.62 \\
~~X030--15 & 3 16 26.1 & $-$55 23 00 & 4.3 & 14.2~~ & 28.3 & 39.6 & 0.97 $\pm$ 0.18 & 0.03 $\pm$ 0.20 \\
~~X304--16 & 3 15 11.4 & $-$55 09 30 & 3.1 & 4.3~~ & 56.5 & 40.8 & 0.94 $\pm$ 0.18 & 0.08 $\pm$ 0.27 \\
~~X019--17 & 3 14 21.8 & $-$55 23 55 & 3.6 & 12.3~~ & 43.0 & 38.8 & 0.93 $\pm$ 0.27 & 0.24 $\pm$ 0.31 \\
~~X029--18 & 3 16 29.9 & $-$55 19 06 & 3.8 & 12.5~~ & 30.1 & 37.0 & 0.89 $\pm$ 0.17 & 0.14 $\pm$ 0.30 \\
~~X039--19 & 3 15 52.9 & $-$55 08 20 & 3.2 & 8.1~~ & 47.5 & 38.0 & 0.88 $\pm$ 0.18 & 1.00~~~~~~~~~~ \\
~~X001--20 & 3 15 20.7 & $-$55 02 33 & 3.6 & 11.3~~ & 46.4 & 37.4 & 0.88 $\pm$ 0.26 & $-$0.07 $\pm$ 0.24 \\
~~X049--21 & 3 15 06.0 & $-$55 09 42 & 3.0 & 4.1~~ & 49.4 & 35.9 & 0.82 $\pm$ 0.21 & $-$0.19 $\pm$ 0.20 \\
~~X211--22 & 3 13 45.5 & $-$55 19 25 & 4.4 & 13.4~~ & 21.5 & 33.8 & 0.81 $\pm$ 0.19 & $-$0.08 $\pm$ 0.32 \\
~~X404--23 & 3 16 48.9 & $-$55 12 40 & 9.3 & 14.1~~ & 10.2 & 29.7 & 0.73 $\pm$ 0.21 & 1.00~~~~~~~~~~ \\
~~X031--24 & 3 15 48.7 & $-$55 22 50 & 4.4 & 10.6~~ & 17.4 & 30.2 & 0.71 $\pm$ 0.18 & $-$0.21 $\pm$ 0.29 \\
~~X050--25 & 3 15 07.6 & $-$55 04 58 & 3.8 & 8.8~~ & 10.0 & 30.1 & 0.70 $\pm$ 0.16 & $-$0.26 $\pm$ 0.23 \\
~~X235--26 & 3 16 31.7 & $-$55 12 27 & 5.2 & 11.7~~ & 14.3 & 26.7 & 0.63 $\pm$ 0.18 & 0.21 $\pm$ 0.45 \\
~~X045--27 & 3 15 10.7 & $-$55 15 23 & 3.5 & 1.6~~ & 30.8 & 26.9 & 0.61 $\pm$ 0.15 & 0.24 $\pm$ 0.33 \\
~~X409--28 & 3 14 26.3 & $-$55 17 41 & 5.2 & 7.4~~ & 22.0 & 26.2 & 0.61 $\pm$ 0.14 & 0.84 $\pm$ 0.85 \\
~~X301--29 & 3 14 36.3 & $-$55 14 04 & 4.1 & 4.9~~ & 20.2 & 26.0 & 0.60 $\pm$ 0.15 & $-$0.21 $\pm$ 0.27 \\
~~X408--30 & 3 14 50.3 & $-$55 19 39 & 4.1 & 6.6~~ & 17.1 & 25.6 & 0.59 $\pm$ 0.16 & 0.65 $\pm$ 0.87 \\
~~X207--31 & 3 13 50.3 & $-$55 13 00 & 4.9 & 11.5~~ & 16.1 & 24.7 & 0.59 $\pm$ 0.15 & $-$0.12 $\pm$ 0.42 \\
~~X040--32 & 3 15 43.1 & $-$55 07 46 & 3.5 & 7.6~~ & 27.5 & 24.9 & 0.58 $\pm$ 0.14 & 0.34 $\pm$ 0.41 \\
~~X028--33 & 3 16 21.8 & $-$55 18 00 & 4.8 & 11.0~~ & 16.5 & 23.8 & 0.56 $\pm$ 0.14 & 0.24 $\pm$ 0.45 \\
~~X250--34 & 3 15 23.3 & $-$55 04 03 & 4.4 & 9.9~~ & 17.4 & 24.0 & 0.56 $\pm$ 0.14 & 0.65 $\pm$ 0.31 \\
~~X011--35 & 3 13 39.9 & $-$55 07 21 & 6.1 & 14.4~~ & 11.2 & 22.7 & 0.56 $\pm$ 0.15 & 0.04 $\pm$ 0.37 \\
~~X032--36 & 3 15 38.7 & $-$55 22 33 & 5.4 & 9.7~~ & 13.4 & 23.2 & 0.54 $\pm$ 0.14 & $-$0.01 $\pm$ 0.43 \\
~~X251--37 & 3 15 31.0 & $-$55 04 43 & 6.0 & 9.5~~ & 16.8 & 22.5 & 0.52 $\pm$ 0.14 & 0.07 $\pm$ 0.67 \\
~~X024--38 & 3 15 34.7 & $-$55 19 27 & 5.4 & 6.7~~ & 12.4 & 22.7 & 0.52 $\pm$ 0.16 & $-$0.11 $\pm$ 0.23 \\
~~X015--39 & 3 13 51.6 & $-$55 18 33 & 6.9 & 12.2~~ & 11.8 & 21.4 & 0.51 $\pm$ 0.14 & $-$0.43 $\pm$ 0.22 \\
~~X236--40 & 3 16 24.0 & $-$55 11 44 & 4.3 & 10.7~~ & 15.9 & 21.3 & 0.50 $\pm$ 0.13 & 0.02 $\pm$ 0.32 \\
~~X234--41 & 3 16 23.7 & $-$55 15 17 & 5.3 & 10.6~~ & 10.7 & 20.7 & 0.48 $\pm$ 0.15 & $-$0.15 $\pm$ 0.41 \\
~~X306--42 & 3 15 50.1 & $-$55 09 15 & 4.5 & 7.2~~ & 17.0 & 20.1 & 0.47 $\pm$ 0.14 & $-$0.22 $\pm$ 0.38 \\
~~X051--43 & 3 15 01.6 & $-$55 03 40 & 6.6 & 10.2~~ & 11.1 & 19.9 & 0.46 $\pm$ 0.14 & 0.93 $\pm$ 1.00 \\
~~X407--44 & 3 14 12.4 & $-$55 25 52 & 7.3 & 14.7~~ & 9.8 & 18.2 & 0.45 $\pm$ 0.14 & $-$0.33 $\pm$ 0.25 \\
~~X233--45 & 3 16 18.9 & $-$55 14 29 & 7.2 & 9.8~~ & 9.8 & 19.3 & 0.45 $\pm$ 0.14 & 0.26 $\pm$ 0.41 \\
~~X109--46 & 3 16 08.1 & $-$55 17 25 & 4.8 & 9.0~~ & 11.9 & 18.3 & 0.43 $\pm$ 0.12 & 0.05 $\pm$ 0.30 \\
~~X213--47 & 3 14 12.1 & $-$55 18 24 & 5.4 & 9.5~~ & 11.7 & 17.9 & 0.42 $\pm$ 0.14 & 1.00~~~~~~~~~~ \\
~~X022--48 & 3 15 03.4 & $-$55 19 06 & 4.4 & 5.4~~ & 12.3 & 18.3 & 0.42 $\pm$ 0.12 & 1.00~~~~~~~~~~ \\
~~X215--49 & 3 14 29.6 & $-$55 16 44 & 5.3 & 6.5~~ & 12.3 & 17.7 & 0.41 $\pm$ 0.12 & 1.00~~~~~~~~~~ \\
~~X264--50 & 3 14 49.1 & $-$55 22 24 & 5.2 & 9.2~~ & 12.0 & 15.8 & 0.37 $\pm$ 0.14 & 1.00~~~~~~~~~~ \\
\noalign{\smallskip}\hline
\end{tabular}
\end{table*}
Table 1 contains the X-ray data for the complete sample of sources
in a circular area with radius of 15$^{\prime}\,$. Within this area we detected 50
sources with a maximum likelihood value ML $\ge$ 9.8. With this
adopted threshold in ML we expect less than one spurious source
over the entire area.
The first column gives an identification code for the X-ray sources: the
first number identifies the sources in our working lists, while the second
one (from 1 to 50) represents the source rank in order of
decreasing X-ray flux. Columns 2 to 5 give
the $\alpha$ (2000) and $\delta$ (2000) positions,
with the one sigma error on each coordinate (in arcsec) derived from the
maximum likelihood fitting, and the off--axis
distance from the center of the X-ray field (in arcmin). The next three
columns give the maximum likelihood value in the hard band, the net counts and
the corresponding 0.5 -- 2.0 keV flux and error.
The flux has been computed assuming the same intrinsic spectrum for all the
sources, i.e. a power law spectrum with a slope $\alpha$ = 1.0, with
no intrinsic absorption and a galactic absorption corresponding to
$N_H = 2.5 \times 10^{20}$. These assumed parameters correspond to the
conversion factor 1 ct/s = 2.26 $\times 10^{-12} $ $erg ~cm^{-2} ~s^{-1}$ at the center of
the field. The last column gives the hardness ratio with its statistical error.
Since most of the detected X-ray sources are too faint to obtain a full
resolution spectrum, we have characterized their spectra using the hardness
ratio technique. The hardness ratio is defined as HR=(H-S)/(H+S),
where S~and~H~are the net counts in the PSPC energy channels 11--41 and
52--201, respectively, corresponding approximately to the energy ranges
0.1--0.4 and 0.5--2.0~keV. The source counts in
each band have been obtained summing all the counts from a circle centered at
the source position. The radius of the extraction circle was 60$^{\prime\prime}\,$. However,
when the extraction circles of two sources overlapped, this radius was reduced
to 30$^{\prime\prime}\,$ in order to avoid contamination from the nearby source.
The resulting net counts
for the instrumental radial vignetting and for the energy dependence
of the ROSAT/PSPC Point Spread Function (Hasinger et al. 1994). The
background counts were estimated in a circular area near the source position,
after excluding the contribution from near--by sources.
Five sources, corresponding to 10\% of the total sample, have a formal
hardness ratio equal to or greater 1.00 (i.e. zero or negative net counts
detected in the soft band). Four of them are at the faint flux limit of the
entire sample.
To eliminate a possible systematic error of a few arcsec in the X-ray
positions, we have cross--correlated the positions of the 50 X-ray sources
within 15$^{\prime}\,$ \ from the center with those of 29 previously known optically
selected AGNs with m$_B \le$ 22.5 in the same area (Zitelli et al. 1992;
Mignoli et al. in preparation). We have found 19 positional coincidences,
with distances smaller than 12$^{\prime\prime}\,$ between X-ray and optical positions, while
less than 0.3 random coincidences are expected on a statistical basis.
The average offset between X-ray and optical positions is about 4$^{\prime\prime}\,$ in right
ascension and less than 1$^{\prime\prime}\,$ in declination, with no additional trend as a
function of position in the field. The one sigma uncertainty for the offset
in each coordinate is of the order of 0.9$^{\prime\prime}\,$.
We have therefore applied this average offset to all the original X-ray
positions and corrected the ML positional errors by adding in quadrature
the uncertainty on the offset. Both the positions and the positional errors
given in Table 1 are the ``corrected'' values.
\begin{figure}
\vspace{0.5cm}
\vspace{8.7cm}
\caption[]{Gray scale representation of the hard ROSAT image. The image
has been slightly smoothed with a gaussian filter with $\sigma$ = 10$^{\prime\prime}\,$.
All the 50 sources detected within a radius of 15$^{\prime}\,$~are labelled.}
\end{figure}
Figure 1 shows a gray scale representation of the hard ROSAT image. The image
has been slightly smoothed with a gaussian with $\sigma$ = 10$^{\prime\prime}\,$. All the
sources detected within a radius of 15$^{\prime}\,$ \ are labelled. The figure shows the
capability of the maximum likelihood algorithm in retrieving pairs of sources
relatively close to each other (see, for example, the two pairs 49--304 and
46--301, for which the distances between the X-ray positions are of the
order~of~50~arcsec). For distances between two X-ray centroids
smaller than $\sim$40$^{\prime\prime}\,$, however, our detection algorithm
is unable to separate the two X-ray sources and will find a single
source (Hasinger et al. 1993), with a position intermediate between the two
true positions. In these cases the resulting X-ray position and its associated
error would not be reliable and this would lead to problems in the optical
identification process.
In order to estimate how many such cases we may have in our list of sources
we have simulated 1000 random
samples with the same number of detected sources and with the same radial
surface densities of sources as the real sample (i.e. higher density in the
inner region of the field and lower density in the outer region). For each
sample we have then computed the number of sources which have a nearby companion
at a distance smaller than the smallest observed distance in the real
sample (i.e. 48$^{\prime\prime}\,$). Figure 2 shows the normalized histogram of the
number of such sources in the 1000 random samples. The figure shows that in only
2\% of the cases there is no pair of sources with a distance smaller than
48~arcsec, while in the central 75\% of the cases (hatched area) the number of
such sources ranges from 4 to 10.
\begin{figure}[t]
\epsfysize=9cm
\epsfbox{rosat5_fig2.ps}
\caption[]{Histogram of the number of sources with a companion at less than
48$^{\prime\prime}\,$ in 1000 simulated samples. In 75\% of the cases the number of these
sources is between 4 and 10 (hatched area).}
\end{figure}
Since each pair is counted twice in this histogram, this means that we may
expect that in about 2--5 cases a
single source in our list may be produced by two different close--by sources, so
that its position may be significantly wrong and we may not be able to find any
reliable optical counterpart within the error box.
Note that this estimate is based on purely geometric considerations; moreover,
since in computing the expected number of close--by sources
we have adopted the observed surface density of sources, we are not
considering in this order of magnitude estimate the cases in which one or both
the sources in a close pair are below the detection limit.
Results from more detailed simulations, which take into account all
the possible effects of source confusion, are discussed in Hasinger et al.
(1998).
\begin{figure}
\epsfysize=9cm
\epsfbox{rosat5_fig3.ps}
\caption[]{Hard X-ray flux versus the off--axis angle for all
the X-ray sources listed in Table 1. The curve shows the
limiting flux as a function of off--axis angle which has been used to
estimate the ``corrected'' observed surface density of sources.
}
\end{figure}
Figure 3 shows the hard X-ray flux versus the off--axis angle for all
the X-ray sources listed in Table 1. Even if we have considered only sources
within 15$^{\prime}\,$ \ from the center, the limiting flux
is not constant over the adopted field of view. Its increase at distances
greater than about 10$^{\prime}\,$ \ is mainly due to the
increase of the width of the point spread function with the off-axis angle.
The curve drawn in the figure shows our estimated limiting flux as a function
of the off--axis angle for the complete sample.
When corrected for the different sensitivity over the field,
and using only the 48 sources with flux greater than the adopted limiting
flux, the estimated observed surface density at
$S_x \ge 4 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ is $272 \pm 40$ sources/sq.deg.
This value is consistent with the surface density
(248 sources/sq.deg.)
derived by Hasinger et al. (1993) from the composite observed log~N -- log~S
relationship using the deeper ROSAT data in the Lockman Hole together
with a number of shallower fields.
\section{The Identification of the X-ray Sources}
\subsection{The Radio Data}
We observed the Marano field at 1.4 and 2.4 GHz
with the Australia Telescope Compact Array (ATCA).
The radio observations were carried out on 1994 January 4, 5, 6 and 7.
Details about the radio observation and data reduction
are given in Gruppioni et al. (1997), where catalogs of 5$\sigma_{local}$
radio sources are given. The radio limits are about 0.2 mJy at both
frequencies. Cross--correlation of
the entire radio and X-ray catalogs has produced four positional coincidences
with a maximum difference of $\sim$10$^{\prime\prime}\,$ between radio
and X-ray positions. All these distances are between 1.1 and 2.4 times the
combined X-ray and radio positional error ($\epsilon$). Fifteen more
radio -- X-ray pairs have distances smaller
than 90$^{\prime\prime}\,$, but none has a distance smaller than 25$^{\prime\prime}\,$. For all
these pairs the distances between radio and X-ray positions are larger than
five times the combined positional error and can
therefore be considered random coincidences.
>From the number of these random pairs we estimate that the expected number of
similarly random pairs within 10$^{\prime\prime}\,$ is 0.20 $\pm$ 0.05.
If, instead of using a radius of 10$^{\prime\prime}\,$,
which can be considered an ``a posteriori'' choice, we use for each X-ray
source a radius corresponding to the 95\% error circle, the estimated
number of random pairs inside the 95\% area becomes 0.40. On this basis we
conclude that probably none and at most one of the observed radio -- X-ray
coincidences is not real.
The four radio -- X-ray pairs correspond to a percentage of radio
detection of 8$\pm$4\% for our X-ray sources at a radio limit
of $\sim$ 0.2 mJy.
Of the same order (10$\pm$5\%) is the percentage of X-ray detections for
our sample of radio sources. Both these percentages
are in good agreement with what has been found by De Ruiter et al.
(1996) from VLA observations in the region of the Lockman Hole.
Table 2 lists these four radio -- X-ray pairs, where the columns are
X-ray number and flux from Table 1, radio number and flux from the 20 cm
catalog in Gruppioni et al. 1997, difference in position ($\Delta$) in arcsec
and normalized to the combined X-ray and radio error ($\Delta$/$\varepsilon$).
Three of these four radio -- X-ray pairs have ben optically identified
(see Section 3.2). For all of them the offset between the radio and the
optical position of the suggested identification is less than 1.8$^{\prime\prime}\,$,
consistent with the combined radio and optical positional errors.
\begin{table}[t]
\caption[]{ The Radio -- X-ray Coincidences}
\begin{tabular}{rc|rc|cc}
\hline\noalign{\smallskip}
\multicolumn{2}{c|}{X-ray~~~~} & \multicolumn{2}{c|}{Radio~~} & \multicolumn{2}{c}{Distance} \\
\#~~~~~~& flux & \# & flux &$\Delta$&$\Delta$/$\varepsilon$\\
\multicolumn{2}{r|}{\hspace{1.2cm}[$\times 10^{-14}$ cgs]} &\hspace{0.8cm}&~~~[mJy]~~~& [arcsec] &\hspace{0.8cm}\\
\noalign{\smallskip}
\hline\noalign{\smallskip}
~~X013--01 & 5.00 & 15 & 158 & 3.8 & 2.0 \\
~~X021--05 & 2.36 & 38 & 1.25 & 2.3 & 1.1 \\
~~X409--28 & 0.61 & ~~30 & 6.32 & 8.2 & 1.5 \\
~~X408--30 & 0.59 & ~~35 & 0.41 & 10.3 & 2.4 \\
\noalign{\smallskip}\hline
\end{tabular}
\end{table}
\subsection{ The Optical Data}
In addition to the ESO 3.6m plates (U, J, F bands) which were already
available and have been used in the past to obtain a complete sample
of optically selected AGNs with $m_B \le 22.0$ (Zitelli et al. 1992),
in the years 1992--1994 we have obtained a set of U, B, V, R
CCD images at the ESO NTT. The V and R images cover $\sim$ 90\% of the circle
with 15$^{\prime}\,$ \ radius and contain all but one of the X-ray sources within this
radius; the U and B images, which have a smaller field of view,
cover a smaller fraction of the 15$^{\prime}\,$~field, but still contain 45 out of
50 X-ray sources. Details about the observations, data reduction and
optical catalogs obtained from these CCD observations will be given
elsewhere (Mignoli et al. in preparation; see also Mignoli 1997).
The CCD limiting magnitudes vary
from field to field, but typically are of the order of 23.5 in U, 25.0
in B, 24.0 in V and R. The typical surface density of objects
at these limits is $\sim$ 60,000 per square degree.
Since the total area covered by the 2$\sigma$~(3$\sigma$) error circles of the
50 X-ray sources corresponds to 5.4~(10.4) sq.arcmin., this implies a total
expected number of $\sim$~90~(175) catalogued objects inside the X-ray error
boxes.
In Table 3 we report the optical data for all the objects in our optical
catalogs within a 3$\sigma$ error box, plus a few interesting objects
at slightly larger distance.
(We recall that, under the assumption that the distribution function of the
positional errors follows a circular normal distribution, the 1$\sigma$,
2$\sigma$ and 3$\sigma$ error boxes correspond to radii equal to 1.51, 2.45
and 3.4 times the error in each coordinate, respectively.)
The first ten columns in the
Table give the X-ray number, the distance in right ascension and declination
between the X-ray source and the optical objects (arcsec), the total distance
both in arcsec and normalized to the X-ray positional error ($\varepsilon_{\rm{x}}$)
as defined in Section 2.1, \hbox{optical B} and R magnitudes and \hbox{U-B},
\hbox{B-V}, \hbox{V-R} colours and a morphological classification
(p for point-like source, e for extended, p/e or e/p for sources classified
differently in the blue and red bands). The morphological classification
is reliable only for objects which are more than $\sim$ 1.5 magnitudes
brighter than the limiting magnitude (Flynn, Gould and Bahcall 1996).
An asterisk in the magnitude columns
means that the magnitudes have been measured from the plates and then
converted to the Johnson system (see Gruppioni, Mignoli and Zamorani 1999).
The next two columns give the likelihood
ratio (LR) computed from the B and R data. These LR values have been computed
following the procedure described by Sutherland and Saunders (1992), which,
differently from previous formulations, has been shown to be valid also
in the case of multiple candidates in the error boxes. For each optical
object its likelihood ratio is defined as:
\begin{equation}
LR = \frac{q(\rm{m,c})\; \em{f}(\rm{x,y})}{n(\rm{m,c})}
\end{equation}
\noindent
where $f$(x,y) is the probability distribution function of the positional
errors, $n$(m,c) is the surface density of ``background'' objects with
magnitude m and type c and $q$(m,c) is the probability distribution function
in magnitude and type of the optical counterparts. With this definition LR
is the ratio between the probability of finding the true optical counterpart
with the observed offset (x,y) from the X-ray position and the observed
magnitude and the probability of finding a similar chance background
object (see Eq. 1 and related discussion in Sutherland and Saunders 1992).
Assuming that the distribution function of the positional errors is
gaussian,
\begin{equation}
f(x,y) = \frac { e^{- \frac {(x^2 + y^2)}{2 \sigma^2} } } {2 \pi \sigma^2}
\end{equation}
\noindent
where x and y are the offsets in right ascension and declination between the
optical and X-ray sources and $\sigma$ is the positional error in each
coordinate (see column 4 in Table 1).
We considered two different types of optical objects, i.e.
point--like and extended. For each of them the observed
$n$(m) has been obtained from our
own CCD data, while $q$(m) has been estimated from the magnitude distribution
of the excess of objects in the 50 X-ray error boxes with respect to the
expected number of ``background'' objects.
The last two columns
give the redshift and the classification of the objects based on our
spectroscopic data. Objects which we consider to be the correct identification
are written in bold face in the last column. When the spectroscopic data
suggest identification with a group or cluster of galaxies, all the galaxies
at the redshift of the cluster are shown in bold face.
Figure 4 shows the finding charts for all the X-ray sources within 15$^{\prime}\,$,
ordered by decreasing X-ray flux, and, when available, the spectrum of the
most likely identification. For completeness we show here also the spectra
of the previously known, optically selected AGNs already published in
Zitelli et al. (1992). For all sources the optical images
(1$^{\prime}\,$$\times$1$^{\prime}\,$) are taken from R CCDs, except for sources X046--03,
X240--09 and X045--27, for which the images from the F plate are shown.
The two circles drawn on the figures correspond to the 68\% and 95\% error
boxes as defined above. The total number of objects in the 95\%
error boxes is $\sim$~140. Spectroscopy
for such a large number of faint objects was obviously not possible, and
therefore we had to decide a strategy for the follow--up spectroscopic
observations of the most promising candidates. Our adopted strategy was the
following:
i. First we have cross--correlated the positions of the 50 X-ray sources
with those of the 29 optically selected AGNs previously known in the same
area, finding 19 positional coincidences (see Section 2.1).
With one exception (X019--17), all these AGNs have at least one of the two LR
values higher than 1.6. We have considered them as
likely identifications (these objects are indicated as {\bf AGN$^{(1)}$}
in the last column of Table 3) and we decided not to
observe spectroscopically other objects in these error boxes, including the
X-ray source X019--17 in which the AGN has a relatively low likelihood ratio. We
note here that our LR values are based only on optical magnitudes and
morphology and do not take into account the spectroscopic information.
At the typical magnitudes of the optical counterparts identified with AGNs
(21 $\le m_B \le$ 23) the ratio between the number of objects classified
as point--like objects in our CCD data and that of broad--line AGNs
(Mignoli and Zamorani 1998) is in the range 10--15. Therefore, the
``a posteriori'' LR, when an AGN is found spectroscopically in any given
error box, would be about
10--15 times higher than that listed in Table 3.
ii. Since AGNs are well known to be the dominant optical counterpart of
faint X-ray sources, at least for $S_x \ge 5 \times 10^{-15} $ $erg ~cm^{-2} ~s^{-1}$ ,
we have then searched our optical catalogs for all the stellar
or slightly fuzzy objects within the remaining 31 X-ray error boxes.
We have then taken spectra for these objects helping us in making a priority
list in each error box with additional information from the colours, the
magnitude and the distance between the objects and the X--centroid (i.e.
the LR value). In addition to
the objects that on the basis of these data were considered to be the most
likely AGN candidates, we gave high priority also to objects which
coincide with a radio source (see Table 2) and to objects with colours
typical of M stars, which are the dominant spectral type of stars found in
faint X-ray surveys. M stars with the highest ratio between X-ray and
optical fluxes and with an X-ray flux equal to our limiting flux are
expected to have at most an optical magnitude of the order of $m_V \sim 19.75
\pm 0.25$ (see Maccacaro et al. 1988),
corresponding to $m_B \sim 21.25 \pm 0.25$. From our own
colour -- colour diagrams (see also Marano et al. 1988) we estimate that the
surface density of M stars with this limiting magnitude is $\sim$ 380 per square
degree, in good agreement (within 15\%) with the predictions
of the Bahcall and Soneira model for the structure of the Galaxy (see,
for example, Ratnatunga and Bahcall 1985).
Following this recipe, we found spectroscopically four stars (two M stars
with $m_b \sim 19.5$ (X030--15 and X028--33), and two very bright
F stars with $m_b \sim 10$ (X046--03 and X045--27)),
eight broad-line AGNs (X012--02, X042--11, X040--32, X032--36, X251--37,
X024--38, X015--39, X306--42) and two radio galaxies (X021--05 and X409--28),
one of which with broad MgII line. The blue magnitudes
of the AGNs and the two radio galaxies are in the range $21.7 - 23.7$.
We also took a spectrum for the relatively bright object ($m_B$ = 21.44), with
LR values greater than 5.5, which is right at the center of the error box of
source X051--43. Its spectrum does
not show any clear evidence for convincing emission or absorption features,
so that, even if we do consider it as the likely counterpart
of the X-ray source, we have not been able to spectroscopically classify it.
The absence of features in the spectrum suggests that it might be a BL Lac
object. If so, it would be a somewhat anomalous BL Lac object. In fact, being
not detected in the radio at a flux limit of about 0.2 mJy, it would be
classified as a radio quiet BL Lac, with a radio to optical spectral index
$\alpha_{ro} <$0.26. None of the X-ray selected BL Lac objects in the Einstein
Medium Sensitivity Survey (Stocke et al. 1991), and just a few in the much
larger samples of ROSAT selected BL Lacs (see, for example, Perlman et al. 1998)
have such a low value of $\alpha_{ro}$.
\begin{table*}
\caption[]{ The Optical Data}
\begin{tabular}{rrrrrrrrrcrrll}
\hline\hline
\\
ID. & $\Delta\alpha$ & $\Delta\delta$ & $\Delta$$^{\prime\prime}\,$/$\displaystyle{\Delta\over\varepsilon_{\rm{x}}}$ &
\multicolumn{1}{c}{~m$_B$} & \multicolumn{1}{c}{m$_R$} & U-B & B-V & V-R & &
\multicolumn{1}{c}{$LR_B$} &\multicolumn{1}{c}{$LR_R$} & \multicolumn{1}{c}{z} & object notes\\
&\multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} & \multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} \\
\hline
X013--01 & 1.2 & -0.0 & 1.2/0.8 & $>$25.00 & 23.17 & & & $>$ 1.13 & e & & 4.73 & & \\
& -4.0 & -1.7 & 4.3/2.8 & 19.96 & 19.61 & -0.55 & -0.09 & 0.44 & p & 1.25 & 1.63 & 1.663 & {\bf AGN$^{(1)}$}\\
\hline
X012--02 & -7.6 & 0.2 & 7.6/2.8 & 21.94 & 20.58 & -0.70 & 0.63 & 0.73 & p & 1.16 & 0.65 & 1.378 & {\bf AGN}\\
\hline
X046--03 & -1.4 & 0.9 & 1.6/1.0 & $\sim$9.3 & & & $\sim$0.5& & & & & 0.00 & {\bf F6IV~star}\\
\hline
X036--04 & -0.4 & -0.3 & 0.5/0.2 & 18.04* & 17.07 & -0.78 & 0.55 & 0.42 & p & 17.86 & 28.60 & 2.531 & {\bf AGN$^{(1)}$}\\
& 0.8 & 5.1 & 5.1/2.1 & $>$23.50* & 23.44 & & $>$-0.67 & 0.73 & e & & 0.29 & & \\
\hline
X021--05 & -2.7 & 0.0 & 2.7/1.5 & 21.79 & 19.28 & 0.66 & 1.48 & 1.03 & e & 6.83 & 3.58 & 0.387 & {\bf RadioGal}\\
\hline
X025--06 & 1.0 & -1.6 & 1.9/1.0 & 21.43 & 20.97 & -0.52 & 0.36 & 0.10 & p & 71.44 & 45.19 & 0.808 & {\bf AGN$^{(1)}$}\\
\hline
X027--07 & 3.0 & 3.3 & 4.4/2.0 & 21.17 & 20.66 & -0.55 & 0.19 & 0.32 & p & 11.89 & 7.52 & 0.636 & {\bf AGN$^{(1)}$}\\
& -3.4 & -5.6 & 6.6/2.9 & 23.34 & 22.87 & -0.86 & 0.29 & 0.18 & e & 0.12 & 0.09 & & \\
\hline
X041--08 & 1.8 & 3.6 & 4.0/1.5 & 21.62 & 21.06 & -0.67 & 0.35 & 0.21 & p & 20.74 & 10.78 & 2.161 & {\bf AGN$^{(1)}$}\\
& -3.5 & -5.7 & 6.7/2.6 & 24.04 & 23.46 & -0.66 & 0.41 & 0.17 & e & 0.08 & 0.08 & & \\
& 0.3 & -7.5 & 7.5/2.9 & 24.20 & $>$23.50 &$>$-0.70 & 0.48 & $<$ 0.22 & e & 0.04 & & & \\
& -8.5 & 2.3 & 8.8/3.4 & 24.15 & $>$23.50 &$>$-0.65 & -0.21 & $<$ 0.86 & e & 0.01 & & & \\
\hline
X240--09 & -0.3 & 4.0 & 4.0/1.0 & 21.81* & 20.87* & -0.71 & 0.52 & 0.42 & p & 17.75 & 9.99 & 0.854 & {\bf AGN$^{(1)}$}\\
& 12.4 & -1.9 & 12.5/3.2 & 22.54* & 21.91* & -0.74 & 0.34 & 0.29 & e & 0.01 & 0.02 & & \\
\hline
X033--10 & 7.2 & 1.4 & 7.3/1.7 & 21.90* & 20.91 & -0.90 & 0.65 & 0.34 & p & 5.70 & 3.21 & 0.983 & {\bf AGN$^{(1)}$}\\
& 1.0 & 12.6 & 12.6/3.0 & $>$23.50* & 21.86 & & $>$ 0.72 & 0.92 & p & & 0.09 & & \\
\hline
X042--11 & -0.8 & 2.4 & 2.5/1.1 & 22.99 & 22.04 & -0.33 & 0.58 & 0.37 &p/e& 19.50 & 5.43 & 1.062 & {\bf AGN}\\
& -6.8 & -3.8 & 7.8/3.3 & $-out-$ & 22.10 & & & 1.00 & e & & 0.04 & & \\
\hline
X043--12 & 1.6 & 3.1 & 3.5/1.5 & 23.80 & 22.61 & -0.32 & 0.17 & 1.02 &e/p& 1.66 & 0.87 & 2.80: & {\bf AGN2}~(?)\\
& 5.3 & -5.8 & 7.9/3.4 & 24.15 & 22.93 & -0.47 & 0.95 & 0.27 & e & 0.01 & 0.02 & & \\
\hline
X023--13 & 0.3 & -0.7 & 0.8/0.3 & 22.24 & 21.99 & -0.45 & -0.10 & 0.35 & p & 36.47 & 19.86 & 1.573 & {\bf AGN$^{(1)}$}\\
& 4.5 & -8.1 & 9.3/3.4 & 23.94 & 21.32 &$>$-0.44 & 1.42 & 1.20 & e & 0.01 & 0.02 & & \\
\hline
X108--14 & 3.6 & 0.6 & 3.6/0.8 & 22.42 & 21.63 & -1.55 & 0.23 & 0.56 & p & 10.04 & 5.47 & 1.374 & {\bf AGN$^{(1)}$}\\
& -4.8 & 3.1 & 5.7/1.3 & 21.30 & 20.36 & 0.05 & 0.52 & 0.42 & e & 2.90 & 1.71 & & \\
\hline
X030--15 & 0.5 & 3.5 & 3.6/0.8 & 24.07 & $>$23.50 & -0.35 & $<$-0.23 & & p & 0.60 & & & \\
& 0.5 & 7.6 & 7.6/1.8 & 19.28 & 16.86 & 1.20 & 1.59 & 0.83 & p & 1.95 & 2.59 & 0.00 & {\bf M~star}\\
& -8.1 & 10.1 & 13.0/3.0 & 23.29 & 22.20 & -0.11 & 0.87 & 0.22 & e & 0.02 & 0.03 & & \\
& -13.0 & -5.7 & 14.2/3.3 & 22.99 & 22.24 & -0.86 & 0.33 & 0.42 & e & 0.01 & 0.01 & & \\
\hline
X304--16 & -1.0 & 1.6 & 1.9/0.6 & 21.53 & 20.73 & -0.74 & 0.39 & 0.41 & p & 39.05 & 21.97 & 1.192 & {\bf AGN$^{(1)}$}\\
& -2.6 & 5.7 & 6.3/2.0 & 24.04 & 23.05 & -0.84 & 0.40 & 0.59 & e & 0.21 & 0.21 & & \\
& 9.8 & -2.2 & 10.1/3.2 & 23.59 & 22.27 & -0.70 & 0.46 & 0.86 & e & 0.02 & 0.03 & & \\
\hline
X019--17 & 0.9 & -2.8 & 2.9/0.8 & 24.34 & 23.47 & -0.90 & 0.78 & 0.09 & e & 0.85 & 0.85 & & \\
& -2.8 & 8.7 & 9.2/2.5 & 23.45 & 21.20 &$>$ 0.05 & 1.35 & 0.90 & p & 0.30 & 0.74 & & \\
& -5.5 & -10.4 & 11.7/3.2 & 22.40* & 21.69 & -0.70 & 0.49 & 0.22 &p/e& 0.12 & 0.03 & 0.614 & {\bf AGN$^{(1)}$}\\
\hline
X029--18 & 3.5 & 4.8 & 6.0/1.6 & 23.67 & 21.94 & -0.10 & 1.11 & 0.62 & e & 0.56 & 1.20 & & \\
& 0.9 & -6.2 & 6.2/1.6 & 21.66 & 21.15 & -1.01 & 0.23 & 0.28 & p & 8.42 & 4.37 & 1.254 & {\bf AGN$^{(1)}$}\\
\hline
X039--19 & -10.5 & 2.2 & 10.7/3.3 & 23.63 & 22.41 & -0.20 & 0.99 & 0.23 & e & 0.01 & 0.02 & ?.??? & low S/N\hfill(A)\\
\hline
X001--20 & -5.4 & -1.8 & 5.7/1.6 & 20.39 & 19.71 & -0.83 & 0.26 & 0.42 & p & 2.31 & 3.92 & 1.353 & {\bf AGN$^{(1)}$}\\
& -8.4 & 7.0 & 10.9/3.0 & 22.72 & 22.73 & -1.00 & -0.09 & 0.08 & p & 0.16 & 0.01 & & \\
\hline
X049--21 & -3.2 & -1.7 & 3.6/1.2 & 23.86 & 22.36 & -0.61 & 0.68 & 0.82 & e & 1.50 & 2.82 & ?.??? & low S/N\\
\hline
X211--22 & -0.2 & -6.9 & 6.9/1.6 & 24.27 & 23.31 & -0.51 & $<$-0.03 & $>$ 0.99 & e & 0.23 & 0.23 & & \\
& 9.3 & 7.5 & 12.0/2.8 & 20.97 & 19.35 & 0.11 & 1.06 & 0.56 & e & 0.07 & 0.04 & 0.180 & early gal\\
& -12.5 & 0.5 & 12.5/2.9 & 22.20 & 20.26 & -0.45 & 0.86 & 1.08 &p/e& 0.23 & 0.07 & 0.281 & {\bf AGN (Sy 1)}\\
& -12.3 & -4.0 & 13.0/3.0 & 24.37 & 22.99 & -0.16 & 0.43 & 0.95 & e & 0.01 & 0.02 & & \\
& -13.5 & 0.1 & 13.5/3.1 & 22.12 & 21.36 & -0.04 & 0.41 & 0.35 & p & 0.12 & 0.10 & 0.00 & B~star\\
& 7.0 & 12.9 & 14.7/3.4 & 24.60 & 23.49 &$>$-1.10 & 0.64 & 0.47 & e & 0.00 & 0.00 & & \\
\hline
\end{tabular}
\end{table*}
\begin{table*}
\begin{tabular}{rrrrrrrrrcrrll}
\hline
\\
ID. & $\Delta\alpha$ & $\Delta\delta$ & $\Delta$$^{\prime\prime}\,$/$\displaystyle{\Delta\over\varepsilon_{\rm{x}}}$ &
\multicolumn{1}{c}{~m$_B$} & \multicolumn{1}{c}{m$_R$} & U-B & B-V & V-R & &
\multicolumn{1}{c}{$LR_B$} &\multicolumn{1}{c}{$LR_R$} & \multicolumn{1}{c}{z} & object notes\\
&\multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} & \multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} \\
\hline
X404--23 & 11.8 & -1.4 & 11.9/1.3 & 24.03 & 22.74 &$>$-0.53 & 0.70 & 0.59 & e & 0.08 & 0.17 & & \\
& 6.0 & 10.9 & 12.4/1.3 & $>$25.00 & 23.25 & & & $>$ 1.05 & e & & 0.07 & & \\
& -9.8 & -10.1 & 14.1/1.5 & 23.67 & 22.54 & -0.38 & 0.87 & 0.26 & e & 0.10 & 0.12 & & \\
& -16.1 & -1.6 & 16.2/1.7 & 24.25 & 22.67 &$>$-0.75 & 0.76 & 0.82 & e & 0.04 & 0.09 & & \\
& 6.2 & -16.5 & 17.6/1.9 & 24.47 & 23.41 &$>$-0.97 & 0.26 & 0.80 & e & 0.03 & 0.03 & & \\
& 8.9 & -16.5 & 18.8/2.0 & 24.89 & 23.18 &$>$-1.39 & 0.70 & 1.01 & e & 0.01 & 0.02 & & \\
& -12.9 & -15.5 & 20.2/2.2 & 23.40 & 22.34 & -0.77 & 0.57 & 0.49 & e & 0.05 & 0.06 & 0.204 & NELG\hfill(G)\\
& -18.0 & -9.3 & 20.3/2.2 & 24.23 & 23.60 & -0.17 & 0.16 & 0.47 & e & 0.02 & 0.01 & & \\
& -3.7 & 21.3 & 21.6/2.3 & 24.32 & 23.47 & -0.66 & 0.41 & 0.44 & e & 0.01 & 0.01 & & \\
& 11.9 & -19.3 & 22.6/2.4 & 23.55 & 21.01 &$>$-0.05 & 1.31 & 1.23 & p & 0.03 & 0.14 & 0.00 & M~star\hfill(J)\\
& 23.3 & 14.2 & 27.3/2.9 & 23.65 & 21.81 &$>$-0.15 & 0.85 & 0.99 & e & 0.00 & 0.01 & & \\
& 24.6 & -12.6 & 27.7/3.0 & 23.80 & 22.20 & -1.18 & 0.75 & 0.85 & e & 0.00 & 0.01 & & \\
& 24.1 & 13.7 & 27.7/3.0 & 23.82 & $>$23.50 &$>$-0.32 & -0.24 & $<$ 0.56 & e & 0.00 & & & \\
& 13.6 & -24.3 & 27.8/3.0 & 23.67 & 23.04 & -0.67 & -0.16 & 0.79 & e & 0.00 & 0.00 & & \\
& 27.8 & -5.3 & 28.3/3.0 & 22.51 & 21.70 & -0.36 & 0.52 & 0.29 & e & 0.00 & 0.01 & 0.157 & Sy2~(?) \hfill(N)\\
& -27.1 & 8.9 & 28.5/3.1 & 23.76 & 21.43 &$>$-0.26 & 1.28 & 1.05 & p & 0.01 & 0.02 & & \\
& -15.7 & 24.5 & 29.1/3.1 & $>$25.00 & 23.35 & & $>$ 0.77 & 0.88 & e & & 0.00 & & \\
& 8.3 & 28.2 & 29.4/3.2 & 24.45 & 21.94 &$>$-0.95 & 1.51 & 1.00 & p & 0.00 & 0.01 & & \\
\hline
X031--24 & 0.0 & 2.8 & 2.8/0.6 & 21.37* & 20.71 & -0.72 & 0.32 & 0.34 & p & 17.50 & 11.07 & 0.409 & {\bf AGN$^{(1)}$}\\
& -1.3 & -3.7 & 3.9/0.9 & $>$23.50* & 21.23 & & $>$ 1.19 & 1.08 & e & & 1.90 & & \\
& 5.8 & 1.0 & 5.9/1.4 & $>$23.50* & 23.18 & & $>$-0.77 & 1.09 & e & & 0.33 & & \\
& -0.9 & -7.8 & 7.8/1.8 & $>$23.50* & 21.71 & & $>$ 1.37 & 0.42 & e & & 0.64 & & \\
& 0.1 & -13.6 & 13.6/3.1 & $>$23.50* & 22.82 & & $>$ 0.21 & 0.47 & e & & 0.01 & & \\
\hline
X050--25 & -0.4 & 0.7 & 0.8/0.2 & 20.77 & 20.39 & -0.72 & 0.04 & 0.34 & p & 18.47 & 13.93 & 1.315 & {\bf AGN$^{(1)}$}\\
& 4.0 & -5.8 & 7.1/1.9 & 24.17 & 23.08 &$>$-0.67 & 0.31 & 0.78 & e & 0.19 & 0.19 & & \\
\hline
X235--26 & -0.9 & -1.7 & 1.9/0.4 & 21.40 & 20.96 & -0.00 & 0.26 & 0.18 & p & 13.94 & 8.82 & 2.536 & {\bf AGN$^{(1)}$}\\
& 2.5 & -1.0 & 2.7/0.5 & 24.92 & 23.42 &$>$-1.42 & $<$ 0.62 & $>$ 0.88 & e & 0.23 & 0.50 & & \\
& -0.5 & -5.2 & 5.2/1.0 & 22.28 & 20.88 & -0.42 & 0.69 & 0.71 & e & 1.54 & 1.24 & & \\
& 5.5 & -9.5 & 11.0/2.1 & 23.37 & 22.14 & -0.62 & 0.57 & 0.66 & e & 0.19 & 0.21 & & \\
& 2.7 & 10.9 & 11.2/2.1 & 24.33 & 23.37 &$>$-0.83 & 0.40 & 0.56 & e & 0.06 & 0.06 & & \\
& -10.5 & -9.7 & 14.3/2.7 & 24.46 & 23.73 & -0.64 & 0.03 & 0.70 & e & 0.01 & 0.01 & & \\
& 7.6 & 14.9 & 16.8/3.2 & $>$25.00 & 22.55 & & $>$ 0.88 & 1.57 & e & & 0.01 & & \\
& 11.6 & -13.4 & 17.7/3.4 & 23.45 & 22.10 & 0.08 & 0.49 & 0.86 & e & 0.01 & 0.01 & & \\
\hline
X045--27 & 2.6 & -9.8 & 10.1/2.9 &$\sim$10.0 & & & $\sim$0.3& & & & & 0.00 & {\bf F5V~star}\\
\hline
X409--28 & 0.1 & -7.5 & 7.5/1.4 & 23.70 & 21.81 & -0.65 & 0.97 & 0.92 & e & 0.36 & 0.79 & 0.957 & {\bf AGN (BLRG)}\\
& -6.5 & 5.0 & 8.2/1.6 & 24.67 & $>$23.50 &$>$-1.17 & $<$ 0.37 & & e & 0.10 & & & \\
& -9.0 & -3.1 & 9.5/1.8 & 24.46 & 23.21 & -0.78 & 0.24 & 1.01 & e & 0.11 & 0.11 & & \\
& 3.5 & -9.3 & 9.9/1.9 & 24.27 & 22.47 &$>$-0.77 & 0.87 & 0.93 & e & 0.09 & 0.32 & & \\
& -6.5 & -8.6 & 10.7/2.0 & $>$25.00 & 22.76 & & & $>$ 1.54 & e & & 0.15 & & \\
& -10.1 & -6.6 & 12.1/2.3 & 23.65 & 22.68 & -0.66 & 0.41 & 0.56 & e & 0.07 & 0.08 & & \\
& 13.5 & 5.6 & 14.6/2.8 & 24.41 & 23.50 & -0.56 & 0.29 & 0.62 & e & 0.01 & 0.00 & & \\
\hline
X301--29 & -2.5 & 0.4 & 2.6/0.6 & 21.13* & 20.41* & -0.88 & 0.40 & 0.32 & p & 19.92 & 10.18 & 1.709 & {\bf AGN$^{(1)}$}\\
& -0.9 & 5.9 & 6.0/1.5 & 23.86 & 22.67 & -0.42 & 0.56 & 0.63 & e & 0.57 & 0.68 & & \\
& -2.7 & -10.4 & 10.8/2.6 & 19.13* & 16.30* & 1.09 & 1.59 & 1.24 & p & 0.32 & 0.25 & & \\
& 4.7 & 11.3 & 12.2/3.0 & 22.79 & 21.13 & -1.38 & 0.96 & 0.70 & e & 0.03 & 0.04 & & \\
\hline
X408--30 & -9.6 & 3.5 & 10.2/2.5 & 24.76 & $>$23.50 &$>$-1.26 & 0.40 & $<$ 0.86 & e & 0.02 & & & \\
& 12.8 & 5.9 & 14.1/3.4 & 24.56 & 23.00 &$>$-1.06 & 0.70 & 0.86 & e & 0.00 & 0.00 & & \\
& 4.6 & 13.8 & 14.6/3.6 & 23.49 & 21.53 & -0.30 & 1.04 & 0.92 & e & 0.00 & 0.01 & 0.814 & NELG\hfill(C)\\
\hline
X207--31 & -5.7 & -0.7 & 5.7/1.2 & 23.37 & 22.19 & -0.38 & 0.64 & 0.54 & e & 0.98 & 1.11 & & \\
& -1.6 & -8.0 & 8.1/1.7 & 23.78 & 22.43 & -0.44 & 0.84 & 0.51 & e & 0.29 & 0.56 & 0.58: & {\bf cluster gal}\\
& -10.0 & 4.8 & 11.1/2.3 & 24.05 & 22.97 &$>$-0.55 & 0.44 & 0.64 & e & 0.05 & 0.11 & & \\
& -6.7 & 10.0 & 12.0/2.5 & 24.58 & 23.55 &$>$-1.08 & 0.45 & 0.58 & e & 0.02 & 0.02 & & \\
& -12.3 & 1.0 & 12.3/2.5 & $>$25.00 & 23.22 & & & $>$ 1.08 & e & & 0.03 & & \\
& -14.0 & -5.8 & 15.2/3.1 & 24.58 & $>$23.50 &$>$-1.08 & $<$ 0.28 & & e & 0.00 & & & \\
& -15.7 & 6.4 & 16.9/3.5 & 23.94 & 22.38 & -0.43 & 0.57 & 0.99 & e & 0.00 & 0.01 & & \\
& -5.8 & 17.3 & 18.2/3.7 & 24.34 & 23.69 & -0.84 & 0.11 & 0.54 & e & 0.00 & 0.00 & & \\
& -17.3 & -6.3 & 18.5/3.8 & 22.68 & 19.65 &$>$ 0.82 & 1.72 & 1.31 & e & 0.00 & 0.00 & 0.584 & {\bf cD gal}\\
\hline
\end{tabular}
\end{table*}
\begin{table*}
\begin{tabular}{rrrrrrrrrcrrll}
\hline
\\
ID. & $\Delta\alpha$ & $\Delta\delta$ & $\Delta$$^{\prime\prime}\,$/$\displaystyle{\Delta\over\varepsilon_{\rm{x}}}$ &
\multicolumn{1}{c}{~m$_B$} & \multicolumn{1}{c}{m$_R$} & U-B & B-V & V-R & &
\multicolumn{1}{c}{$LR_B$} &\multicolumn{1}{c}{$LR_R$} & \multicolumn{1}{c}{z} & object notes\\
&\multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} & \multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} \\
\hline
X040--32 & 6.0 & 2.6 & 6.6/1.9 & 23.37 & 22.31 & -0.68 & 0.57 & 0.49 & p & 1.30 & 0.99 & 1.204 & {\bf AGN}\\
& -6.1 & 4.8 & 7.8/2.2 & 24.53 & $>$23.50 & -0.82 & -0.30 & $<$ 1.33 & e & 0.08 & & & \\
& 8.3 & 1.5 & 8.4/2.4 & 23.26 & 21.99 & -0.74 & 0.52 & 0.75 & e & 0.21 & 0.27 & 0.804 & NELG\\
& 9.0 & 4.1 & 9.8/2.8 & 23.61 & 22.27 & 0.41 & 0.52 & 0.82 & e & 0.04 & 0.08 & & \\
\hline
X028--33 & -3.1 & -0.0 & 3.1/0.6 & $>$25.00 & $>$23.50 &$<$-1.77 & & & p & & & & \\
& 4.3 & -0.3 & 4.3/0.9 & 23.20 & 22.20 & -0.78 & 0.40 & 0.60 & e & 1.35 & 1.53 & & \\
& -5.5 & -1.8 & 5.8/1.2 & 19.16 & 17.22 & 1.06 & 1.29 & 0.65 & p & 3.64 & 3.59 & 0.00 & {\bf M~star}\\
& -1.8 & -10.7 & 10.8/2.3 & 21.95 & 20.98 & -0.61 & 0.36 & 0.61 & e & 0.23 & 0.19 & 0.657 & NELG\\
& 11.1 & -4.4 & 11.9/2.5 & 23.88 & 23.32 & -0.79 & 0.28 & 0.28 & e & 0.06 & 0.03 & & \\
& 12.2 & 8.0 & 14.6/3.0 & 23.89 & 23.01 & -0.43 & 0.10 & 0.78 & p & 0.02 & 0.01 & & \\
& 16.0 & 1.8 & 16.1/3.4 & 24.25 & 21.71 &$>$-0.75 & 1.26 & 1.28 & e & 0.00 & 0.01 & & \\
\hline
X250--34 & 7.1 & 0.6 & 7.1/1.6 & 24.30 & 23.59 &$>$-0.80 & $<$ 0.00 & $>$ 0.71 & e & 0.22 & 0.16 & & \\
& -8.9 & 5.6 & 10.5/2.4 & 23.29 & 22.94 &$>$ 0.21 & 0.08 & 0.27 & e & 0.13 & 0.10 & & \\
& -9.1 & -6.0 & 10.9/2.5 & 24.32 & 23.68 &$>$-0.82 & $<$ 0.02 & $>$ 0.62 & e & 0.04 & 0.02 & & \\
& 6.8 & -8.9 & 11.2/2.6 & $>$25.00 & 23.18 & & & $>$ 1.12 & e & & 0.03 & & \\
\hline
X011--35 & 0.6 & -0.8 & 1.0/0.2 & 21.94 & 20.33 & -0.20 & 1.14 & 0.47 & e & 1.80 & 2.03 & 0.189 & abs.gal\hfill(A)\\
& 1.7 & 2.5 & 3.1/0.5 & 24.49 & 23.75 &$>$-0.99 & $<$ 0.19 & $>$ 0.55 & e & 0.37 & 0.23 & & \\
& 0.8 & 6.3 & 6.4/1.0 & 23.61 & 22.40 & -0.32 & 0.63 & 0.58 & e & 0.43 & 0.81 & & \\
& -7.8 & 5.5 & 9.5/1.6 & 21.95 & 20.63 & -0.67 & 0.76 & 0.56 & e & 0.54 & 0.44 & 0.391 & {\bf NELG}\hfill(D)\\
& -7.6 & -5.8 & 9.6/1.6 & 23.99 & $>$23.50 &$>$-0.49 & -0.19 & $<$ 0.68 & e & 0.22 & & & \\
& -1.0 & -10.2 & 10.2/1.7 & 24.03 & 21.34 &$>$-0.53 & 1.34 & 1.35 & e & 0.10 & 0.36 & 0.586 & early gal\hfill(F)\\
& 7.3 & -7.4 & 10.4/1.7 & 24.02 & 23.42 & -0.97 & -0.05 & 0.65 & p & 0.10 & 0.10 & & \\
& -11.9 & 0.2 & 11.9/1.9 & 23.63 & 22.94 &$>$-0.13 & 0.39 & 0.30 & e & 0.11 & 0.14 & & \\
& -6.0 & -12.0 & 13.4/2.2 & 22.52 & 21.29 & -0.39 & 0.64 & 0.59 & e & 0.09 & 0.13 & ?.??? & low S/N\hfill(I)\\
& -10.0 & -10.9 & 14.8/2.4 & 23.45 & 23.63 &$>$ 0.05 & 0.05 & -0.23 & e & 0.07 & 0.02 & & \\
& 7.4 & -16.9 & 18.5/3.0 & 24.72 & 22.17 &$>$-1.22 & 1.41 & 1.14 & p & 0.00 & 0.02 & & \\
& -15.0 & 13.4 & 20.1/3.3 & 22.00 & 20.15 &$>$ 1.00 & 1.14 & 0.71 & e & 0.01 & 0.01 & 0.390 & {\bf early+[OII]}\hfill(L)\\
& -4.8 & -21.2 & 21.7/3.5 & 24.02 & 22.66 &$>$-0.52 & 0.90 & 0.46 & e & 0.00 & 0.00 & & \\
& -19.5 & 10.9 & 22.3/3.6 & $-out-$ & 22.86 & & & 0.93 & e & & 0.00 & & \\
& 8.9 & -22.3 & 24.0/3.9 & 22.90 & 20.57 &$>$ 0.60 & 1.31 & 1.02 & e & 0.00 & 0.00 & 0.390 & {\bf early gal}\hfill(O)\\
\hline
X032--36 & 5.3 & -3.1 & 6.1/1.1 & 24.43 & 23.19 &$>$-0.93 & 0.51 & 0.73 & e & 0.28 & 0.28 & & \\
& -3.8 & 9.4 & 10.1/1.9 & 24.24 & 22.81 &$>$-0.74 & 0.87 & 0.56 & e & 0.09 & 0.20 & & \\
& 4.4 & 9.2 & 10.3/1.9 & 21.26 & 20.61 & -0.15 & 0.34 & 0.31 & p & 2.27 & 1.44 & 0.00 & B~star\\
& 0.5 & 12.4 & 12.4/2.3 & 23.36 & 21.98 & -0.56 & 0.79 & 0.59 & p & 0.54 & 0.37 & 1.190 & {\bf AGN}\\
& 10.6 & 9.0 & 13.9/2.6 & $>$25.00 & 22.88 & & $>$ 1.09 & 1.03 & e & & 0.04 & & \\
& -0.2 & -14.3 & 14.3/2.6 & 24.21 & 23.46 &$>$-0.71 & 0.66 & 0.09 & p & 0.02 & 0.02 & & \\
& -8.9 & -13.4 & 16.1/3.0 & 24.21 & 23.43 &$>$-0.71 & 0.22 & 0.56 & e & 0.01 & 0.01 & & \\
& 16.3 & 1.6 & 16.4/3.0 & $>$25.00 & 23.27 & & & $>$ 1.03 & e & & 0.01 & & \\
\hline
X251--37 & -6.1 & 4.5 & 7.6/1.3 & 21.72 & 20.68 & -0.20 & 0.70 & 0.34 & p & 5.70 & 3.20 & 2.710 & {\bf AGN}\\
& 4.3 & 6.9 & 8.1/1.3 & 24.34 & 23.40 &$>$-0.84 & 0.57 & 0.37 & e & 0.17 & 0.17 & & \\
& 3.5 & 12.4 & 12.9/2.1 & 24.61 & $>$23.50 &$>$-1.11 & $<$ 0.31 & & e & 0.03 & & & \\
& -13.7 & 0.8 & 13.7/2.3 & 20.03 & 18.40 & 0.28 & 1.07 & 0.56 & e & 0.23 & 0.14 & 0.091 & early gal\\
& -12.3 & 8.0 & 14.7/2.4 & 23.95 & 22.80 &$>$-0.45 & 0.81 & 0.34 & e & 0.04 & 0.05 & & \\
\hline
X024--38 & 0.8 & -0.0 & 0.8/0.1 & 22.19 & 20.76 & -0.37 & 0.77 & 0.66 & p & 9.29 & 8.72 & 1.430 & {\bf AGN}\\
& 3.6 & -3.6 & 5.1/0.9 & 17.94 & 16.82 & 0.02 & 0.45 & 0.67 & p & 3.39 & 5.06 & 0.00 & B~star\\
& 0.4 & 9.3 & 9.3/1.7 & 23.21 & 22.23 & -0.24 & 0.73 & 0.25 & p & 0.75 & 0.57 & 0.276 & NELG\\
& 8.5 & 10.1 & 13.2/2.4 & 22.76 & 21.26 & 0.24 & 0.88 & 0.62 & e & 0.06 & 0.09 & & \\
& 2.4 & 15.8 & 16.0/3.0 & 23.82 & 23.54 &$>$-0.32 & 0.50 & -0.22 & e & 0.01 & 0.01 & & \\
& -17.4 & -1.8 & 17.5/3.2 & 24.35 & 22.65 &$>$-0.85 & 0.95 & 0.75 & e & 0.00 & 0.01 & & \\
& 17.8 & -2.1 & 17.9/3.3 & 14.59 & 13.36 & -0.54 & 0.23 & 1.00 & p & 0.15 & 0.06 & 0.00 & A~star\\
& -19.1 & -1.1 & 19.1/3.5 & 23.95 & $>$23.50 &$>$-0.45 & -0.34 & $<$ 0.79 & e & 0.00 & & & \\
& -10.1 & 16.9 & 19.7/3.6 & 16.99 & 15.68 & 0.29 & 0.42 & 0.89 & p & 0.00 & 0.00 & 0.00 & G~star\\
\hline
X015--39 & -5.4 & -5.8 & 7.9/1.1 & 23.27 & 22.01 & -0.46 & 0.69 & 0.57 & p & 1.04 & 0.79 & 0.500 & {\bf AGN (Sy 1)}\\
& 13.1 & -8.1 & 15.4/2.2 & $>$25.00 & 23.22 & & & $>$ 1.08 & e & & 0.03 & & \\
& -17.4 & -2.8 & 17.6/2.5 & 23.48 & 22.73 & -0.68 & 0.15 & 0.60 & e & 0.04 & 0.03 & & \\
& 7.2 & -17.2 & 18.7/2.7 & 24.64 & 23.43 &$>$-1.14 & $<$ 0.34 & $>$ 0.87 & e & 0.01 & 0.01 & & \\
& -17.3 & -7.5 & 18.8/2.7 & 23.85 & 22.62 & -0.79 & 0.21 & 1.02 & e & 0.01 & 0.02 & ?.??? & low S/N\\
\hline
\end{tabular}
\end{table*}
\begin{table*}
\begin{tabular}{rrrrrrrrrcrrll}
\hline
\\
ID. & $\Delta\alpha$ & $\Delta\delta$ & $\Delta$$^{\prime\prime}\,$/$\displaystyle{\Delta\over\varepsilon_{\rm{x}}}$ &
\multicolumn{1}{c}{~m$_B$} & \multicolumn{1}{c}{m$_R$} & U-B & B-V & V-R & &
\multicolumn{1}{c}{$LR_B$} &\multicolumn{1}{c}{$LR_R$} & \multicolumn{1}{c}{z} & object notes\\
&\multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} & \multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} \\
\hline
X236--40 & 5.6 & -0.3 & 5.6/1.3 & 22.23* & 21.64 & -1.00 & 0.24 & 0.35 & p & 6.38 & 3.47 & 1.140 & {\bf AGN$^{(1)}$}\\
& -7.7 & 0.5 & 7.7/1.8 & $>$23.50* & 22.00 & & $>$ 0.11 & 1.39 & e & & 0.57 & & \\
& -8.7 & -2.1 & 8.9/2.1 & 24.55 & 23.73 &$>$-1.05 & 0.19 & 0.63 & e & 0.07 & 0.06 & & \\
& -12.0 & -3.0 & 12.4/2.9 & $>$25.00 & 23.19 & & $>$ 0.39 & 1.42 & e & & 0.01 & & \\
& -6.2 & 12.3 & 13.8/3.2 & 24.30 & 22.64 & -0.60 & 0.65 & 1.01 & e & 0.00 & 0.01 & & \\
\hline
X234--41 & 0.9 & 1.2 & 1.5/0.3 & 23.38 & 22.18 & 0.17 & 0.77 & 0.43 & e & 1.58 & 1.78 & & \\
& -0.5 & 4.1 & 4.1/0.8 & 23.83 & 23.29 & -0.54 & -0.03 & 0.57 & e & 0.73 & 0.41 & & \\
& 1.5 & -4.1 & 4.3/0.8 & 23.72 & 22.58 & 0.41 & 0.75 & 0.39 & e & 0.71 & 0.86 & & \\
& 2.3 & -5.6 & 6.1/1.1 & 23.83 & 22.51 & -0.69 & 0.50 & 0.82 & e & 0.51 & 0.62 & & \\
& -1.1 & -6.4 & 6.5/1.2 & 23.67 & 22.08 & -0.24 & 0.39 & 1.20 & e & 0.47 & 0.88 & & \\
& -8.5 & 2.8 & 9.0/1.7 & 23.18 & 22.84 & -0.65 & 0.03 & 0.31 & e & 0.39 & 0.28 & & \\
& 11.4 & -6.3 & 13.1/2.5 & 23.56 & 22.48 & -0.78 & 0.37 & 0.71 & e & 0.05 & 0.09 & & \\
& 0.2 & -14.7 & 14.7/2.8 & $>$25.00 & 23.22 & & & $>$ 1.08 & e & & 0.01 & & \\
& 0.0 & 16.2 & 16.2/3.0 & 23.51 & 22.60 & -0.80 & 0.24 & 0.67 & e & 0.01 & 0.01 & & \\
& -4.8 & -17.6 & 18.2/3.4 & 23.10 & 21.46 & -0.34 & 0.77 & 0.87 & e & 0.00 & 0.01 & & \\
& -6.3 & 18.0 & 19.0/3.6 & 24.22 & $>$23.50 & -0.72 & 0.14 & $<$ 0.58 & e & 0.00 & & & \\
& 17.3 & 8.8 & 19.4/3.6 & 18.93 & $<$18.02 & -0.15 & 0.55 & $>$ 0.36 & p & 0.01 & 0.00 & 0.00 & G~star\\
\hline
X306--42 & 3.1 & -3.6 & 4.7/1.0 & 24.08 & $>$23.50 & -0.74 & 0.11 & $<$ 0.47 & e & 0.44 & & & \\
& -3.6 & 5.6 & 6.6/1.5 & 22.62 & 22.04 & -0.60 & 0.34 & 0.24 & p & 3.18 & 1.23 & 1.065 & {\bf AGN}\\
& -8.0 & 2.1 & 8.2/1.8 & 24.03 & 23.29 & -0.25 & 0.02 & 0.72 & e & 0.15 & 0.15 & & \\
& 10.4 & -5.4 & 11.7/2.6 & 23.37 & 22.47 & -0.60 & 0.36 & 0.54 & e & 0.08 & 0.09 & & \\
& -2.0 & 12.2 & 12.3/2.7 & $>$25.00 & 23.40 & & $>$ 0.34 & 1.26 & p & & 0.02 & & \\
& 10.4 & 11.8 & 15.7/3.4 & 22.03 & 21.19 & -0.10 & 0.54 & 0.30 & e & 0.01 & 0.01 & 0.078 & NELG\\
\hline
X051--43 & 1.1 & -1.1 & 1.5/0.2 & 21.44 & 20.60 & -0.12 & 0.48 & 0.36 & p & 9.00 & 5.70 & ?.??? & {\bf Bl~Lac}~(?)\\
& -6.3 & 6.9 & 9.4/1.4 & 24.45 & 23.55 &$>$-0.95 & $<$ 0.15 & $>$ 0.75 & e & 0.13 & 0.10 & & \\
& -9.7 & 5.2 & 11.0/1.7 & 23.39 & $>$23.50 & -0.50 & -0.05 & $<$-0.06 & e & 0.27 & & & \\
& -0.6 & 13.9 & 13.9/2.1 & 24.28 & 22.68 &$>$-0.78 & 0.78 & 0.82 & e & 0.04 & 0.09 & & \\
& -13.1 & 4.8 & 14.0/2.1 & 23.71 & 23.59 &$>$-0.21 & -0.41 & 0.53 & e & 0.07 & 0.03 & & \\
& 3.8 & -17.4 & 17.8/2.7 & 24.22 & $>$23.50 &$>$-0.72 & $<$-0.08 & & e & 0.01 & & & \\
& -12.2 & -14.5 & 19.0/2.9 & 20.23 & 19.20 & -0.13 & 0.60 & 0.43 & e & 0.04 & 0.01 & 0.096 & Starburst\\
& -11.7 & 16.6 & 20.3/3.1 & 23.86 & 23.51 &$>$-0.36 & -0.30 & 0.65 & e & 0.01 & 0.00 & & \\
& -6.1 & 21.2 & 22.0/3.3 & 23.00 & 21.52 &$>$ 0.50 & 0.81 & 0.67 & p & 0.01 & 0.01 & & \\
\hline
X407--44 & -3.3 & -4.7 & 5.7/0.8 & 21.41* & 20.65 & -0.94 & 0.19 & 0.57 & p & 5.68 & 3.59 & 1.821 & {\bf AGN$^{(1)}$}\\
& 14.1 & 0.9 & 14.2/2.0 & $>$23.50* & 21.08 & & & $>$ 1.42 & p & & 0.67 & & \\
& -18.2 & -16.0 & 24.2/3.3 & 22.73* & 20.99 & -0.92 & 0.93 & 0.81 & e & 0.00 & 0.00 & & \\
\hline
X233-45 & 4.9 & -5.8 & 7.6/1.1 & 24.86 & 23.72 & -1.30 & $<$ 0.56 & $>$ 0.58 & e & 0.09 & 0.11 & & \\
& -8.4 & -2.5 & 8.8/1.2 & 24.83 & 23.33 & -1.29 & 0.71 & 0.79 & e & 0.07 & 0.14 & & \\
& -4.0 & -8.8 & 9.6/1.3 & 22.55 & 21.86 & -0.82 & 0.19 & 0.50 & e & 0.29 & 0.48 & ?.??? & low S/N\\
& -9.2 & 5.9 & 10.9/1.5 & 24.69 & 21.76 &$>$-1.19 & 1.64 & 1.29 & e & 0.06 & 0.37 & & \\
& -11.0 & 3.2 & 11.5/1.6 & 24.86 & 23.10 &$>$-1.36 & 0.57 & 1.19 & e & 0.04 & 0.08 & & \\
& -13.3 & -1.8 & 13.4/1.9 & 22.51 & 21.70 & -0.66 & 0.25 & 0.56 & e & 0.12 & 0.20 & 1.180 & {\bf AGN}\\
& -12.0 & -11.9 & 16.9/2.4 & 21.29 & 18.62 & 1.19 & 1.51 & 1.16 & e & 0.49 & 0.21 & & \\
& -16.1 & 10.4 & 19.2/2.7 & 24.49 & $>$23.50 & -0.49 & $<$ 0.19 & & e & 0.01 & & & \\
& 6.3 & 20.7 & 21.6/3.0 & 22.41 & 21.12 & -0.54 & 0.58 & 0.71 & e & 0.01 & 0.01 & & \\
& -17.6 & -17.0 & 24.5/3.4 & 24.50 & 23.34 & -0.69 & $<$ 0.20 & $>$ 0.96 & e & 0.00 & 0.00 & & \\
\hline
X109-46 & -1.9 & -0.2 & 1.9/0.4 & 23.56 & 22.59 & -0.52 & 0.42 & 0.55 & p & 2.25 & 0.58 & & \\
& 2.6 & 3.6 & 4.4/0.9 & 23.63 & 23.10 & -0.52 & 0.29 & 0.24 & e & 0.80 & 0.44 & & \\
& -6.2 & 5.1 & 8.0/1.7 & 21.70 & 20.56 & -0.54 & 0.75 & 0.39 & e & 0.74 & 0.60 & 0.269 & Starburst\hfill(C)\\
& 2.5 & 7.6 & 8.0/1.7 & 24.60 & 22.96 &$>$-1.10 & 1.04 & 0.60 & e & 0.11 & 0.36 & & \\
& 2.1 & -9.2 & 9.4/2.0 & 24.30 & 21.36 &$>$-0.80 & 1.44 & 1.50 & p & 0.10 & 1.51 & 0.00 & M~star\hfill(E)\\
& 4.0 & -9.8 & 10.6/2.2 & 22.79 & 20.06 &$>$ 0.71 & 1.57 & 1.16 & p & 0.71 & 0.79 & 0.00 & M~star\hfill(F)\\
& 7.5 & 9.1 & 11.8/2.5 & 24.78 & $>$23.50 &$>$-1.28 & $<$ 0.48 & & e & 0.02 & & & \\
& 0.5 & -12.7 & 12.7/2.7 & 24.75 & 23.44 &$>$-1.25 & 0.83 & 0.48 & e & 0.01 & 0.02 & & \\
& -4.9 & 14.7 & 15.5/3.2 & 22.78 & 21.83 & -0.83 & 0.25 & 0.70 & e & 0.01 & 0.01 & 0.626 & NELG\hfill(I)\\
\hline
X213--47 & -8.5 & -13.4 & 15.8/2.9 & 24.70 & $>$23.50 &$>$-1.20 & $<$ 0.40 & & e & 0.00 & & & \\
& -13.8 & -13.8 & 19.5/3.5 & 24.51 & 23.54 &$>$-1.01 & 0.25 & 0.72 & e & 0.00 & 0.00 & & \\
\hline
\end{tabular}
\end{table*}
\begin{table*}
\begin{tabular}{rrrrrrrrrcrrll}
\hline
\\
ID. & $\Delta\alpha$ & $\Delta\delta$ & $\Delta$$^{\prime\prime}\,$/$\displaystyle{\Delta\over\varepsilon_{\rm{x}}}$ &
\multicolumn{1}{c}{~m$_B$} & \multicolumn{1}{c}{m$_R$} & U-B & B-V & V-R & &
\multicolumn{1}{c}{$LR_B$} &\multicolumn{1}{c}{$LR_R$} & \multicolumn{1}{c}{z} & object notes\\
&\multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} & \multicolumn{1}{c}{~~[$^{\prime\prime}\,$]} \\
\hline
X022--48 & -1.7 & -1.3 & 2.1/0.5 & 22.21 & 19.86 & 0.72 & 1.40 & 0.95 & e & 3.25 & 3.36 & 0.32: & {\bf Interactive~gal}\\
& -1.6 & -4.6 & 4.9/1.1 & 23.31 & 22.06 & -0.20 & 0.74 & 0.51 & e & 1.30 & 1.47 & & companion\\
& 5.9 & -6.0 & 8.5/2.0 & 24.04 & $>$23.50 &$>$-0.54 & -0.03 & $<$ 0.57 & e & 0.12 & & & \\
& 8.9 & -7.9 & 11.9/2.7 & 22.33 & 21.00 & -0.38 & 0.66 & 0.67 & e & 0.09 & 0.07 & 0.474 & NELG\\
& -3.3 & 11.5 & 12.0/2.8 & 21.90 & 20.70 & -0.38 & 0.64 & 0.56 & e & 0.08 & 0.07 & 0.389 & early+[OII]\\
& -10.0 & -9.6 & 13.9/3.2 & 23.76 & 22.55 & -0.25 & 0.71 & 0.50 & e & 0.01 & 0.01 & & \\
& -15.4 & -0.6 & 15.4/3.5 & 24.51 & $>$23.50 &$>$-1.01 & $<$ 0.21 & & e & 0.00 & & & \\
& -6.5 & -14.9 & 16.3/3.7 & 21.03 & 20.01 & 0.01 & 0.64 & 0.38 & p & 0.02 & 0.01 & 0.00 & F/G star\\
\hline
X215--49 & 5.4 & -7.4 & 9.2/1.7 & 23.12 & 22.44 & -0.69 & 0.18 & 0.50 & e & 0.37 & 0.42 & & \\
& -0.1 & -14.3 & 14.3/2.7 & $>$25.00 & 23.32 & & & $>$ 0.98 & e & & 0.01 & & \\
& 3.7 & 14.9 & 15.4/2.9 & 22.58 & 22.19 & -0.78 & 0.19 & 0.20 & e & 0.02 & 0.03 & 1.053 & strong\hfill[OII] gal~(C)\\
& -3.1 & 16.4 & 16.6/3.1 & 23.62 & 22.47 & -0.60 & 0.59 & 0.56 & e & 0.01 & 0.01 & & \\
& 17.6 & 0.0 & 17.6/3.3 & 24.35 & 22.93 &$>$-0.85 & 0.77 & 0.65 & e & 0.00 & 0.00 & & \\
\hline
X264--50 & 0.4 & 3.5 & 3.5/0.7 & 24.44 & $>$23.50 &$>$-0.94 & $<$ 0.14 & & e & 0.45 & & & \\
& -2.1 & -7.2 & 7.5/1.4 & 21.84 & 20.77 & -0.42 & 0.49 & 0.58 & e & 0.89 & 0.73 & 0.568 & {\bf Starburst}\\
& 3.1 & -9.1 & 9.6/1.8 & 23.46 & 22.20 &$>$ 0.04 & 0.66 & 0.60 & e & 0.31 & 0.36 & & \\
& -7.6 & -7.5 & 10.7/2.0 & 24.60 & $>$23.50 &$>$-1.10 & 0.10 & $<$ 1.00 & e & 0.05 & & & \\
& -9.9 & -8.7 & 13.2/2.5 & 23.24 & 22.96 & -0.41 & -0.03 & 0.31 & p & 0.14 & 0.02 & & \\
& 17.8 & 2.1 & 17.9/3.4 & 23.53 & 22.06 & -0.17 & 0.64 & 0.83 & e & 0.00 & 0.01 & & \\
\hline
\end{tabular}
\end{table*}
The LR values for most of these 15 suggested identifications are greater
than 1 and are the highest in their error boxes. The only exceptions are
the broad--line radio--galaxy identified with the source X409--28 and the AGN
identified with the source X032--36, whose LR values are of the order of 0.5.
In the latter error box there is also a brighter stellar object with
higher LR values. However, its spectrum shows that it is a B~type star
and its identification with the X-ray source can therefore be excluded
on the basis of the $f_x/f_v$ ratio (Maccacaro et~al. 1988).
A few other objects have been observed spectroscopically in these 15
error boxes (see last column in Table 3) and most of them have been found to be
narrow emission line galaxies (NELG) or starburst galaxies. In all cases
their LR value is significantly smaller than that of the best candidate and
therefore we consider them unlikely to be associated to the
X-ray sources. The estimated surface densities of AGNs with
$ m_B \le$ 23.5 ($\sim$ 325 per square degree, Mignoli and Zamorani 1998)
is similar to the surface density of M stars with $m_V < $ 20. From the
sum of the two surface densities we estimate that only
$\sim$ 0.7 random coincidences in the
95\% error circles of the 31 X-ray sources are expected for AGNs and M stars
in the magnitude ranges covered by these suggested identifications.
iii. For the 16 error boxes which at this stage were still without
a reliable optical identification, we have taken spectra also of objects
classified as extended. At the typical magnitudes of the optical
counterparts ($m_B >$ 22.0) the surface density of objects classified as
extended in our optical images is higher than that of stellar objects. The
ratio between the numbers of extended and stellar objects in our CCD
catalogue is $\sim$ 2 at $m_B \sim$ 22.25 and $\sim$ 5 at $m_B \sim$ 23.25.
Therefore, the large number of extended objects
makes more difficult to deal with the problem of random coincidences and to find
convincing optical identifications. Each of these sixteen error boxes is now
discussed in some detail, in order of decreasing X-ray flux.
{\bf X043--12}: the 95\% error box contains only one object, classified
as extended in the blue band and as point-like in the V and R bands. Its noisy
spectrum shows two narrow lines well coincident with Ly$\alpha$ and CIV at z
= 2.80. It appears to be a high redshift analogue of the low--z type--2
AGNs, similar to the QSO~2 at z = 2.35 found by Almaini et al. (1995)
in the optical follow--up of other deep ROSAT fields. Although a higher quality
spectrum would be needed to confirm the nature of this object, given its
relatively large LR values, we consider it as the correct identification.
{\bf X039--19} and {\bf X049--21}: these are the brightest X-ray sources in our sample without any
suggested spectroscopic identification. The 95\% error box of X039--19 does not
contain any object in our CCD data, while a faint ($m_B \sim 24$), extended
object is present near the center of the error box of X049--21. Its relatively
large LR values suggest that it might be the correct optical identification.
However, both the spectrum of this object (shown in figure 4) and the
spectrum of a similarly faint object just outside the error box of X039--19
(object A in figure 4) have extremely low S/N and do not allow any
spectroscopic classification. Moreover, both X-ray sources have a close--by
source at less than one arcmin (see figure 1), so that it is possible that
their X-ray positions are not as well determined as those of the other
sources with similar X-ray flux.
{\bf X211--22}: we took spectra for three objects, all of them just outside
the 95\% radius. The brightest one is an early type
galaxy at z = 0.180, while the other two, separated by about 2 arcsec
from each other,
are a Sey 1 galaxy (classified as point--like in the B band and as extended
in the R band) at z = 0.281 and a B star. We consider the Sey 1 galaxy as the
most likely identification.
{\bf X404--23}: its large error box makes the identification difficult. The
3$\sigma$ error box contains 18 sources and all
of them have low LR values. We took spectra for three
objects, finding a NELG galaxy (object G) and a Sey 2 galaxy (object N)
at different redshifts, while the third one (object J) is a faint
($m_B = 23.74$) M star. None of them is a
convincing identification. Visual inspection of the X-ray image suggests
the presence of two different sources, at about 40$^{\prime\prime}\,$ from each other,
which our detection algorithm has not been able to separate. This is also
supported by the large positional error resulting from the maximum likelihood
fit. Since the two sources appear to have approximately the same number
of counts, it is likely that both of them are very close to or just
below our X-ray detection threshold.
\begin{figure*}[p]
\epsfysize=24.5cm
\vspace{-1.0cm}
\vspace{22.cm}
\caption[]{Finding charts for the X-ray sources with 68\% and 95\% error boxes
and spectra of the most likely identifications.}
\end{figure*}
\begin{figure*}[p]
\epsfysize=24.5cm
\vspace{-1.0cm}
\vspace{22.cm}
\addtocounter{figure}{-1}
\caption[]{continued}
\end{figure*}
\begin{figure*}[p]
\epsfysize=24.5cm
\vspace{-1.0cm}
\vspace{22.cm}
\addtocounter{figure}{-1}
\caption[]{continued}
\end{figure*}
\begin{figure*}[p]
\epsfysize=24.5cm
\vspace{-1.0cm}
\vspace{22.cm}
\addtocounter{figure}{-1}
\caption[]{continued}
\end{figure*}
\begin{figure*}[p]
\epsfysize=24.5cm
\vspace{-1.0cm}
\vspace{22.cm}
\addtocounter{figure}{-1}
\caption[]{continued}
\end{figure*}
\begin{figure*
\epsfysize=24.5cm
\vspace{-1.0cm}
\vspace{22.cm}
\addtocounter{figure}{-1}
\caption[]{continued}
\end{figure*}
{\bf X408--30}: the radio source coinciding with X408-30 and indicated with
a cross in figure 4 is not associated to any optical object
(see Gruppioni, Mignoli and Zamorani 1999). If the radio and X-ray sources
are really associated, then the optical counterpart of the X-ray source
is fainter than our optical limit. The brightest object in
the field (object C), but well outside the 95\% error box, is a
NELG at z = 0.814.
{\bf X207--31}: all the objects in the error box are extended. We took
spectra for two galaxies, one ($m_R$ = 22.43) just outside the 1$\sigma$ error
box and the second one ($m_R$ = 19.65) just outside the 3$\sigma$ error box.
Both of them have z $\sim$ 0.58; the second galaxy, which is the brightest
and reddest of all the surrounding galaxies appears to be a cD in a cluster.
(Figure 4 shows the spectrum of this galaxy).
Just below our maximum likelihood threshold for X-ray detection, at about
two arcmin from X207--31, there is an other X-ray source in a position where an
overdensity of faint galaxies is clearly seen. Three of these galaxies
have the same redshift as those in X207--31. We therefore identify X207--31
with a cluster, probably interacting with a second cluster at a distance of
about 1.5 Mpc.
{\bf X250--34}: the error box contains only faint ($m_B \ge$ 23.2,
$m_R \ge$ 22.9) extended objects, none of which with a large LR. No spectrum
was taken for any object.
{\bf X011--35}: the density of objects in the error box is twice as
high as the average density in our catalog. We took spectra for the
six brightest objects ($m_R <$ 21.5) finding three galaxies at z $\sim$ 0.39
(objects D, L and O in figure 4), one absorption line galaxy at z = 0.189
(object A; this is the object closest to the X-ray position),
one early--type galaxy at z = 0.586 (object F).
The redshift of this galaxy is about the same as that of the
clusters discussed in connection with X207--31. The angular distance between
X011--35 and X207--31 ($\sim$5 arcmin) corresponds to about 4 Mpc, suggesting
the existence of a large scale structure at this redshift. The sixth
spectrum (object I) has a very low S/N and no redshift was derived.
The spectrum in figure 4 is the spectrum of the galaxy closest to the
X--position among those at z $\sim$ 0.39 (D). On the basis of these results,
although with some possible ambiguity, we identify the X-ray source with
a group of galaxies.
{\bf X234--41}: also in this error box the density of faint extended
objects is twice as high as the average density. In particular,
in the inner 10$^{\prime\prime}\,$ there are six extended objects with $ 23 \le m_B \le 24 $,
while about one would be expected.
Although
no spectrum was taken, because of the faintness of the objects, we tentatively
identify this source with a group or cluster of galaxies. The bright object
just outside the 3$\sigma$ error box is a G star with a too large X-ray to
optical ratio to be associated with the X-ray source.
{\bf X233--45}: we took spectra of the two brightest objects
($m_B \sim$ 22.5) within 15$^{\prime\prime}\,$ from the X-ray position. Both of them are
classified as extended. One spectrum is very blue, with low S/N and no redshift
was determined; the second object is an AGN at z = 1.180, which we consider
to be the identification. No spectrum is available for the more distant
and brighter extended object, whose colours suggest a low redshift
elliptical galaxy.
{\bf X109--46}: the error box contains three relatively bright
objects ($m_R <$ 21.5), two of which are point--like and one extended. The
two stellar objects are two M stars (objects E and F in figure 4),
while the extended one is a starburst galaxy at z = 0.269 (object C).
The two M stars appear to be too faint for being the counterpart of the
X-ray source. The starburst galaxy might be considered a likely candidate.
We note, however, that very close to the X-ray position, there are two faint,
blue objects with LR values larger than that of the starburst galaxy. One of
these two objects, classified as point--like, is a good candidate
for being an AGN. Waiting for spectra for these objects, we consider
this source not identified yet.
{\bf X213--47}: no object appears in our optical catalogs within the
95\% error box, although a very faint object ($m_R \ge$ 24.0) is barely
detected in two different exposures of the R CCDs at $\sim$ 2 arcsec from
the X-ray position.
{\bf X022--48}: the noisy spectrum of the galaxy close to the center
of the X-ray error box shows a narrow emission line which, if interpreted
as [OII], corresponds to z = 0.32. Both the blue and the red CCDs show clear
signs of interaction with a fainter, blue extended object located about 3$^{\prime\prime}\,$
south. Also on the basis of the LR values, we consider this complex of
interacting galaxies as a likely identification. The two galaxies just outside
the 2$\sigma$ error box have different redshifts, while the brighter, more
distant point--like object is an F/G star.
{\bf X215--49}: no obvious candidate is contained within the 95\% error
box. The spectrum of the galaxy just outside of it (object C)
shows a strong [OII] emission line and MgII2800 absorption line at z = 1.053.
Its LR is small and we do not consider this object as the correct
identification.
{\bf X264--50}: there is one object in the error box with LR $\sim$ 1.
It is a relatively bright ($m_R$ = 20.77) starburst galaxy at z = 0.568.
Despite the presence of a much fainter
($m_R \sim $ 23) blue, point-like source which may be an AGN candidate,
we consider the galaxy as a possible identification.
\section{ Discussion }
\subsection {Summary of identifications and not identified sources}
On the basis of the criteria discussed above, we have found reliable
spectroscopic identifications for 41 sources (82\% of the total),
indicated in bold face in the last column of Table 3.
For one more source (X234--41) the identification is very likely to be with
a faint cluster of galaxy, although no redshift is available. These 42
reliable identifications are 33 AGNs (including the two radio galaxies and
the BL Lac candidate; 79\% of the identified sources), 2 galaxies, 3 groups or
clusters of galaxies and 4 stars. Except for the higher fraction of unidentified
sources (see discussion below), the identification content of this sample
is in excellent agreement with what has been found, at a similar flux
limit, in the much deeper PSPC and HRI surveys in the Lockman field (Hasinger
et al. 1998, Schmidt et al. 1998).
Figure 5 shows the expected and
observed cumulative distributions of the distances normalized to the error
on each coordinate between the optical counterparts and the X-ray sources.
In the observed distribution we have excluded the three sources
identified with groups or clusters of galaxies, because for these objects
it is more difficult to unambiguously define the optical position. The excellent
agreement between the two distributions shows the goodness of our
derived ML estimates for the positional uncertainties, at least when the
X-ray position is not affected by confusion problems (see below).
\begin{figure}[t]
\epsfysize=9cm
\epsfbox{rosat5_fig5.ps}
\caption[]{ Cumulative observed distribution (dashed curve) of the distances
normalized to the error on each coordinate between the optical counterparts
and the X-ray sources. Here we have excluded the three sources identified
with groups or clusters of galaxies. The solid curve shows the expected
distribution.}
\end{figure}
If we divide our sample into two equally populated sub--samples as a
function of flux (S$> 6.5 \times 10^{-15}$ and
S$< 6.5 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ ), we find that
the percentage of identifications remains approximately constant (88\% and
80\% in the high and low flux sub--samples, respectively).
AGNs are the dominant class of objects in both sub--samples (90\% of the
optical identifications in the high flux sub--sample and 65\% in the low flux
sub--sample), while the few identifications with clusters and galaxies are
all in the low flux sub--sample. With the two extreme assumptions that none
or all of the still unidentified objects are AGNs we derive that the
percentage of AGNs among {\bf all} the X-ray sources in our sample
is comprised between 66\% and 82\%.
\begin{figure}[t]
\epsfysize=9cm
\epsfbox{rosat5_fig6.ps}
\caption[]{ X-ray flux versus $m_R$ magnitudes for all the sources in our
sample and for the AGNs identified in the
Lockman Field (Schmidt et al. 1998). The two straight dashed lines,
corresponding to constant X-ray to optical ratios, show the locus of this
plane (-0.6 $< log~f_x/f_v < $ 0.6) which contains most of the identified
AGNs in both samples. For a description of the upper limits, both in optical
and in X-ray, see text.}
\end{figure}
Figure 6 shows the X-ray flux versus $m_R$ magnitudes for the sources in our
sample and for the AGNs identified in the Lockman Field (Schmidt et al. 1998).
The two bright F stars ($m_B \sim$ 10) are not shown here.
The two straight dashed lines, corresponding to constant X-ray to optical
ratios, show the approximate locus of this plane which contains most of the
identified AGNs, both in our sample and in the Lockman Field.
This range in X-ray to optical fluxes correspond to
-0.6 $< log~f_x/f_v < $ 0.6, where $ log~f_x/f_v $ is defined as in
Maccacaro et al. (1988). For any given X-ray flux the range in magnitude
is about $\pm$ 1.5, corresponding to about $\pm$ a factor of 4
with respect to the average X-ray to optical ratio. Despite the much higher
X-ray flux limit and, correspondingly, much brighter limiting magnitudes,
also most ($\sim$ 90\%) of the EMSS AGNs lie in the same band.
For the three sources identified with groups or clusters of galaxies we
have plotted the magnitude of the brightest galaxies. The two X-ray
sources identified with galaxies have an X-ray to optical ratio close to the
lower limit of those of AGNs.
The limits for the eight unidentified sources
are plotted at the magnitude of the brightest object within the 95\% error
box which is not excluded from being the optical counterpart on the
basis of the available spectra. Three of these upper limits lie outside
the band shown in the figure and three more are very close to the upper
bound of this band. If associated with AGNs, they would have an X-ray to
optical ratio significantly higher than the average. As shown in the figure,
6 such AGNs have been identified in the Lockman field; four of them are
classical broad line objects, while the other two, showing only narrow lines,
have been classified as AGNs on the basis of the presence of [NeV]$\lambda$3426
emission (Schmidt et al. 1998). The HRI arcsec positions in the Lockman field
has made these identifications with such faint optical objects possible.
In our case, with the 95\% error radius for these sources ranging from
8$^{\prime\prime}\,$ to 24$^{\prime\prime}\,$, similar identifications are significantly more difficult.
Alternatively, we can not exclude that some of these sources may have
a ``wrong'' position because of confusion. In Section 2 we have estimated
that in about 2--5 cases a single source in our list may be produced by two
different close--by sources, so that its position may be significantly wrong
and we may not be able to find any reliable optical counterpart within the
error box. The much more detailed simulations performed by
Hasinger et al. (1998) show that in a PSPC survey like ours up to
almost 20\% of the detected sources with $S_x < 1 \times 10^{-14}$ $erg ~cm^{-2} ~s^{-1}$
(corresponding to about 7 sources) may appear at a detected position more
than 15$^{\prime\prime}\,$ away from the true position because of confusion. From these
considerations it follows that only higher resolution, deep X-ray data
(e.g. with AXAF and XMM) can fully clarify the situation for these sources.
\begin{figure}[t]
\epsfysize=9cm
\epsfbox{rosat5_fig7.ps}
\caption[]{ Hardness ratio versus X-ray flux for the sources in the Marano
field (panel a) and in the Lockman field (panel b). The symbols for the various
classes of objects are shown in panel a.
}
\end{figure}
Some additional clues on the nature of the unidentified sources can be
obtained from the analysis of the hardness ratio. Figure 7 shows the
hardness ratio versus X-ray flux for the sources in the Marano
field (panel a) and in the Lockman field (panel b).
Analysis of this figure shows the following results:
i. Most of the sources in both fields occupy well defined bands in HR. On
average, the observed hardness ratio values in the Lockman field are smaller
than those in the Marano field. The reason for this shift in HR is
essentially due to the lower $N_H$ value in the Lockman field, so that the
observed spectra appear softer. When converted to energy spectral indices,
under the assumption of a single power law with the galactic $N_H$,
the HR band in the Marano field corresponds to the range
0.5 $\le \alpha_x \le $ 2.0. Approximately the same range in $\alpha$
is derived for the HR band in the Lockman field assuming $N_H \sim 1 \times
10^{20}$, i.e. approximately twice the value of atomic hydrogen
derived from measurements in the 21 cm line (Lockman et al. 1986).
A similar $N_H$ value was derived
by Hasinger et al. (1993) when fitting the average spectrum
of all the X-ray sources in the Lockman field. This excess
of X-ray absorption is consistent with the column density of ionized gas
in the galactic disk derived from analyses of the dispersion measures
of pulsars (Reynolds 1989). The presence of this additional column of
ionised gas should be taken into account when analyzing ROSAT data,
particularly in directions of low $N_H$ as, for example, in the Lockman field.
ii. In both samples there are about (15--20)\% of the sources which appear
to have hard or absorbed spectra, with HR values close to 1. These sources
appear to be reasonably well separated from the others, especially in the
Lockman field where, because of the higher S/N ratio, the errors in HR are
smaller. The fraction of sources without optical identification is significantly
higher among these hard sources than among sources with ``normal'' X-ray
spectrum. For example, in the Lockman field the unidentified sources
are 3 out of 8 for sources
with HR $>$ 0.6, to be compared with zero out of 42 for sources with
HR $<$ 0.6; in the Marano field the corresponding fractions of unidentified
sources in the same HR ranges are 6/12 and 2/38.
The eleven identifications (Lockman and
Marano fields together) among these hard sources comprise 2 galaxies
(X264--50 and X022--48), and 9 AGNs (4 broad line AGNs, 2 radio galaxies, 1
possible BL Lac, and 2 narrow line objects classified as AGNs on the basis of
the presence of [NeV]$\lambda$3426). This shows that AGNs
are the dominant population also in this range of HR. However, contrary
to what happens for objects with ``normal'' X-ray spectra, a relatively large
fraction of them does not show the prominent broad lines typical of
classical quasars. The absence of broad lines coupled with the
presence of X-ray absorption are consistent with the standard unified models
for AGNs. It is interesting to note that the two narrow line objects
with [NeV]$\lambda$3426 have two of the highest X-ray to optical ratios
in the Lockman field sample. Since the current optical limits for the
unidentified sources in the Marano field correspond to X-ray to optical ratios
similar to those of these sources, it is possible that at least some of our
hard, unidentified sources belong to the same class of objects.
iii. In both samples, but especially in the Marano field, the hard sources tend
to be more numerous at low flux levels. While this is likely to be, at least
in part, a real effect, as shown by the identification content of these
sources discussed above, there is also an obvious selection effect
which favours hard spectra near the detection limit. At these fluxes, because
of the combined effect of confusion and statistical fluctuations on small
number of counts, a not--negligible fraction of the detected sources has
measured fluxes significantly higher than the true fluxes (see Hasinger et al.
1998). Since the sample is defined in the ROSAT hard band, the measured
hardness ratio for some of these sources would be biased toward large
values and therefore the observed fraction of hard sources is probably
higher than the real one.
\subsection{Optically and X-ray selected AGNs}
As mentioned in Section 1, in the same area covered by the X-ray survey we
have conducted in the past years a search for optically selected AGNs, using
multi-colour data from plates taken at the ESO 3.6m telescope.
(Zitelli et al. 1992). This survey has later been extended to fainter
magnitudes using CCD data (Mignoli et al. in preparation) and has produced,
so far,
spectroscopic data for 29 optically selected broad--line AGNs inside
the ROSAT area. Ten of these have not been detected in the X-ray data and
their corresponding X-ray upper limits are shown in Figure 6. Three of these
AGNs have X-ray upper limits outside the band shown in Figure 6,
in the area of very low X-ray to optical ratio.
On the basis of the available CCD and spectroscopic data, which do not cover
the entire ROSAT area of about 0.2 sq.deg., Mignoli and Zamorani (1998)
estimate surface densities of $\sim$ 185 and 140 AGNs per sq.deg. with
$m_B < 22.5$ and $22.5 < m_B < 23.5$, respectively. These estimates
are in good agreement with the predictions obtained by Zamorani (1995)
on the basis of reasonable extrapolations from counts at slightly
brighter magnitudes. Limiting ourselves at the classical broad--line AGNs
and merging together the X-ray and the optically selected samples, we
have a total of 35 such objects with $m_B \le$ 22.6 corresponding to a
surface density of 178 $\pm$ 30 broad--line AGNs per sq.deg.
This is the highest reported surface density for these objects so far
at this magnitude.
Taking into account that, as mentioned above, the optically selected
sample is not complete over the entire ROSAT area, this density
is significantly higher than the estimate of about 125 AGNs
per sq.deg. with $m_B \le 22.6$ recently obtained in the Deep Multicolor Survey,
on the basis of 53 spectroscopically confirmed AGNs (Kennefick et al.
1997). Our estimated surface densities for $m_B < 22.5$ and
$22.5 < m_B < 23.5$, correspond to $36 \pm 6$ and $27 \pm 5$ AGNs inside the
ROSAT area. Since in the same magnitude ranges ROSAT has detected 23
and 6 broad--line AGNs respectively,
we conclude that the ``efficiency'' of AGN selection with X-ray
exposures reaching about $4 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ is
$\sim$ 65\% and $\sim$ 20\% in the two magnitude ranges.
\begin{figure}[t]
\epsfysize=9cm
\epsfbox{rosat5_fig8.ps}
\caption[]
U-B versus B-V for all the AGNs in our field.
The different symbols represent AGNs detected in X-ray and already present
in the optical sample (solid circles), AGNs detected in X-ray and not
present in the optical sample (empty circles), AGNs not detected in X-ray
(triangles). The small dots represent all the point--like objects in our
CCD data with $m_B \le$ 22.5. The dotted curve shows the approximate locus
occupied by stars in this plane.
}
\end{figure}
Having colours for all the X-ray selected AGNs, we can also estimate how many of
them would have been missed by a purely optical selection. Figure 8
shows, with different symbols (see figure caption) all known AGNs in our field.
The small dots represent all the point--like objects in our
CCD data with $m_B \le$ 22.5. The dotted curve shows the approximate locus
occupied by stars in this plane. This figure shows that
two X-ray selected AGNs (X251--37 at z = 2.71 and X042--11 at z = 1.062) are
inside the star locus and therefore would have not been easily selected as AGN
candidates on the basis of this diagram.
Moreover, two X-ray selected AGNs (X233--45 at z = 1.180, and X409--28,
the broad line radio galaxy at z = 0.958) are classified as extended by our
morphological algorithm and two more (X043--12, the AGN2 at z = 2.80, and
X211--22 at z = 0.281) have uncertain classification (i.e. p/e in Table 3).
Since the colour-colour area occupied by faint AGNs contains a number of
extended objects which is about ten times higher than that of the point--like
ones, also these four objects would have not been easily selected as AGN
candidates by purely optical data. Moreover, the location of these objects in
the magnitude -- redshift plane is not the same as that of AGNs which are more
easily selected optically; five have $m_B \ge$ 22.2 and only one has
$m_B <$ 22.2, while the corresponding numbers for all the other AGNs with
redshift in this field are 11 and 25. The same difference is also visible
as a function of the absolute magnitude: five of the objects which would
have been more difficult to detect from the optical data have $M_B \ge$ -22.4
and only one has $M_B <$ -22.4, while the corresponding numbers for all the
other AGNs are 14 and 22. This clearly shows that not only colour--colour
selection of AGN candidates among point--like objects can be significantly
incomplete at faint apparent magnitudes, but also the incompleteness increases
at faint absolute magnitudes. As a consequence, if these not so stellar AGNs
were really missed in faint optical surveys, this would introduce a bias
in the derived optical luminosity function, which would appear flatter
than the real one. Finally, we note also that a not negligible number of
X-ray selected AGNs are significantly redder in B-V than the bulk of AGNs
(see figure 8). Also this has to be taken into account in devising
the most efficient optical selection criteria for faint AGN candidates.
\section{ Conclusion}
We have presented the X-ray data and the optical identifications for
a deep ROSAT PSPC observation in the Marano field. Careful statistical
analysis of multi--colour CCD data in the error boxes of the 50 X-ray sources
detected in the inner region of the ROSAT field (15$^{\prime}\,$ \ radius) has
led to the identification of 42 sources, corresponding to 84\% of the
X-ray sample. These 42 reliable identifications are 33 AGNs (including
two radio galaxies and one BL Lac candidate; 79\% of the identified sources),
2 galaxies, 3 groups or clusters of galaxies and 4 stars. Except for the
higher fraction of unidentified sources, the identification content of this
sample is in excellent agreement with what has been found, at a similar flux
limit, in the much deeper PSPC and HRI surveys in the Lockman field (Hasinger
et al. 1998, Schmidt et al. 1998).
With simple simulations we have shown that in a few cases the reason for
not having found an optical identification can be due to the fact that a
single X-ray source in our list may be produced by
two different close--by sources, so that its detected position may be
significantly wrong. Most of the unidentified sources have a large ratio
of X-ray to optical fluxes and harder than average X-ray spectra. Since
most of the identified objects with these characteristics in our field and
in the Lockman field are AGNs, we conclude that most of the sources with
good position determination but without identification are likely to be AGNs.
Most of the sources in the Marano and Lockman fields occupy well defined
bands in the plane hardness ratio versus X-ray flux. These bands correspond
to the same energy spectral index range 0.5 $\le \alpha_x \le $ 2.0,
only if the effective X-ray absorbing column in the Lockman field is
about twice the value of atomic hydrogen derived from measurements in the
21 cm line. In both samples there are about (15--20)\% of the sources which
appear to have hard or absorbed spectra, with HR values close to 1 and,
especially in the Marano field, they tend to be more numerous at low flux
levels. However, since there is an obvious selection effect which favours hard
spectra near the detection limit, it is difficult to quantify the reality
of this effect. The fraction of sources without optical
identification is significantly higher among these hard sources than among
sources with ``normal'' X-ray spectrum. The eleven identifications (Lockman and
Marano fields together) among these hard sources (9 AGNs and 2 galaxies)
show that AGNs are the dominant population also in this range of HR.
However, contrary to what happens for objects with ``normal'' X-ray spectrum,
a relatively large fraction of them does not show the prominent broad lines
typical of classical quasars.
Finally, comparing the optically and X-ray selected samples of AGNs in the
same area, we estimate that the ``efficiency'' of AGN selection with X-ray
exposures reaching about $4 \times 10^{-15}$ $erg ~cm^{-2} ~s^{-1}$ is
$\sim$ 65\% and $\sim$ 20\% in the magnitude ranges $m_B < 22.5$ and
$22.5 < m_B < 23.5$, respectively. On the other hand, a not negligible
fraction of the X-ray selected AGNs would have not been easily selected
as AGN candidates on the basis of purely optical criteria, either because
of colours similar to those of normal stars or because of morphological
classification not consistent with point--like sources.
Moreover, the location of these objects in the
magnitude -- redshift plane is not the same as that of AGNs which are more
easily selected optically. They tend to be fainter in terms of both apparent
and absolute magnitudes. As a consequence, if these not so stellar AGNs
were really missed in faint optical surveys, this would introduce a bias
in the derived optical luminosity function, which would appear flatter
than the real one. Finally, we note also that a not negligible number of
X-ray selected AGNs are significantly redder in B-V than the bulk of AGNS.
Also this has to be taken into account in devising
the most efficient optical selection criteria for faint AGN candidates.
\begin{acknowledgements}
The ROSAT project is supported by the Bunderministerium f\"ur Forschung und
Technologie (BMFT), by the Science and Engineering Research Council (SERC)
and by the National Aeronautics and Space Administration (NASA). This work
was supported in part by NASA grants NAG5-1531 (M.S.), NAG8-794, NAG5-1649 and
NAGW-2508 (R.B. and R.G). G.H. acknowledges the DARA grant FKZ 50 OR 9403 5;
G.Z. acknowledges partial support by the Italian Space Agency (ASI) under
ASI contract ARS-96-70 and by the Italian Ministry for University and Research
(MURST) under grant Cofin98-02-32.
\end{acknowledgements}
|
\section*{Acknowledgments}
I would like to thank J.H.~K\"uhn for the fruitful collaboration on this
subject and the organizers of WIN 99 for the very nice atmosphere during the
workshop. This work was supported by the {\it Schweizer Nationalfond}.
\section*{References}
|
\section{Introduction}
Many different techniques have been proposed
for the detection of dark matter particles which could
make up the halo of our Galaxy. Among the different
possibilities, the detection of a neutrino flux by
means of neutrino telescopes represents certainly
an interesting tool, which is already at the level of
imposing some (mild) constraint on the particle physics
properties of the neutralino, the most interesting and
studied dark matter candidate \cite{refer}. This
particle is present in all the supersymmetric extensions
of the standard model as a linear combination
of the superpartners of the neutral gauge and higgs fields.
In the present paper we will perform our calculations
in the minimal supersymmetric extension of the standard
model (MSSM), for a definition of which we refer to Ref.\cite{ICTP}
and to the references therein quoted.
\section{Up--going muons from neutralino annihilation in the Earth}
The neutrino flux has origin from neutralino pair annihilation inside
the Earth where these dark matter particles
can be accumulated after having been captured by gravitational
trapping. The differential flux is
\begin{equation}
\Phi^0_{\stackrel{(-)}{\nu_\mu}} (E_\nu) \equiv
\frac{dN_{\stackrel{(-)}{\nu_\mu}}}{dE_\nu} =
\frac{\Gamma_A}{4\pi d^2} \sum_{F,f}
B^{(F)}_{\chi f}\frac{dN_{f {\stackrel{(-)}{\nu_\mu}}}}{dE_\nu} \, ,
\label{eq:fluxnu}
\end{equation}
where $\Gamma_A$ denotes the annihilation rate,
$d$ is the distance of the detector from the source (i.e. the
center of the Earth), $F$ lists the
neutralino pair annihilation final states,
$B^{(F)}_{\chi f}$ denotes the branching ratios into
heavy quarks, $\tau$ lepton and gluons in the channel $F$.
The spectra $dN_{f {\stackrel{(-)}{\nu_\mu}} }/dE_{\nu}$
are the differential distributions of the (anti) neutrinos generated
by the $\tau$ and by hadronization of quarks
and gluons and the subsequent semileptonic decays of the
produced hadrons. For details, see for instance
Refs. \cite{ICTP,noi_nuflux,altri_nuflux}. Here we only
recall that the annihilation rate depends, through
its relation with the capture rate of neutralinos in the Earth,
on some astrophysical parameters, the most relevant of which
is the local density $\rho_l$.
The neutrino flux is produced in the inner part of the Earth \cite{ICTP}
and propagates toward a detector
where it can be detected as a flux of up--going muons,
as a consequence of neutrino--muon conversion inside
the rock that surrounds the detector. A double differential
muon flux can be defined as
\begin{eqnarray}
& & \frac{d^2 N_\mu}{d E_\mu d E_\nu} = \\
& & N_A \int_0^\infty dX \int_{E_\mu}^{E_\nu} d E'_\mu
g(E_\mu, E'_\mu; X) \; S(E_\nu, E_\mu) \nonumber \, ,
\end{eqnarray}
where $N_A$ is the Avogadro's number,
$g(E_\mu,E'_\mu; X)$ is the survival
probability that a muon of initial energy $E'_\mu$
will have a final energy $E_\mu$ after propagating
along a distance $X$ inside the rock and
\begin{equation}
S(E_\nu, E_\mu) = \sum_i
\Phi_i(E_\nu) \frac{d \sigma_i(E_\nu,E'_\mu)}{d E'_\mu}
\end{equation}
where $i = \nu_\mu, \bar\nu_\mu$
and $d \sigma_{{\stackrel{(-)}{\nu_\mu}}} (E_\nu,E'_\mu) / d E'_\mu$ is
the charged current cross--section for the
production of a muon of energy $ E'_\mu$ from
a neutrino (antineutrino) of energy $E_\nu$.
\FIGURE[t]{
\epsfig{figure=yieldx.ps,width=6.5cm,bbllx=50bp,bblly=200bp,bburx=520bp,bbury=650bp,clip=}
\caption{Muon response function $d N_\mu / d\log x$ vs.
the parent neutrino fractional energy $x = E_\nu / m_\chi$ for
neutralino annihilation in the Earth. Different curves refer to
different neutralino masses : $m_\chi = 50$ GeV (solid),
$m_\chi = 80$ GeV (dotted), $m_\chi = 120$ GeV (shot--dashed),
$m_\chi = 200$ GeV (long--dash), $m_\chi = 500$ GeV (dot--dashed).}
}
A useful quantity for the discussion in the following Sections is
the muon response function
\begin{equation}
\frac{d N_\mu}{d E_\nu} = \int_{E^{\mathrm{th}}}^{E_\nu}
d E_\mu \; \frac{d^2 N_\mu}{d E_\mu d E_\nu}
\end{equation}
where $E^{\mathrm{th}}$ is minimal energy for detection
of up--going muons. For SuperKamiokande and MACRO,
$E^{\mathrm{th}} \simeq 1.5$ GeV \cite{oscill_exp}.
The muon response
function indicates the neutrino energy range
that is mostly responsile for the up--going muon signal.
Fig. 1 shows a few examples of it, plotted as
functions of the variable $x = E_\nu/m_\chi$, where
$m_\chi$ denotes the neutralino mass. Fig. 1 shows that
the maximum of the muon reponse happens for neutrino
energies of about $E_\nu \simeq (0.4 - 0.6) \; m_\chi$, with
a half width which extends from $E_\nu \simeq 0.1\; m_\chi$
to $E_\nu \simeq 0.8 \; m_\chi$.
Finally, the total flux of up--going muons is defined as
\begin{equation}
\Phi_\mu = \int_{E^{\mathrm th}}^{m_\chi}
d E_\nu \; \frac{d N_\mu}{d E_\nu}
\end{equation}
\FIGURE[t]{
\epsfig{figure=flux_earth.ps,width=6.5cm,bbllx=50bp,bblly=200bp,bburx=520bp,bbury=650bp,clip=}
\caption{Flux of up--going muons $\Phi_\mu^{\mathrm{Earth}}$ from
neutralino annihilation in the Earth, plotted as a function of
$m_\chi$. The solid line denotes the present upper limit \cite{MACRO}.
Different neutralino compositions are shown with different symbols:
crosses for gauginos, open circles for higgsinos and dots for mixed
neutralinos.}
}
The natural background for these kind of
searches is represented by the flux of up--going
muons originated by the atmospheric neutrino flux.
Experimentally one searches, inside a small angular
cone around the center of the Earth,
for a statistically significant up--going
muon excess over the muons of atmospheric $\nu_\mu$ origin.
No excess has been found so far and therefore, an upper
limit on $\Phi_\mu$ can be derived. Fig. 2 shows the
present most stringent upper limit obtained by the
MACRO Collaboration \cite{MACRO}. In the same figure
the theoretical calculations of $\nu_\mu$ for a scan
of the supersymmetric parameter space are also displayed.
The plot refers to $\rho_l = 0.3$ GeV cm$^{-3}$
and is obtained by a variation of the MSSM parameters
in the following ranges:
$20\;\mbox{GeV} \leq M_2 \leq 500\;\mbox{GeV}$,
$20\;\mbox{GeV} \leq |\mu| \leq 500\;\mbox{GeV}$,
$80\;\mbox{GeV} \leq m_A \leq 1000\;\mbox{GeV}$,
$100\;\mbox{GeV} \leq m_0 \leq 1000\;\mbox{GeV}$,
$-3 \leq {\rm A} \leq +3,\; 1 \leq \tan \beta \leq 50$.
For further details
of the calculation, we refer to Ref. \cite{ICTP}. The comparison
of the scatter plot with the experimental upper limit would
imply that a fraction of the supersymmetric configuration
could be excluded. However, a variation of the value
of $\rho_l$ inside its range of uncertainty can lower the
theoretical prediction by about a factor of 3 \cite{ICTP}. As a consequence,
we can conservatively consider that only a small fraction of the susy
configurations can be potentially in conflict with the experimental
upper limit, when no oscillation effect on the neutrino signal
is assumed.
\section{Neutrino oscillation effect on the up--going muon signal}
The recent data on the atmospheric neutrino deficit indicate that
the $\nu_\mu$ may oscillate, either into $\nu_\tau$ or into
a sterire neutrino $\nu_s$ \cite{oscill_exp,oscill_the}. If this
is the case, also the $\nu_\mu$ produced by neutralino annihilations
would undergo an oscillation process. The energies involved in both
atmospheric and neutralino--produced neutrinos are the same.
The baseline of oscillation of the two neutrino components is
different, since atmospheric neutrinos cross the entire Earth,
while neutrinos produced by neutralino annihilation travel
from the central part of the Earth to the detector
(we recall once more that neutralinos annihilate in the core of the
Earth). On the basis of the features of the $\nu_\mu$ oscillation
which are required to fit the experimental data on atmospheric
neutrinos \cite{oscill_exp,oscill_the}, we expect that also
the neutrino flux from dark matter annihilation would be
affected. In the next Sections we will explicitely discuss
the $\nu_\mu \rightarrow \nu_\tau$ and the
$\nu_\mu \rightarrow \nu_s$ cases, in a two neutrino mixing
scenario \cite{Ellis}.
\subsection{$\nu_\mu \rightarrow \nu_\tau$ vacuum oscillation}
\FIGURE[t]{
\epsfig{figure=psurv_vacuum.ps,width=6.5cm,bbllx=50bp,bblly=200bp,bburx=520bp,bbury=650bp,clip=}
\caption{$\nu_\mu$ survival probability in the case of
$\nu_\mu \rightarrow \nu_\tau$ oscillation. The solid line refers to
$\sin^2 (2\theta_v) = 1$, the dashed line is for $\sin^2 (2\theta_v) = 0.8$.
In both cases, $\Delta m^2 = 5\cdot 10^{-3}$ eV $^{-2}$.}
}
\FIGURE[t]{
\epsfig{figure=flux_vacuum.ps,width=6.5cm,bbllx=50bp,bblly=200bp,bburx=520bp,bbury=650bp,clip=}
\caption{Scatter plot of the ratio
$(\Phi_\mu)^{\rm VAC}_{\rm oscill}/\Phi_\mu$ vs. the neutralino
mass $m_\chi$. $(\Phi_\mu)^{\rm VAC}_{\rm oscill}$ is the up--going
muon flux in the case of $\nu_\mu \rightarrow \nu_\tau$ oscillation,
while $\Phi_\mu$ is the corresponding flux in the case of no oscillation.}
}
In the case of $\nu_\mu \rightarrow \nu_\tau$ oscillation,
the $\nu_\mu$ flux is reduced because of oscillation, but
we have to take into account also that neutralino annihilation
can produce $\nu_\tau$ which in turn can oscillate into
$\nu_\mu$ and contribute to the up--going muon flux.
The $\nu_\tau$ flux can be calculated as discussed in Sect. 1
for the $\nu_\tau$ flux, and it turns out to be always
a relatively small fraction of the $\nu_\mu$ flux.
The muon neutrino flux at the conversion region can therefore
be expressed as
\begin{eqnarray}
\Phi_{{\stackrel{(-)}{\nu_\mu}}} (E_\nu) &=&
\Phi^0_{{\stackrel{(-)}{\nu_\mu}}}\;
P^{\mathrm{vac}} ({{\stackrel{(-)}{\nu_\mu}}} \rightarrow
{{\stackrel{(-)}{\nu_\mu}}}) \nonumber \\
& + &
\Phi^0_{{\stackrel{(-)}{\nu_\tau}}}\;
P^{\mathrm{vac}} ({{\stackrel{(-)}{\nu_\tau}}} \rightarrow
{{\stackrel{(-)}{\nu_\mu}}})
\end{eqnarray}
where the vacuum survival probability is
\begin{eqnarray}
& & P^{\mathrm{vac}} ({{\stackrel{(-)}{\nu_\mu}}} \rightarrow
{{\stackrel{(-)}{\nu_\mu}}})
\;\; = \;\; \\
& & 1 - \sin^2(2\theta_v)\sin^2
\left (
\frac{1.27 \Delta m^2 (\mathrm{eV}^2) R(\mathrm{Km})}
{E_\nu (\mathrm{GeV})} \nonumber
\right )
\label{vac}
\end{eqnarray}
where $\Delta m^2$ is the mass square difference of the
two neutrino mass eigenstates, $\theta_v$ is the mixing angle
in vacuum and $R$ is the Earth's radius.
Fig. 3 shows the survival probability for two different values of
the neutrino oscillation parameters. Smaller (larger) values of
$\Delta m^2$ have the effect of shifting the curves to the left (right).
Comparing Fig. 1 with Fig. 3, we notice that the reduction of the
up--going muon flux is stronger when there is matching between the
the energy
$E_\nu^1 \simeq 5.2 \cdot 10^{-3} \Delta m^2 (\mathrm{eV}^2)$
of the first (from the right) minimum of the
survival probability and the energy
$E_\nu \simeq 0.5 m_\chi$
which is responsible for most of the muon response in the detector.
This implies that a maximum reduction of the signal could
occur for neutralino masses of the order of
$m_\chi (\mathrm{GeV}) \simeq 10^4 \Delta m^2 (\mathrm{eV}^2)$.
The $\nu_\tau \rightarrow \nu_\mu$ oscillation makes the
reduction of the muon flux less severe, but it is not able
to completely balance the reduction effect because the
original $\nu_\tau$ flux at the source is sizeably smaller than
the $\nu_\tau$ flux. Therefore, the overall effect of
the neutrino oscillation is to reduce the up--going muon
signal. This effect is summarized in Fig. 4, where the
ratio between the up--going muon signals in the presence
and in the absence of oscillation are plotted as a function
of the neutralino mass. The susy parameter space has been
varied in the same ranges quoted for Fig. 2. We notice
that the strongest effect is present for light neutralinos,
since in this case the muon flux is mostly produced from
neutrinos whose energy is in the range of maximal suppression
for the oscillation phenomenon. The effect is between
0.5 and 0.8 for $m_\chi \lsim 100$ GeV.
On the contrary, the fluxes for larger masses are less
affected, and the reduction is less than about
20\% for $m_\chi \gsim 200$ GeV.
\subsection{$\nu_\mu \rightarrow \nu_s$ matter oscillation}
\FIGURE[t]{
\epsfig{figure=psurv_matter.ps,width=6.5cm,bbllx=50bp,bblly=200bp,bburx=520bp,bbury=650bp,clip=}
\caption{$\nu_\mu$ survival probability in the case of
$\nu_\mu \rightarrow \nu_s$ oscillation, for $\sin^2 (2\theta_v) = 0.8$
and $\Delta m^2 = 5\cdot 10^{-3}$ eV $^{-2}$. The solid line refers to
neutrinos, the dashed line is for antineutrinos.}
}
\FIGURE[t]{
\epsfig{figure=flux_matter.ps,width=6.5cm,bbllx=50bp,bblly=200bp,bburx=520bp,bbury=650bp,clip=}
\caption{Scatter plot of the ratio
$(\Phi_\mu)^{\rm MAT}_{\rm oscill}/\Phi_\mu$ vs. the neutralino
mass $m_\chi$. $(\Phi_\mu)^{\rm MAT}_{\rm oscill}$ is the up--going
muon flux in the case of $\nu_\mu \rightarrow \nu_s$ oscillation,
while $\Phi_\mu$ is the corresponding flux in the case of no oscillation.}
}
In the case of $\nu_\mu \rightarrow \nu_s$ oscillation,
the neutrino flux is simply
\begin{equation}
\Phi_{{\stackrel{(-)}{\nu_\mu}}} (E_\nu) =
\Phi^0_{{\stackrel{(-)}{\nu_\mu}}}\;
P^{\mathrm{mat}} ({{\stackrel{(-)}{\nu_\mu}}} \rightarrow
{{\stackrel{(-)}{\nu_\mu}}})
\end{equation}
and no $\nu_\mu$ regeneration is possible from the
sterile neutrino. In this case, the effective potential
of $\nu_\mu$ and $\nu_s$ inside the Earth are different
and we have to solve the evolution equation for propagation
in the core and in the mantle. Neutrinos
(produced in the center of the Earth) cross once half
of the core and once the mantle. By considering both
core and mantle as of constant density, we can express the
survival probability as \cite{akhmedov, Kim}
\begin{eqnarray}
& & P^{\mathrm{mat}} ({{\stackrel{(-)}{\nu_\mu}}} \rightarrow
{{\stackrel{(-)}{\nu_\mu}}}) \,\,=\,\, \\
& & \left [
U(\theta_c) D(\phi_c) U^\dagger (\theta_c - \theta_m)
D(\phi_m) U^\dagger (\theta_m)
\right ]_{\mu\mu} \nonumber
\label{mat}
\end{eqnarray}
where $U$ is the $2 \times 2$
neutrino mixing matrix, $\theta_a$
($a=c,m$ for core and mantle, respectively) are
the effective mixing angles in matter and they are related
to the vacuum mixing angle $\theta_v$ as
\begin{equation}
\sin^2 (2\theta_a) = \frac{\sin^2 (2\theta_v) \xi_a^2}
{\left [
{(\xi_a \cos(2\theta_v) + 1)^{2} +
\xi_a^2 \sin^2 (2\theta_v)}
\right ]}
\end{equation}
with $\xi_a = \Delta m^2 / (2E_\nu V_a)$;
$V_a = \pm G_F N_n^a / \sqrt{2}$ is the matter
potential in a medium of number density $N_n^a$
for neutrinos ($+$) and antineutrinos ($-$).
In Eq.(3.4), $D$ is the evolution
matrix $D_{ij}(\phi_a) = \delta_{ij} d_j^a$,
where $d_1^a = 1$, $d_2^a = \exp(i \phi_a)$ and
\begin{equation}
\phi_a = V_a R_a
\left [
(\xi_a \cos(2\theta_v) + 1)^2 +
\xi_a^2 \sin^2 (2\theta_v)
\right ]^{1/2}
\end{equation}
In Fig. 5 an example of the $\nu_\mu$ and $\bar \nu_\mu$
survival probability is given for representative
values in the range allowed by the fits on the
atmospheric neutrino data \cite{oscill_exp,oscill_the}:
$\Delta m^2 = 5\cdot 10^{-3}$ eV$^2$
and $\sin^2 (2\theta_v) = 0.8$. {}From Fig. 5
and the previous discussion relative to Fig. 3, we
expect that in the case of $\nu_\mu \rightarrow \nu_s$
the reduction of the muon signal is significantly
less severe than in the case of $\nu_\mu \rightarrow \nu_\tau$.
In fact, in these case the minima of the survival probability
occur for lower neutrino energies, and threfore the
oscillation can affect only muon fluxes originated by very
light neutralinos. This is manifest in Fig. 6, where
the ratio of the up--going muon fluxes in presence and
absence of oscillation are shown. In this case, the reduction
of the signal is always less than 30\%. This maximal reduction
occurs for neutralino masses lower that about 80 GeV. For
larger masses, the up--going muon flux is almost unaffected.
\section{Conclusions}
We have discussed the effect on the up--going muon
signal from neutralino annihilation in the Earth, in the
case that the $\nu_\mu$ flux produced by neutralinos would oscillate
as indicated by the data on the atmospherice neutrino deficit.
While the experimental upper limit is, at present, practically
not affected by the possibility of neutrino
oscillation \cite{MACRO}, the
theoretical predictions are reduced in the presence of
oscillation. With the oscillation parameters deduced
{}from the fits on the atmospheric neutrino data,
the effect is always larger for lighter neutralinos.
In the case of $\nu_\mu \rightarrow \nu_\tau$
the reduction is between 0.5 and 0.8 for
$m_\chi \lsim 100$ GeV and less than about
20\% for $m_\chi \gsim 200$ GeV.
In the case of $\nu_\mu \rightarrow \nu_s$,
the reduction of the signal is up to 30\% for neutralino
masses lower that about 80 GeV and smaller than 10\% for
heavier neutralinos.
\acknowledgments
I wish to thank Sandro Bottino for very stimulating and
interesting discussions about the topic of this paper.
This work was supported by DGICYT under grant number
PB95--1077 and by the TMR network grant ERBFMRXCT960090 of
the European Union.
|
\section*{Acknowledgments}
We owe a lot to M.Mezard and we are very happy to thanks him.
We are also very grateful to W.Kob and F.Sciortino for their suggestions and
to A.Cavagna and I.Giardina for useful discussions.
B.C. would like to thank the Physics Department of Rome University
'La Sapienza', where this work was partially developed during her PhD.
\end{section}
\newpage
|
\section{Introduction}
For many dissipative systems one has the exponential time decay
of correlations. This result was established for various models
and by using various approximations, see for ex. \cite{Loui}.
For certain models, in particular for the spin-boson Hamiltonian,
also a regime with the oscillating behavior was found
\cite{Leggett1}, \cite{Leggett2}, \cite{AcKoVol}.
The presence of such a regime is very important in the ivestigation
of quantum decoherence.
The aim of this note is to show that for the model of particle interacting
with quantum field, in particular for the polaron model,
one can have not only the standard exponential decay
but also the non-exponential decay (as some powers of time) of correlations.
We investigate the model describing interaction of non-relativistic particle
with quantum field.
This model is widely studied in elementary particle physics,
solid state physics, quantum optics, see for example
\cite{Bog}-\cite{Feynman}.
We consider the simplest case in which matter is represented by a
single particle, say an electron, whith position and momentum
$q=(q_{1}, q_{2}, q_{3})$ and
$p=(p_{1}, p_{2}, p_{3})$ satisfying the commutation relations
$ [q_{j}, p_{n} ] = i \delta_{jn}$. The electromagnetic field is described by
boson operators
$a(k)=(a_{1}(k), a_{2}(k), a_{3}(k));a^{\dag}(k)=
(a^{\dag}_{1}(k), \ldots ,a^{\dag}_3(k))$
satisfying the canonical commutation relations
$[a_{j}(k),a_{n}^{\dag}(k')]=\delta_{jn}\delta(k-k')$.
The Hamiltonian of a free non relativistic atom interacting with a quantum
electromagnetic field is
\begin{equation}\label{1.1}
H=H_0+\lambda H_I=\int\omega(k)a^{\dag}(k)a(k)dk+{\frac{1}{2}}\,p^2+\lambda
H_I
\end{equation}
where $\lambda$ is a small constant, $\omega (k)$ is the dispersion law
of the field,
\begin{equation}\label{1.2}
H_I=\int d^3k\left(g(k)
p\cdot a^{\dag}(k) e^{-ikq} +
\overline g(k)p\cdot a(k)e^{ikq} \right) + h.c.
\end{equation}
Here $p\cdot a(k)=\sum_{j=1}^{3}p_j a_j(k)$, $p^2=\sum_{j=1}^{3}p_j^2$,
$a^{\dag}(k)a(k)=\sum_{j=1}^{3}a^{\dag}_j(k)a_j(k)$, $kq=\sum_{j=1}^{3}k_j
q_j$.
For the polaron model the Hamiltonian has the form
$$
H=\int\omega(k)a^{\dag}(k)a(k)dk+{\frac{1}{2}}\,p^2+\lambda
\int d^3k\left(g(k)
a^{\dag}(k) e^{-ikq} +
\overline g(k) a(k)e^{ikq} \right)
$$
It is different from (\ref{1.1}), (\ref{1.2}) by a momentum $p$
in the interaction Hamiltonian. For the analysis of this paper
this difference is not important.
In the present paper we will use the method
for the approximation of the quantum mechanical evolution
that is called the stochastic limit method, see for example
\cite{AcKoVol}, \cite{VanHove}-\cite{AcLuVo}.
The general idea of the stochastic limit is to make the time rescaling
$t\to t/\lambda^2$ in the solution of the Schr\"o\-din\-ger equation in
interaction picture $U^{( \lambda )}_t=e^{itH_0} e^{-itH}$,
associated to the Hamiltonian $H$, i.e.
$$
{ \frac{\partial}{ \partial t} } U^{( \lambda )}_t
=-i { \lambda } H_I (t) \ U_t^{(\lambda )}
$$
with $H_I(t)=e^{it H_0}H_Ie^{-itH_0}$.
We get the rescaled equation
$$
{ \frac{\partial}{ \partial t} } U^{( \lambda )}_{t/\lambda^2}
=- {\frac{i}{\lambda}} H_I (t/\lambda^2) \
U_{t/\lambda^2}^{(\lambda )}
$$
and one wants to study the limits, in a topology to be specified,
\begin{equation}\label{1.6}
\lim_{\lambda\to0}U^{(\lambda)}_{t/\lambda^2}= U_t;\qquad
\lim_{\lambda\to0}{\frac{1}{\lambda}}\,H_I\left({\frac{t}{\lambda^2}}\right)
=H_t
\end{equation}
We will prove that $U_t$ is the solution of the equation
\begin{equation}\label{1.7}
\partial_t U_t\,=-iH_t U_t\quad;\qquad U_0=1
\end{equation}
The interest of this limit equation is in the fact that many problems
become explicitly integrable.
The stochastic limit of the model (\ref{1.1})-(\ref{1.2})
has been considered in \cite{AcLu}, \cite{AcLuVo}, \cite{Gou96},
\cite{Ske96}, \cite{q-alg}, \cite{tmf}.
After the rescaling $t\to t/\lambda^2$ we consider the
simultaneous limit $\lambda\to 0$, $t\to\infty$ under the condition
that $\lambda^2 t$ tends to a constant (interpreted as a new {\it slow scale}
time). This limit captures the main
contributions to the dynamics in a regime, of {\it long times and small
coupling\/} arising from the cumulative effects, on a large time scale,
of small interactions ($\lambda\to 0$). The physical idea is that,
looked from the slow time scale of the atom, the field looks like a very
chaotic object: a {\it quantum white noise}, i.e. a $\delta$-correlated
(in time) quantum field $b_j^{\dag}(t,k), b_j(t,k)$ also called a {\it master
field}. If one introduces the dipole approximation the master field is the
usual boson Fock white noise. Without the dipole approximation the master
field is described by a new type of commutation relations
of the following form
\cite{AcLuVo}
\begin{equation}\label{1.8}
b_j(t,k)p_n=(p_n-k_n)b_j(t,k)
\end{equation}
\begin{equation}\label{1.9}
b_j(t,k)b_n^{\dag}(t',k')=2\pi\delta(t-t')
\delta\left(\omega(k)-kp+\frac{1}{2}k^2\right)
\delta(k-k')\delta_{jn}
\end{equation}
Such quantum white noises can be treated as an operator deformation of quantum
Boltzmann commutation relations.
Recalling that $p$ is the particle momentum, we see that
the relation (\ref{1.8})
shows that the particle and the master field are not independent even at a
kinematical level. This is what we call {\it entanglement}.
The relation (\ref{1.9}) is a generalization of the algebra of free
creation--annihilation operators with commutation relations
$$A_iA^{\dag}_j=\delta_{ij}$$
and the corresponding statistics becomes a generalization of the
Boltzmannian (or Free) statistics. This generalization is due to
the fact that the right hand side is
not a scalar but an operator (a function of the atomic momentum). This
means that the relations (\ref{1.8}), (\ref{1.9}) are {\it module commutation relations}.
For any fixed value $\bar p$ of the atomic momentum we get a copy of the
free (or Boltzmannian) algebra. Given the relations
(\ref{1.8}), (\ref{1.9}), the
statistics of the master field is uniquely determined by the condition
$$
b_j(t,k) \Psi = 0
$$
where $\Psi $ is the vacuum of the master field, via a module
generalization of the free Wick theorem, see \cite{q-alg}.
In Section 2 the dynamically $q$-deformed commutation relations
(\ref{2.5}), (\ref{2.6}), (\ref{2.4}) are
obtained and the stochastic limit for collective operators is evaluated.
In Section 3 the stochastic limit of the evolution equation is found.
In Section 4 the non-exponential decay for vacuum vector
in the polaron model is investigated.
\section{Deformed commutation relations}
In this section we reproduce in the brief form the notations and
the main results of the work \cite{q-alg}.
In order to determine the limit (\ref{1.6}) one rewrites
the rescaled interaction Hamiltonian in terms of some rescaled fields
$a_{\lambda,j}(t,k)$:
$$
{\frac{1}{\lambda}}\,H_I\left({\frac{t}{\lambda^2}}\right)=
\int d^3k p(\overline g(k) a_\lambda(t,k)+ g(k)a^{\dag}_\lambda(t,k)) + h.c.
$$
where
$$
a_{\lambda,j}(t,k):={\frac{1}{\lambda}}\,e^{i{\frac{t}{\lambda^2}}\,H_0}
e^{ikq}a_j(k)e^{-i{\frac{t}{\lambda^2}}\,H_0}=
\frac{1}{\lambda}e^{-i{\frac{t}{\lambda^2}}\,
\left(\omega(k)-kp+\frac{1}{2}k^2\right)}e^{ikq}a_j(k)
$$
It is now easy to prove
that operators $a_{\lambda,j}(t,k)$ satisfy the following $q$--deformed
module relations,
$$
a_{\lambda,j}(t,k)a^{\dag}_{\lambda,n}(t',k')=
$$
\begin{equation}\label{2.5}
=a^{\dag}_{\lambda,n}(t',k')a_{\lambda,j}(t,k)
\cdot q_\lambda(t-t',kk')+
{\frac{1}{\lambda^2}}\,q_\lambda\left(t-t',\omega(k)-kp+\frac{1}{2}k^2\right)
\delta(k-k')\delta_{jn}
\end{equation}
\begin{equation}\label{2.6}
a_{\lambda,j}(t,k)p_n=(p_n-k_n)a_{\lambda,j}(t,k)
\end{equation}
where
\begin{equation}\label{2.3}
q_\lambda(t-t',x)=e^{-i{\frac{t-t'}{\lambda^2}}\,x}
\end{equation}
is an oscillating exponent.
This shows that the module $q$--deformation of the commutation relations arise
here as a result of the dynamics and are not put artificially {\it ab
initio}.
For a discussion of $q$-deformed commutation relations see for example
\cite{AreVo}.
Now let us suppose that the master field
\begin{equation}\label{2.7}
b_j(t,k)=\lim_{\lambda\to0}a_{\lambda,j}(t,k)
\end{equation}
exist. Then it is natural to conjecture that its algebra shall be
obtained as the stochastic limit ($\lambda\to0$) of the
algebra (\ref{2.5}), (\ref{2.6}).
Notice that
the factor $q_\lambda(t-t',x)$ is an oscillating exponent and one easily sees
that
\begin{equation}\label{2.8}
\lim_{\lambda\to0}q_\lambda(t,x)=0\ ,
\qquad\lim_{\lambda\to0}{\frac{1}{\lambda^2}}\,
q_\lambda(t,x)=2\pi\delta(t)\delta(x)
\end{equation}
Thus it is natural
to expect that the limit of (\ref{2.6})
is
\begin{equation}\label{2.9}
b_j(t,k)p_n=(p_n-k_n)b_j(t,k)
\end{equation}
and the limit of (\ref{2.5}) gives the module free relation
\begin{equation}\label{2.10}
b_j(t,k)b_n^{\dag}(t',k')=2\pi\delta(t-t')
\delta\left(\omega(k)-kp+\frac{1}{2}k^2\right)
\delta(k-k')\delta_{jn}
\end{equation}
Operators $a_{\lambda,j}(t,k)$ also obey the relation
\begin{equation}\label{2.4}
a_{\lambda,j}(t,k)a_{\lambda,n}(t',k')=a_{\lambda,n}(t',k')a_{\lambda,j}(t,k)
q_\lambda^{-1}(t-t',kk')
\end{equation}
In what follows we will not write indexes $j$, $n$ explicitly.
The
relation (\ref{2.4}) should disappear after the limit, see \cite{q-alg}.
In fact, if the relation
(\ref{2.4}) would survive in the limit then,
because of (\ref{2.8}), it should give
$b(t,k)b(t',k')=0$, hence also $b^{\dag}(t,k)b^{\dag}(t',k')=0$, so all the
$n$--particle vectors with $n\geq2$ would be zero.
\section{ Evolution equation}
Let us find stochastic differential equation for the model we consider.
In the introduction we claimed that the stochastic limit for the
Shr\"odinger equation in interaction picture will have the form (\ref{1.7}):
$\partial_t U_t\,=-iH_t U_t$.
But in this equation both $H_t$ and $U_t$ are distributions.
We need to regularize this product of distributions.
In the present section we will make the following regularization:
roughly speaking we replace $H_t$ by $H_{t+0}+const$.
We investigate the evolution operator in interaction picture
$U^{(\lambda)}_t$.
We start with the equation
$$U^{(\lambda)}_{t+dt}=\biggl(1+(-i\lambda)\int^{t+dt}_tH_I(t_1)
dt_1+
$$
$$
+(-i\lambda)^2\int^{t+dt}_tdt_1\int^{t_1}_tdt_2H_I(t_1)
H_I(t_2)+\dots\biggr)U^{(\lambda)}_t
$$
where $dt>0$.
We get for $dU^{(\lambda)}_t=U^{(\lambda)}_{t+dt}-U^{(\lambda)}_t$
$$
dU^{(\lambda)}_t=\biggl((-i\lambda)\int^{t+dt}_tH_I(t_1)dt_1+
(-i\lambda)^2\int^{t+dt}_tdt_1\int^{t_1}_tdt_2H_I(t_1)H_I(t_2)+
\dots\biggr)U^{(\lambda)}_t
$$
Let us make the rescaling $t\to t/\lambda^2$ in this perturbation theory
series. We get
$$dU^{(\lambda)}_{t/\lambda^2}=\biggl((-i)\int^{t+dt}_tdt_1{1\over
\lambda}\,H_I\left({t_1\over\lambda^2}\right)+
$$
\begin{equation}\label{rescalseries}
+(-i)^2\int^{t+dt}_t
dt_1\int^{t_1}_tdt_2\,{1\over\lambda}\,H_I\left({t_1\over\lambda^2}
\right)\,{1\over\lambda}\,
H_I\left({t_2\over\lambda^2}\right)+\dots\biggr)U^{(\lambda
)}_{t/\lambda^2}
\end{equation}
To find the stochastic differential equation we need
to collect all the terms of
order $dt$ in the perturbation theory
series (\ref{rescalseries}).
Terms of order $dt$ are contained only in the first two terms of these
series. Let us investigate the first two terms.
For the first term of the perturbation theory we get
\begin{equation}\label{first_term}
\int^{t+dt}_tdt_1{1\over\lambda}\,H_I\left({t_1\over\lambda^2}
\right)=
\int^{t+dt}_tdt_1
\int dk\left(\overline g(k)(2p+k)a_\lambda(t_1,k)+g(k)a^{\dag}_\lambda
(t_1,k)(2p+k)\right)
\end{equation}
In the stochastic limit $\lambda\to0$ this term gives us
$$\int dk\left(\overline g(k)(2p+k)dB(t,k)+g(k)dB^{\dag}(t,k)(2p+k)\right)
$$
where the stochastic differential $dB(t,k)$
is the stochastic limit of the field $a_\lambda(t,k)$
in the time interval $(t,t+dt)$:
$$
dB(t,k)=\lim_{\lambda\to 0}
\int_{t}^{t+dt}d\tau\, a_\lambda(\tau,k)=\int_{t}^{t+dt}d\tau\, b(\tau,k)
$$
We will prove that the stochastic differential
$dB(t,k)$ and the evolution operator
$U_t$ are free
independent. In the bosonic case independence would result in the relation
$[dB(t,k), U_t]=0$.
From this relation follows that
$\langle X\, dB(t,k) U_t\rangle =0$ for arbitrary observable $X$.
In the case of Boltzmannian statistics we get the same relation:
the (free) independence
means that roughly speaking $dB(t,k)$ kills all creations in $U_t$.
We have the following
{\it Lemma.\qquad}{\sl
The stochastic differental $dB(t,k)$
and the evolution operator $U_t$ are free independent.
This means that for an arbitrary observable $X$
$$
\langle X\, dB(t,k) U_t\rangle =0\quad \forall X
$$
Here $\langle\cdot\rangle$ is the stochastic limit of the
vacuum expectation of boson field
(that acts as a conditional expectation on momentum of quantum particle $p$).
}
We will prove this result by analizing of the perturbation theory series.
We have
$$
\langle X\, dB(t,k) U_t\rangle=\lim_{\lambda\to 0}
\langle X_{\lambda} \int_{t}^{t+dt}d\tau\, a_\lambda(\tau,k)
\biggl(1+(-i)\int^{t}_0 dt_1\, {1\over\lambda}\,
H_I\left({t_1\over\lambda^2}\right)
+ \dots+
$$
$$ +(-i)^n\int^{t}_0 dt_1\dots \int^{t_{N-1}}_0 dt_N\,
{1\over\lambda}\,H_I\left({t_1\over\lambda^2}\right)\dots
{1\over\lambda}\,H_I\left({t_n\over\lambda^2}\right)+\dots
\biggr)\rangle
$$
where ${1\over\lambda}\,H_I\left({t_k\over\lambda^2}\right)$
is given by the formula (\ref{first_term}).
Here $\lim_{\lambda\to 0}X_{\lambda}=X$.
Let us analize the $N$-th term of perturbation theory.
The $N$-th term of perturbation theory
is the linear combination of the following terms
(we omit integration over $k$, $k_n$)
$$
\langle X_{\lambda} \int_{t}^{t+dt}d\tau\, a_\lambda(\tau,k)
\int^{t}_0 dt_1\dots \int^{t_{N-1}}_0 dt_N\,
a^{\varepsilon_1}_\lambda(t_1,k_1)\dots
a^{\varepsilon_N}_\lambda(t_N,k_N)\rangle
$$
Let us shift $a_\lambda(\tau,k)$ to the right using dynamically $q$-deformed
relations. In the following we will use notions of the work \cite{q-alg}.
Let us enumerate annihilators in the product
$
a^{\varepsilon_1}_\lambda(t_1,k_1)\dots
a^{\varepsilon_N}_\lambda(t_N,k_N)
$
as $a_\lambda(t_{m_j},k_{m_j})$, $j=1,\dots J$, and enumerate
creators as $a^{\dag}_\lambda(t_{m'_j},k_{m'_j})$, $j=1,\dots I$, $I+J=N$.
This means that if $\varepsilon_m=0$ then
$a^{\varepsilon_m}_\lambda(t_m,k_m)=a_\lambda(t_{m_j},k_{m_j})$
for $m=m_j$ (and the analogous condition for $\varepsilon_m=1$).
We will use the following recurrent relation for correlator
(analogous formula was proved in the work \cite{q-alg}):
$$
\langle X_{\lambda} \int_{t}^{t+dt}d\tau\,
\int^{t}_0 dt_1\dots \int^{t_{N-1}}_0 dt_N\,
a_\lambda(\tau,k)
a^{\varepsilon_1}_\lambda(t_1,k_1)\dots
a^{\varepsilon_N}_\lambda(t_N,k_N)\rangle=
$$
$$
=
\sum_{j=1}^{I}\delta ( k- k_{m'_j})
\langle X_{\lambda} \int_{t}^{t+dt}d\tau\,
\int^{t}_0 dt_1\dots \int^{t_{N-1}}_0 dt_N\,
a^{\varepsilon_1}_\lambda(t_1,k_1)\dots
{\widehat a_\lambda^{\dag}(t_{m'_j},k_{m'_j})} \dots
a^{\varepsilon_N}_\lambda(t_N,k_N)\rangle
$$
$$
\frac{1}{\lambda^2}
q_{\lambda}\left( \tau - t_{m'_j},\omega (k) -k p +\frac{1}{2}k^2\right)
\prod_{m_i>m'_j}
q_{\lambda}^{-1}\left( \tau - t_{m'_j},k k_{m_i}\right)
\prod_{m'_i>m'_j}
q_{\lambda}\left( \tau - t_{m'_j},k k_{m'_i}\right)
$$
\begin{equation}\label{recur}
\prod_{m_i<m'_j}q_{\lambda}^{-1}(\tau - t_{m_i},k k_{m_i})
\prod_{m'_i<m'_j}q_{\lambda}(\tau - t_{m'_i},k k_{m'_i})
\end{equation}
Here the notion ${\widehat a_\lambda^{\dag}}$ means that we omit the operator
$a_\lambda^{\dag}$ in this product.
The right hand side of the equation (\ref{recur}) is equal to
$$
\sum_{j=1}^{I}\delta ( k- k_{m'_j})
\langle X_{\lambda}
\int^{t}_0 dt_1\dots \int^{t_{N-1}}_0 dt_N\,
a^{\varepsilon_1}_\lambda(t_1,k_1)\dots
{\widehat a_\lambda^{\dag}(t_{m'_j},k_{m'_j})} \dots
a^{\varepsilon_N}_\lambda(t_N,k_N)\rangle
$$
$$
\frac{1}{\lambda^2}
q_{\lambda}\left( - t_{m'_j},\omega (k) -k p +\frac{1}{2}k^2\right)
\prod_{m_i>m'_j}
q_{\lambda}^{-1}\left( - t_{m'_j},k k_{m_i}\right)
\prod_{m'_i>m'_j}
q_{\lambda}\left( - t_{m'_j},k k_{m'_i}\right)
$$
$$
\prod_{m_i<m'_j}q_{\lambda}^{-1}( - t_{m_i},k k_{m_i})
\prod_{m'_i<m'_j}q_{\lambda}( - t_{m'_i},k k_{m'_i})
$$
$$
\int_{t}^{t+dt}d\tau\,
q_{\lambda}\left( \tau,\omega (k) -k p +\frac{1}{2}k^2\right)
\prod_{m_i>m'_j}
q_{\lambda}^{-1}\left( \tau ,k k_{m_i}\right)
\prod_{m'_i>m'_j}
q_{\lambda}\left( \tau ,k k_{m'_i}\right)
$$
$$
\prod_{m_i<m'_j}q_{\lambda}^{-1}(\tau ,k k_{m_i})
\prod_{m'_i<m'_j}q_{\lambda}(\tau ,k k_{m'_i})
$$
(we use that $q_{\lambda}$ is an exponent). The first three lines of this
formula do not depend on $\tau$ and the last two lines do not depend
on $t_1,\dots,t_N$. Therefore the stochastic limits for these values
can be made independently (the limit of product is equal to the
product of limits). It is easy to see that the stochastic limit for
the multiplier that depends on $\tau$ (of the last two lines)
is equal to zero. This finishes the proof of the Lemma.
The second term of the perturbation theory series is equal (up to terms
of order $(dt)^2$)
$$\int^{t+dt}_tdt_1\int^{t_1}_tdt_2{1\over\lambda}\,H_I
\left({t_1\over\lambda^2}\right){1\over\lambda}\,H_I\left({t_2
\over\lambda^2}\right)=$$
$$=\int^{t+dt}_tdt_1\int^{t_1}_tdt_2\int dk|g(k)|^2(2p+k)^2
{1\over\lambda^2}\,e^{-i{t_1-t_2\over\lambda^2}\,\left(\omega(k)-
kp+{1\over2}\,k^2\right)}$$
due to $q$--module relations on $a_\lambda(t,k)$, $p$.
Performing integration over $t_1$, $t_2$ and using the formulas
$$
\int_{t}^{t+dt}dt_1\,\int_{t}^{t_1}dt_2\,
{1\over\lambda^2}\, e^{-i\frac{t_1-t_2}{\lambda^2}x}=
\int_{t}^{t+dt}dt_1\, \int_{-t_1/\lambda^2}^{0}d\tau\,e^{i\tau x}
$$
$$
\int^0_{-\infty}dte^{itx}=\frac{-i}{x-i0}
=\pi\delta(x)-i\ P.P.{1\over x}$$
we get for the second term
$$
-i dt\int dk|g(k)|^2(2p+k)^2
{1\over\omega(k)-kp+{1\over2}\,k^2-i0}
$$
Let us denote
$$(g|g)_-(p)=
-i \int dk|g(k)|^2(2p+k)^2
{1\over\omega(k)-kp+{1\over2}\,k^2-i0}=
$$
$$
=\int dk|g(k)|^2(2p+k)^2\left(\pi\delta\left(\omega(k)-kp+{1
\over2}\,k^2\right)-i\ P.P.{1\over\omega(k)-kp+{1\over2}\,k^2}\right)
$$
Combining all the terms of order $dt$
we get the following result:
{\it Theorem.}{\sl\qquad
The stochastic differential
equation for $U_t=\lim\limits_{\lambda\to0}U^{(\lambda)}_{t/\lambda^2}$
have a form
\begin{equation}\label{stocheq}
{dU_t}\,=\biggl(-i\int dk\left(\overline g(k)(2p+k)dB(t,k)+g
(k)dB^{\dag}(t,k)(2p+k)\right)-dt\,(g|g)_-(p) \biggr)U_t
\end{equation}
}
The equation (\ref{stocheq}) can be rewritten in the language of
distributions as
\begin{equation}\label{distrib}
\frac{dU_t}{dt}\,=\biggl(-i\int dk\left(\overline g(k)(2p+k)b(t,k)+g
(k)b^{\dag}(t,k)(2p+k)\right)-\,(g|g)_-(p) \biggr)U_t
\end{equation}
Here we uderstand the singular product of distributions $b(t,k)U_t$
in the sense that (\ref{distrib}) is equivalent to (\ref{stocheq}).
We have to stress that
$dB(t,k)=\int_{t}^{t+dt}d\tau\, b(\tau,k)\ne b(t,k)dt$
and we can not obtain (\ref{distrib}) dividing (\ref{stocheq})
by $dt$.
\section{Non-exponential decay}
Let us investigate the behavior of $\langle U_t\rangle$ using
stochastic differential equation (\ref{stocheq}). We get
$$
\langle dU_t\rangle=\langle\biggl((-i)\int dk\overline
g(k)(2p+k)dB(t,k)-dt\,(g|g)_-(p) \biggr)U_t\rangle
$$
Using the free independence of $dB(t,k)$ and $U_t$ we get
$$
{d\over dt}\,\langle U_t\rangle=\langle{d\over dt}\,U_t\rangle=-
(g|g)_-(p)\langle U_t\rangle
$$
Because $U_0=1$, we have the solution
$$\langle U_t\rangle=e^{-t(g|g)_-(p)}$$
In this section we calculate the matrix element $\langle X|U_t|X\rangle$
where $X=f(p)\otimes\Phi$ in the momentum representation and $\Phi$ is
the vacuum vector for the master field. This matrix element is equal to
\begin{equation}\label{smeared}
\langle X|U_t|X\rangle=\int dp|f(p)|^2e^{-t(g|g)_-(p)}
\end{equation}
We investigate the polaron model when $\omega(k)=1$.
For this choice of $\omega(k)$ we get
$$
\omega(k)-kp+\frac{1}{2}k^2=1-\frac{1}{2} p^2 +\frac{1}{2}(k-p)^2
$$
One can expect non-exponential relaxation when
$\hbox{ supp }f(p)\subset \{|p|<\sqrt{2}\}$.
In this case $\hbox{Re}(g|g)_-(p)=0$ and there is no dumping.
All decay in this case is due to interferention.
We will use the approximation
$\hbox{diam supp }g(k)\gg\hbox{ diam supp }f(p)$.
Physically this means that the particle is more localized in momentum
representation than the field. This assumption seems natural because the
field's degrees of freedom are fast and the particles one are slow.
Under this assumption we can estimate the matrix element (\ref{smeared}).
We will prove that in this case there will be polynomial decay.
For $|p|<\sqrt{2}$ we get
$$
(g|g)_{-}(p)=-i\int dk\, |g(k)|^2 (2p+k)^2
\frac{1}{1-\frac{1}{2}p^2 +\frac{1}{2}(k-p)^2}=
$$
$$
=-2i \int dk\, |g(k)|^2 -i \left(I_1+I_2\right);
$$
$$
I_1=\left(-2+10p^2\right) \int dk\, |g(k)|^2
\frac{1}{1-\frac{1}{2}p^2 +\frac{1}{2}(k-p)^2}
$$
$$
I_2=6 \int dk\, |g(k)|^2 p(k-p)
\frac{1}{1-\frac{1}{2}p^2 +\frac{1}{2}(k-p)^2}
$$
Here only $I_1$ and $I_2$ depend on $p$ and therefore can interfere.
Let us find the asymptotics of $(g|g)_{-}(p)$ on $p$
(we investigate the case when $p$ is a small parameter).
We will use the following assumption on $g(k)$: let $g(k)$
be a very smooth function. This means that
$|g(k)|^2=\lambda F(\lambda k)$, $F(k)>0$
is compactly supported smooth function, $\lambda$ is a small parameter.
Let us consider the Taylor expansion on the small parameter $p$
$$
\lambda F(\lambda k)=\lambda F(\lambda (k-p)) +
\lambda^2\sum_i p_i \frac{\partial}{\partial k_i} F(\lambda (k-p)) +\dots.
$$
We get that $\lambda F(\lambda (k-p))$ is a leading term with respect to
$\lambda$. Taking $\lambda\to 0$ we get that we can use
$|g(k-p)|^2$ instead of $|g(k)|^2$ in the formulas for $I_1$ and $I_2$
for sufficiently smooth $g(k)$.
Let us calculate $I_1$ a d $I_2$. Using assumptions considered above we get
$$
I_1=\left(-2+10p^2\right) \int dk\, |g(k-p)|^2
\frac{1}{1-\frac{1}{2}p^2 +\frac{1}{2}(k-p)^2}=
$$
$$
=\left(-2+10p^2\right)
\int dk\, |g(k)|^2
\frac{1}{1+\frac{1}{2}k^2}
-p^2
\int dk\, |g(k)|^2
\frac{1}{\left(1+\frac{1}{2}k^2\right)^2}
$$
$$
I_2=pQ,\qquad Q= 6 \int dk\, |g(k)|^2 k
\frac{1}{1+\frac{1}{2}k^2}
$$
We get
$$
(g|g)_{-}(p)=
-2i \int dk\, |g(k)|^2 +
2i \int dk\, |g(k)|^2\frac{1}{1+\frac{1}{2}k^2}
-iA p^2 -i pQ;
$$
\begin{equation}\label{A}
A=10\int dk\, |g(k)|^2
\frac{1}{1+\frac{1}{2}k^2}-
\int dk\, |g(k)|^2
\frac{1}{\left(1+\frac{1}{2}k^2\right)^2}
\end{equation}
We get for $X(t)=\langle X|U_t|X\rangle$
$$
X(t)=\int dp|f(p)|^2e^{-t(g|g)_-(p)}=
$$
$$
= e^{it2\left(\int dk\, |g(k)|^2 -
\int dk\, |g(k)|^2\frac{1}{1+\frac{1}{2}k^2} \right)}
\int dp|f(p)|^2 e^{it\left(Ap^2+pQ\right)}
$$
Let us estimate this integral for $f(p)=e^{-Bp^2}$, $B>>1$.
Let us consider for simplicity the case $Q=0$
(for example $g(k)$ is spherically symmetric).
In this case the integral is equal to
$$4\pi\int^\infty_0dp \, p^2 e^{-Bp^2}e^{iAtp^2}
=\left(\frac{\pi}{B- i At}\right)^{\frac{3}{2}}
$$
We get that for large $t$ the decay of the matrix element
$X(t)=\langle X|U_t|X\rangle$ is proportional to
$\left(At\right)^{-\frac{3}{2}}$ where $A$ is the functional of the cut-off
function given by (\ref{A}).
To conclude, in this paper
we obtain that in the polaron model for some (symmetric and very smooth)
cut-off functions we have the
polynomial relaxation, the matrix element being proportional to
$t^{-\frac{3}{2}}$. The dependence on the parameter $B$
that corresponds to the size of the support of the
smearing function $f(p)$ of quantum particle in the momentum space
for large $t$ is not important. We can say that particles with large
momentum decay exponentially and the particles with small momentum
decay as $t^{-\frac{3}{2}}$ and the decay for large $t$ does not depend
on the smearing function $f(p)$.
{\bf Acknowlegements}
\medskip
S.Kozyrev and I.Volovich are grateful to Centro Vito Volterra
where this work was started for kind hospitality.
This work was partially supported by INTAS 96-0698
and RFFI-9600312 grants.
I.V. Volovich is supported in part by a fellowship
of the Italian Ministry of Foreign Affairs organized by the Landau
Network-Centro Volta.
|
\section{Introduction}
A key challenge of modern cosmology
is understanding the nature of the primordial
fluctuations that eventually led to the formation of
large scale structure in our universe. One possibility is
that the fluctuations were
generated in a period of inflation
prior to the radiation dominated era of the hot big bang.
As inflation ended the fluctuations would then have been imprinted as
initial conditions for the radiation era on scales far beyond
the Hubble radius. The second possibility
is that the structure was generated through some
causal mechanism operating within the standard big bang radiation and
matter eras.
In this paper we focus on the first option, that the fluctuations
were imprinted early in the radiation era as
linear fluctuations in the metric and in the matter and radiation content.
For several good reasons the possibility that the primordial
perturbations were adiabatic has been the focus of most interest to date.
If the relative abundances of different particle species
were determined directly from the Lagrangian
describing local physics, one would expect those abundance ratios to
be spatially constant because all regions of the universe would
share an identical early history, independent of
the long wavelength perturbations. The
stress-energy present in the universe would then be characterized on large
scales by
a single, spatially uniform equation of state. Such
fluctuations are termed adiabatic and are the simplest possibility
for perturbing the matter content and the geometry of the universe.
They are also naturally predicted by the simplest inflationary
models \cite{infpert}, although there also exist
more complex models giving other types of
perturbations \cite{nongauss}.
For a recent discussion see ref.~\cite{peenew}.
Nevertheless, there is no {\it a priori} reason why the situation
could not be more complicated with abundance ratios varying from
place to place. Perturbations of this sort are known as
{\it isocurvature,} or sometimes {\it entropy,}
perturbations. Most studies of isocurvature perturbations have
examined the possibility that the primordial
perturbations were entirely isocurvature with
a vanishing primordial adiabatic component and have sought
to explore whether such pure
isocurvature models could explain the observed structure in the
universe \cite{pjepa}--\cite{bond}.
In this paper we adopt a more phenomenological approach,
which we believe is now warranted by the prospect of
upcoming precision measurements of the CMB anisotropy on
small scales. Ground based and balloon borne telescopes
and the MAP and PLANCK satellites will
provide very detailed measurements of the
primordial fluctuations \cite{mapp,plnck}.
Most work on how
to interpret this data has focused on parameter
estimation starting from the assumption that the initial
perturbations were generated by an inflationary model
specified by a small number of undetermined
free parameters \cite{paramest}. Adiabatic perturbations, characterized
by an amplitude and spectral index, as well as tensor
fluctuations also characterized by two parameters
are usually assumed, and based on these assumptions a host
of cosmological parameters, such as $\Omega _{total},$
$\Omega _{b},$ $\Omega _{\Lambda },$ $h,$ $N_\nu ,$
are to be inferred from the observed
CMB multipole moments. While the prospects of such measurements
are beguiling, they rely heavily on assumptions regarding the
form of
the primordial perturbations. We feel that those assumptions are
worth checking against the data using
an approach that assumes neither inflation nor any other
favorite theoretical model.
For learning about the fundamental physics responsible
for structure formation,
determining whether the primordial fluctuations
were in fact Gaussian and adiabatic is at least as
important and interesting
as measuring the values of cosmological
parameters.
For this purpose the relevant question is not whether
primordial isocurvature perturbations offer
a viable alternative to adiabatic perturbations,
which has been the focus of prior work,
but rather how and to what extent can observations constrain
the presence of isocurvature modes. Rather than
considering a particular isocurvature model,
it seems appropriate to consider the most
general primordial perturbation possible
without adding new physics to the hot big bang era.
In other words, we assume a universe
filled with neutrinos, photons, leptons and baryons,
and a cold dark matter (CDM) component and then try to
place constraints on the amplitudes of all
possible perturbation modes.
As mentioned above, we shall need to make a
simplifying assumption
to limit the parameter
space of possible perturbations
to reasonable size. We shall assume that
the fluctuations were indeed primordial---that is,
that the perturbation modes were excited by
physics operating at a very early epoch preceding the
hot radiation dominated era and that
no new dynamics influenced the perturbations
at late times.
Here `late' means
well before decoupling, so that whatever decaying
modes were excited earlier had a chance to
decay before influencing the observed structure in our
cosmic microwave sky.
This assumption excludes cosmic
defect models (i.e., strings, global textures, etc.)~%
\cite{shellard} of structure formation
where a detailed understanding of the dynamics of
the order parameter field is required to determine
the unequal time correlations at late times \cite{pst,durrer}.
How should the most general perturbation be characterised?
This question has historically generated some
confusion in the literature.
The standard terminology refers to `growing,' `decaying,' and
`gauge' modes, the latter referring to modes affected by
general coordinate transformations preserving the chosen gauge condition.
In this paper as in most work on cosmological
perturbations we shall use synchronous gauge. The variables in
other common gauges (e.g., Newtonian gauge) are merely linear
combinations of the synchronous gauge variables and their time derivatives.
In synchronous gauge,
the two gauge modes for scalar perturbations are easily identified.
The first corresponds to time-independent spatial reparametrizations of the
constant time hypersurfaces. In this mode the metric perturbation
is constant and the matter is unperturbed. The second corresponds to
deformations of these hypersurfaces through a spatially
dependent shift in the time direction. For the latter gauge mode,
which diverges at early times,
both the matter variables and the metric are perturbed.
Indeed this mode may be regarded
as a shift in the time of the big bang singularity.
The remaining modes are then defined modulo the two gauge modes.
We characterise the remaining modes as either `regular' or `singular,'
according to their behavior as the time since the big bang tends to
zero.
This terminology is preferable to the standard one because it includes
constant modes such as the neutrino velocity mode we shall discuss.
In this paper we shall treat only the `regular' modes (i.e., those
regular up to the gauge modes).
There are several reasons for this.
Singular modes are necessarily decaying as one proceeds
forward in time away from the big bang, so if they are present
at some very
early time with linear amplitude so that a perturbative treatment
is valid they quickly become irrelevant. A second reason is that
even if the perturbation amplitude is small
the perfect fluid approximation breaks down at early times
for the singular modes. Higher moments in the Boltzmann hierarchy become
progressively more important as one goes back in time,
and specifying the initial
conditions involves specifying an infinite number of constants.
The perfect fluid approximation seems essential
to a simple specification of primordial initial perturbations.
Of course, decaying modes might be produced by
some late time physics (such as cosmic defects) operating well
after the big bang, but our point here is that it would appear very difficult
to characterise them without introducing explicit source terms.
In contrast, the `regular' modes are completely
characterised by specifying the leading terms in a power series expansion
in conformal time of the low moments of the phase space density---that is,
the density and velocity of the fluids.
Having characterised the regular perturbations by the leading terms
in the power series for the metric and fluid densities and velocities,
any quadratic observable (e.g., the matter power spectrum or the
cosmic microwave anisotropy power spectrum) is then
completely determined by a primordial power spectral matrix,
which
rather than a single function of $k$ as it would be for growing mode
adiabatic perturbations is
a $5\times 5,$ real, symmetric matrix function of $k.$
The off-diagonal elements establish correlations between
the modes. As long as only quadratic observables are considered,
no assumption of Gaussianity is required.
The counting arises as follows. Each cosmological fluid is described by
two first order equations, so each fluid introduces two new perturbation modes.
However, in synchronous gauge, as mentioned, there is one gauge mode affecting
the fluid perturbations.
For a single fluid there is therefore just one regular,
growing mode perturbation, and no physical (i.e., non-gauge) decaying mode.
A convenient way to deal with the gauge mode is to identify it with
the velocity of the cold dark matter. By a coordinate choice this may be
chosen initially to be zero. If it is assumed that
the cold dark matter couples to the other fluids only via
gravity, there is no scattering term
to consider, and with coordinates chosen so that the velocity
is initially zero, it remains so for all times.
If we now introduce photons and baryons, four new modes arise.
The first is an adiabatic decaying mode. There are also the
baryon isocurvature and cold dark matter isocurvature modes, where
the initial conditions contain equal and opposite perturbations of
the photon density and the baryon or cold dark matter densities, respectively.
The fourth new mode is that in which the photon and baryon fluids
have a relative velocity that diverges at early times and the
fluid approximation breaks down. So far we have three regular modes.
Let us now introduce neutrinos into the picture. We shall imagine we
are setting up the perturbations after neutrino decoupling ($T\sim $MeV,
$t \sim $ seconds).
For our purposes there is no distinction between the different neutrino species
nor between neutrinos and antineutrinos, since we are only interested in how
perturbations in the neutrino fluid affect cosmological observations
of the density and microwave background today. Two new
perturbation modes are introduced, the first a neutrino isocurvature density
perturbation and the second a neutrino isocurvature velocity perturbation.
In the latter, we can arrange the neutrino and photon-baryon
fluids to have equal and opposite momentum density. In the approximation that
we ignore the collision term coupling
neutrinos to the photon-baryon fluid, which is valid
after neutrino decoupling, and in the small time limit
there is no divergence in this mode.
The neutrino isocurvature velocity perturbation may
be considered as primordial as long as one takes primordial to
mean `generated after one second' but well
before photon-baryon decoupling. Of course similar remarks apply to the
cold dark matter, baryon isocurvature or neutrino isocurvature perturbations.
Namely if we go back far enough in cosmic history, where the various
conservation laws for baryon or lepton number break down, or where
the cold dark matter was initially
generated or reached thermal equilibrium,
then a description of the form we are using would also
break down. Since one second is still quite early, and
certainly well before
photon-baryon decoupling, we regard it admissible to consider
the possibility of `primordial' neutrino velocity perturbations.
Possible mechanisms for generating them shall be
discussed in a separate paper \cite{ltter}.
A primeval baryon isocurvature model (PBI)
was introduced by Peebles \cite{pjepa}
in which a universe with just baryons, radiation, and neutrinos
is assumed and primordial perturbations in the baryon-to-photon
ratio are assumed. Since at early times the baryons contribute
negligibly to the total density, such perturbations lack an
adiabatic, or curvature, component at early times.
Compared to an $\Omega =1$ CDM model PBI gives:
(1) lower small-scale relative peculiar velocities,
(2) greater large-scale flow velocities,
(3) earlier reionization,
and (4) earlier galaxy and star formation \cite{cen}.
The consequences of PBI and comparison with observations
have been studied by a number of
authors \cite{yo,sugi,gor,chiba,hu,nnewa}.
Bond and Efstathiou \cite{bond,bondb} have considered a CDM
isocurvature model in which the CDM-to-photon ratio
varies spatially. A possible mechanism for generating
such perturbations arises in models with axion dark matter
in which scale invariant fluctuations of the axion field
$A(x)$ are converted into density fluctuations after
the axion field acquires a mass in the QCD phase
transition \cite{axenides}.
Under the assumption that quantum fluctuations during
inflation impart a scale free spectrum of fluctuations
to the axion field, so that $\delta A(k)\sim k^{-3/2},$
and that $\delta A(k)\ll \bar A,$ a scale-free spectrum
of Gaussian fluctuations in the axion-to-photon ratio
is generated. The resulting spectrum
of density fluctuations today has the same power
law on large scales as for adiabatic fluctuations with
a scale free spectrum $P_\rho (k)\sim k^1,$ but
compared to adiabatic, scale-free perturbations
the turnover to $P_\rho (k)\sim k^{-3}$ power
law behavior on small scales occurs on a larger scale
for the isocurvature case.
For the amplitude of density perturbations
normalized on small scales (for example, using
$\sigma _8),$ this gives about thirty times more
power in the matter perturbations
on large scales and entails an excessive
CMB anisotropy on large angles. This work
has been extended to consider mixtures of
CDM isocurvature and adiabatic fluctuations,
low density universes, and more recent CMB
data \cite{stompor,gouda}.
The neutrino isocurvature modes discussed here allow
for a spatially varying relative density of photons and neutrinos and
a relative velocity between the photon and
neutrino components as well. For the neutrino isocurvature density
mode, the total density perturbation vanishes but the
relative density of neutrinos and photons varies
spatially. On superhorizon scales, the neutrinos and photons
evolve similarly but upon entering the horizon
the neutrinos free stream, developing nonvanishing
higher moments of the neutrino phase
space density $F_{\nu \ell } (k),$ while
because of Thomson scattering the photons continue
to behave much as a perfect fluid.
These distinct behaviors subsequently
generate perturbations in the total density.
For the neutrino isocurvature velocity mode, the
rest frames of the neutrino and the photon fluids do
not coincide. The relative velocities are such
that initially the perturbation in the total
momentum density vanishes. If this last condition
were not satisfied, the metric perturbation
generated by this mode would diverge at early
times, rendering the mode a singular mode, which
according to the discussion above
should be excluded. But the lack of divergence
owing to this cancellation makes
this an admissible regular perturbation mode.
The neutrino isocurvature mode solutions are implicit in
the work of Rebhan and Schwarz\cite{nnewb} and of
Challinor and Lasenby\cite{nnewc}, but their implications
were not explored.
A possible obstacle to exciting neutrino isocurvature modes
arises at early times, from processes in which neutrinos are
generated or scattered. If the neutrino chemical potentials
vanish, so that there are in each generation
precisely as many neutrinos as antineutrinos,
a spatially varying relative density can only be
established at a temperature sufficiently low
that the processes turning photons into
neutrino-antineutrino pairs and vice versa
had already been frozen out---that is, below a few MeV.
Nonvanishing chemical potentials for the various
neutrino species can protect variations in the
ratio $(\rho _\nu /\rho _\gamma )$ from
erasure by processes involving $\nu \bar \nu $
annihilation, and observational
constraints that would rule out neutrino chemical
potentials of the required magnitude are lacking.
$L$-violating processes mediated through sphalerons,
unsuppressed at temperatures above the
electroweak phase transition, can readjust the
neutrino chemical potentials
$\mu _{\nu _e},$ $\mu _{\nu _\mu },$
and $\mu _{\nu _\tau }$ and, moreover,
can convert lepton asymmetries in the neutrino sector into
baryon asymmetry, but if the neutrino chemical potentials
satisfy $\mu _{\nu _e}+\mu _{\nu _\mu }+\mu _{\nu _\tau }=0,$
then no tendency favoring
sphalerons over anti-sphalerons is introduced and the net effect
of these electroweak processes vanishes. Recall that the
neutrino overdensity is proportional to
${\mu _{\nu _e}}^2+{\mu _{\nu _\mu }}^2+{\mu _{\nu _\tau }}^2.$
Similarly, for the neutrino isocurvature velocity mode
scattering of neutrinos with other
components dampens this mode at very early times, and
for this mode to be relevant there must exist a process capable of
exciting it after this dampening effect has frozen out.
There exist many candidate mechanisms that might generate these
neutrino isocurvature modes. The neutrino density isocurvature
mode could be generated during inflation if the theory included a
light scalar field carrying lepton number, with mass much smaller
than the Hubble constant during inflation.\cite{nnewd}
During inflation this field would be excited but would contribute
negligibly to the density of the universe. After inflation,
when the Hubble constant fell below the mass of the scalar field
it would oscillate and decay, producing a lepton asymmetry. We
would expect the neutrino chemical potential to be
proportional to the scalar field value, so the most natural possibility
would be Gaussian, scale invariant perturbations in the neutrino chemical
potential, with the density perturbation in the neutrinos being proportional
to the square of the chemical potential.
The neutrino velocity mode could have been excited by the decay of
relics such as cosmic strings, walls, or superstrings, surviving
until after the neutrinos had decoupled and then decaying into
neutrinos. The neutrino fluid produced in such processes would
be perturbed both in its density and
velocity, and these perturbations would be isocurvature in
character \cite{ltter}.
Another possibility for exciting the modes arises from magnetic
fields frozen into the plasma whose stress gradients impart a
velocity to the photon-lepton-baryon plasma but not to the
neutrino component.
\def{\bf x}{{\bf x}}
\def{\bf k}{{\bf k}}
\def{\cal H}{{\cal H}}
\section{Identifying the Modes}
We now turn to identifying the possible perturbation
modes using synchronous gauge with the
line element
\begin{eqnarray}
ds^2=a^2(\tau )\cdot \Bigl[
-d\tau ^2+\bigl( \delta _{ij}+h_{ij}\bigr) ~dx^idx^j~
\Bigr] .
\end{eqnarray}
For spatial dependence of a given wavenumber ${\bf k} ,$ we define
\begin{eqnarray}
h_{ij}({\bf k} ,\tau)=e^{i{\bf k} \cdot {\bf x} }\cdot
\left[ \hat k_i~ \hat k_j ~h({\bf k} ,\tau)+
\left( \hat k_i~ \hat k_j-{1\over 3} \delta _{ij}\right)
6\eta ({\bf k} ,\tau) \right] .
\end{eqnarray}
Our discussion generally follows the
notation of ref. \cite{ma}.
The linearized Einstein equations are
\begin{eqnarray}
k^2\eta -{1\over 2}{\cal H} \dot h &=&
{~}(4\pi Ga^2)~\delta {T^0}_0,\nonumber\\
k^2\dot \eta &=&
{~}(4\pi Ga^2)~(\bar \rho +\bar p)\theta ,\nonumber\\
\ddot h+2{\cal H} \dot h-2k^2\eta &=& -(8\pi Ga^2)~\delta {T^i}_i,\nonumber\\
\ddot h+6\ddot \eta +2{\cal H} (\dot h+6\dot \eta )-2k^2\eta &=&
-(24\pi Ga^2)(\bar \rho +\bar p)\sigma
\end{eqnarray}
where we define $(\bar \rho +\bar p)\theta =ik^j~{T^0}_j$
and $(\bar \rho +\bar p)\sigma =-(\hat k_i\hat k_j-{1\over 3}
\delta _{ij}){T_i}^j.$
We define ${\cal H} =\dot a/a.$
With only a single fluid, $\theta $ is simply the
divergence $(\nabla \cdot {\bf v}).$
Assuming $N$ fluids, labeled $(J=1,\ldots ,N),$ one may
rewrite the above as
\begin{eqnarray}
k^2\eta -{1\over 2}{\cal H} \dot h &=&
-{3\over 2}{\cal H} ^2\sum _J\Omega _J\delta _J,\nonumber\\
k^2\dot \eta &=&
{~~}{3\over 2}{\cal H} ^2\sum _J\Omega _J(1+w_J)\theta _J,\nonumber\\
\ddot h+2{\cal H} \dot h-2k^2\eta &=& -9{\cal H} ^2\sum _J\Omega _J~c_{sJ}^2~\delta _J,\nonumber\\
\ddot h+6\ddot \eta +2{\cal H} (\dot h+6\dot \eta )-2k^2\eta &=&
-12{\cal H} ^2\Omega _\nu \sigma _\nu
\label{apad}
\end{eqnarray}
where $w_J={p_J/\rho _J}$ and $c_{s}^2=\partial p_J/\partial \rho _J.$
In the last equation only the neutrino contribution to the
anisotropic stress is included. At early times only the neutrino
contribution is relevant, but later when the photons and baryons begin
to decouple the photon quadrupole moment must also be included.
For the photons, the equations of motion at early times
before the baryonic component of the fluid was significant are
\begin{eqnarray}
&&\dot \delta _\gamma +{4\over 3}\theta _\gamma +{2\over 3}\dot h=0,\nonumber\\
&&\dot \theta _\gamma -{1\over 4}k^2\delta _\gamma =0.
\end{eqnarray}
Similarly, for the neutrinos, after neutrino decoupling at temperatures
of $\sim$ 1 MeV, we have
\begin{eqnarray}
&&\dot \delta _\nu +{4\over 3}\theta _\nu +{2\over 3}\dot h=0,\nonumber\\
&&\dot \theta _\nu -{1\over 4}k^2\delta _\nu +k^2\sigma _\nu =0,\nonumber\\
&&\dot \sigma _\nu ={2\over 15}
\left[ 2\theta _\nu +\dot h+6\dot \eta \right] -{3\over 10} k F_{\nu 3}.
\end{eqnarray}
where $\sigma_\nu = F_{\nu 2}/2$ is the quadrupole moment of the neutrino
phase space density and $F_{\nu l}$ is the $l$-th multipole,
in the notation used below.
The photon and neutrino evolution equations above
differ only in the
presence of an anisotropic stress term $\sigma_\nu.$ Photons,
because of their frequent scattering by charged leptons
and baryons, at early times are unable to develop
a quadrupole moment in their velocity distribution.
Neutrinos, on the other hand, develop significant
anisotropic stresses upon crossing the horizon.
The addition of an extra degree of freedom that would result
from the equation for
$\dot \sigma _\nu $ is avoided by setting
$\sigma _\nu=0$ at $\tau =0.$\footnote{If we were to
consider
the physical decaying mode with $\eta \propto \tau ^{-1},$ we
would find
this condition cannot be satisfied. As discussed above,
decaying modes are inevitably associated with the
breakdown of the
fluid approximation at early times.}
For a CDM component, the equations of motion are
\begin{eqnarray}
&&\dot \delta _c+\theta _c+{1\over 2}\dot h=0,\nonumber\\
&&\dot \theta _c+{\cal H} \theta _c=0.
\end{eqnarray}
At later times, when $\Omega _b$ becomes comparable to
$\Omega _\gamma ,$ the photon equations of motion
are modified as follows:
\begin{eqnarray}
\dot \delta _\gamma
+{4\over 3}\theta _\gamma +{2\over 3}\dot h&=&0,\nonumber\\
\dot \theta _\gamma -{k^2\over 4}\delta _\gamma +an_e\sigma _T
(\theta _\gamma -\theta _b)&=&0.
\label{apagaa}
\end{eqnarray}
Here $a$ is the scale factor, $n_e$ is the electron density, and
$\sigma _T$ is the Thomson scattering cross section.
The baryon equations of motion are
\begin{eqnarray}
\dot \delta _b
+\theta _b+{1\over 2}\dot h&=&0,\nonumber\\
\dot \theta _b+{\cal H} \theta _b +{4\over 3}{\Omega _\gamma \over \Omega _b}
an_e\sigma _T
(\theta _b-\theta _\gamma )&=0&.
\label{apagbb}
\end{eqnarray}
At early times, the characteristic time for the synchronization of the
photon and baryon velocities $t_{b\gamma }\approx 1/(n_e\sigma _T)$
is small compared to the expansion time $t_{exp}\approx a\tau $ and to the
oscillation period $t_{osc}\approx (a\tau /k).$ (We use units where
the speed of light $c=1$.)
Any deviation of
$(\theta _\gamma -\theta _b)$ from zero rapidly decays away.
This may be seen by subtracting the second equation of \ref{apagbb}~from
that of \ref{apagaa}~and regarding
${\cal H} \theta _b+{1\over 3}k^2\delta _\gamma $ as a forcing
term. In the limit $\sigma _T\to \infty ,$ one obtains
$\theta _b=\theta _\gamma $---in other words, a tight coupling
between the baryons and the photons. Therefore,
at early times we may set $\theta _b=\theta _\gamma =
\theta _{\gamma b}.$
In the tight coupling approximation the evolution
equation for $\theta _{\gamma b}$ is obtained by
adding the second equation of \ref{apagaa}~multiplied by
${4\over 3}\Omega _\gamma $ to the second equation of \ref{apagbb} ~multiplied by
$\Omega _b,$ so that the scattering terms cancel giving
\begin{eqnarray}
\left( {4\over 3}\Omega _\gamma +\Omega _b\right)
\dot \theta _{\gamma b}
=-\Omega _b{\cal H}\theta _{\gamma b}+{1\over 3}\Omega _\gamma k^2\delta _\gamma
\end{eqnarray}
and the baryon and photon density contrasts evolve according to
\begin{eqnarray}
\dot \delta _b&=&-\theta _{\gamma b}-{1\over 2}\dot h,\nonumber\\
\dot \delta _\gamma &=&-{4\over 3}\theta _{\gamma b}-{2\over 3}\dot h.
\end{eqnarray}
In the absence of baryon isocurvature perturbations,
$\delta _\gamma ={4\over 3}\delta _b,$
making one of these equations redundant, but both equations are required
for the most general type of perturbation.
Although an excellent approximation early on, the
tight coupling assumption breaks down at later times
as the photons and baryons decouple, and for a more accurate
description the two equations in \ref{apagaa}~for the
time derivatives of the monopole and dipole moments
of the velocity of the photon phase space distribution
must be replaced by the following
infinite hierarchy of equations for the time
derivatives of higher order moments
of the photon distribution function as well:
\begin{eqnarray}
\dot \delta _\gamma &=&
-{4\over 3}\theta _\gamma -{2\over 3}\dot h,\nonumber\\
\dot \theta _\gamma &=&k^2\left( {\delta _\gamma \over 4}
-{F_{\gamma 2}\over 2}\right)
+an_e\sigma _T(\theta _b-\theta _\gamma ),\nonumber\\
\dot F_{\gamma 2}&=&
{8\over 15}\theta _\gamma -{3\over 5}kF_{\gamma 3}+
{4\over 15}\dot h
+{8\over 5}\dot \eta
-{9\over 10}an_e\sigma _TF_{\gamma 2},\nonumber\\
\dot F_{\gamma \ell }&=&
{k\over 2\ell +1}\left[ {\ell }F_{\gamma (\ell -1)} -
{(\ell +1)}F_{\gamma (\ell +1)}
\right] -an_e\sigma _TF_{\gamma \ell }
\end{eqnarray}
where $\ell \ge 3$ for the last equation.
Initially, as $\tau \to 0,$ $F_{\gamma \ell }=0$
for $\ell \ge 2.$ If this condition were relaxed
an infinite number of decaying modes would appear.
The fluid approximation for the photons, eqn.~\ref{apagaa},
is obtained using only the first two of the above
equations combined with the approximation $F_{\gamma 2}=0.$
To describe the neutrinos during and after horizon
crossing requires a Boltzman hierarchy for
$\delta _\nu ,$ $\theta _\nu ,$ $F_{\nu \ell }$
identical to the one above except that the Thomson
scattering terms are omitted. This assumes that
neutrino masses are irrelevant.
Before solving the Einstein equations
\ref{apad}, we first identify the two gauge modes
resulting from the residual gauge freedom remaining
within synchronous gauge. A general infinitesimal
gauge transformation considered to linear order
is a coordinate transformation $x^\mu \to
x^{\prime \mu }(x)$ where $x^\mu =x^{\prime \mu }+
\epsilon ^\mu (x).$ From the transformation
rule for tensors
\begin{eqnarray}
{g^\prime }_{\mu \nu }(x')=
{\partial x^\xi \over \partial x^{\prime \mu }}
{\partial x^\eta \over \partial x^{\prime \nu }}
g_{\xi \eta }(x)
\end{eqnarray}
it follows that
\begin{eqnarray}
\delta g_{\mu \nu }=\epsilon ^\xi ~{\partial g_{\mu \nu }^{(0)}\over
\partial x^\xi }+{\epsilon ^\xi }_{,\mu }~g_{\xi \nu }^{(0)}
+g_{\mu \xi }^{(0)}~{\epsilon ^\xi }_{,\nu }
\label{apai}
\end{eqnarray}
where $g_{\mu \nu }^{(0)}$ is the unperturbed, zeroth-order
metric. Applying \ref{apai}~to the metric
$ds^2=a^2(\tau ) \Bigl[ -(1-h_{00})d\tau ^2+2h_{0i}~d\tau dx^i
+(\delta _{ij}+h_{ij})dx^idx^j\Bigr] ,$ where
\begin{eqnarray}
\tau =\tau ^\prime +T({\bf x} ,\tau ), \quad
{\bf x} ={\bf x} '+{\bf S}({\bf x} ,\tau )
\end{eqnarray}
where $S$ and $T$ are regarded as infinitesimal, one obtains
\begin{eqnarray}
\delta h_{00}&=&-{2\dot a\over a}T({\bf x} ,\tau )-2\dot T({\bf x} ,\tau ),\nonumber\\
\delta h_{0i}&=&-T_{,i}({\bf x} ,\tau )+\dot S_i({\bf x} ,\tau ),\nonumber\\
\delta h_{ij}&=&S_{i,j}+S_{j,i}+{2\dot a\over a}\delta _{ij}T.
\end{eqnarray}
For the density contrast for a perfect fluid component
labeled by $\alpha ,$ one obtains $\delta (\delta _\alpha )
=-3(1+w_\alpha ){\cal H} T.$ After a gauge transformation
the velocity is shifted by $\delta {\bf v}=-\dot S({\bf x} ,\tau )$
for all components.
The synchronous gauge condition $\delta h_{00}=0$ implies
that $T({\bf x} ,\tau )=A({\bf x} )/a(\tau ),$ and $\delta h_{0i}=0$
implies that
\begin{eqnarray}
{\bf S}({\bf x} ,\tau )=B({\bf x} )+\nabla A({\bf x} )\int {d\tau \over a(\tau )}.
\end{eqnarray}
Therefore
\begin{eqnarray}
\delta h_{ij}={2\dot a\over a^2}A({\bf x} )\delta _{ij}+2A_{,ij}
\int {d\tau \over a(\tau )}+B_{i,j}+B_{j,i}.
\end{eqnarray}
For a radiation-dominated universe [with $a(\tau )=\tau $]
considering for the moment only a single wavenumber, we find that%
\footnote{Note that the gauge mode $A$
disagrees with eqns.~[94] and [95] of ref.~\cite{ma},
which are incorrect.}
\begin{eqnarray}
\delta h_{ij}=\left[ \left( {2\over \tau ^2}\delta _{ij}
-2k_ik_j\ln (\tau )\right) A({\bf k} )+
2B~k_ik_j \right] e^{i{\bf k} \cdot {\bf x} },
\end{eqnarray}
or
\begin{eqnarray}
h&=&A\left[ {6\over \tau ^2}-2k^2\ln (\tau )\right] +2B,\quad
\eta =A\left[ {-1\over \tau ^2}\right].
\label{apao}
\end{eqnarray}
To obtain a more accurate power series expansion close to
the time of matter-radiation equality and to consider the baryon and
CDM isocurvature modes, we assume a scale factor evolution
for a universe filled with matter and radiation:
$a(\tau )=\tau +\tau ^2.$ At matter-radiation equality
$a_{eq}=1/4$ and $\tau _{eq}=(\sqrt{2}-1)/2.$ Since
${\cal H} =(2\tau +1)/(\tau ^2+\tau )$ has a pole at
$\tau =-1,$ the resulting power series solutions are expected
to diverge beyond the unit circle.
For the matter-radiation universe, with $a(\tau )=\tau +\tau ^2$
eqn. \ref{apao}~is modified to become
\begin{eqnarray}
h&=&A\left[ {6(1+2\tau )\over \tau ^2(1+\tau )^2}
-2k^2\ln \left({\tau \over 1+\tau }\right) \right] +2B,\quad
\eta =-A\left[ {(1+2\tau )\over \tau ^2(1+\tau )^2}\right].
\end{eqnarray}
We now present the power series solutions.
$R_\gamma $ and $R_\nu $ represent the fractional contribution
of photons and neutrinos to the total density at early times
deep within the
radiation dominated epoch. (We ignore possible effects due to
nonvanishing neutrino masses). We also assume that we are
considering the perturbations after the
annihilation of electrons and positrons, so that the latter have
dumped their energy into the photon background.
For $N_\nu$ species of massless neutrinos we define
$R ={7\over 8} N_{\nu} ({4\over 11})^{4\over 3}$ and we have
$R_\gamma = (1+R)^{-1}$ and $R_\nu =R(1+R)^{-1}.$
\vskip 25pt
\noindent
{\it Adiabatic Growing Mode.}
\begin{eqnarray}
h&=&{1\over 2}k^2\tau ^2,\nonumber\\
\eta &=& 1-{5+4R_\nu \over 12(15+4R_\nu )}k^2\tau ^2,\nonumber\\
\delta _c&=&-{1\over 4}k^2\tau ^2,\nonumber\\
\delta _b&=&-{1\over 4}k^2\tau ^2,\nonumber\\
\delta _\gamma &=&-{1\over 3}k^2\tau ^2,\nonumber\\
\delta _\nu &=&-{1\over 3}k^2\tau ^2,\nonumber\\
\theta _c&=&0,\nonumber\\
\theta _{\gamma b}&=&-{1\over 36}k^4\tau ^3,\nonumber\\
\theta _\nu &=&-{1\over 36}\left[ {23+4R_\nu \over 15 +4R_\nu }\right]
k^4\tau ^3,\nonumber\\
\sigma _\nu &=&{2\over 3(12+R_\nu )}k^2\tau ^2.
\end{eqnarray}
\vskip 25pt
\noindent
{\it Baryon Isocurvature Mode.}
\begin{eqnarray}
h&=&4{\Omega _{b,0}} \tau -6{\Omega _{b,0}} \tau ^2,\nonumber\\
\eta &=&-{2\over 3}{\Omega _{b,0}} \tau +{\Omega _{b,0}} \tau ^2 ,\nonumber\\
\delta _c&=&-2{\Omega _{b,0}} \tau +3{\Omega _{b,0}} \tau ^2,\nonumber\\
\delta _b&=&1-2{\Omega _{b,0}} \tau +3{\Omega _{b,0}} \tau ^2,\nonumber\\
\delta _\gamma &=&-{8\over 3}{\Omega _{b,0}} \tau +4{\Omega _{b,0}} \tau ^2,\nonumber\\
\delta _\nu &=&-{8\over 3}{\Omega _{b,0}} \tau +4{\Omega _{b,0}} \tau ^2,\nonumber\\
\theta _c&=&0,\nonumber\\
\theta _{\gamma b}&=&-{1\over 3}{\Omega _{b,0}} k^2\tau ^2,\nonumber\\
\theta _\nu &=&-{1\over 3}{\Omega _{b,0}} k^2\tau ^2,\nonumber\\
\sigma _\nu &=&{-2{\Omega _{b,0}} \over 3(2R_\nu +15)}k^2 \tau ^3.
\end{eqnarray}
There is no regular baryon velocity mode because of the
tight coupling of the baryons to the photons.
\vskip 25pt
\noindent
{\it CDM Isocurvature Mode.}
\begin{eqnarray}
h&=&4{\Omega _{c,0}} \tau -6{\Omega _{c,0}} \tau ^2,\nonumber\\
\eta &=&-{2\over 3}{\Omega _{c,0}} \tau +{\Omega _{c,0}} \tau ^2 ,\nonumber\\
\delta _c&=&1-2{\Omega _{c,0}} \tau +3{\Omega _{c,0}} \tau ^2,\nonumber\\
\delta _b&=&-2{\Omega _{c,0}} \tau +3{\Omega _{c,0}} \tau ^2,\nonumber\\
\delta _\gamma &=&-{8\over 3}{\Omega _{c,0}} \tau +4{\Omega _{c,0}} \tau ^2,\nonumber\\
\delta _\nu &=&-{8\over 3}{\Omega _{c,0}} \tau +4{\Omega _{c,0}} \tau ^2,\nonumber\\
\theta _c&=&0,\nonumber\\
\theta _{\gamma b}&=&-{1\over 3}{\Omega _{c,0}} k^2\tau ^2,\nonumber\\
\theta _\nu &=&-{1\over 3}{\Omega _{c,0}} k^2\tau ^2,\nonumber\\
\sigma _\nu &=&{-2{\Omega _{c,0}} \over 3(2R_\nu +15)}k^2 \tau ^3.
\end{eqnarray}
The CDM velocity mode may be identified with the gauge mode.
This can be seen from the equation for the CDM
velocity which is decoupled from the other equations.
$\theta _{CDM}$ behaves as $\tau ^{-1}$ at early times.
If there were several species of CDM, however, a new
physical, non-gauge relative velocity mode would arise,
which would be divergent at early times.
\vskip 25pt
\noindent
{\it Neutrino Isocurvature Density Mode.}
\begin{eqnarray}
h&=&{{\Omega _{b,0}} R_\nu \over 10R_\gamma }k^2\tau ^3,\nonumber\\
\eta &=&-{R_\nu \over 6(15+4R_\nu )}k^2\tau ^2,\nonumber\\
\delta _c&=&{-{\Omega _{b,0}} R_\nu \over 20R_\gamma }k^2\tau ^3,\nonumber\\
\delta _b&=&{1\over 8}{R_\nu \over R_\gamma }k^2\tau ^2,\nonumber\\
\delta _\gamma &=&-{R_\nu \over R_\gamma }
+{1\over 6}{R_\nu \over R_\gamma }k^2\tau^2,\nonumber\\
\delta _\nu &=&1-{1\over 6}k^2\tau ^2,\nonumber\\
\theta _c&=&0,\nonumber\\
\theta _{\gamma b}&=&-{1\over 4}{R_\nu \over R_\gamma }k^2\tau +
{3{\Omega _{b,0}} R_\nu \over 4R_\gamma ^2}k^2\tau ^2,\nonumber\\
\theta _\nu &=&{1\over 4}k^2\tau ,\nonumber\\
\sigma _\nu &=&{1\over 2(15+4R_\nu )}k^2\tau ^2.
\end{eqnarray}
Physically, one starts with a uniform energy density,
with the sum of the neutrino and photon densities unperturbed.
When a mode enters the horizon, the photons behave
as a perfect fluid, while the neutrinos free stream,
creating nonuniformity in the energy density, pressure,
and momentum density, thus sourcing
metric perturbations.
\vskip 25pt
\noindent
{\it Neutrino Isocurvature Velocity Mode.}
\begin{eqnarray}
h&=&{3\over 2}{\Omega _{b,0}} {R_\nu \over R_\gamma }k\tau ^2,\nonumber\\
\eta &=&-{4R_\nu \over 3(5+4R_\nu )}k\tau +
\left( {-{\Omega _{b,0}} R_\nu \over 4R_\gamma }+{20R_\nu \over
(5+4R_\nu )(15+4R_\nu )}\right) k\tau ^2,\nonumber\\
\delta _c&=&-{3{\Omega _{b,0}} \over 4}{R_\nu \over R_\gamma }k\tau ^2,\nonumber\\
\delta _b&=&{R_\nu \over R_\gamma }k\tau -
{3{\Omega _{b,0}} R_\nu (R_\gamma +2)\over 4R_\gamma ^2}k\tau ^2,\nonumber\\
\delta _\gamma &=&{4\over 3}{R_\nu \over R_\gamma }k\tau -
{{\Omega _{b,0}} R_\nu (R_\gamma +2)\over R_\gamma ^2}k\tau ^2,\nonumber\\
\delta _\nu &=&-{4\over 3}k\tau - {{\Omega _{b,0}} R_\nu \over R_\gamma }k\tau ^2,\nonumber\\
\theta _c&=&0,\nonumber\\
\theta _{\gamma b}&=&-{R_\nu \over R_\gamma }k+{3{\Omega _{b,0}} R_\nu \over R_\gamma ^2}k\tau +
{R_\nu \over R_\gamma }\left(
{3{\Omega _{b,0}} \over R_\gamma }-{9{\Omega _{b,0}} ^2\over R_\gamma ^2}
\right) k\tau ^2+
{R_\nu \over 6R_\gamma }k^3\tau ^2,\nonumber\\
\theta _\nu &=&k-{(9+4R_\nu )\over 6(5+4R_\nu )}k^3\tau ^2,\nonumber\\
\sigma _\nu &=&{4\over 3(5+4R_\nu )}k\tau
+{16R_\nu \over (5+4R_\nu )(15+4R_\nu )}k\tau ^2,\nonumber\\
F_{\nu 3}&=&{4\over 7(5+4R_\nu )}k^2\tau ^2.
\end{eqnarray}
Here the neutrinos and photons start with uniform total
density and uniform density ratio but with relative velocities
matched so that initially the total momentum density vanishes.
If the relative momenta were not perfectly matched, the metric
perturbation generated would diverge at early times, as in the
adiabatic decaying mode. But because of the perfect match, a
divergence at early times is avoided.
It is also possible, at least in principle, to consider
regular modes with higher moments of
$F_{\nu \ell }$ with $\ell \ge 3$ excited initially,
as was considered in ref. \cite{nnewb}; however, it is difficult
to envision plausible mechanisms for exciting these higher
moment modes.
\vskip 14pt
\noindent
{\bf Newtonian Potentials.} In this paper we have used
synchronous gauge, but for completeness we give the
form of the Newtonian potentials for the regular modes
presented above. The Newtonian potentials $\phi $ and
$\psi $ are related to the synchronous variables as
follows:
\begin{eqnarray}
\phi &=&{1\over 2k^2}\bigl[ \ddot h+6\ddot \eta
+{\cal H}(\dot h+6\dot \eta )\bigr] ,\nonumber\\
\psi &=& \eta -{{\cal H} \over 2k^2}[\dot h+6\dot \eta ].
\end{eqnarray}
We define the Newtonian potentials according to the
convention $ds^2=a^2(\tau )[-d\tau ^2(1+2\phi )+
dx^i~dx^j~\delta _{ij}(1-2\psi )].$
We now give the Newtonian potentials to leading order.
For the growing adiabatic mode
\begin{eqnarray}
\phi &=&{10\over (15+4R_\nu )},\nonumber\\
\psi &=&{10\over (15+4R_\nu )}.
\end{eqnarray}
For the neutrino isocurvature density mode
\begin{eqnarray}
\phi &=&{-2R_\nu \over (15+4R_\nu )},\nonumber\\
\psi &=&{R_\nu \over (15+4R_\nu )}.
\end{eqnarray}
For the neutrino isocurvature velocity mode
\begin{eqnarray}
\phi &=&{-4R_\nu \over (15+4R_\nu )}k^{-1}\tau ^{-1},\nonumber\\
\psi &=&{4R_\nu \over (15+4R_\nu )}k^{-1}\tau ^{-1}.
\end{eqnarray}
The potentials for the baryon
isocurvature mode are
\begin{eqnarray}
\phi &=&{(4R_\nu -15){\Omega _{b,0}} \over 2(15+2R_\nu )}\tau ,\nonumber\\
\psi &=&{-(4R_\nu -15){\Omega _{b,0}} \over 6(15+2R_\nu )}\tau ,
\end{eqnarray}
and for the CDM isocurvature mode are
\begin{eqnarray}
\phi &=&{(4R_\nu -15){\Omega _{c,0}} \over 2(15+2R_\nu )}\tau ,\nonumber\\
\psi &=&{-(4R_\nu -15){\Omega _{c,0}} \over 6(15+2R_\nu )}\tau .
\end{eqnarray}
It is curious that the Newtonian potential diverges
at early times for the neutrino isocurvature velocity mode
while in synchronous gauge there is no singularity.
It appears that synchronous gauge is a more physical
gauge and that Newtonian gauge is inadequate for modes
based on anisotropic stresses. The dimensionless
Ricci curvature $a^2(\tau )\tau ^2R$ is nonsingular at early times.
In any case, synchronous gauge is more physical in the sense that
its evolution is {\it local} whereas in Newtonian gauge the evolution
of the shape of the constant cosmic time hypersurfaces depends on how
matter behaves infinitely far away (because the gauge choice is defined in
terms of the {\it scalar-vector-tensor} decomposition,
which is nonlocal). The divergence
of the Newtonian potentials arises from the need to warp the surfaces
of constant cosmic time so as to put the metric in a spatially
isotropic form; however, there is nothing at all physical about this
particular form. The neutrino isocurvature velocity mode
was excluded in ref. \cite{nnewb} because of the behavior of the
Newtonian potentials at early times; however, when a physical
phenomemon can be described in a nonsingular manner in some gauge,
its singularity in another gauge should be regarded as a
coordinate singularity.
\begin{figure}
\centerline{\psfig{file=nidspectra.ps,width=3.in}}
\caption{
{\bf CMB Anisotropy for the Neutrino Isocurvature Density Mode.}
We plot $\ell (\ell +1)c_\ell /2\pi $ for the
neutrino isocurvature density mode
(the dashed curves) for initial power spectra
$P_{\delta _\nu }\sim k^\alpha $ where $\alpha =-3.0,\ldots ,-2.4$
increasing in increments of $0.1$ from bottom to top at
large $\ell .$ The adiabatic growing mode (the solid curve) with a
scale-invariant spectrum is included for comparison.
All curves are normalized to COBE. For the lowest curve the
variations in the photon-to-neutrino ratio
obey a scale-invariant initial power spectrum.
}
\end{figure}
\begin{figure}
\centerline{\psfig{file=nivspectra.ps,width=3.in}}
\caption{
{\bf CMB Anisotropy for the Neutrino Isocurvature Velocity Mode.}
We plot $\ell (\ell +1)c_\ell /2\pi $ for the neutrino isocurvature
velocity mode (the dashed curves) for initial power spectra
$P_{v_\nu }\sim k^\alpha $ with $\alpha =-3.0,\ldots ,-2.0$
increasing in increments of $0.2$ from bottom to top
at large $\ell .$ $\alpha =-2.0$ corresponds to
a white noise initial power spectrum in the divergence
of the velocity field, possibly resulting from a large
number of small explosions.
}
\end{figure}
\section{Discussion}
We have identified five nondecaying modes
corresponding to each wavenumber: an adiabatic growing mode,
a baryon isocurvature mode, a CDM isocurvature mode,
a neutrino density isocurvature mode, and
a neutrino velocity isocurvature mode.
Under the assumption that the primordial perturbations
are small enough so that the linear theory suffices,
two-point derived observables are completely determined by the
generalization of the power spectrum given by the
correlation matrix
\begin{eqnarray}
\langle A_m(k)A_n(-k)\rangle \end{eqnarray}
where the indices $(m, n=1,\ldots, 5)$ label the modes,
independently of whether or not
the primordial fluctuations were Gaussian.
In Figs.~1 and 2 the shape of the CMB moments for
the neutrino isocurvature modes are indicated.
In this computation the cosmological parameters
$H_0=50{\rm km~s}^{-1},$
$\Omega _b=0.05,$ $\Omega _c=0.95$ were assumed.
In a companion paper, we examine prospects for
constraining the amplitudes of these modes using
upcoming MAP and PLANCK data.
\vskip 12pt
\noindent
{\bf Acknowledgements:} We would like to thank David Spergel
for useful discussions at an early stage of this project.
The CMB spectra were computed using a modified
version of the code CMBFAST written by Uros Seljak and Matias
Zaldarriaga. We would like to thank Anthony Challinor, Anthony
Lasenby, and Dominik Schwarz for useful comments.
This work was supported in part
by the UK Particle Physics and Astronomy Research
Council. KM was supported by the Commonwealth Scholarship
Commission in the UK.
|
\section{Introduction}
Despite the impressive success of the Standard Model, few are convinced that
it is the final theory of particle interactions. For example, the
supersymmetric modification of the Standard Model yields a very promising
framework in which we are able to understand the stability of the
electroweak scale. The Minimal Supersymmetric Standard Model (MSSM) provides
a plethora of new phenomenological predictions which range from new charged
and colored particles actively searched for in accelerators, to cold dark
matter candidates, to new CP-violating phenomena such as the electric dipole
moments of the neutron and electron which are generated if the additional
CP-violating phases in the MSSM are non-zero. In this work, we study in
detail the predictions of the MSSM for the electric dipole moment of the
mercury atom and derive the constraints on the MSSM phases from the
experimental limits on $d_{Hg}$.
The null experimental results for the electric dipole moments (EDMs) of the
electron, neutron, heavy atoms and diatomic molecules \cite
{nEDM,mEDM,eEDM,molEDM} can in general place very strong constraints on the
CP-violating sector of a new theory and probe energy scales which are
inaccessible for direct observations at colliders \cite{KL}. In general, the
relevant contribution to the dipole moments at scales of $\sim$1 GeV can be
parameterized in terms of effective operators of different dimensions
suppressed by corresponding powers of a high scale $M$ where these operators
were generated:
\begin{equation}
{\cal L}_{eff}=\sum_{n\geq 4} \frac{c_{ni}}{M^{n-4}}{\cal O}^{(n)}_{i},
\end{equation}
Here ${\cal O}_{i}^{(n)}$ are operators of dimension $n$, with its field
content, Lorentz structures, etc., denoted by $i$. The fields relevant for the
low-energy dynamics of interest are gluons, the three light quarks, the
electron, and the electromagnetic field. This general form is independent of
the particular construction of the new theory, and the details of a given
model enter only through the values of the coefficients $c_{ni}/M^{n-4}$.
In the MSSM, the number of operators which can generate an EDM is
considerably smaller than in the generic case. In fact, all four-fermion
operators are numerically insignificant. They can be generated in the MSSM
only with additional factor of order $(m_{q}/M_{SUSY})^{2}$ modulo possible
nontrivial flavor structure of the soft-breaking sector. Here we assume the
minimal scenario with flavor-blind breaking of supersymmetry and therefore
we can safely drop all four-fermion CP-violating operators. Hence,
the relevant part of the effective Lagrangian at the scale of 1 GeV contains
the theta term, the three-gluon Weinberg operator, the EDMs of quarks and
electron and the color EDMs (CEDMs) of quarks,
\begin{eqnarray}
{\cal L}_{eff} &=&\theta \frac{g_{s}^{2}}{32\pi ^{2}}G_{\mu \nu }^{a}
\mbox{$\tilde{G}$}_{\mu \nu }^{a}+w\frac{g_{s}^{3}}{6}f^{abc}G_{\mu \nu }^{a}
\mbox{$\tilde{G}$}_{\nu \alpha }^{b}G_{\alpha \mu }^{c} \label{eq:eff} \\
&&+i\sum_{i=u,d,s}\frac{d_{i}}{2}\bar{q}_{i}F_{\mu \nu }\mbox{$\sigma$}_{\mu
\nu }\mbox{$\gamma_{5}$}q_{i}+i\sum_{i=u,d,s}\frac{\tilde{d}_{i}}{2}\bar{q}
_{i}g_{s}t^{a}G_{\mu \nu }^{a}\mbox{$\sigma$}_{\mu \nu }\mbox{$\gamma_{5}$}
q_{i}+i\frac{\tilde{d}_{e}}{2}\bar{e}F_{\mu \nu }\mbox{$\sigma$}_{\mu \nu }
\mbox{$\gamma_{5}$}e. \nonumber
\end{eqnarray}
We will assume here that the PQ mechanism of $\theta$-relaxation \cite{PQ}
eliminates $\theta\sim O(1)$ and sets $\theta$ to $\theta_{eff}$ at the
minimum of
the axion potential. When both the CEDMs and Weinberg operator
are absent, the value of $\theta_{eff}$ is exactly zero. However, nonzero $w$
and $\tilde d_i$ induce a linear term in the axion potential, and the
effective value of $\theta$ is different from zero. This value leads to an
additional contribution to the EDM of the neutron, usually ignored in the
literature.
The coefficients in front of the operators in Eq. (\ref{eq:eff}) can be
calculated for any given model of CP-violation and then evolved down to
the low-energy scale, using standard renormalization group techniques.
In the MSSM, in particular, one can compute
effective Lagrangian (\ref{eq:eff}) for any given point in the
supersymmetric parameter space.
Then, to get the final predictions for EDMs,
one has to take various matrix elements for these operators over hadronic,
nuclear and atomic states \cite{KL,KKZ,KKY,KK}. In most cases this is a
source of major uncertainty, especially when hadronic physics is involved.
The exception is the case of a paramagnetic atom, in which the
EDM is generated by
the electron EDM $d_{e}$, and where the effects of
nuclear CP-odd moments induced
by the rest of the operators in (\ref{eq:eff}) can be safely neglected.
The EDM of $^{205}$Tl is extremely sensitive to $d_{e}$ due to a
very large relativistic enhancement factor $c\sim -600$, which relates the
EDM of the atom with $d_e$, $d_{Tl}=c d_e$. The experimental bound on
the EDM of the thallium atom \cite{eEDM}, combined with good
stability of atomic calculations (see \cite{KL} and
references therein), leads to the
following limit on the EDM of the electron:
\begin{equation}
d_{e}<4\cdot 10^{-27}e\cdot cm. \label{de}
\end{equation}
Therefore, the calculation of $d_{e}$ in the MSSM
gives the most reliable limits
on CP-violating phases. It is clear, however, that the electron EDM limit
alone cannot exclude the possibility of large CP-violating phases. This is
because $d_{e}$, as any other coefficient in Eq. (\ref{eq:eff}), is in
general a function of {\em several} CP-violating phases, and mutual
cancellations are possible. This is what happens, for example, in the MSSM
with the minimal number of parameters in the soft-breaking sector (see recent
works \cite{Kane,FO1}). In the MSSM, it is well known that there are two
independent CP-violating phases, $\theta_{\mu }$ and $\theta _{A}$,
associated with the supersymmetric Higgs mass parameter $\mu $ and the soft
supersymmetry breaking trilinear parameter $A$. The calculation of the
relevant one loop diagrams determines $d_{e}$ as a function of these two
phases. If the phases are small, $d_{e}$ is simply a linear combination of
$\theta _{A}$ and $\theta _{\mu }$. Therefore even a constraint as strong as
that given in (\ref{de}) leaves a band on the $\theta _{A}$--$\theta _{\mu
}$ plane, along which a cancellation occurs and the phases are not
constrained. In general, a second constraint could be expected to lift
this degeneracy and place a strong constraint on both phases. It has
been common to use the limit on the neutron EDM as this second
constraint. Although there are large uncertainties in
the calculation of the neutron EDM, as we argue below, when the limit on
the neutron EDM is used,
cancellations in the electron EDM occur in many of the same regions as
cancellations in the neutron EDM. Therefore, one is led to the conclusion
that large phases are still possible.
In what follows, we critically reexamine the reliability of
the calculation of the EDM of the neutron in the MSSM. We demonstrate that
this calculation is subject to very large hadronic uncertainties, which
makes the extraction of the limits on CP-violating phases in MSSM
tenuous. Instead, we propose that useful limits may be obtained from the
limits on the EDM of the mercury atom. This arises from the T-odd
nucleon-nucleon interaction in the MSSM, induced mainly due to the CEDMs of
light quarks. This interaction gives rise to an EDM of the mercury atom by
inducing the Schiff moment of mercury nucleus. We demonstrate that the
degree of QCD uncertainties related to this calculation is in fact
smaller than in the case of $d_n$ and that it is possible to calculate the
T-odd nucleon-nucleon interaction as a function of the different MSSM
phases. As an example, we proceed with the calculation of the EDM of the
mercury atom in one specific point of the supersymmetric parameter space
where all squark, gaugino masses, $|\mu|$ and $|A|$ parameters are set
equal. This ``pilot'' calculation demonstrates the sensitivity of
$d_{Hg}$ to the CP-violating phases of MSSM. We find in this case that
$d_{Hg}$ provides somewhat better limits on CP-violating phases than $d_e$.
We proceed further and
combine mercury EDM and electron EDM constraints to exclude most of the
parameter space on
$\theta_A$--
$\theta_\mu$ plane in this toy example. Finally, we consider more realistic
constraints over the supersymmetric plane when supersymmetry
breaking scalar and gaugino masses are unified at the GUT scale. In this
case, we find that the limits on CP-violating phases obtained from
$d_{Hg}$ is no longer more restrictive than $d_e$,
as the RG evolution of soft-breaking parameters makes squarks and gluino
heavier than sleptons, charginos and neutralinos.
The combined limits are still very powerful as the cancellation of
different supersymmetric contributions typically occur in different regions
of parameter space.
\section{The Neutron EDM in the MSSM}
Limits on the neutron EDM are commonly used to set constraints on new CP
violating interactions. In particular, the upper limit to $d_{n}$ is often
used to limit the size of the CP-violating phases in the MSSM \cite{ko}.
The current experimental limit on the EDM of the neutron is
\begin{equation}
d_{n}<1.1\cdot 10^{-25}e\cdot cm,
\end{equation}
Indeed, the EDM of the neutron receives contribution from all operators
listed above in Eq. (\ref{eq:eff}) except $d_{e}$. However, there is a
complication in using the neutron EDM as compared to the electron EDM, due to
QCD uncertainties which make the extraction of the limits on CP-violating
phases in the fundamental Lagrangian problematic. We demonstrate two aspects
of this problem below.
The most straightforward contribution to the EDM of the neutron is due to
the quark EDM operators. It is usually estimated using nonrelativistic
SU(6) quark model. The result,
\begin{equation}
d_{n}=\frac{4}{3}d_{d}-\frac{1}{3}d_{u}, \label{eq:naive}
\end{equation}
can be compared, in fact, with the model calculations \cite{MP} and lattice
simulations of light quark tensor charges in the nucleon \cite{ADHK}. The
matrix elements for the tensor charges of the nucleon are defined by
\begin{equation}
\langle N|{\bar{\psi}_{q}}\sigma _{\mu \nu }\psi _{q}|N\rangle =\delta q
\overline{N}\sigma _{\mu \nu }N,
\end{equation}
whereas the axial charges are defined by
\begin{equation}
\langle N|{\bar{\psi}_{q}}\gamma _{\mu }\gamma _{5}\psi _{q}|N\rangle
=\Delta q\overline{N}~\gamma _{\mu }\gamma _{5}N
\end{equation}
In the Na\"{\i }ve quark model, both Lorentz structures correspond to the
spin of a nonrelativistic quark. In this case $\delta u=\Delta u=-1/3$,
$ \delta d=\Delta d=4/3$, $\delta s=\Delta s=0$, yielding eq.
(\ref{eq:naive}). Note that isospin symmetry gives us
$(\Delta u)_{n}=(\Delta
d)_{p}$, etc. However, as argued in \cite{EF}, since it appears
that the contribution to the nucleon spin from the strange quark ($\Delta
_{s}$) is non-vanishing \cite{EMC},
the na\"{\i }ve quark model may not be sufficient to describe the
quark EDM contribution to the neutron EDM. While it is not the axial charges
which need to be considered for the calculation of the neutron EDM, but
rather the tensor charges, the
departure of the axial charge values from their NQM values indicates that
more realistic (non-NQM) values of the tensor charges ($\delta q$) must be
used. According to calculations based on Lattice QCD \cite{ADHK}, the
tensorial charges of up and down quarks in the proton
should be read as $\delta u\sim 0.8$ and $\delta d\sim -0.23$.
This means that the naive nonrelativistic formula predicts the EDM of the
neutron due to the quark EDMs to be 1.5-1.7 times larger than the lattice
result. Slightly different values of $\delta u\sim 1.1$ and $\delta d\sim
-0.4$ can be derived from the SU(3) chiral quark soliton model \cite{MP}.
The tensor charge
of the strange quark is found to be consistent with zero in both methods
\cite{MP,ADHK}. This is due to the fact that the $\bar{s}\mbox{$\sigma$}
_{\mu \nu }s$ operator is odd under charge conjugation which must result in
the Zweig-type suppression of this matrix element over the neutron state.
Even with the usual $m_{s}/m_{d}$ enhancement of this operator, it is
unlikely to be important. This does not exclude other possible CP-violating
operators involving the $s$-quark, CEDM or generic four-fermion operators, as
their contributions to the EDM of the neutron can be significant \cite
{KKZ,HPs}. Departures from the predictions of the non-relativistic quark
model were recently considered in \cite{bartl}.
Unfortunately, the quantitative evaluation of the remaining contributions to
the neutron EDM is complicated due to of our lack of knowledge about strong
interaction dynamics at 1 GeV and below. Typically, one resorts to Naive
Dimensional Analysis (NDA) \cite{NDA}, formulated within the
constituent quark framework. This method is, however, only
an order of magnitude estimate to be used when other methods of calculation
fail to produce an answer. When the problem of estimating $d_N$ due to
$\tilde d_{u,d}$ is considered, there are several possible answers in the
literature:
\begin{eqnarray}
d_N\simeq e0.7(\tilde d_u + \tilde d_d)&
{\rm Ref.} \cite{KK}\label{cle}\\
d_N\sim\fr{eg_s}{4\pi}(O(1)\tilde d_u + O(1)\tilde d_d) &{\rm NDA, Ref. }
\cite{Duff}\label{duffe}
\end{eqnarray}
We have chosen a normalization where $g_s$ is included
in the definition of the operator in (\ref{eq:eff}) and
correspondingly include an
additional $g_s$ in the estimate (\ref{duffe}).
The first result is based on a combination of chiral perturbation
theory and QCD sum rules. The latter estimate
is derived with the use of NDA \footnote{We note that the estimate in
\cite{Barbieri} is suppressed by an additional factor of $g_s/4\pi$.}. For
a realistic choice of the strong coupling constant at the scale of $1$ GeV,
$g_s\sim
\sqrt{0.5\cdot 4\pi}=2.5$, the overall numerical coefficient in eq.
(\ref{duffe}) is about 3.6 times smaller than in (\ref{cle}). Estimates
based on NDA imply that for natural relations among
coefficients $d_{i}/e\sim \tilde{d}_{i}$, the effects of color EDMs on the
electric dipole moment of the neutron are negligible and the result can
indeed be approximated by the linear combination of EDMS of quarks.
In fact, it is possible to show that the CEDMs can lead to a substantially
larger contribution to the neutron EDM than some of the
predictions based on
NDA. The easiest
way to see that CEDMs can be numerically important is to calculate the
effective $\theta$-term induced by CEDMs in the presence of the PQ
symmetry and then use the result for $d_N(\theta)$.
This value, $\theta_{eff}(\tilde d_i)$ can be calculated
within the current algebra approach, in a manner similar to the calculation of the
vacuum topological susceptibility \cite{SVZ,BU,Posp}. The dynamically
induced theta term can be expressed in the following compact form:
\begin{equation}
\theta_{eff}=-\frac{m^2}{2}\left(\frac{\tilde d_u}{m_u}+ \frac{\tilde d_d}
{m_d}+\frac{\tilde d_s}{m_s}\right). \label{eq:theta}
\end{equation}
Here, $m^2$ is the ratio of the quark-gluon condensate to the quark
condensate. It is known to good accuracy from QCD sum rules
\cite{SR} that,
\begin{equation}
m^2=\frac{\langle 0|g \bar q (G\sigma)q|0\rangle}{\langle 0| \bar q
q|0\rangle}\simeq 0.8\mbox{GeV}^2. \label{eq:m0}
\end{equation}
The accuracy of the estimate (\ref{eq:theta}) is of order
$m^2_{\pi,K}/m^2_{\eta^{\prime}}$, which is acceptable for our purposes.
If no interference with other terms is expected, then the expression (\ref
{eq:theta}) must be less than the current limit on $\theta$, extracted from
the same neutron EDM data. Using the fact that in the simplest variant of the
MSSM, $\tilde d_d/m_d=\tilde d_s/m_s$, and assuming for a moment that this
is the only contribution to the EDM of the neutron, one can obtain the
following, quite stringent, level of sensitivity for the CEDM:
\begin{equation}
\tilde d_d<10^{-25} cm. \label{eq:thelim}
\end{equation}
This fact alone suggests that CEDMs may contribute significantly to
the EDM of the neutron, typically at the level of the prediction (9) and
an order of magnitude above NDA
predictions. Remarkably, the main uncertainty in the limit
(\ref{eq:thelim}) comes not
from the calculation of $\theta(CEDM)$, but rather from the
principal difficulties
in calculating $d_N(\theta)$. In the standard approach \cite{CDVW}, the
chiral loop diagram is used to estimate $d_N(\theta)$. This loop is
logarithmically divergent in the exact chiral limit and therefore is
distinguished from the rest of the contributions. For realistic values of the
parameters, however, this logarithm is not large and other contributions can
be equally important. This makes the whole calculation problematic even
in predicting the sign of the $\theta$ term contribution to $d_N$.
Besides $d_N(\theta(CEDM))$, one should also consider direct CEDM-induced
contributions to the EDM of the neutron which can be computed within the same
chiral loop approach \cite{KK}. Combining different contributions, we can
symbolically write the result for the EDM of the neutron in the following
form:
\begin{equation}
d_N\simeq 0.8 d_d- 0.23 d_u + e\left[\tilde d_u \left(c_1\ln\frac{\Lambda
}{m_\pi}+c_2\right)+\tilde d_d \left(c_3\ln\frac{\Lambda}{m_\pi}
+c_4\right)+ \tilde d_s\left(c_5\ln\frac{\Lambda}{m_K}+c_6\right)\right].
\label{eq:symbolic}
\end{equation}
The coefficients $c_1$, $c_3$ and $c_5$ were estimated in Ref. \cite{KK} to
be $c_1\ln(m_\rho/m_\pi)=c_3\ln(m_\rho/m_\pi)\simeq 0.7$ and
$c_5\simeq 0.1$. The cutoff parameter $\Lambda$
corresponds to scales where chiral perturbation theory breaks down, that is,
of order $m_\rho$. In the exact chiral limit, $m_\pi,~m_K\rightarrow 0$, and
the logarithmic terms dominate. In practice, however, the logarithmic terms
are numerically not distinguished from the coefficients $c_2$, $c_4$ and
$c_6$, which are {\em a priori} comparable with $c_1$, $c_3$ and $c_5$ and are
not calculable in this approach. It is clear then that these terms can
change both the value and the signs of different contributions to $d_N$.
Therefore, although in principle very important as an order of magnitude
estimate, Eq. (\ref
{eq:symbolic}) fails to provide $d_N$ as a known function of individual
$\tilde d_i$-contributions and, ultimately, of different CP-violating phases.
As emphasized in Ref. \cite{W},
the NDA estimate of $d_n(\theta)$ essentially reproduces the calculation
of Ref. \cite{CDVW}. The source of the disagreement in the case of
$d_n(CEDMs)$ can be traced to the problem of estimating the CP-odd
$\pi^+pn$--vertex, proportional to the
matrix element $\langle p|\bar u g_s(G\mbox{$\sigma$})d|n\rangle$.
In Ref. \cite{KK} this
matrix element was estimated to be -1.5 GeV$^2$ and is essentially
proportional to the quark-gluon condensate parameter $m^2\sim 0.8{\rm
GeV}^2$ (\ref{eq:m0}).
On the other hand, it can be shown
that NDA suggests for this matrix element a value of order $4\pi f_\pi^2\sim
{\rm GeV}^2/(4\pi)$, i.e. one order of magnitude smaller.
This difference is related to the fact that the NDA assumes
nonrelativistic quarks whose chromomagnetic interactions are suppressed,
whereas QCD sum rules use more realistic descriptions of hadronic
properties in terms of vacuum quark--gluonic condensates.
To summarize this discussion, the extraction of reliable limits on the
CP-violating phases in the MSSM from the EDM of the neutron is difficult and
uncertain. Even the best estimates of $d_n$
based on the ``chiral logarithm''
approach \cite{KK}, bear a large degree of uncertainty and
cannot produce a precise prediction for $d_n$ as a function of the
CP-violating phases. Useful limits are still available from the electron
EDM; however, the magnitude of the phases is not terribly constrained on
this basis alone, due to cancellations in the various MSSM contributions
to $d_e$. Fortunately, the EDM of the neutron is not the only source of
information about CP-violation in the strongly-interacting sector.
Limits on T-violating nuclear forces are provided by experiments
aimed at the detection of the EDM of paramagnetic atoms, among which the
EDM of $^{199}$Hg atom is the most constraining. In what follows, we
will discuss the constraints these limits provide, both alone and in
conjunction with the electron EDM limits.
\section{CP-violating nucleon-nucleon interaction in MSSM}
The limits on T-odd nuclear forces extracted from the atomic experiments are
in general very important for particle physics \cite{KL}. In the case of
diamagnetic atoms, the most impressive limit is obtained for the EDM of
$^{199}$Hg \cite{mEDM}:
\begin{equation}
d_{Hg}<9\cdot 10^{-28} e \cdot cm. \label{eq:Hglim}
\end{equation}
The electric screening of the electric dipole moments of the atom's
constituents is violated by the finite size of the nucleus and can be
conveniently expressed by the Schiff moment $S$, which
parametrizes the effective interaction between the electron and
nucleus of spin ${\bf I}$, $V_{eff}=-eS({\bf I \nabla})\delta({\bf r})$
\cite{KL}.
Atomic calculations derive the atomic EDM as a function of $S$ and
translate the experimental result (\ref{eq:Hglim}) into the limit on the
Schiff moment of the nucleus:
\begin{eqnarray}
d_{Hg}=S\cdot 3.2\cdot 10^{-18}\mbox{fm}^{-2} \nonumber \\
S<2.8\cdot 10^{-10}e\mbox{fm}^3. \label{eq:Sexp}
\end{eqnarray}
The Schiff moment of the nucleus can be induced either due to the Schiff moment of
the valence nucleons or due to the breaking of time invariance in the
nucleon-nucleon interaction, the latter being enhanced by the collective
effects in the nucleus. The calculation of the Schiff moment of the nucleus,
originating from various $\bar NN \bar N^{\prime}i\gamma_5N^{\prime}$
interactions was done in the single particle approximation with square-well
and Woods-Saxon potentials \cite{KSF}. The results show that the Schiff
moment of mercury is primarily sensitive to the $\bar pp \bar n i\gamma_5 n$
interaction. If we parameterize the coefficient in front of this interaction
as $\xi G_F/\sqrt{2}$, the nuclear calculation \cite{KSF} provides us with the
following value for $S$:
\begin{equation}
S=-1.8\cdot 10^{-7}\xi e \cdot\mbox{fm}^3. \label{eq:Sth}
\end{equation}
Combined with Eq. (\ref{eq:Sexp}), it gives the following constraint on
$\xi$:
\begin{equation}
\xi<1.9\cdot 10^{-3}.
\end{equation}
Questions concerning the
calculation of the strength of $\bar pp \bar n i\gamma_5 n$ interaction
induced by different operators was considered in (\ref{eq:eff}),
\cite{KL,KKY,KK}. The effective theta term, the Weinberg three-gluon
operator and the CEDMs of quarks can generate this interaction.
Numerically, the contributions provided by the CEDMs of up and down quarks
are the most important and we concentrate our analysis on them, trying to
incorporate the effect of $\tilde d_s$ as well.
\begin{figure}[thb]
\begin{center}
\epsfig{file=F1.eps,height=3in}
\vspace{-1.5in}
\caption{Pseudoscalar meson exchange diagrams, inducing
$\bar pp \bar n i \gamma_5 n $
interaction. } \ \end{center}
\vspace{-0.3in}\end{figure}
Following \cite{KKY,KK}, we approximate the T-violating nucleon-nucleon
interaction by pseudoscalar exchange, as shown in Fig 1. In the limit of
exact chiral symmetry this exchange has the power-like singularity $
m_\pi^{-2}$, to be compared with the logarithmic singularity in the case
of the EDM
of the neutron. The CP violation resides in proton-meson vertex which can be
calculated with QCD sum rules and current algebra
techniques. The CP-conserving meson-neutron vertex is sufficiently well
known from SU(3)-relations in baryon octet decay amplitudes and from the
axial charges of nucleons. If only $\tilde d_u$ and $\tilde d_d$ are
present, the pion exchange dominates $\eta$ exchange by a factor
$m_\eta^2/m_\pi^2\simeq 16$. In the MSSM, though, the strange quark CEDM is
enhanced relative to that of the down quark by a factor $m_s/m_d$ and
$\eta$ meson exchange is not {\em a priori} negligible. In the chiral
approach, CP-violating vertices of interest can be reduced to the following
set of matrix elements:
\begin{eqnarray}
\bar g_{\pi pp} =\frac{\tilde d_u + \tilde d_d}{4f_\pi} \left(
\langle p|\bar u
g_s(G\mbox{$\sigma$})u - \bar d g_s(G\mbox{$\sigma$})d |p\rangle \right) +
\nonumber\\
\frac{\tilde
d_u - \tilde d_d}{4f_\pi}\left(\langle p|\bar u g_s(G\mbox{$\sigma$})u
+\bar d g_s(G \mbox{$\sigma$})d |p\rangle -
m^2\langle p|\bar u u +\bar d d |p\rangle\right) \nonumber \\
\bar g_{\eta pp} =-\frac{\tilde d_s}{\sqrt{3}f_\pi}
\left (\langle p|\bar s g_s(G\mbox{$\sigma$})s|p\rangle -
m^2\langle p|\bar s s|p\rangle\right)
\label{eq:matrel}
\end{eqnarray}
Here $m^2$ is the ratio of quark-gluon condensate to quark condensate
introduced earlier in Eqs. (\ref{eq:theta}) and (\ref{eq:m0}).
At this point our results are already slightly different from
\cite{KKZ,KKY,Z}. Namely, we have
included additional contributions related to the fact that
the octet combination of color EDM operators has the quantum numbers of the
$\pi^0$ and $\eta$ fields which can therefore be produced from the vacuum.
$\pi^0$, for example,
can be ``rescattered'' on the nucleon with an amplitude proportional to
$(m_d+m_u)\langle N|\bar uu +\bar dd|N\rangle$. As a result, the diagram
shown in Fig. 2 is responsible for a contribution directly proportional to
$m^2$ which is effectively of the same order as the direct contribution
considered in
\cite{KKY,KK}.
\begin{figure}[thb]
\begin{center}
\epsfig{file=F1a.eps,height=1.5in}
\caption{Additional contribution to the $\bar g_{\pi NN}$, proportional
to the nucleon sigma term} \ \end{center}
\vspace{-0.3in}\end{figure}
Further calculation relies on QCD sum rules \cite{SVZ} and low-energy
theorems in QCD. Matrix elements from
$q g_s(G\mbox{$\sigma$})q$ operators were evaluated in \cite{KKZ,KKY,Z}:
\begin{equation}
\langle p|\bar q g_s(G\mbox{$\sigma$})q|p\rangle\simeq \frac{5}{3} m^2
\langle p|\bar qq|p\rangle. \label{eq:gluonium}
\end{equation}
The matrix
elements over the proton can be obtained from baryon mass splittings and
pion-nucleon scattering data. Here we take the following values for the
matrix elements of $\bar qq$ over the nucleon \cite{Z}:
\begin{equation}
\langle p|\bar uu|p\rangle\simeq 4.8;\;\;\langle p|\bar dd |p\rangle\simeq
4.1;\;\; \langle p|\bar ss|p\rangle\simeq 2.8 \label{eq:me}
\end{equation}
These values of $\bar qq$ matrix elements correspond to the choice
$m_u=4.5$ MeV, $m_d=9.5$ MeV and $m_s=175$ MeV. The values of these
matrix elements, together with the factorization formula (\ref{eq:gluonium}),
suggest that T-odd
nucleon-nucleon forces are primarily
sensitive to $\tilde d_u-\tilde d_d$ and insensitive to $\tilde
d_u+\tilde d_d$, simply because the contribution to $\bar g_{\pi pp}$
proportional to $\tilde d_u+\tilde d_d$ in (\ref{eq:matrel}) is relatively
suppressed by
\begin{equation}
\frac{\bar g_{\pi pp}(\tilde d_u+\tilde d_d)}
{\bar g_{\pi pp}(\tilde d_u-\tilde d_d) } \simeq
\frac{2\langle p|\bar u u - \bar d d
|p\rangle}{\langle p|\bar u u + \bar d d|p\rangle} \sim 0.2
\label{+/-}
\end{equation}
In this sense, the contribution furnished by $\theta_{eff}$ is numerically
insignificant because $\bar g_{\pi pp}$ generated by $\theta$ is also
proportional to $\langle p|\bar u u - \bar d d |p\rangle$.
Thus, these simple considerations suggest that due to the
numerical dominance of the
triplet combination of color EDM operators, the final answer for
$\xi$ takes the following form:
\begin{equation}
\xi=G_F^{-1}\frac{3 g_{\pi pp}m_0^2}{f_\pi m_\pi^2} (\tilde d_d-\tilde
d_u-0.012\tilde d_s), \label{eq:xi}
\end{equation}
We can see that the contribution of the strange quark CEDM is numerically
suppressed, mainly due to the additional smallness of $\eta NN$
CP-conserving interaction as compared to $g_{\pi NN}$.
Combining equations (\ref{eq:Sexp}), (\ref{eq:Sth}) and (\ref{eq:xi}), we
arrive at the following prediction for the EDM of the mercury atom:
\begin{equation}
d_{Hg}=-(\tilde d_d-\tilde d_u-0.012\tilde d_s)\times 3.2 \cdot 10^{-2} e,
\label{eq:EDMhg}
\end{equation}
where the the numerical coefficient $3.2 \cdot 10^{-2}$ corresponds to the
choice of light quark masses given above.
Using the experimental limits (\ref{eq:Hglim}), we deduce the very strong
constraint on the following combinations of the CEDMs of quarks:
\begin{equation}
| \tilde d_d-\tilde d_u-0.012\tilde d_s | < 3.0\cdot 10^{-26} cm.
\label{lim}
\end{equation}
It is important to note that the quark
EDM operators cannot induce a large value for $S$. They do not induce the
$\bar ni\gamma_f n \bar pp$ interaction, and their contribution to the
Schiff moment of the nucleus is associated only with electric dipole
moment of the external valence nucleon \cite{KL}. Current limits on $d_{Hg}$
are only sensitive to quark EDMs larger than $ 10^{-24}e\cdot cm$ and thus
these operators can be safely neglected.
Similarly, the potential contribution from the three gluon operator $GG\tilde
G$ to $d_{Hg}$ are small.
We rely here on the QCD sum rule estimates \cite{Volodia}, showing no
significant
contribution from this operator to the T-odd nucleon-nucleon forces and
thus to the EDM of mercury.
Finally, we would like to comment on the accuracy of the predictions
(\ref{eq:EDMhg}) and (\ref{lim}), distinguishing between the error
in the overall coefficients and the errors in the relative coefficients
of $\tilde d_i$-proportional contributions. The uncertainties of the atomic
calculations of $d_{Hg}(S)$ and nuclear calculations of $S(\xi)$
mostly affects the overall coefficients. Although the uncertainty in the overall
coefficient can be significant \cite{KL}, it
is acceptable for our purpose as it influences only the width of the
allowed region in $\theta_\mu-\theta_A$-plane. What is more important,
however, is that the prediction of the relative coefficients in front of
individual $\tilde d_i$ in eqs.(\ref{eq:EDMhg}) and (\ref{lim}) can be
done in a more reliable way and we estimate that the accuracy of keeping
the triplet combination $\tilde d_d-\tilde d_u$ and neglecting
$\tilde d_d+\tilde d_u$, eq. (\ref{+/-}), is at the level of 20\%.
In effect, it makes the constraints imposed by $d_{Hg}$ much more useful than
those provided by $d_n$. Another advantage of the approach for calculating
$d_{Hg}$ and $d_n$, developed in refs. \cite{KKY,KK,KL} and applied here,
is that it reduces the error from the poor knowledge of the light quark masses.
Indeed, even in the case of the na\"{\i }ve formula for the EDM of the
neutron, $d_n\simeq (4d_d-d_u)/3$, the individual quark EDM contributions
are proportional to $m_{u,d}$ which are known to 50\%. In the present
approach, the answer for $\xi$ is ultimately proportional to a linear
combination of $m_i\langle 0|\bar qq|0\rangle$, which can be rewritten
as $f_{\pi}^2m_\pi^2$ times the function which depends only on the
{\em ratio} of light quark masses, known to much better
accuracy than the masses themselves.
\section{The limits on the MSSM CP-violating phases}
In previous work \cite{FO1}, limits on the neutron and electron dipole
moments were used to constrain the two independent phases (of $\mu$ and
$A$) in the MSSM assuming that all the terms in the Higgs potential and all
gaugino masses are real and that all of the $A$-parameters are equal at the
GUT scale and share a common phase. In absolute terms, the phases are not
overly constrained, $\theta_\mu \stackrel{<}{{}_\sim} 0.3$, for $\theta_A
\simeq
\pi/2$. The reason for the lax limits, are several cancellations in the
various contributions to the EDMs. Furthermore, in some regions of parameter
space, these cancellations occur simultaneously for the electron and neutron
EDMs.
As we argued above, there are several reasons to suspect that the limit due
to the neutron EDM must be treated with caution. Instead, we have argued
that the limit coming from the EDM of Hg is the result of a ``cleaner"
calculation and carries fewer QCD uncertainties. In what follows, we will
explore in detail the limits on the two phases using the constraint based on
the EDM of $^{199}$Hg (\ref{lim}) derived above. We will compare these
constraints on the phases to that obtained from the electron EDM. As we will
see, the cancellations in the EDMs do not always occur at the same points in
parameter space. To demonstrate the importance of the mercury EDM limit, we
first consider a SUSY model with a single mass scale. We then present
general results which assume gaugino and sfermion mass universality at the
GUT scale.
Following \cite{FOS,FO}, we analyze the limits on $\theta_A$ and $\theta_\mu$
for different values of supersymmetric parameters. To demonstrate the
sensitivity of the mercury EDM to a common scale of the supersymmetric masses
with arbitrary and uncorrelated phases, we choose $m_{\tilde f}\simeq
M_{\lambda_i}\simeq |\mu|\simeq |A_k|$ at the electroweak scale and take $
\tan\beta=2$. In Figures 3a and 3b, we show the sensitivity to the EDM of the
mercury atom for the cases of $\sin\theta_A=1,~\sin\theta_\mu=0$ and $
\sin\theta_A=0,~\sin\theta_\mu=1$. At this particular point of the
supersymmetric parameter space all of the calculations are significantly
simplified. When all soft-breaking parameters are
sufficiently heavy, close to the TeV scale, the chargino and gluino
propagators can be simply expanded in $v_1/M$ or $v_2/M$ and only the zeroth
and first order terms in the expansions need be kept. If needed, for lower
values of gaugino masses, the results can be generalized to include all
effects of mixing in the gluino and chargino sectors.
\begin{figure}
\normalsize
\begin{center}
\mbox{\epsfxsize=100mm\epsffile{r3.eps}}
\vspace{1cm}
\mbox{\epsfxsize=100mm\epsffile{r4.eps}}
\vspace{1cm}
\end{center}
\caption{\label{fig:mssm}The sensitivity of the EDMs of mercury and electron
to the scale of the soft-breaking parameters with a) maximal phase of
$A$ ($\theta_A=\pi/2$, $\theta_\mu=0$), and b) maximal phase of
$\mu$ ($\theta_A=0$, $\theta_\mu=\pi/2$). We've taken
$|\mu|=|A|=m_{\tilde Q}=m_{\tilde U}=m_{\tilde
D}=M_{\lambda_i}\equiv M$. The horizontal line is the current
experimental limit.
}
\end{figure}
The calculation of the chromoelectric dipole moments of quarks in MSSM was
done in the series of papers \cite{Duff,KZ,IN}. When the CEDMs of quarks are
induced by $\theta_A$, (as in Fig 3a), the result is dominated by gluino
exchange, with very small corrections coming from $\lambda_1$-exchange:
\begin{eqnarray}
\tilde d_d=-\eta\frac{m_d|A|\sin\theta_A}{16\pi^2M^3} \left(\frac{5g_3^2}{18}-
\frac{g_1^2}{108}\right) \\
\tilde d_u=-\eta\frac{m_u|A|\sin\theta_A}{16\pi^2M^3} \left(\frac{5g_3^2}{18}+
\frac{g_1^2}{54}\right). \nonumber
\label{eq:phiA}
\end{eqnarray}
Here $\eta$ denotes the renormalization group factor which reflects the
QCD evolution of the color EDM from the weak scale to 1 GeV. When the color
EDM operator is defined as in eq. (\ref{eq:eff}), its anomalous dimension
is negative and small so that the overall renormalization of $\tilde d_i$ is
not important. The alternative definition of color EDM operator,
frequently occurring
in literature, is $\frac{1}{2}\tilde{d}'\bar{q}t^{a}G_{\mu \nu }^{a}
\mbox{$\sigma$}_{\mu \nu }\mbox{$\gamma_{5}$} q$, where $g_s$ is included
in $\tilde{d}'$. Defined this way, this operator acquires a
renormalization factor roughly proportional to $g_s(1{\rm
GeV})/g_s(M_Z)\simeq 2$ which is smaller than the value 3.3 quoted
in \cite{Duff}. This is because in Refs. \cite{Duff,IN} a very large
coupling constant at low energies, $\alpha_s\simeq 2$, is used. There is,
however, an important numerical contribution to $\eta$ which reflects the
suppression of light quark masses at the high energy scale,
$m_d(M_Z)/m_d({\rm 1GeV})$. We choose
to use low energy values for $m_u$ and $m_d$, 4.5 and 9.5 MeV,
and include the quark mass RG factor into $\eta$.
For the scale $M$ of order $M_Z$, $\eta$ is
numerically close to 0.35 and is mainly due to the suppression of the quark
masses at the high energy scale. This suppression factor was omitted
in Ref. \cite{Duff} where $m_u(M_Z)=8$ MeV is used.
Combining all numerical factors, we obtain the following value for the EDM
of $^{199}$Hg:
\begin{equation}
d_{Hg}=e\cdot 1.5\cdot10^{-2}\frac{5\mbox{$\alpha$}_3}{72\pi} \frac{
(m_d-m_u-0.012m_s)|A|\sin\theta_A}{M^3}\simeq 2\cdot 10^{-27}
\left(\frac{{\rm 1 TeV}}{M}\right)^2~e\cdot cm,
\end{equation}
where we simply take $|A|=M$. We see from Fig. 3a that the mercury EDM
places a constraint on $M$, $M \stackrel{>}{{}_\sim} 1.5$ TeV.
In the other case, with $\sin\theta_A=0,~\sin\theta_\mu=1$, we have to include
$\lambda_2$-higgsino and $\lambda_1$-higgsino exchanges as well, so that
the result for the CEDMs, (shown in Fig. 3b), is as follows:
\begin{eqnarray}
\tilde d_d=\eta\frac{m_d|\mu|\tan\beta\sin\theta_\mu}{16\pi^2M^3} \left(\frac{
5g_3^2}{18}+\frac{g_2^2}{8}+\frac{g_1^2}{216}\right) \\
\tilde d_u=\eta\frac{m_u|\mu|\cot\beta\sin\theta_\mu}{16\pi^2M^3} \left(\frac{
5g_3^2}{18}+\frac{g_2^2}{8}+\frac{7g_1^2}{216}\right). \nonumber
\end{eqnarray}
As a result, the contribution of the up quark relative to that of the down
quark is suppressed by $m_u/(m_d\tan^2\beta)$. Numerically, the gluino
exchange diagram dominates again with less than a 10\% contribution coming
from $\lambda_2$-higgsino exchange. In this case, the limit is somewhat
stronger giving, $M \stackrel{>}{{}_\sim} 3$ TeV.
As one can see, the EDM of mercury is sensitive to the scale of
supersymmetric masses as high as 1.5-3 TeV. This can be compared with the
sensitivity of the EDM of the electron, which we calculate at the same point
of the supersymmetric parameter space, taking the slepton masses equal
to the squark masses:
\begin{eqnarray}
d_e & = & \frac{m_e|A|\sin\theta_A}{16\pi^2M^3}\frac{g_1^2}{12} \\
d_e& = & \frac{m_e|\mu|\tan\beta\sin\theta_\mu}{16\pi^2M^3}
\left(\frac{5g_2^2}{24}+
\frac{g_1^2}{24}\right).
\nonumber
\label{eq:EDMe}
\end{eqnarray}
The limits based on the electron EDM, for the two cases
considered, are weaker as can be seen from Fig. 3a and 3b
where the limit on $M$ is 0.4 and 1.7 TeV.
There is also the possibility
of destructive interference between two contributions induced by the
CP-violating phases. Again, we choose the supersymmetric parameters to be
equal and fix them to be in the range from $250$ -- $750$ GeV.
Fig. 4a-4c
show the combined exclusion plots. The two bands correspond
to the parts of the parameter space where the mercury or electron (Tl)
constraints are lifted by the cancellation of different supersymmetric
contributions. The allowed area lies on the intersection of these two
bands. We observe that the band corresponding to the mercury EDM
constraint has a different slope than that of the electron EDM, mainly
because $d_{Hg}$ is by far more sensitive to $\theta_A$. We observe that
{\em both} phases are sufficiently constrained for the low values of
$M$.
\begin{figure}[hbtp]
\vspace{0cm}
\begin{center}
\hspace*{-1.3cm}\normalsize \mbox{\epsfxsize=70mm\epsffile{F4a.eps}}
\hspace*{2cm}\normalsize \mbox{\epsfxsize=70mm\epsffile{F4b.eps}}
\vspace{-4.3cm}
\large
\hspace{-8.5cm} $\frac{\theta_A}{\pi}$ \hspace{-10cm} $\frac{\theta_A}{\pi}$
\vspace{3.5cm}
\normalsize
\vspace*{-3cm}
\hspace*{.1cm}$d_{Hg}$ \hspace*{8.1cm} $d_{Hg}$\hspace*{5.2cm}
\vspace*{2.4cm}
\vspace*{-2cm}
\hspace*{2.2cm}$d_{e}$ \hspace*{9cm} $d_{e}$
\vspace*{1.4cm}
\hspace{-1cm} $\theta_\mu/\pi$
\hspace{-10.3cm} $\theta_\mu/\pi$
\end{center}
\hspace*{1.5cm}\normalsize Fig. 4a: $M$ =250 GeV
\hspace*{4.6cm} Fig. 4b: $M$ =500 GeV
\vspace{1cm}
\end{figure}
\newpage
\begin{figure}
\begin{center}
\normalsize \mbox{\epsfxsize=70mm\epsffile{F4c.eps}}
\vspace{-4.4cm}
\large
\hspace{-7.5cm} $\frac{\theta_A}{\pi}$
\normalsize
\vspace*{+1cm}
\hspace*{.1cm}$d_{Hg}$ \hspace*{4.6cm}
\hspace*{2.2cm}$d_{e}$
\vspace*{-2cm}
\vspace{3.7cm}\hspace{0.5cm} $\theta_\mu/\pi$
\vspace{1cm}
\normalsize Fig. 4c: $M$ =750 GeV
\end{center}
\caption{
Combined, $d_{Hg}$ and $d_e$, constraints on the supersymmetric phases $
\theta_A/\pi$ and $\theta_\mu/\pi$ for different values of $M$. Allowed
area is on the intersection of two bands.}
\end{figure}
\section{EDMs in mSUGRA and Cosmological Constraints}
We now consider the constraints on $C\!P$ violating phases in
mSUGRA-like models, i.e. models with unified gaugino and sfermion
masses. We recall that to one loop, the phase of $\mu$ does not
evolve with scale, but the phases of $A_u, A_d$ and $A_e$ must be run
separately from the unification scale to low energies. We follow the
analysis of \cite{FO1,FO}, but with two changes. First, we replace
constraints from the neutron electric dipole moment with limits from
the EDM of Hg, discussed above. Second, we include recent results on
the effect of coannihilations of neutralinos with staus on the
neutralino relic density \cite{EFO}. The latter has the effect of
weakening the cosmological upper bound on the gaugino masses. This is
demonstrated in Fig.~\ref{fig:rd}, where the light shading indicates
the region of the $\{m_0,m_{1\!/2}\}$ plane which yields a neutralino relic
abundance in the cosmologically preferred range $0.1\le\Omega_{\widetilde\chi}\, h^2\le0.3$.
The upper limit of the light shaded region crosses below the
line $m_{\tilde \chi}=m_{{\widetilde \tau}_{\scriptscriptstyle\rm R}}$ at $m_{1\!/2}\sim1400{\rm \, Ge\kern-0.1em V}$; for greater $m_{1\!/2}$, either
the relic density violates the upper bound $\Omega_{\widetilde\chi}\, h^2\le 0.3$ (which follows
from a lower limit of $12 {\rm \, G\kern-0.1em yr}$ on the age of the universe) or the
lightest supersymmetric particle is a stau, leading to an unacceptable
abundance of charged dark matter.
Here we've taken $\tan\beta=2$, but the light shaded region is quite
insensitive to $\tan\beta$ for the values of $\tan\beta$ we consider, as well as
insensitive to the
phase of $\mu$\cite{FO}. For
comparison, the dashed lines demarcate the inferred cosmologically
preferred region if one ignores the effects of neutralino-slepton
coannihilation. Whereas in \cite{FO1}, the constraint $\Omega_{\widetilde\chi}\, h^2\le0.3$
yielded an upper bound of $450{\rm \, Ge\kern-0.1em V}$ on $m_{1\!/2}$, we now have to consider
larger values of $m_{1\!/2}$. However, we will see that this does not
effect the upper bound on $\theta_\mu$.
\begin{figure}[thb]
\begin{center}
\epsfig{file=rd2c.eps,height=3.5in}
\caption{\label{fig:rd}The light-shaded area is the cosmologically preferred
region with \mbox{$0.1\leq\Omega_{\widetilde\chi}\, h^2\leq 0.3$}. The dashed line shows
the location of the cosmologically preferred region if one ignores
the light sleptons. In the dark shaded region the LSP is the
${\tilde \tau}_R$, leading to an unacceptable abundance of charged
dark matter. Also shown as a dotted line is the contour $m_{\chi^{\pm}}=95{\rm \, Ge\kern-0.1em V}$.}
\end{center}
\vspace{-0.1in}\end{figure}
In contrast to the results of the previous section, we find that in
mSUGRA-like models, constraints from the electron EDM are typically
more restrictive than those from the EDM of Hg. This difference
arises because in models with gaugino masses unified at the GUT scale,
the gluino tends to be considerably heavier than the neutralino and
charginos, and this suppresses the contribution to $d_{Hg}$ from the
quark chromoelectric dipole moments due to gluino exchange. We recall
that cancellations between the chargino and neutralino exchange
contributions to the electron EDM allow for large values of
$\theta_\mu$ \cite{FO,IN,FO1}. A similar effect also applies in the case of the
Hg EDM, where cancellations can occur between the gluino exchange and
neutralino and chargino exchange contributions to the quark
chromoelectric dipole moments. The power of combining the electron and
Hg limits lies in the fact that for fixed $\theta_\mu$ and $\theta_A$, the
cancellations in
the electron and Hg dipole moments occur for different, and often
non-overlapping, ranges in $m_{1\!/2}$. Thus the combined limits are
stronger than either limit alone.
Following \cite{FO,FO1}, we compute the electron and Hg EDMs in mSUGRA
as a function of $\theta_\mu,\theta_A$ and $m_{1\!/2}$ for fixed $A_0,m_0$ and $\tan\beta$.
In Fig.~\ref{fig:bedm}a-c we display the
minimum value of $m_{1\!/2}$ required to bring both the electron and Hg
EDMs below their experimental limits, for $\tan\beta=2$ and $m_0=130{\rm \, Ge\kern-0.1em V}$. We
exclude points which violate the current LEP2 chargino and slepton mass
bounds \cite{expt}. The EDMs are computed on a 40x40 grid in
$\{\theta_\mu,\theta_A\}$, and features smaller than the grid size are not
significant. Although the dependence of the EDMs on $m_{1\!/2}$ is not
monotonic, there is still a minimum value of $m_{1\!/2}$, due to cancellations,
which is permitted. In the zone labeled ``I'', $m_{1\!/2}^{\rm min}<200{\rm \, Ge\kern-0.1em V}$,
while the zones labeled ``II'', ``III'', ``IV'' and ``V'' correspond to
$200{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm min}<300{\rm \, Ge\kern-0.1em V}$, $300{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm min}<450{\rm \, Ge\kern-0.1em V}$,
$450{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm min}<600{\rm \, Ge\kern-0.1em V}$ and $m_{1\!/2}^{\rm min}>600{\rm \, Ge\kern-0.1em V}$,
respectively. Comparing with Fig.~\ref{fig:rd}, we see that values of
$m_{1\!/2}$ larger than about $600{\rm \, Ge\kern-0.1em V}$ are cosmologically excluded for this
value $m_0$. Therefore, region V corresponds to an excluded region in the
phase plane. Of course, for this value of $\tan\beta$, the current Higgs mass bound
requires enormous sfermion masses $\gg 1{\rm \, Te\kern-0.1em V}$, which are cosmologically
prohibited. We've chosen to plot our results for $\tan\beta=2$ in order to compare to
our previous results \cite{FO,FO1}. Qualitatively similar conclusions
apply for larger $\tan\beta$, which we summarize at the end of this section.
Figure 2 of Ref.~\cite{FO1} displays the corresponding contours to our
Fig.~\ref{fig:bedm}a-c, but imposing only the constraint from the
electron EDM\footnote{In \cite{FO1} we take $m_0=100{\rm \, Ge\kern-0.1em V}$, rather than
$130{\rm \, Ge\kern-0.1em V}$; however, taking $m_0=130{\rm \, Ge\kern-0.1em V}$ makes only a small change
in the displayed
contours and a slight reduction in the upper bound on $\theta_\mu$. }.
Note that in \cite{FO1} we do not include a contour corresponding to
$450{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm min}<600{\rm \, Ge\kern-0.1em V}$, as this region would be
cosmologically excluded in the absence of coannihilations of
neutralinos with sleptons, whose effects were not included in
\cite{FO1}. The effect of including the Hg EDM bounds is
particularly significant at large $A_0$, where the cancellations are
enhanced and the bounds on $\theta_\mu$ are weakest. Here the widths of the
allowed region in $m_{1\!/2}$ at fixed $\theta_A$ and $\theta_\mu$ are narrowest,
leaving less opportunity for overlap between the ranges allowed by the
electron and Hg EDMs, respectively. Indeed, for $A_0=1.5{\rm \, Te\kern-0.1em V}$, the
upper bound on $\theta_\mu$ is reduced from $\sim0.3\pi$, in the case of the
electron EDM alone, to $\sim0.18\pi$, combining the two constraints,
and, further, the width of the region in $\theta_\mu$ is considerably
narrowed. The reduction in the bound on $\theta_\mu$ is minimal for small
$A_0$, where the bounds on $\theta_\mu$ are strongest. However, notice that
the $m_{1\!/2}^{\rm min}$ at the largest allowed values of $\theta_\mu$ is
shifted from less than $200{\rm \, Ge\kern-0.1em V}$ in the case of the electron EDM alone
to between 200 and 300${\rm \, Ge\kern-0.1em V}$ for the combined bound, in the case
$A_0=300{\rm \, Ge\kern-0.1em V}$. For $A_0=1{\rm \, Te\kern-0.1em V}$ and $1.5{\rm \, Te\kern-0.1em V}$, $m_{1\!/2}^{\rm min}$ lies
above $300{\rm \, Ge\kern-0.1em V}$ at the largest $\theta_\mu$.
\begin{figure}
\vspace*{-2.3in}
\begin{minipage}{6.0cm}
\hspace*{-1in}
\epsfig{file=eha.eps,height=6in}
\end{minipage}
\hspace*{0.3in}
\begin{minipage}{6.0cm}
\epsfig{file=ehb.eps,height=6in}
\end{minipage}\hfill
\vspace{-2.3in}
\begin{minipage}{6.0cm}
\hspace*{-1in}
\epsfig{file=ehc.eps,height=6in}
\end{minipage}
\hspace*{1.0in}
\begin{minipage}{6.0cm}
\vspace*{0.65in}
\epsfig{file=ehd.eps,height=3.5in}
\end{minipage}\hfill
\vspace{-0.7in}
\caption{\label{fig:bedm}Contours of $m_{1\!/2}^{\rm min}$, the minimum $m_{1\!/2}$
required to bring both the electron and Hg EDMs below their
respective experimental bounds, for $\tan\beta=2, m_0=130{\rm \, Ge\kern-0.1em V}$, and
a)$A_0=300{\rm \, Ge\kern-0.1em V}$, b)$A_0=1000{\rm \, Ge\kern-0.1em V}$ and c)$A_0=1500{\rm \, Ge\kern-0.1em V}$. The central
light zone labeled ``I'' has $m_{1\!/2}^{\rm min}<200{\rm \, Ge\kern-0.1em V}$, while the
zones labeled ``II'', ``III'', and ``IV'' correspond to
\hbox{$200{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm min}<300{\rm \, Ge\kern-0.1em V}$}, $300{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm
min}<450{\rm \, Ge\kern-0.1em V}$, $450{\rm \, Ge\kern-0.1em V}<m_{1\!/2}^{\rm min}<600{\rm \, Ge\kern-0.1em V}$ and $m_{1\!/2}^{\rm
min}>600{\rm \, Ge\kern-0.1em V}$, respectively. Zone V is therefore cosmologically
excluded. Panel d) shows the allowed region in the ``trunk'' for
$A_0=300{\rm \, Ge\kern-0.1em V}, m_0=200{\rm \, Ge\kern-0.1em V}$. The light shaded regions ``A'' and
``B'' are permitted, whereas the dark shaded region is
cosmologically excluded (see the text).}
\end{figure}
We note that the larger values of $m_{1\!/2}$ which neutralino-slepton
coannihilation permit do not increase the maximum $\theta_\mu$, for this
value of $m_0$. This is
because, as we see from Fig.~\ref{fig:bedm}, the region of mutual
cancellations happens to lie at lower $m_{1\!/2}$, between 300 and
400${\rm \, Ge\kern-0.1em V}$. The widths of the allowed regions in $m_{1\!/2}$ are typically
between 50 and 80${\rm \, Ge\kern-0.1em V}$ for the lightest shaded zones in
Fig.~\ref{fig:bedm}b and \ref{fig:bedm}c and greater than 80${\rm \, Ge\kern-0.1em V}$
almost everywhere in Fig.~\ref{fig:bedm}a. Larger $m_{1\!/2}$ does,
however, widen the allowed swath in $\theta_\mu$, by the region labeled
``IV'' in Fig.~\ref{fig:bedm}. It helps in particular at small
$\theta_\mu$, where the electron EDM can be beaten down sufficiently by
taking heavy gaugino masses, without resorting to cancellations
between different contributions. At large $m_{1\!/2}$, the Hg EDM
typically provides little constraint on $\theta_\mu$ due to the heaviness of
the gluinos. As $m_0$ is increased, the regions of cancellation
shift, and the maximum value of $\theta_\mu$ slowly decreases.
\begin{figure}[thb]
\begin{center}
\vspace*{-0.2in}
\epsfig{file=tanb.eps,height=3.5in}
\caption{\label{fig:thvtb}The maximum values of $\theta_\mu$ allowed by
cosmology
and both the electron and Hg EDMs, as a function of $\tan\beta$,
for $m_0=100{\rm \, Ge\kern-0.1em V}$ (thick lines) and $m_0=200{\rm \, Ge\kern-0.1em V}$ (thin lines)
and for $A_0 =300, 1000$ and $1500{\rm \, Ge\kern-0.1em V}$.}
\end{center}
\end{figure}
These effects are enhanced at larger $m_0$ and $m_{1\!/2}$, in the
cosmologically allowed ``trunk'' which lies on top of the
$\tilde\tau_R$ LSP region (see Fig.~\ref{fig:rd}). The allowed region
narrows as $m_0$ increases, and the $\Omega_{\widetilde\chi}\, h^2=0.3$ contour crosses the
line $m_{\tilde\tau_R}=m_{\tilde \chi}$ and gives an upper bound on $m_{1\!/2}$ and
$m_0$ at $m_{1\!/2}\sim1400{\rm \, Ge\kern-0.1em V}, m_0\sim300{\rm \, Ge\kern-0.1em V}$. The trunk region yields
much larger sparticle masses than are cosmologically permitted in the
absence of coannihilations, and this can suppress the contributions to
the electric dipole moments sufficiently so that significant
cancellations between the various contributions are not necessary.
For low $A_0$, where the bounds on $\theta_\mu$ are tightest, the bounds on
$\theta_\mu$ are somewhat relaxed in the trunk area. In
Fig.~\ref{fig:bedm}d, we display the allowed region in the
$\{\theta_\mu,\theta_A\}$ plane for $A_0=300{\rm \, Ge\kern-0.1em V}, m_0=200{\rm \, Ge\kern-0.1em V}$. For this value
of $m_0$, $m_{1\!/2}$ is cosmologically restricted to lie between $850{\rm \, Ge\kern-0.1em V}$
and $950{\rm \, Ge\kern-0.1em V}$. In the light region labeled ``A'', the EDMs are below
the experimental limits for all $850{\rm \, Ge\kern-0.1em V}\lem_{1\!/2}\le950{\rm \, Ge\kern-0.1em V}$, while in
the regions labeled ``B'', only part of this range of $m_{1\!/2}$
satisfies the EDM constraints. The dark regions at large $|\theta_\mu|$
require $m_{1\!/2}>950{\rm \, Ge\kern-0.1em V}$ to satisfy the EDM bounds, and so these regions
are cosmologically excluded, as they yield a stau LSP. The upper bound
on $\theta_\mu/\pi$ is relaxed to $\sim 0.055$. Taking $m_{1\!/2}$ and $m_0$ at
their maximal values allows $\theta_\mu/\pi$ up to about 0.1.
For large $A_0$, where $\theta_\mu$ can take its maximal values, the bound
on $\theta_\mu$ does not weaken in the trunk region. As above, this is due
to the fact that at larger $\theta_\mu$, cancellations are still required to
bring the EDMs below their experimental limits, and the regions
of cancellations occur at lower $m_{1\!/2}$. Even taking $m_{1\!/2}$ and $m_0$
at their largest cosmologically permitted values does not allow for
$\theta_\mu$ larger than the bounds in Fig.~\ref{fig:bedm}b,c. Further, since
the regions of low $m_{1\!/2}$ are cosmologically forbidden at large $m_0$,
the bounds on $\theta_\mu$ at large $A_0$ actually decrease for large $m_0$.
Thus the presence of the coannihilation trunk region does not increase
the overall combined cosmology/EDM bound on the phase $\theta_\mu$.
Lastly, we plot in Fig.~\ref{fig:thvtb} the maximum value of $\theta_\mu$
allowed by the electron and Hg electric dipole moments and the upper
limit on $\Omega_{\widetilde\chi}\, h^2$, as a function of $\tan\beta$. The thick lines are for $m_0=100{\rm \, Ge\kern-0.1em V}$,
while the thin lines are for $m_0=200{\rm \, Ge\kern-0.1em V}$ and show the effect on the bounds
described above as one moves into the trunk region.
\section{Conclusions}
We have shown that the calculation of the EDM of the neutron as the function
of different MSSM phases is problematic due to large uncertainties
related to the contributions of the color EDMs.
This is in contrast to the electric dipole moment
of the mercury atom, induced by the T-odd nucleon-nucleon interaction. In
the chiral limit the coefficient $\xi$, characterizing the strength of
T-odd forces has
a power-like singularity $\sim m_\pi^{-2}$, whereas $d_N \sim \log
m_\pi^{2}$ in the same chiral approach.
It is apparent that the $\pi^0$ and $
\eta$ exchange diagrams dominate both parametrically and numerically and
therefore yield a very good approximation to the magnitude of the T-odd
interaction. The final result is proportional to $(\tilde d_d-\tilde
d_u-0.012\tilde d_s)\times 3.2 \cdot 10^{-2} e$ and can be further
developed in terms of CP-violating phases of the MSSM.
There are two serious problems with the calculation of the T-odd nuclear
forces due to the effective interaction (\ref{eq:eff}) with the coefficients
provided by the MSSM. First is the status of the factorization in Eq. (\ref
{eq:gluonium}), related with the low-energy theorem in $0^+$ channel.
Following Refs. \cite{KKZ,KKY,KK}, we have have taken $\langle p|\bar q g_s(G
\mbox{$\sigma$})q|p\rangle\simeq 1.3 {\rm GeV}^2 \langle p|\bar qq|p\rangle$.
We note that a designated sum rule calculation of this quantity and/or its
simulation on lattice is highly desirable for it is the main source of
uncertainties in the calculation of T-odd nuclear forces. The second
potentially troublesome point is the effective negative sign between the
$\tilde d_d$ and $\tilde d_s$ contributions. Although the numerical
suppression in front of $\tilde d_s$ is quite strong and $d_d$ dominates,
destructive interference is still possible in both cases,
$\theta_A\neq 0$ and $
\theta_\mu \neq 0$.
In this paper, we have considered first
a very specific
part of the supersymmetric parameter space, when all squark, slepton and
gaugino masses were chosen to be equal. The theoretical prediction for $
d_{Hg}$ exhibits remarkable sensitivity to the scale of soft-breaking mass
parameters as high as 1.5-3 TeV. When the scale is fixed below 1 TeV, $d_{Hg}$
limits {\em both} phases. The constraints on the CP-violating supersymmetric
phases, obtained in this way are the strongest constraints so far.
We have also considered the combined constraints from the Hg and electron
EDMs in the mSUGRA, when all supersymmetry breaking gaugino masses,
soft scalar masses, and soft trilinear terms are separately unified at the
GUT scale. In this case, the sensitivity to the Hg EDM is weakened due to
the relative size of the gluino mass. Nevertheless, the results are as
strong or stronger (particularly when $|A|$ is large and the limits are
weakest) than the combined results from $d_e$ and $d_N$. The improvement in
the limit is due to the fact that cancellations among the contributions to
the EDMs occur at slightly different regions of the SUSY parameter space.
\section{Acknowledgments}
M.P. would like to thank I.B. Khriplovich, A. Ritz, A.I. Vainshtein and A.R.
Zhitnitsky for numerous important discussions. The work of M.P. and K.O. was
supported in part by DOE grant DE--FG02--94ER--40823. The work of T.F. was
supported in part by DOE grant DE--FG02--95ER--40896 and in part by the
University of Wisconsin Research Committee with funds granted by the
Wisconsin Alumni Research Foundation.
|
\section{Introduction}
The existence of the large--scale clustering of galaxies had already
been well established by the early 1970's mainly due to the pioneering
work of {\scite{totsuji:correlation}} and {\scite{peebles:nature}} who
showed that the two--point correlation function for galaxies in the
Lick and Zwicky catalogues was positive and had the power-law form
$\xi(r)\propto{r^{-1.8}}$ on scales $\lesssim10\ifmmode{h^{-1}{\rm Mpc}}\else{$h^{-1}{\rm Mpc}$}\fi$. Their result
was later extended to three--dimensional galaxy catalogues as well.
Although the clustering of galaxies is now a well--known fact, a
complete description of clustering which includes its geometrical
features has so far eluded researchers. This is perhaps due to the
fact that the galaxy density field which we observe appears to be
strongly non--Gaussian. A Gaussian random field is uniquely described
by its power spectrum $P(k)$, or its two--point correlation function
$\xi(r)$ since $\xi(r)$ and $P(k)$ form a Fourier transform pair.
This is no longer true for a non--Gaussian field for which $\xi$ must
be complemented by other statistical descriptors which are sensitive
to the structure of matter on large scales. In the so-called
``standard model'' of structure formation, an initially Gaussian
density distribution becomes non--Gaussian due to mode-coupling and
the resultant build up of phase correlations during the non--linear
regime. These phase correlations give rise to the amazing diversity
of form, which is characteristic of a highly evolved distribution of
matter, and is often referred to as being cellular, filamentary,
sheet--like, network--like, sponge--like, a cosmic web etc. Most of
these descriptions are based either on a visual appearance of
large--scale structure or on the presence of features that are absent
in the reference Gaussian distribution which, by definition, is
assumed to be featureless. Gravitational instability, for example,
may cause CDM--like Gaussian initial perturbations to evolve towards a
density field that percolates at a higher density threshold, i.e.\ at
a lower filling factor, than a Gaussian field {\cite{melott:cluster}}.
Such distributions display greater connectivity and are sometimes
referred to as being ``network--like'' {\cite{yess:universality}}.
In order to come to grips with the rich textural possibilities
inherent in large--scale structure, a number of geometrical indicators
of clustering have been proposed in the past including minimal
spanning trees {\cite{barrow:minimalspanning}}, the genus curve
{\cite{gott:sponge}}, percolation theory
{\cite{zeldovich:maximum,shandarin:percolation}} and shape analysis
(\pcite{sathyaprakash:morphology,sahni:approximation} and references
therein). A major recent advance in our understanding of
gravitational clustering has been associated with the application of
Minkowski functionals (MFs) to cosmology {\cite{mecke:robust}}. The
four MFs $V_0,\ldots,V_3$ provide an excellent description of the
geometrical properties of a collection of point objects (galaxies) or,
alternatively, of continuous distributions such as density fields in
large--scale structure or brightness contours in the cosmic microwave
background. The scope and descriptive power of the MFs is enhanced by
the fact that both percolation analysis and the genus curve are
members of the family. Additionally, as demonstrated by
{\scite{sahni:shapefinders}}, ratios of MFs provide us with an
excellent ``shape statistic'' with which one may attempt to quantify
the morphology of large scale structure, including the shapes of
individual superclusters and voids. Spurred by the success of MFs in
quantifying the geometrical properties of large--scale structure we
apply the MFs to scale--invariant N--body simulations of gravitational
clustering, with an attempt to probe both the global properties and
the individual ``bits and pieces'' that might make up the ``cosmic
web'' {\cite{bond:filament}}.
\section{Method}
\subsection{Minkowski functionals}
Minkowski functionals (named after {\pcite{minkowski:volumen}}) were
introduced into cosmology by {\scite{mecke:robust}} employing the
generalized Boolean grain model. This model associates a body with a
point process (in our case given by the location of galaxies or
clusters) by decorating each point with a ball of radius $r$. The
union set of the covering balls is then studied morphologically,
whereby the radius of the ball serves as a diagnostic parameter
probing the spatial scale of the body.
In this paper we shall use the excursion set approach which is
applicable to continuous fields (that may be constructed from point
processes). {\scite{schmalzing:beyond}} pointed out how to apply
Minkowski functionals to isocontours of continuous fields, where the
contour level (the threshold) is employed as diagnostic parameter.
The excursion set approach inherits two diagnostic parameters, because
we may also vary the smoothing scale used in constructing the
continuous field.
In three dimensions there exist four Minkowski functionals $V_\mu$,
$\mu=0,1,2,3$. They provide a complete and unique description of a
pattern's global morphology in the sense of Hadwiger's theorem
{\cite{hadwiger:vorlesung}}. While reducing the information contained
in the full hierarchy of correlation functions, this small set of
numbers incorporates correlations of arbitrary order and therefore
provides a complementary look on large--scale structure. The
geometric interpretations of all Minkowski functionals in three
dimensions are summarized in Table~\ref{tab:minkowski}.
We calculate global Minkowski functionals of the isodensity contours
of the density field as described in Appendix~\ref{app:minkowski}.
Furthermore, we separately calculate the partial Minkowski functionals
of each isolated part of the isodensity contour. Since the total
isodensity contour is the union of all its parts, the global
functionals are given as sums of the partial functionals at the same
threshold. This follows the spirit of {\scite{mecke:robust}}, where
partial Minkowski functionals are introduced to measure local
morphology for the generalized Boolean grain model.
Partial Minkowski functionals (PMF) offer the possibility of probing
the morphology of individual objects, or the object's environmental
morphological properties, respectively. We expect that this concept
will be more powerful when applied to continuous fields at high
spatial resolution so that the details of structures are not smoothed
out. Their application to point processes also delivers more direct
information. PMF provide a wide range of possibilities for
morphological studies which we shall explore in a forthcoming paper.
\subsection{Shapefinders}
One important task of morphological statistics consists in quantifying
strongly non--Gaussian features such as filaments and pancakes. Given
the four Minkowski functionals, we aim at reducing their morphological
information content to two measures of planarity and filamentarity,
respectively, as has been done for example with various geometrical
quantities {\cite{mo:statistical}}, moments of inertia
{\cite{babul:filament}}, and cumulants of counts--in--cells
{\cite{luo:shape}}.
Recently, {\scite{sahni:shapefinders}} proposed a set of
shapefinders derived from Minkowski functionals. One starts from
the three independent ratios of Minkowski functionals that have
dimension of length. Requiring that they yield the radius $R$ if
applied to a ball, we define
\begin{equation}
\text{Thickness }T:=\frac{V_0}{2V_1},\qquad
\text{Width }W:=\frac{2V_1}{{\pi}V_2},\qquad
\text{Length }L:=\frac{3V_2}{4V_3}.
\end{equation}
By the isoperimetric inequalities~(\ref{eq:isoperimetric}), we have
$L{\ge}W{\ge}T$ for any convex body.
Going one step further, {\scite{sahni:shapefinders}} also define
dimensionless shapefinders by
\begin{equation}
\label{eq:shapefinders}
\text{Planarity }{\mathcal{P}}:=\frac{W-T}{W+T},\qquad
\text{Filamentarity }{\mathcal{F}}:=\frac{L-W}{L+W}.
\end{equation}
Some examples are in order.
\subsection{Simple examples}
Let us consider some simple families of convex bodies in
three--dimensional space that can take both filamentary and planar
shape.
A spheroid with two axes of length $r$ and one axis of length
$\lambda{r}$ has Minkowski functionals
\begin{equation}
V_0= \frac{4\pi}{3}r^3\lambda ,\qquad
V_1= \frac{\pi}{3}r^2\left(1+f(1/\lambda)\right) ,\qquad
V_2= \frac{2}{3}r\left(\lambda+f(\lambda)\right) ,\qquad
V_3= 1,
\label{eq:spheroid}
\end{equation}
where\footnote{Note that for arguments $x>1$, one can use the relation
$i\arccos{x}=\ln\left(x+\sqrt{x^2-1}\right)$ to recover an explicitly
real--valued expression.}
\begin{equation}
f(x)=\frac{\arccos{x}}{\sqrt{1-x^2}}.
\end{equation}
By varying the parameter $\lambda$ from zero to infinity, we can
change the morphology of the spheroid from a filament to a pancake via
a spherical cluster.
A different way of deforming a filament into a pancake goes via
generic triaxial ellipsoids; thereby one of the smaller axes of a
strongly prolate spheroid is increased in size until it matches the
larger axis and an oblate spheroid has been reached. An integral
expression can be found in {\cite{sahni:shapefinders}}.
Yet another transition from prolate to oblate shape is provided by
cylinders of radius $r$ and height $\lambda{r}$. Here, the Minkowski
functionals are given by
\begin{equation}
V_0=\pi r^3\lambda ,\qquad
V_1=\frac{\pi}{3}r^2(1+\lambda) ,\qquad
V_2=\frac{1}{3}r(\pi+\lambda) ,\qquad
V_3=1.
\label{eq:cylinder}
\end{equation}
A Blaschke diagram, that is a plot of planarity ${\mathcal{P}}$ and
filamentarity ${\mathcal{F}}$, summarizing the morphological properties of these
simple convex bodies is shown in Figure~\ref{fig:blaschke.simple}.
\section{A set of $N$--body simulations}
\subsection{Description}
We start from a family of initial power law spectra $P(k)\propto{k}^n$
with $n\in\{-2,-1,0,+1\}$ set before an Einstein--de~Sitter background
($\Omega$=1, $\Lambda$=0). We conduct numerical experiments using a
PM code (consult {\pcite{melott:controlled}} for details). Four sets
of phases were used for each model, making a total of 16 simulation
runs. Each run consists of $128^3$ particles sampled at an epoch well
in the non--linear regime. This epoch is chosen such that the scale
of non--linearity $k_{\text{nl}}$, defined in terms of the evolved
spectrum
\begin{equation}
\sigma^2_{k_{\text{nl}}} = \int_0^{k_{\text{nl}}}{\mathrm{d}}^3k P(k) = 1,
\end{equation}
is equal to eight in units of the fundamental mode of the simulation
box. By using the stage $k_{\text{nl}}=8$, we make sure that
structure is already sufficiently developed on scales much larger than
the simulation's resolution, while it is not yet influenced by
boundary effects.
Using a cloud--in--cell kernel, these particles were put onto a
$256^3$ grid, which is the maximum value an ordinary workstation can
tackle with acceptable time and memory consumption. Subsequently, the
density field was smoothed with a Gaussian kernel
$\propto\exp\left(-x^2/2\lambda^2\right)$, where $x$ is the distance
in mesh units and the width $\lambda$ is set to 3. Tests have shown
that this value both leads to a reasonably smooth field, and preserves
at least some detail on smaller scales. Throughout the article, we
re--scale the density to the density contrast $\delta$, ranging from
$-1$ to infinity with zero mean.
\subsection{The global field}
The global Minkowski functionals calculated from the density fields
described in the previous section are shown in
Figure~\ref{fig:epoch.f}. The four different line styles correspond
to the different spectral indices. Figure~\ref{fig:epoch.f.rescaled}
shows the Minkowski functionals for the same set of models, but
instead of the density threshold $\delta$, the rescaled threshold
$\nu$ is used as the $x$--axis. $\nu$ is calculated from the volume
Minkowski functional $v_0$, that is the filling factor $f$, as
described by {\scite{gott:quantitative}}. Essentially, its use forces
exact Gaussian behavior of the volume $v_0$, by the implicit
connection
\begin{equation}
v_0(\delta)=\frac{1}{2}-\frac{1}{2}\Phi\left(\frac{\nu}{\sqrt{2}}\right).
\end{equation}
Thus the deviations from Gaussianity that are due to changes in the
one--point probability distribution function are removed, and
deviations due to higher--order correlations are emphasized.
Obviously, the global functionals clearly discriminate between the
various models. However, in order to make this statement more
quantitative, let us take a closer look at the individual coherent
objects composing the isodensity contours.
\subsection{The largest objects}
At intermediate thresholds, the excursion sets consist of numerous
isolated objects. We identify them by grouping adjacent occupied grid
cells into one object, where adjacent means that the cells have a
common face. Since the Minkowski functionals of the global field are
calculated by integrating over quantities that can be assigned to
individual grid cells, the partial Minkowski functionals of each
object can be obtained at no extra cost once the cells belonging to
each object have been identified.
Several plots in Figures~\ref{fig:clusters.abcd},
{\ref{fig:clusters.ijkl}}, {\ref{fig:clusters.uvwx}}, and
{\ref{fig:clusters.qrst}} illustrate the behavior of the Minkowski
functionals of these objects.
Obviously, the contribution of smaller objects to the volume is almost
negligible compared to the largest one. Note that in all figures, the
mean and standard deviation over all four realizations are shown
instead of the individual curves. It is worth noting that the
variance is largest in the $n=-2$ and $-1$ models, which are dominated
by structures on large scales and hence show the strongest sample
variance.
The models with various initial spectral indices $n$ show
qualitatively a similar behavior. At small filling factors (high
density thresholds) the two largest clusters give negligible
contribution to each of the global characteristics. Then, at
percolation transition the largest cluster quickly becomes the
dominant structure in terms of volume, area, and integrated mean
curvature. The second largest cluster also grows at the percolation
threshold but just a little and then quickly diminishes. The
percolation transition is clearly marked in all three characteristics
of the largest cluster by their sudden growth. However, this
transition does not happen at a well--defined threshold. Instead,
clusters gradually merge into the largest objects as the threshold is
decreased (the filling factor grows). This continuous transition has
also been observed using percolation analysis, i.e.\ the zeroth
Minkowski functional alone
{\cite{shandarin:topology,shandarin:detection,klypin:percolation,sahni:probing}}
and is explained by the finite size of the sample.
Nevertheless, the percolation transitions happen within fairly narrow
ranges of the filling factor that are clearly distinct for different
models in question: The filling factors are approximately
$0.03\pm0.01,0.07\pm0.015,0.11\pm 0.015,0.14\pm0.015$ in the
$n=-2,-1,0,+1$ models, respectively. It is remarkable that all
percolation transitions occur at smaller filling factors than in
Gaussian fields (about 0.16) indicating that even in the most
hierarchical model ($n=+1$) the structures tend to be more connected
than in the ``structureless'' Gaussian field.
{\scite{pauls:hierarchical}} showed positive correlation with networks
based on the same phases all the way to $n=+3$. This confirms the
conclusion of {\scite{yess:universality}}: The universality of the
network structures results from the evolution of Gaussian initial
conditions through gravitational instability. The Euler
characteristic of the largest cluster also marks the percolation
threshold but in a different manner: before percolation it is zero and
after percolation it becomes negative in every model, however, in the
$n=0$ and $n=+1$ models it grows to a small positive peak before
becoming negative. All global functionals have no particular features
at the percolation threshold.
\subsection{Small objects}
As an example, the Blaschke diagram for the model $n=-2$ is shown in
Figure~\ref{fig:blaschke.abcd}. The distributions for the other models
look qualitatively very similar and the average quantities for other
models are presented in Figures~\ref{fig:mass.x} and
{\ref{fig:mass.y}}. Figure~\ref{fig:mass.x} shows that most of the
small objects are either spherical or slightly planar (two largest
dots in Figure~\ref{fig:blaschke.abcd}). There is also a considerable
number of elongated clusters with filamentarities from 0.1 to 0.5. In
some cases filamentarity reaches large values $\sim1$. On the
contrary planarity is much weaker, it hardly reaches the value of 0.2
(which is partly a consequence of the smoothing). There is a hint of
a small correlation between filamentarity and planarity: the objects
with the largest filamentarity also tend to have larger planarity.
Figures~\ref{fig:mass.x} and {\ref{fig:mass.y}} display the
shapefinders as functions of the cluster mass. The curves give
averages over the realizations of each model. Apparently, the signal
for filamentarity is much stronger than for planarity, regardless of
the model which is in full agreement with
Figure~\ref{fig:blaschke.abcd}. The planarity and filamentarity
distributions qualitatively look very similar except the amplitude.
Small objects ($5\times10^{-6}\lesssim{m}\lesssim5\times10^{-4}$)
display stronger planarity and filamentarity for models with more
power on large scales. However, for greater masses
($m\gtrsim5\times10^{-4}$) the situation is reversed: the less
large-scale power the greater the filamentarity and planarity. If the
former seems to be natural and was expected, the latter has been
unexpected. Both the planarity and filamentarity monotonically grow
and reach their maxima at largest clusters: ${\mathcal{P}}_m\approx0.1$ and
${\mathcal{P}}_m\approx0.5$ in all models. As expected the largest objects
possess the largest planarities and filamentarities, but the
independence of the maxima from the model again was unexpected.
Figures~\ref{fig:blaschke.x} and {\ref{fig:blaschke.y}} show the
histograms for the shapefinders ${\mathcal{P}}$ and ${\mathcal{F}}$, respectively. They
clearly show the large difference in the total number of structures:
the more power on small scales the greater the abundance of clusters.
These figures are in general agreement with the
Figures~\ref{fig:mass.x} and {\ref{fig:mass.y}}.
\section{Conclusion and Outlook}
Global Minkowski functionals do discriminate
(Figures~\ref{fig:epoch.f} and {\ref{fig:epoch.f.rescaled}}). Only
the $n=-2$ model shows slight drawbacks as far as robustness is
concerned, but that is due to the method's sensitivity to large--scale
features of the smoothed density field. The total area, mean
curvature, and Euler characteristic are sensitive to abundances of the
structures and are easy to interpret: the more power on small scales
(greater $n$) the more abundant structures and therefore the greater
the amplitude of the curve.
Even more valuable information is obtained from looking at the
Minkowski functionals of the largest coherent object at each threshold
(Figures~\ref{fig:clusters.abcd}, {\ref{fig:clusters.ijkl}},
{\ref{fig:clusters.uvwx}}, and {\ref{fig:clusters.qrst}}). All four
Minkowski functionals of the largest cluster clearly consistently
detect the percolation transition. Two points are worth stressing: 1)
in all models the percolation transition happens at smaller filling
factors than in the ``structureless'' Gaussian fields and 2) the more
power on the large scales (i.e. the smaller $n$) the smaller the
filling factor at percolation. Both conclusions confirm the results of
{\scite{yess:universality}} about the universality of the network
structures in the power law models with $n\le1$. The results of
{\scite{pauls:hierarchical}} present evidence that this should be
expected all the way up to $n=+3$; at $n=+4$ mode coupling effects
from smaller scales should begin to fully disrupt the network
structure.
Small objects, on the other hand, give different results. Their
abundance discriminates well, but is already determined by the
difference in the Euler characteristic as well as by the total area
and mean curvature of the whole contour (Figures~\ref{fig:epoch.f} and
{\ref{fig:epoch.f.rescaled}}) that are also sensitive to the abundance
of structures. The morphology of small objects as measured by
shapefinders shows little differences between models so far
(Figures~\ref{fig:mass.x}, {\ref{fig:mass.y}}, {\ref{fig:blaschke.x}}
and {\ref{fig:blaschke.y}}). Both the maximum average planarity
(${\mathcal{P}}\approx0.1$) and the maximum average filamentarity
(${\mathcal{F}}\approx0.5$) are reached in the most massive non-percolating
objects. None of the model showed ribbon--like objects characterized
by both large planarity and large filamentarity. We may speculate that
the smaller objects are ones which formed earlier, are more nonlinear,
and therefore more decoupled from initial conditions.
However, all models used Gaussian initial conditions, and evolve under
the influence of gravity. Hence the similar morphology of the clumps
may point towards universal behavior. Note that things such as string
wakes might produce totally different results.
The grouping and measurement techniques used in this study may be less
accurate for small objects than for large clusters. It is worth
trying to study the morphology of small objects by applying more
accurate methods of measuring partial Minkowski functionals such as a
Boolean grain model (\pcite{schmalzing:diplom}, and a follow--up to
this article) or the interpolation method of
{\scite{novikov:minkowski}} generalized to three dimensions.
\section*{Acknowledgments}
JS wishes to thank Martin Kerscher for interesting discussions and
valuable comments. This work is part of a project in the
``Sonderforschungsbereich 375--95 f\"ur Astroteilchenphysik'' of the
Deutsche Forschungsgemeinschaft. ALM and SFS acknowledge support of
the NSF--EPSCoR program and the GRF program at the University of
Kansas. SFS acknowledges support from TAC in Copenhagen.
|
\section*{Introduction}
Suppose $X$ is a right Hilbert module over a $C^*$\nobreakdash- algebra $A$.
If $X$ also carries a left action of $A$ as adjointable operators on $X_A$,
we call $X$ a {\em Hilbert bimodule\/} over $A$.
In \cite{pimsner}, Pimsner associated with every such bimodule $X$ a $C^*$\nobreakdash-
algebra ${\mathcal O}_X$, which we shall call the {\em Cuntz-Pimsner algebra\/} of $X$,
and showed that every crossed product by $\field{Z}$
and every Cuntz-Krieger algebra can be realized as ${\mathcal O}_X$ for suitable $X$.
He also commented that the algebras ${\mathcal O}_X$ include the crossed products by $\field{N}$;
that is, for each endomorphism $\alpha$ of a $C^*$\nobreakdash- algebra
$A$ there is a bimodule $X = X(\alpha)$ such that ${\mathcal O}_X$ is
canonically isomorphic to the semigroup crossed product
$A\rtimes_\alpha\field{N}$ of \cite{cuntz82,stacey}.
The work in this paper is motivated by the following observation,
which also serves as our primary example.
Suppose $\beta$ is an action of a discrete semigroup $P$
as endomorphisms of a $C^*$\nobreakdash- algebra $A$.
For each $s\in P$ let $X_s := X(\beta_s)$ be the bimodule canonically associated
with the endomorphism $\beta_s$.
Then the family $X = \{X_s: s\in P\}$ admits an associative multiplication
$(x,y)\in X_s\times X_t \mapsto xy\in X_{ts}$ which implements
isomorphisms $X_s\otimes_A X_t \to X_{ts}$;
we call a family with this structure a {\em product system of Hilbert bimodules\/}.
(In this example $X$ is a product system over the opposite semigroup $P^o$.)
Such families generalize the product systems of
\cite{dinhjfa,dinhjot,fowrae,fowler},
where the fibers $X_s$ are complex Hilbert spaces (bimodules over $\field{C}$).
To each product system $X$ we associate a generalized Cuntz-Pimsner
algebra ${\mathcal O}_X$. When $X$ is the product system associated with
the semigroup dynamical system $(A,P,\beta)$, ${\mathcal O}_X$ is canonically
isomorphic to the semigroup crossed product $A\rtimes_\beta P$.
Moreover, if we ``twist'' $X$ by a multiplier $\omega:P\times P\to\field{T}$,
then the corresponding Cuntz-Pimsner algebra is isomorphic to
the twisted semigroup crossed product $A\rtimes_{\beta,\omega} P$.
Our construction applies even when $A$ is nonunital provided each
endomorphism $\beta_s$ extends to the multiplier algebra $M(A)$.
The aim of this paper is to take a first step towards analyzing
the Cuntz-Pimsner algebra of a product system $X$.
Following Pimsner \cite{pimsner},
we begin by studying the structure of its Toeplitz extension ${\mathcal T}_X$.
This algebra is universal for {\em Toeplitz representations\/} of $X$;
these are multiplicative maps whose restriction to each fiber $X_s$
is a Toeplitz representation in the sense of \cite{fowrae2}.
Our results generalize those of \cite{fowrae} for product systems
of Hilbert spaces; indeed, much of the paper is devoted to
adapting the methods of \cite{fowrae} to the bimodule setting.
Thus our basic assumptions about the underlying semigroup $P$ are
as in \cite{fowrae}:
to allow our analysis to extend beyond the totally-ordered case,
we assume that $P$ is the positive cone of a group $G$
such that $(G,P)$ is quasi-lattice ordered in the sense of Nica \cite{nica}.
The class of such $(G,P)$ includes all direct sums and free products
of totally ordered groups.
We also impose a covariance condition,
called {\em Nica covariance\/}, on Toeplitz representations of $X$.
This means that the universal $C^*$\nobreakdash- algebra ${\mathcal T}_{\cv}(X)$
which we analyze is in general a quotient of ${\mathcal T}_X$.
However, if $(G,P)$ is totally-ordered,
then Nica-covariance is automatic,
and hence ${\mathcal T}_{\cv}(X)$ is the same as ${\mathcal T}_X$.
Our main goal is to characterize the faithful representations
of ${\mathcal T}_{\cv}(X)$.
We accomplish this by embedding ${\mathcal T}_{\cv}(X)$
in a certain twisted semigroup crossed product $B_P \textstyle{\rtimes_{\tau, X}} P$
(Theorem~\ref{theorem:subalgebra}), and then characterizing
its faithful representations
(Theorem~\ref{theorem:faithfulness of representations}).
When $P = \field{N}$, ${\mathcal T}_{\cv}(X)$ is precisely the Toeplitz algebra of
the Hilbert bimodule $X_1$ (the fiber over $1\in\field{N}$),
and our Theorem~\ref{theorem:faithfulness of representations}
reduces to \cite[Theorem~2.1]{fowrae2}.
In fact, the analysis in \cite{fowrae2} was motivated
by our preliminary work on this paper.
We would like to point out in particular how
the stronger result \cite[Theorem~3.1]{fowrae2}
arose from our investigations into product systems,
for it serves as a good illustration of the usefulness
of Nica covariance.
Suppose $Z$ is an orthogonal direct sum
$\bigoplus_{\lambda\in\Lambda} Z^\lambda$ of Hilbert bimodules.
Let $G$ be the free group on $\Lambda$,
let $P$ be the subsemigroup of $G$ generated by $\Lambda$,
and let $X$ be the unique product system over $P$
whose fiber over $\lambda$ is $Z^\lambda$.
Then ${\mathcal T}_{\cv}(X)$ is canonically isomorphic
to the Toeplitz algebra of the bimodule $Z$,
and \cite[Theorem~3.1]{fowrae2} follows from our
Theorem~\ref{theorem:faithfulness of representations}.
The main application of \cite[Theorem~3.1]{fowrae2}
was to establish the simplicity of the graph algebras
associated with certain infinite directed graphs
\cite[Corollary~4.3]{fowrae2}.
Although here we confine our applications
to twisted semigroup crossed products,
we anticipate that our results will also give interesting
information about ${\mathcal O}_X$ when each of the fibers of $X$
arise from infinite directed graphs.
We begin in Section~\ref{section:prelims} by giving a
brief review of Hilbert bimodules,
their representations, and their $C^*$\nobreakdash- algebras.
In Section~\ref{section:product systems} we introduce
product systems of Hilbert bimodules,
discuss their representations, and define
the algebras ${\mathcal T}_X$ and ${\mathcal O}_X$.
In Section~\ref{section:crossed products} we associate with each twisted
semigroup dynamical system $(A,P,\beta,\omega)$ a product system
$X = X(A,P,\beta,\omega)$ whose Cuntz-Pimsner algebra ${\mathcal O}_X$
is the twisted semigroup crossed product $A\rtimes_{\beta,\omega} P$.
We show that the Toeplitz algebra of $X(A,P,\beta,\omega)$
also has a crossed product structure, and this motivates the definition
of a ``Toeplitz'' crossed product ${\mathcal T}(A\rtimes_{\beta,\omega} P)$
in which the endomorphisms are implemented not by isometries,
but rather by partial isometries.
In Section~\ref{section:crossed products2} we generalize the notion
of twisted crossed product by replacing the multiplier
$\omega$ by a product system $X$ of Hilbert bimodules.
This extends the philosophy developed in \cite{fowrae}
that one should regard product systems as noncommutative cocycles.
Hence given an action $\beta$ of $P$ as endomorphisms of a $C^*$\nobreakdash-
algebra $C$, we consider $(C,P,\beta,X)$ as a twisted
semigroup dynamical system,
and we define a twisted crossed product $C\rtimes_{\beta,X} P$.
In Section~\ref{section:covariance} we assume that $(G,P)$
is quasi-lattice ordered, and we discuss the notion
of Nica covariance for a Toeplitz representation.
As illustrated in \cite[Example~1.3]{fowler}
using product systems of Hilbert spaces,
when $(G,P)$ is not a total order
it is possible that the $C^*$\nobreakdash- algebra ${\mathcal T}_{\cv}(X)$
which is ``universal'' for such representations
may admit representations which are not the integrated form
of a Nica-covariant Toeplitz representation.
To avoid this pathology
we adapt the methods of \cite{fowler} to our setting:
we define the notion of a product system being
{\em compactly aligned\/}, and show that ${\mathcal T}_{\cv}(X)$
is truly universal when $X$ is compactly aligned
(Proposition~\ref{prop:preserve covariance}).
We show that $X$ is compactly aligned
if the left action of $A$ on each fiber $X_s$ is by compact
operators (Proposition~\ref{prop:CA conditions});
it follows that the product systems $X(A,P,\beta,\omega)$
associated with twisted semigroup dynamical systems
are compactly aligned.
In Section~\ref{section:system} we consider a certain
$C^*$\nobreakdash- subalgebra $B_P$ of $\ell^\infty(P)$
which is invariant under left translation $\tau:P\to\End(\ell^\infty(P))$.
As in \cite{lacarae,fowrae}, covariant representations
of the twisted system $(B_P,P,\tau,X)$
are in one-one correspondence with Toeplitz representations of $X$
which are Nica-covariant (Proposition~\ref{prop:Lpsi}),
and hence ${\mathcal T}_{\cv}(X)$ embeds naturally as a subalgebra
of $B_P \textstyle{\rtimes_{\tau, X}} P$ (Theorem~\ref{theorem:subalgebra}).
When the left action of $A$ on each fiber $X_s$ is by compact operators,
${\mathcal T}_{\cv}(X)$ is all of $B_P \textstyle{\rtimes_{\tau, X}} P$.
In Section~\ref{section:faithful} we prove our main result,
Theorem~\ref{theorem:faithfulness of representations},
which characterizes the faithful
representations of $B_P\rtimes_{\tau,X} P$
under the assumption that $X$ is compactly aligned
and $(B_P,P,\tau,X)$ satisfies a certain amenability hypothesis.
In Section~\ref{section:amenability} we give conditions on $(G,P)$
which ensure that $(B_P,P,\tau,X)$ is amenable.
In particular, $(B_P,P,\tau,X)$ is amenable
if $X$ is compactly aligned and
$(G,P)$ is a free product $*(G^\lambda,P^\lambda)$
with each $G^\lambda$ amenable (Corollary~\ref{cor:amenability}).
Finally, in Section~\ref{section:applications} we apply our
Theorem~\ref{theorem:faithfulness of representations}
to the product system
$X(A,P,\beta,\omega)$ of Section~\ref{section:crossed products}.
When $(G,P)$ is a total order,
$B_P \textstyle{\rtimes_{\tau, X}} P$ is isomorphic to the Toeplitz crossed product
${\mathcal T}(A\rtimes_{\beta,\omega} P)$;
in general $B_P \textstyle{\rtimes_{\tau, X}} P$ is a certain quotient ${\mathcal T}_{\cv}(A\rtimes_{\beta,\omega} P)$
which also has a crossed product structure,
and Theorem~\ref{theorem:crossed products}
characterizes its faithful representations.
Applying this to the twisted system $(B_P,P,\tau,\omega)$,
we show that ${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega} P)$ is universal
for partial isometric representations of $P$ which are {\em bicovariant\/}
(Proposition~\ref{prop:universal bicovariant}),
and we obtain a characterization of its faithful representations
(Theorem~\ref{theorem:bicovariant})
which is particularly nice
when $P$ is the free semigroup on infinitely many generators
(Theorem~\ref{theorem:free}).
The author thanks Iain Raeburn for the many helpful discussions
while this research was being conducted.
\section{Preliminaries}\label{section:prelims}
Let $A$ be a separable $C^*$\nobreakdash- al\-ge\-bra.
A {\em Hilbert bimodule over $A$\/} is a right Hilbert $A$-module $X$
together with a ${}^*$\nobreakdash- homomorphism $\phi:A\to{\mathcal L}(X)$
which is used to define a left action of $A$ on $X$ via $a\cdot x := \phi(a)x$
for $a\in A$ and $x\in X$.
A {\em Toeplitz representation} of $X$
in a $C^*$\nobreakdash- al\-ge\-bra $B$ is a pair $(\psi,\pi)$
consisting of a linear map $\psi:X\to B$
and a homomorphism $\pi:A\to B$ such that
\begin{align*}
\psi(x\cdot a) &= \psi(x)\pi(a), \\
\psi(x)^*\psi(y)&= \pi(\langle x,y \rangle_A),\ \text{ and} \\
\psi(a\cdot x) &= \pi(a)\psi(x)
\end{align*}
for $x,y\in X$ and $a\in A$.
Given such a representation, there is
homomorphism $\pi^{(1)}:{\mathcal K}(X)\to B$ which satisfies
\begin{equation}\label{eq:pione}
\pi^{(1)}(\Theta_{x,y}) = \psi(x)\psi(y)^*
\qquad\text{for all $x,y\in X$,}
\end{equation}
where $\Theta_{x,y}(z) := x\cdot\langle y,z \rangle_A$ for $z\in X$;
see \cite[p.~202]{pimsner},\cite[Lemma~2.2]{kpw}, and \cite[Remark~1.7]{fowrae2}
for details.
We say that $(\psi,\pi)$ is {\em Cuntz-Pimsner covariant\/}
if
\[
\pi^{(1)}(\phi(a)) = \pi(a)
\qquad\text{for all $a \in \phi^{-1}({\mathcal K}(X))$.}
\]
The {\em Toeplitz algebra\/} of $X$ is the $C^*$\nobreakdash- al\-ge\-bra ${\mathcal T}_X$
which is universal for Toeplitz representations of $X$
\cite{pimsner,fowrae2}, and the {\em Cuntz-Pimsner algebra\/}
of $X$ is the $C^*$\nobreakdash- al\-ge\-bra ${\mathcal O}_X$ which is universal for
Toeplitz representations which are Cuntz-Pimsner covariant
\cite{pimsner,dpz,kpw,ms, ms2,fmr}.
Every right Hilbert $A$-module $X$ is essential, in the sense that
$X$ is the closed linear span of elements $x\cdot a$.
We say that a Hilbert bimodule $X$ is {\em essential\/}
if it is also essential as a left $A$-module; that is, if
\[
X = \clsp\{\phi(a)x: a\in A, x\in X\}.
\]
When $X$ is essential,
two applications of the Hewitt-Cohen Factorization Theorem allow us to write
any $x\in X$ as $\phi(a)y\cdot b$ for some $y\in X$ and $a,b\in A$.
Hence if $(a_i)$ is an approximate identity for $A$,
then
\begin{equation}\label{eq:approximate identity}
\lVert x - x\cdot a_i\rVert \to 0
\quad\text{and}\quad
\lVert x - \phi(a_i)x\rVert \to 0
\qquad\text{for all $x\in X$.}
\end{equation}
\section{Product systems of Hilbert bimodules}
\label{section:product systems}
For each $n\ge 1$ the $n$-fold internal tensor product
$X^{\otimes n} := X\otimes_A \dotsb \otimes_A X$ has a natural structure
as a Hilbert bimodule over $A$; see \cite[ Section~2.2]{ms2} for details.
The following definition,
based on Arveson's continuous tensor product systems over $(0,\infty)$ \cite{arv},
generalizes the collection $\{X^{\otimes n}: n\in\field{N}\}$
to semigroups other than $\field{N}$.
\begin{definition}\label{defn:ps}
Suppose $P$ is a countable semigroup with identity $e$
and $p:X\to P$ is a family of Hilbert bimodules over $A$.
Write $X_s$ for the fibre $p^{-1}(s)$ over $s\in P$,
and write $\phi_s:A\to{\mathcal L}(X_s)$ for the homomorphism which
defines the left action of $A$ on $X_s$.
We say that $X$ is a {\em (discrete) product system over $P$\/} if $X$ is a semigroup,
$p$ is a semigroup homomorphism,
and for each $s,t\in P\setminus\{e\}$ the map
$(x,y) \in X_s \times X_t \mapsto xy \in X_{st}$
extends to an isomorphism of the Hilbert bimodules
$X_s\otimes_A X_t$ and $X_{st}$.
We also require that $X_e = A$
(with its usual right Hilbert module structure and $\phi_e(a)b = ab$ for $a,b\in A$),
and that the multiplications $X_e\times X_s\to X_s$
and $X_s\times X_e \to X_s$ satisfy
\begin{equation}\label{eq:multiply}
ax = \phi_s(a)x,\qquad xa = x\cdot a
\qquad \text{for $a\in X_e$ and $x\in X_s$.}
\end{equation}
\end{definition}
\begin{remark}\label{remark:essential}
Multiplication $X_e \times X_s \to X_s$ will not induce an isomorphism
$X_e\otimes_A X_s \to X_s$ unless $X_s$ is essential as a left $A$-module.
\end{remark}
\begin{remark}
The associativity of multiplication in $X$ implies that
$\phi_{st}(a) = \phi_s(a)\otimes_A 1^t$ for all $a\in A$;
that is, $\phi_{st}(a)(xy) = (\phi_s(a)x)y$
for all $x\in X_s$ and $y\in X_t$.
\end{remark}
\begin{remark} It is possible that some of the $X_s$ may be zero.
\end{remark}
\begin{definition}\label{defn:psrep}
Suppose $B$ is a $C^*$\nobreakdash- al\-ge\-bra and $\psi:X\to B$;
write $\psi_s$ for the restriction of $\psi$ to $X_s$.
We call $\psi$ a {\em Toeplitz representation\/} of $X$ if
\textup{(1)} For each $s\in P$, $(\psi_s, \psi_e)$
is a Toeplitz representation of $X_s$, and
\textup{(2)} $\psi(xy) = \psi(x)\psi(y)$ for $x,y\in X$.
\noindent If in addition each $(\psi_s,\psi_e)$ is Cuntz-Pimsner covariant,
we say that $\psi$ is {\em Cuntz-Pimsner covariant\/}.
\end{definition}
\begin{remark}\label{remark:toeplitz rep}
By \cite[Remark~1.1]{fowrae2}, every Toeplitz representation $\psi$ is contractive;
moreover, if the homomorphism $\psi_e:A\to B$ is isometric, then so is $\psi$.
Also, since we are assuming \eqref{eq:multiply}, a map $\psi:X\to B$
is a Toeplitz representation if it satisfies both (2) and
\textup{(1')} $\psi_s(x)^*\psi_s(y) = \psi_e(\langle x,y \rangle_A)$
whenever $s\in P$ and $x,y\in X_s$.
\end{remark}
\begin{notation}
We write $\psi^{(s)}$ for the homomorphism of ${\mathcal K}(X_s)$ into $B$
which corresponds to the pair $(\psi_s,\psi_e)$,
as in \eqref{eq:pione}; that is,
\[
\psi^{(s)}(\Theta_{x,y}) = \psi_s(x)\psi_s(y)^*
\qquad\text{for all $x,y\in X_s$.}
\]
\end{notation}
\subsection*{The Fock representation}
Let $F(X)$ be the right Hilbert $A$-module
\[
F(X) := \bigoplus_{s\in P} X_s.
\]
By this we mean the following: as a set,
$F(X)$ is the subset of $\prod_{s\in P} X_s$
consisting of all elements $(x_s)$ for which
$\sum_{s\in P} \langle x_s, x_s \rangle_A$ is summable in $A$;
that is, for which $\sum_{s\in F} \langle x_s, x_s \rangle_A$
converges in norm as $F$ increases over the finite subsets of $P$.
We write $\oplus x_s$ for $(x_s)$ to indicate that the above series is summable.
The right action of $A$ is given by $(\oplus x_s)\cdot a := \oplus (x_s \cdot a)$,
and the inner product by
$\langle \oplus x_s, \oplus y_s \rangle_A
:= \sum_{s\in P} \langle x_s, y_s \rangle_A$.
The algebraic direct sum $\bigodot_{s\in P} X_s$ is dense in $F(X)$.
Suppose $P$ is left-cancellative. Then for any $x\in X$ and $\oplus x_t\in F(X)$
we have $p(xx_s) = p(xx_t)$ if and only if $s = t$,
so there is an element $(y_s) \in\prod X_s$ such that
\[
y_s = \begin{cases} xx_t & \text{if $s = p(x)t$} \\
0 & \text{if $s\notin p(x)P$;} \end{cases}
\]
we write $(xx_t)$ for $(y_s)$.
Since
$\langle xx_s, xx_s \rangle_A \le \lVert x \rVert^2 \langle x_s, x_s \rangle_A$
for each $s\in P$,
the series $\sum \langle xx_s, xx_s \rangle_A$ is summable.
It is routine to check that
\[
l(x)(\oplus x_s) := \oplus xx_s
\qquad\text{for $\oplus x_s\in F(X)$}
\]
determines an adjointable operator $l(x)$ on $F(X)$;
indeed, the adjoint $l(x)^*$ is zero on any summand $X_s$
for which $s\notin p(x)P$, and on $X_{p(x)t} = \clsp X_{p(x)}X_t$
it is determined by the formula
$l(x)^*(yz) = \langle x, y\rangle_A\cdot z$ for $y\in X_{p(x)}$ and $z\in X_t$.
It follows that $l:X\to{\mathcal L}(F(X))$
is a Toeplitz representation, called the {\em Fock representation of $X$\/}.
The homomorphism $l_e:A\to{\mathcal L}(F(X))$ is simply the diagonal left action of $A$; that is,
$l_e(a) = \oplus \phi_s(a)$. Since $\phi_e$ is just left multiplication on $X_e = A$,
it is isometric, and hence so is $l_e$;
by Remark~\ref{remark:toeplitz rep}, $l$ is isometric.
\begin{prop}\label{prop:toeplitz algebra}
Let $X$ be a product system over $P$ of Hilbert $A$--$A$ bimodules.
Then there is a $C^*$\nobreakdash- al\-ge\-bra ${\mathcal T}_X$,
called the {\em Toeplitz algebra\/} of $X$,
and a Toeplitz representation $i_X:X\to{\mathcal T}_X$, such that
\textup{(a)} for every Toeplitz representation $\psi$ of $X$,
there is a homomorphism $\psi_*$ of ${\mathcal T}_X$ such that
$\psi_*\circ i_X = \psi$; and
\textup{(b)} ${\mathcal T}_X$ is generated as a $C^*$\nobreakdash- al\-ge\-bra by $i_X(X)$.
\noindent The pair $({\mathcal T}_X, i_X)$ is unique up to canonical isomorphism,
and $i_X$ is isometric.
\end{prop}
\begin{proof} It is straightforward to translate
the proof of \cite[Proposition~1.3]{fowrae2} to this setting.
\end{proof}
\begin{prop}\label{prop:CP algebra}
Let $X$ be a product system over $P$ of Hilbert $A$--$A$ bimodules.
Then there is a $C^*$\nobreakdash- al\-ge\-bra ${\mathcal O}_X$,
called the {\em Cuntz-Pimsner algebra\/} of $X$,
and a Toeplitz representation $j_X:X\to{\mathcal O}_X$ which is Cuntz-Pimsner covariant,
such that
\textup{(a)} for every Cuntz-Pimsner covariant Toeplitz representation $\psi$ of $X$,
there is a homomorphism $\psi_*$ of ${\mathcal O}_X$ such that
$\psi_*\circ j_X = \psi$; and
\textup{(b)} ${\mathcal O}_X$ is generated as a $C^*$\nobreakdash- al\-ge\-bra by $j_X(X)$.
\noindent The pair $({\mathcal O}_X, j_X)$ is unique up to canonical isomorphism.
\end{prop}
\begin{remark}
Although the universal map $i_X:X\to{\mathcal T}_X$ is always isometric,
it is quite possible that $X$ might not admit any nontrivial
Toeplitz representations which are Cuntz-Pimsner covariant,
in which case ${\mathcal O}_X$ is trivial.
\end{remark}
\begin{proof}[Proof of Proposition~\ref{prop:CP algebra}]
With $({\mathcal T}_X,i_X)$ as in Proposition~\ref{prop:toeplitz algebra},
let ${\mathcal I}$ be the ideal in ${\mathcal T}_X$ generated by
\[
\{ i_X(a) - i_X^{(s)}(\phi_s(a)) : s\in P, a \in \phi_s^{-1}({\mathcal K}(X)) \}.
\]
Define ${\mathcal O}_X := {\mathcal T}_X/{\mathcal I}$ and $j_X := q\circ i_X$, where $q:{\mathcal T}_X \to {\mathcal O}_X$
is the canonical projection. Obviously $j_X$ is a Toeplitz representation
which generates ${\mathcal O}_X$,
and it is Cuntz-Pimsner covariant because $j_X^{(s)} = q\circ i_X^{(s)}$.
If $\psi$ is another Cuntz-Pimsner covariant Toeplitz representation,
then the homomorphism $\psi_*$ of ${\mathcal T}_X$ satisfies
\[
\psi_*(i_X(a) - i_X^{(s)}(\phi_s(a))) = \psi(a) - \psi^{(s)}(\phi_s(a)) = 0
\]
whenever $\phi_s(a) \in {\mathcal K}(X_s)$,
and hence $\psi_*$ descends to the required homomorphism of ${\mathcal O}_X$
(also denoted $\psi_*$).
\end{proof}
\begin{prop}\label{prop:ps over N}
Let $X$ be a product system over $\field{N}$ of Hilbert $A$--$A$ bimodules.
Then ${\mathcal T}_X$ is canonically isomorphic to the Toeplitz algebra ${\mathcal T}_{X_1}$
of the Hilbert bimodule $X_1$.
If the left action on each fiber is isometric,
or if the left action on each fiber is by compact operators,
then ${\mathcal O}_X$ is canonically isomorphic to ${\mathcal O}_{X_1}$.
\end{prop}
\begin{proof}
Let $i_X:X\to{\mathcal T}_X$ be universal for Toeplitz representations of $X$,
and define $\mu := (i_X)_1 : X_1 \to {\mathcal T}_X$
and $\pi := (i_X)_0 : A = X_0 \to {\mathcal T}_X$.
Since $(\mu,\pi)$ is a Toeplitz representation of $X_1$,
we get a homomorphism $\mu\times\pi:{\mathcal T}_{X_1} \to {\mathcal T}_X$.
To construct the inverse of $\mu\times\pi$,
let $(i_{X_1}, i_A)$ be the universal Toeplitz representation of $X_1$ in ${\mathcal T}_{X_1}$,
and fix $n\ge 1$. By \cite[Proposition~1.8(1)]{fowrae2},
there is a linear map $\psi_n:X_n\to{\mathcal T}_{X_1}$ which satisfies
\[
\psi_n(x_1\dotsm x_n) := i_{X_1}(x_1)\dotsm i_{X_1}(x_n)
\qquad\text{for all $x_1, \dots, x_n\in X_1$,}
\]
and then $(\psi_n, i_A)$ is a Toeplitz representation of $X_n$.
Defining $\psi_0 := i_A$ thus gives a Toeplitz representation $\psi:X\to{\mathcal T}_{X_1}$,
and it is routine to check that $\psi_*:{\mathcal T}_X\to{\mathcal T}_{X_1}$ is the inverse of $\mu\times\pi$.
Now let $i_X:X\to{\mathcal O}_X$ be universal for
Cuntz-Pimsner covariant Toeplitz representations of $X$.
As above, we get a homomorphism $\mu\times\pi:{\mathcal O}_{X_1}\to{\mathcal O}_X$.
To construct the inverse, we let $(i_{X_1},i_A):(X_1,A)\to {\mathcal O}_{X_1}$
be universal and define a Toeplitz representation $\psi:X\to{\mathcal O}_{X_1}$
as before; we need to check that $\psi$ is Cuntz-Pimsner covariant
under each of the hypotheses on the left action.
By definition $(\psi_1,\psi_0)$ is Cuntz-Pimsner covariant, so we use induction.
Assume that $(\psi_n,\psi_0)$ is Cuntz-Pimsner covariant for some $n\ge 1$,
and suppose $a\in A$ acts compactly on the left of $X_{n+1}$;
that is, $\phi(a)\otimes_A 1^n \in{\mathcal K}(X_{n+1})$.
Since the left action is isometric on each fiber,
by \cite[Lemma~7.2]{fmr} we have $\phi(a)\otimes_A 1^{n-1}\in{\mathcal K}(X_n)$;
hence $\psi^{(n)}(\phi(a)\otimes_A 1^{n-1}) = \psi_0(a)$.
But \cite[Lemma~7.5]{fmr} gives
$\psi^{(n+1)}(\phi(a)\otimes_A 1^n) = \psi^{(n)}(\phi(a)\otimes_A 1^{n-1})$,
so $(\psi_{n+1},\psi_0)$ is Cuntz-Pimsner covariant.
Now suppose that $A$ acts by compact operators on each fiber.
By representing ${\mathcal O}_{X_1}$ faithfully on a Hilbert space ${\mathcal H}$
we can assume that $\psi$ is a Toeplitz representation of $X$ on ${\mathcal H}$.
Assuming again that $(\psi_n,\psi_0)$ is Cuntz-Pimsner covariant for some $n\ge 1$,
\cite[Lemma~1.6]{fmr} gives
$\overline{\psi_0(A){\mathcal H}} \subseteq \clsp(\psi_n(X_n){\mathcal H})$.
Let $x\in X_n$, and express $x = y\cdot a$ with $y\in X_n$ and $a\in A$.
Since $(\psi_1,\psi_0)$ is Cuntz-Pimsner covariant and $\phi(a)\in{\mathcal K}(X_1)$, we have
\[
\psi_n(x) = \psi_n(y)\psi_0(a) = \psi_n(y)\psi^{(1)}(\phi(a)).
\]
Now $\phi(a)$ can be approximated by a finite sum $\sum \Theta_{x_i,y_i}$,
hence $\psi_n(x)$ can be approximated by a finite sum
$\psi_n(y)\psi(x_i)\psi(y_i)^* = \psi_{n+1}(yx_i)\psi(y_i)^*$.
Thus
\[
\overline{\psi_0(A){\mathcal H}}
\subseteq \clsp(\psi_n(X_n){\mathcal H})
\subseteq\clsp(\psi_{n+1}(X_{n+1}){\mathcal H}),
\]
and $(\psi_{n+1}, \psi_0)$ is Cuntz-Pimsner covariant by \cite[Lemma~1.6]{fmr}.
\end{proof}
\begin{definition}\label{defn:nondegenerate}
Let $X$ be a product system over $P$ of Hilbert $A$--$A$ bimodules.
A Toeplitz representation $\psi:X\to B$ is {\em nondegenerate\/}
if the induced homomorphism $\psi_*:{\mathcal T}_X\to B$ is nondegenerate.
\end{definition}
\begin{lemma}\label{lemma:nondegenerate}
Suppose each fiber $X_s$ is essential as a left $A$-module.
Then a Toeplitz representation $\psi:X\to B$ is nondegenerate
if and only if the homomorphism $\psi_e:A\to B$ is nondegenerate.
\end{lemma}
\begin{proof}
Let $(a_i)$ be an approximate identity for $A = X_e$.
By \eqref{eq:approximate identity},
$i_X(a_i)$ is an approximate identity for ${\mathcal T}_X$,
and the result follows.
\end{proof}
\section{Crossed products twisted by multipliers}
\label{section:crossed products}
Our main examples of product systems come from $C^*$\nobreakdash- dynamical systems.
Suppose $\beta$ is an action of $P$ as endomorphisms of $A$
such that $\beta_e$ is the identity endomorphism.
We will assume that each $\beta_s$ is {\em extendible\/};
that is, that each $\beta_s$ extends to a strictly continuous endomorphism
$\overline{\beta_s}$ of $M(A)$.
For $P$ the positive cone of a totally ordered abelian group,
Adji has shown that extendibility is necessary to define a reasonable
crossed product $A\rtimes_\beta P$ \cite{adji}.
In this section we will consider crossed products which are twisted by
a multiplier $\omega$ of $P$; that is, by a function
$\omega:P\times P\to\field{T}$ which satisfies
$\omega(e,e) = 1$ and
\[
\omega(r,s)\omega(rs,t) = \omega(r,st)\omega(s,t)
\qquad\text{for all $r,s,t\in P$.}
\]
We call $(A,P,\beta,\omega)$ a {\em twisted semigroup dynamical system\/}.
\begin{definition}\label{defn:omega crossed product}
Let $B$ be a $C^*$\nobreakdash- al\-ge\-bra.
A function $V:P\to B$ is called an
{\em $\omega$\nobreakdash- representation\/} of $P$ if
\begin{equation}\label{eq:omegarep}
V_sV_t = \omega(s,t)V_{st}
\qquad\text{for all $s,t\in P$.}
\end{equation}
If in addition each $V_s$ is an isometry (resp. partial isometry),
$V$ is called {\em isometric\/} (resp. {\em partial isometric\/})
$\omega$\nobreakdash- representation.
A {\em covariant representation\/} of $(A,P,\beta,\omega)$ on a Hilbert space ${\mathcal H}$
is a pair $(\pi,V)$ consisting of a nondegenerate representation $\pi:A\to B({\mathcal H})$
and an isometric $\omega$\nobreakdash- representation $V:P\to B({\mathcal H})$ such that
\begin{equation}\label{eq:piVcovariance}
\pi(\beta_s(a)) = V_s\pi(a)V_s^*
\qquad\text{for all $s\in P$ and $a\in A$.}
\end{equation}
A {\em crossed product\/} for $(A,P,\beta,\omega)$ is a triple
$(B,i_A,i_P)$ consisting of a $C^*$\nobreakdash- al\-ge\-bra $B$,
a nondegenerate homomorphism $i_A:A\to B$, and a map $i_P:P\to M(B)$
such that
\textup{(a)} if $\sigma$ is a nondegenerate representation of $B$,
then $(\sigma\circ i_A, \overline\sigma\circ i_P)$
is a covariant representation of $(A,P,\beta,\omega)$;
\textup{(b)} for every covariant representation $(\pi,V)$,
there is a representation $\pi\times V$ such that
$(\pi\times V)\circ i_A = \pi$ and $\overline{\pi\times V}\circ i_P = V$;
and
\textup{(c)} $B$ is generated as a $C^*$\nobreakdash- al\-ge\-bra by
$\{i_A(a)i_P(s): a\in A, s\in P\}$.
\end{definition}
After establishing the existence of a crossed product,
it is easily seen to be unique up to canonical isomorphism;
we denote the crossed product algebra $A\rtimes_{\beta,\omega} P$.
We will construct a product system $X = X(A,P,\beta,\omega)$
over the opposite semigroup $P^o$,
and show that its Cuntz-Pimsner algebra ${\mathcal O}_X$
is a crossed product for $(A,P,\beta,\omega)$.
Moreover, we will show that the Toeplitz algebra of this product system
also has a crossed product structure:
it will be universal for pairs $(\pi,V)$ satisfying \eqref{eq:piVcovariance}
in which $\pi$ is a nondegenerate representation of $A$
and $V$ is a partial isometric $\omega$\nobreakdash- representation
such that
\begin{equation}\label{eq:Vpartialisometry}
\text{$V_s^*V_s\pi(a) = \pi(a)V_s^*V_s$ for all $s\in P$ and $a\in A$.}
\end{equation}
We call such a pair $(\pi,V)$ a {\em Toeplitz covariant representation\/} of $(A,P,\beta,\omega)$,
and write ${\mathcal T}(A\rtimes_{\beta,\omega} P)$ for the
corresponding universal $C^*$\nobreakdash- al\-ge\-bra, called the {\em Toeplitz crossed product\/}
of $(A,P,\beta,\omega)$.
For each $s\in P$ let
\[
X_s := \{s\}\times \overline{\beta_s}(1)A,
\]
and give $X_s$ the structure of a Hilbert bimodule over $A$ via
\[
(s,x)\cdot a := (s,xa),
\quad
\langle (s,x),(s,y) \rangle_A := x^*y,
\]
and
\[
\phi_s(a)(s,x) := (s,\beta_s(a)x).
\]
Let $X = \bigsqcup_{s\in P} X_s$, let $p(s,x) := s$,
and define multiplication in $X$ by
\[
(s,x)(t,y) := (ts, \overline{\omega(t,s)}\beta_t(x)y )
\qquad\text{for $x\in\overline{\beta_s}(1)A$ and $y\in\overline{\beta_t}(1)A$.}
\]
\begin{lemma}\label{lemma:crossed product ps}
$X = X(A,P,\beta,\omega)$ is a product system over the opposite semigroup $P^o$.
For each $s\in P$, the fiber $X_s$ is essential as a left $A$-module,
and the left action of $A$ on $X_s$ is by compact operators.
\end{lemma}
\begin{proof} Let $(s,x)\in X_s$ and $(t,y) \in X_t$.
If $x = \overline{\beta_s}(1)a$ and $y = \overline{\beta_t}(1)b$, then
\[
\beta_t(x)y
= \beta_t(\overline{\beta_s}(1)a)\overline{\beta_t}(1)b
= \overline{\beta_t}(\overline{\beta_s}(1))\beta_t(a)b
= \overline{\beta_{ts}}(1)\beta_t(a)b,
\]
and hence the product $(s,x)(t,y)$ belongs to $X_{ts}$.
Letting $a$ vary over an approximate identity for $A$,
this product converges in norm to $\overline{\beta_{ts}}(1)b$,
so the set of products $(s,x)(t,y)$ has dense linear span in $X_{ts}$.
Hence to see that multiplication induces an isomorphism $X_s \otimes_A X_t \to X_{ts}$,
it suffices to check that it preserves the inner product of any pair of elementary
tensors:
\begin{align*}
\langle (s,x)\otimes_A (t,y),
& (s,x')\otimes_A (t,y') \rangle_A \\
& = \langle (t,y), \phi_t(\langle (s,x), (s,x') \rangle_A)(t,y')\rangle_A \\
& = \langle (t,y), \phi_t(x^*x')(t,y') \rangle_A \\
& = \langle (t, y), (t, \beta_t(x^*x')y') \rangle_A \\
& = y^*\beta_t(x^*x')y' \\
& = \langle (ts, \overline{\omega(t,s)}\beta_t(x)y),
(ts, \overline{\omega(t,s)}\beta_t(x')y') \rangle_A \\
& = \langle (s,x)(t,y), (s,x')(t,y') \rangle_A.
\end{align*}
Multiplication is associative since
\begin{align*}
((s,x)(t,y))(r,z)
& = (ts, \overline{\omega(t,s)}\beta_t(x)y)(r,z) \\
& = (rts, \overline{\omega(r,ts)}\beta_r(\overline{\omega(t,s)}\beta_t(x)y)z) \\
& = (rts, \overline{\omega(rt,s)}\beta_{rt}(x)\overline{\omega(r,t)}\beta_r(y)z) \\
& = (s,x)(rt, \overline{\omega(r,t)}\beta_r(y)z) \\
& = (s,x)((t,y)(r,z)).
\end{align*}
If $(a_i)$ is an approximate identity for $A$, then for each $a\in A$ and $s\in P$
we have
$\lim \phi(a_i)(s,\overline{\beta_s}(1)a)
= \lim (s,\beta_s(a_i)a) = (s,\overline{\beta_s}(1)a)$,
so each $X_s$ is essential. If $a\in A$, then by writing $a = bc^*$
with $b,c\in A$, we see that
$\phi_s(a) = \Theta_{(s,\beta_s(b)),(s,\beta_s(c))} \in{\mathcal K}(X_s)$
is compact.
\end{proof}
\begin{lemma}\label{lemma:strict convergence}
Let $i_X:X\to{\mathcal T}_X$ be universal for Toeplitz representations of $X$,
and let $(a_i)$ be an approximate identity for $A$.
Then for each $s\in P$, $i_X(s,\beta_s(a_i))$ converges strictly in $M{\mathcal T}_X$.
\end{lemma}
\begin{proof}
Since each fiber $X_t$ is essential, any vector $\xi\in X_t$ can be written in the form
$\xi = \phi_t(a)\eta\cdot b$ with $a,b\in A$ and $\eta\in X_t$.
But then $i_X(\xi) = i_X(e,a)i_X(\eta)i_X(e,b)$,
and since elements of the form $i_X(\xi)$ generate ${\mathcal T}_X$ as a $C^*$\nobreakdash- al\-ge\-bra,
the result follows from the calculations
\begin{equation}\label{eq:strict1}
i_X(s,\beta_s(a_i))i_X(e,a)
= i_X(s, \beta_s(a_i)a) \to i_X(s, \overline{\beta_s}(1)a)
\end{equation}
and
\begin{equation}\label{eq:strict2}
i_X(e,a)i_X(s,\beta_s(a_i))
= i_X(s,\beta_s(aa_i)) \to i_X(s,\beta_s(a)).
\end{equation}
\end{proof}
Define $i_A:A\to{\mathcal T}_X$ by $i_A(a) := i_X(e,a)$,
and define $i_P:P\to M{\mathcal T}_X$ by
$i_P(s) := \lim i_X(s,\beta_s(a_i))^*$.
\begin{prop}\label{prop:crossed product}
${\mathcal T}_X$ and ${\mathcal O}_X$ are canonically isomorphic
to $A\rtimes_{\beta,\omega} P$ and ${\mathcal T}(A\rtimes_{\beta,\omega} P)$, respectively.
More precisely, $({\mathcal T}_X, i_A, i_P)$ is a Toeplitz crossed product for $(A,P,\beta,\omega)$,
and, with $q:{\mathcal T}_X\to{\mathcal O}_X$ the canonical projection,
$({\mathcal O}_X, q\circ i_A, \overline q\circ i_P)$ is a crossed product for $(A,P,\beta,\omega)$.
\end{prop}
\begin{proof}
Taking $s = e$ in \eqref{eq:strict1} and \eqref{eq:strict2},
shows that $i_A(a_i)$ converges strictly to the identity in $M({\mathcal T}_X)$;
that is, $i_A$ is nondegenerate.
For condition (a) of a Toeplitz crossed product,
suppose $\sigma$ is a nondegenerate representation of ${\mathcal T}_X$;
we must show that $(\pi,V) := (\sigma\circ i_A, \overline\sigma\circ i_P)$
is a Toeplitz covariant representation of $(A,P,\beta,\omega)$.
First note that $\pi$ is nondegenerate since $\sigma$ and $i_A$ are.
Equation \eqref{eq:strict2} gives
\begin{equation}\label{eq:AP}
i_A(a)i_P(s)^* = i_X(s,\beta_s(a))
\qquad\text{for all $a\in A$ and $s\in P$,}
\end{equation}
so
\begin{align*}
i_P(s)i_A(a)i_P(s)^*
& = \lim i_X(s,\beta_s(a_i))^*i_X(s,\beta_s(a)) \\
& = \lim i_X(e,\beta_s(a_i^*a))
= i_X(e, \beta_s(a)) = i_A(\beta_s(a)),
\end{align*}
and applying $\overline\sigma$ gives $V_s\pi(a)V_s^* = \pi(\beta_s(a))$.
In particular
\[
i_P(s)i_P(s)^*
= \lim i_P(s)i_A(a_i)i_P(s)^*
= \lim i_A(\beta_s(a_i)) = \overline{i_A}(\overline{\beta_s}(1))
\]
is a projection, so $i_P(s)$, and hence $V_s$, is a partial isometry.
To establish \eqref{eq:Vpartialisometry},
take any $a\in A$, write $a = bc^*$ with $b,c\in A$, and compute:
\begin{equation}\label{eq:domain commutes}
\begin{split}
i_A(bc^*)i_P(s)^*i_P(s)
& = \lim i_X(s,\beta_s(bc^*))i_X(s,\beta_s(a_i))^*
\qquad \text{(by \eqref{eq:AP})} \\
& = \lim i_X(s,\beta_s(b))i_X(e,\beta_s(c^*))i_X(s,\beta_s(a_i))^* \\
& = \lim i_X(s,\beta_s(b)) \bigl( i_X(s,\beta_s(a_i)i_X(e,\beta_s(c)) \bigr)^* \\
& = \lim i_X(s,\beta_s(b)) i_X(s,\beta_s(a_i c))^* \\
& = i_X(s,\beta_s(b)) i_X(s,\beta_s(c))^*.
\end{split}
\end{equation}
Taking adjoints, interchanging $b$ and $c$,
and applying $\overline\sigma$ gives
$V_s^*V_s\pi(a) = \pi(a)V_s^*V_s$.
For every $s,t\in P$ we have
\begin{align*}
i_P(t)^*i_P(s)^*
& = (\lim_i j_X(t,\beta_t(a_i)))(\lim_j j_X(s,\beta_s(a_j))) \\
& = \lim_i \lim_j j_X(st, \overline{\omega(s,t)}\beta_s(\beta_t(a_i))\beta_s(a_j)) \\
& = \lim_i j_X(st, \overline{\omega(s,t)}\beta_{st}(a_i)) \\
& = \overline{\omega(s,t)}i_P(st)^*;
\end{align*}
taking adjoints and applying $\overline\sigma$ gives
$V_sV_t = \omega(s,t)V_{st}$.
This completes the proof of condition (a).
For condition (b),
suppose $(\pi,V)$ is a Toeplitz covariant representation
of $(A,P,\beta,\omega)$ on a Hilbert space ${\mathcal H}$.
Define $\psi:X\to B({\mathcal H})$ by
\[
\psi(s,x) := V_s^*\pi(a).
\]
Since $\pi$ is nondegenerate and
$\pi(a) = \pi(\beta_e(a)) = V_e\pi(a)V_e^*$ for all $a\in A$,
$V_e$ is a coisometry. Since $V_e^2 = \omega(e,e)V_e = V_e$,
we deduce that $V_e = 1$.
Thus
\begin{align*}
\psi(s,x)^*\psi(s,y)
& = \pi(x)^*V_sV_s^*\pi(y)
= \pi(x^*\overline{\beta_s}(1)y) \\
& = V_e^*\pi(x^*y)
\qquad\qquad\text{(since $y\in\overline{\beta_s}(1)A$ and $V_e = 1$)} \\
& = \psi(e,x^*y)
= \psi(\langle (s,x), (s,y) \rangle_A),
\end{align*}
and since we also have
\begin{align*}
\psi(s,x)\psi(t,y)
& = V_s^*\pi(x)V_t^*\pi(y) & \\
& = V_s^*\pi(x)V_t^*V_tV_t^*\pi(y) & \text{($V_t$ is a partial isometry)} \\
& = V_s^*V_t^*V_t\pi(x)V_t^*\pi(y) & \text{(by \eqref{eq:Vpartialisometry})} \\
& = (V_tV_s)^*\pi(\beta_t(x))\pi(y) & \\
& = \overline{\omega(t,s)}V_{ts}^*\pi(\beta_t(x)y) & \\
& = \overline{\omega(t,s)}\psi(ts,\beta_t(x)y) & \\
& = \psi((s,x)(t,y)),&
\end{align*}
$\psi$ is a Toeplitz representation of $X$.
Let $\pi\times V$ be the representation
$\psi_*:{\mathcal T}_X\to B({\mathcal H})$. Then
\[
(\pi\times V)\circ i_A(a) = \psi_*\circ i_X(e,a) = \psi(e,a) = V_e^*\pi(a) = \pi(a),
\]
and
\begin{align*}
\overline{\pi\times V}\circ i_P(s)\pi(a)
& = \psi_*(i_P(s)i_A(a)) & \\
& = \psi_*(i_X(s,\beta_s(a^*))^*) & \text{(by \eqref{eq:AP})} \\
& = \psi(s,\beta_s(a^*))^*
= \pi(\beta_s(a))V_s & \\
& = V_s\pi(a)V_s^*V_s
= V_sV_s^*V_s\pi(a)
= V_s\pi(a); &
\end{align*}
since $\pi$ is nondegenerate, this implies that
$\overline{\pi\times V}\circ i_P = V$,
as required.
For condition (c), simply note that
$i_A(a)i_P(s) = i_X(s, \overline{\beta_s}(1)a^*)^*$,
and elements of this form generate ${\mathcal T}_X$.
We now show that $({\mathcal O}_X, q\circ i_A, \overline q\circ i_P)$
is a crossed product for $(A,P,\beta,\omega)$.
Since $i_A$ and $q$ are nondegenerate, so is $q\circ i_A$.
If $\rho$ is a nondegenerate representation of ${\mathcal O}_X$,
then $\sigma := \rho\circ q$ is a nondegenerate representation
of ${\mathcal T}_X$.
Hence $(\pi,V) := (\rho\circ q \circ i_A, \overline\rho\circ\overline q \circ i_P)
= (\sigma \circ i_A, \overline\sigma\circ i_P)$ is a Toeplitz covariant representation
of $(A,P,\beta,\omega)$.
To see that each $V_s$ is an isometry,
let $b,c\in A$. Since $q\circ i_X$ is Cuntz-Pimsner covariant,
\eqref{eq:domain commutes} gives
\begin{align*}
q\circ i_A(bc^*)\overline q \circ i_P(s)^*\overline q \circ i_P(s)
& = q\circ i_X(s,\beta_s(b)) q\circ i_X(s,\beta_s(c))^* \\
& = (q\circ i_X)^{(s)}(\Theta_{(s,\beta_s(b)),(s,\beta_s(c))}) \\
& = (q\circ i_X)^{(s)}(\phi_s(bc^*)) \\
& = q\circ i_X(e,bc^*) = q\circ i_A(bc^*).
\end{align*}
Since $q\circ i_A$ is nondegenerate,
this implies that $\overline q\circ i_P(s)$, and hence $V_s$, is an isometry.
This gives condition (a) for a crossed product.
Condition (c) is obvious, so it remains only to verify (b).
Suppose $(\pi,V)$ is a covariant representation of $(A,P,\beta,\omega)$ on a Hilbert space ${\mathcal H}$,
and define $\psi(s,x) := V_s^*\pi(x)$ as before. We have already seen that
$\psi$ is a Toeplitz representation of $X$, and it is Cuntz-Pimsner covariant since,
for any $b,c\in A$,
\begin{align*}
\pi^{(s)}(\phi_s(bc^*))
& = \pi^{(s)}(\Theta_{(s,\beta_s(b)),(s,\beta_s(c))})
= \psi(s,\beta_s(b))\psi(s,\beta_s(c))^* \\
& = V_s^*\pi(\beta_s(b))\pi(\beta_s(c^*))V_s
= V_s^*V_s\pi(bc^*)V_s^*V_s = \pi(bc^*).
\end{align*}
Defining $\pi\times V := \psi_* :{\mathcal O}_X\to B({\mathcal H})$ gives condition (c).
\end{proof}
\section{Crossed products twisted by product systems}
\label{section:crossed products2}
Multipliers of $P$ correspond to product systems over $P$ of one-dimensional
Hilbert spaces: given a multiplier $\omega$,
one defines multiplication on $P\times\field{C}$ by $(s,z)(t,w) := (st, \omega(s,t)zw)$.
In this section we consider twisted semigroup dynamical systems
in which the multiplier $\omega$ is replaced by a product system $X$ of Hilbert bimodules,
and we construct a crossed product which is ``twisted by $X$''.
For this, we first need to see how semigroups of endomorphism
arise from Toeplitz representations of product systems.
\begin{prop}\label{prop:alpha}
\textup{(1)} Let $X$ be a Hilbert bimodule over $A$, and suppose $(\psi,\pi)$
is a Toeplitz representation of $X$ on a Hilbert space ${\mathcal H}$.
Then there is a unique endomorphism
$\alpha = \alpha^{\psi,\pi}$ of $\pi(A)'$ such that
\begin{equation}\label{eq:alpha}
\alpha(S)\psi(x) = \psi(x)S
\qquad\text{for all $S\in\pi(A)'$ and $x\in X$,}
\end{equation}
and such that $\alpha(1)$ vanishes on $(\psi(X){\mathcal H})^\perp$.
\textup{(2)} Let $X$ be a product system over $P$ of Hilbert $A$--$A$ bimodules
in which each fiber $X_s$ is essential as a left $A$-module.
Let $\psi$ be a nondegenerate Toeplitz representation of $X$ on a Hilbert space ${\mathcal H}$,
and let $\alpha^\psi_s$ be the endomorphism $\alpha^{\psi_s,\psi_e}$ above.
Then $\alpha^\psi:P\to\End(\psi_e(A)')$ is a semigroup homomorphism,
and $\alpha^\psi_e$ is the identity endomorphism.
\end{prop}
\begin{proof} (1) The uniqueness of $\alpha$ is obvious.
By \cite[Proposition~2.69]{rw}, there is a unital homomorphism
$S\in\pi(A)' \mapsto 1\otimes_A S\in\Ind\pi({\mathcal L}(X))' \subseteq B(X\otimes_A{\mathcal H})$
determined by
\[
1\otimes_A S(x\otimes_A h) = x\otimes_A Sh
\qquad\text{for $x\in X$ and $h\in{\mathcal H}$.}
\]
Let $U:X\otimes_A{\mathcal H} \to {\mathcal H}$ be the isometry which satisfies
$U(x\otimes_A h) = \psi(x)h$ (see the proof of \cite[Proposition~1.6(1)]{fowrae2}),
and define
\[
\alpha(S) := U(1\otimes_A S)U^*
\qquad\text{for all $S\in\pi(A)'$.}
\]
Then $\alpha$ is a homomorphism, and $\alpha(1) = UU^*$ vanishes on $(\psi(X){\mathcal H})^\perp$.
If $S\in\pi(A)'$ and $x\in X$, then for any $h\in{\mathcal H}$ we have
\[
\alpha(S)\psi(x)h
= U(1\otimes_A S)(x\otimes_A h)
= U(x\otimes_A Sh) = \psi(x)Sh,
\]
giving \eqref{eq:alpha}.
Since $\pi(a)\psi(x)h = \psi(\phi(a)x)h$,
the space $\clsp\{\psi(x)h: x\in X, h\in{\mathcal H}\}$ reduces $\pi$;
hence for any $S\in\pi(A)'$ and $a\in A$,
both $\alpha(S)\pi(a)$ and $\pi(a)\alpha(S)$ vanish on $(\psi(X){\mathcal H})^\perp$.
This and
\begin{align*}
\alpha(S)\pi(a)\psi(x)h
& = \alpha(S)\psi(\phi(a)x)h
= \psi(\phi(a)x)Sh \\
& = \pi(a)\psi(x)Sh
= \pi(a)\alpha(S)\psi(x)h
\end{align*}
show that $\alpha(\pi(A)') \subseteq \pi(A)'$.
\textup{(2)}
Let $s,t\in P$, and suppose $x\in X_s$ and $y\in X_t$.
Vectors of the form $xy$ have dense linear span in $X_{st}$;
since $X_t$ is essential, this holds even when $s = e$ (see Remark~\ref{remark:essential}).
Since
\begin{align*}
\alpha^\psi_s(\alpha^\psi_t(S))\psi_{st}(xy)
& = \alpha^\psi_s(\alpha^\psi_t(S))\psi_s(x)\psi_t(y) \\
& = \psi_s(x)\alpha^\psi_t(S)\psi_t(y)
= \psi_s(x)\psi_t(y)S
= \psi_{st}(xy)S,
\end{align*}
we deduce that
\begin{equation}\label{eq:intertwine}
\alpha^\psi_s\circ\alpha^\psi_t(S)\psi_{st}(z) = \psi_{st}(z)S
\qquad\text{for all $S\in\psi_e(A)'$ and $z\in X_{st}$.}
\end{equation}
Once we show that
$\alpha^\psi_s\circ\alpha^\psi_t(1) = \alpha^\psi_{st}(1)$,
it follows from the uniqueness of $\alpha^\psi_{st}$
that $\alpha^\psi_s\circ\alpha^\psi_t = \alpha^\psi_{st}$.
From \eqref{eq:intertwine} we see that
$\alpha^\psi_s\circ\alpha^\psi_t(1) \ge \alpha^\psi_{st}(1)$.
Suppose that $\alpha^\psi_s\circ\alpha^\psi_t(1)f = f$;
we will show that $\alpha^\psi_{st}(1)f = f$, which
will complete the proof.
Since $f$ is in the range of $\alpha^\psi_s(1)$, it can be approximated
by a finite sum $\sum_i \psi_s(x_i)g_i$. Then
\[
f \doteq \alpha^\psi_s\circ\alpha^\psi_t(1)\sum_i \psi_s(x_i)g_i
= \sum_i \psi_s(x_i)\alpha^\psi_t(1)g_i.
\]
Now each $\alpha^\psi_t(1)g_i$ can be approximated by a finite
sum $\sum_j \psi_t(y_{ij})h_{ij}$, and then
\[
f
\doteq \sum_{i,j} \psi_s(x_i)\psi_t(y_{ij})h_{ij}
= \sum_{i,j} \psi_{st}(x_iy_{ij})h_{ij}.
\]
Thus $f$ can be approximated arbitrarily closely by a vector in the range
of $\alpha^\psi_{st}(1)$, and hence $\alpha^\psi_{st}(1)f = f$.
Since each $X_t$ is essential, the assumption that $\psi$ is nondegenerate
implies that $\psi_e$ is a nondegenerate representation of $A$.
Since $\alpha^\psi_e(S)\psi_e(a)h = \psi_e(a)Sh = S\psi_e(a)h$
for all $a\in A$ and $h\in{\mathcal H}$, we have $\alpha^\psi_e(S) = S$
for all $S\in\psi_e(A)'$.
\end{proof}
Consider a twisted semigroup dynamical system $(C, P, \beta, X)$
in which $C$ is a separable $C^*$\nobreakdash- al\-ge\-bra,
$\beta:P\to\End C$ is an action of the semigroup $P$
as extendible endomorphisms of $C$,
and $X$ is a product system over $P$ of Hilbert $A$--$A$ bimodules.
We assume that $\beta_e$ is the identity endomorphism,
and that each fiber $X_s$ is essential as a left $A$-module.
\begin{definition}\label{defn:covariant pair}
A {\em covariant representation\/} of $(C,P, \beta, X)$ on a Hilbert space ${\mathcal H}$
is a pair $(L, \psi)$ consisting of a nondegenerate representation
$L:C\to B({\mathcal H})$ and a nondegenerate Toeplitz representation $\psi:X\to B({\mathcal H})$
such that
\textup{(i)} $L(C) \subseteq \psi_e(A)'$, and
\textup{(ii)} $L\circ\beta_s = \alpha^\psi_s\circ L$ for $s\in P$.
\end{definition}
\begin{definition}\label{defn:crossed product}
A {\em crossed product\/} for $(C, P, \beta, X)$ is a triple
$(B, i_C, i_X)$ consisting of a $C^*$\nobreakdash- al\-ge\-bra $B$,
a nondegenerate homomorphism $i_C:C\to M(B)$, and a nondegenerate Toeplitz representation
$i_X:X\to M(B)$ such that
(a) there is a faithful nondegenerate representation $\sigma$ of $B$
such that $(\overline\sigma\circ i_C, \overline\sigma\circ i_X)$
is a covariant representation of $(C, P, \beta, X)$;
(b) for every covariant representation $(L,\psi)$ of $(C, P, \beta, X)$,
there is a representation $L\times\psi$ of $B$
such that $(\overline{L\times\psi})\circ i_C = L$
and $(\overline{L\times\psi})\circ i_X = \psi$;
(c) the $C^*$\nobreakdash- al\-ge\-bra $B$ is generated by
$\{i_C(c)i_X(x): c\in C,\ x\in X\}$.
\end{definition}
\begin{remark}
If each fiber $X_s$ has a finite basis $\{u_{s,1},\dots, u_{s,n(s)}\}$
(in the sense that $x = \sum_k u_{s,k}\cdot\langle u_{s,k}, x\rangle_A$
for every $x\in X_s$), it is not hard to show that (a) is equivalent
to asking that $i_C(c)i_X(a) = i_X(a)i_C(c)$ for all $c\in C$ and $a\in A=X_e$,
and that
\[
i_C(\beta_s(c)) = \sum_k i_X(u_{s,k})i_C(c)i_X(u_{s,k})^*
\qquad\text{for all $s\in P$ and $c\in C$.}
\]
In this case,
$(\overline\sigma\circ i_C, \overline\sigma\circ i_X)$
will be a covariant representation of $(C, P, \beta, X)$
for {\em every\/} nondegenerate representation $\sigma$ of $B$;
however, as demonstrated in \cite[Example~2.5]{fowrae}
for product systems of Hilbert spaces,
in general one cannot expect this to be the case.
\end{remark}
\begin{prop}\label{prop:existence of cp}
If $(C, P, \beta, X)$ has a covariant representation, then it has
a crossed product $(C\textstyle{\rtimes_{\beta,X}} P, i_C, i_X)$
which is unique in the following sense:
if $(B, i_C', i_X')$ is another crossed product for $(C, P, \beta, X)$,
then there is an isomorphism $\theta: C\textstyle{\rtimes_{\beta,X}} P \to B$
such that $\overline\theta\circ i_C = i_C'$ and $\theta\circ i_X = i_X'$.
\end{prop}
\begin{remark} When $X$ is the product system $P\times\field{C}$
with multiplication given by a multiplier $\omega$,
it is not hard to see that $C\textstyle{\rtimes_{\beta,X}} P$ is precisely the crossed product
$C\rtimes_{\beta,\omega} P$ defined in the previous section.
If $C$ is unital and $A = \field{C}$, then $C\textstyle{\rtimes_{\beta,X}} P$ is the crossed product
defined in \cite[Section~2]{fowrae}.
\end{remark}
\begin{proof}[Proof of Proposition~\ref{prop:existence of cp}]
If $S$ is a set of pairs $(L,\psi)$ consisting of maps $L:C\to B({\mathcal H}_{L,\psi})$
and $\psi:X\to B({\mathcal H}_{L,\psi})$, then $(\oplus L,\oplus\psi)$ is a covariant
representation of $(C,P,\beta,X)$ if and only if each $(L,\psi)$ is.
The main point here is that the value of $\alpha^{\oplus\psi}_s$
on an element of $(\oplus\psi)_e(A)'$ of the form $\oplus L(c)$
is $\oplus\alpha^\psi_s(L(c))$.
Suppose $(L,\psi)$ is a nondegenerate covariant representation
on a separable Hilbert space ${\mathcal H}$; that is, the $C^*$\nobreakdash- al\-ge\-bra
\[
{\mathcal U} := C^*(\{ L(c)\psi(x): c\in C, x\in X\})
\]
acts nondegenerately on ${\mathcal H}$.
We will identify the multiplier algebra of ${\mathcal U}$
with the concrete $C^*$\nobreakdash- al\-ge\-bra
\[
M({\mathcal U}) = \{S\in B({\mathcal H}): ST, TS\in{\mathcal U}\text{ for every $T\in{\mathcal U}$}\}.
\]
We claim that $L(C)\cup\psi(X)\subseteq M({\mathcal U})$.
For this, it suffices to check
that multiplying a generator $L(c)\psi(x)$ of ${\mathcal U}$ on either the left
or the right by an operator of the form $L(d)$, $\psi(y)$, or $\psi(y)^*$
yields another element of ${\mathcal U}$.
Certainly $L(d)L(c)\psi(x) = L(dc)\psi(x)\in{\mathcal U}$
and $L(c)\psi(x)\psi(y) = L(c)\psi(xy)\in{\mathcal U}$,
and since
\begin{equation}\label{eq:psiL}
\psi(y)L(c) = \alpha^\psi_{p(y)}(L(c))\psi(y) = L(\beta_{p(y)}(c))\psi(y),
\end{equation}
we also have $\psi(y)L(c)\psi(x) = L(\beta_{p(y)}(c))\psi(yx)\in{\mathcal U}$
and $L(c)\psi(x)L(d) = L(c\beta_{p(x)}(d))\psi(x)\in{\mathcal U}$.
Writing $c = c_1^*c_2$ with $c_1, c_2\in C$ gives
\[
\psi(y)^*L(c)\psi(x) = (L(c_1)\psi(y))^*L(c_2)\psi(x) \in {\mathcal U}.
\]
Finally, to see that $L(c)\psi(x)\psi(y)^*\in{\mathcal U}$, we use a trick from \cite{adji}.
Let $(c_i)$ be an approximate identity for $C$;
we claim that
\begin{equation}\label{eq:norm convergence}
L(c)\psi(x)L(c_i) \stackrel{\lVert\ \rVert}\longrightarrow L(c)\psi(x).
\end{equation}
Since $L$ is nondegenerate, $L(c)\psi(x)L(c_i)$
converges strongly to $L(c)\psi(x)$.
On the other hand, using \eqref{eq:psiL} we see that
$L(c)\psi(x)L(c_i)
= L(c\beta_{p(x)}(c_i))\psi(x)$
converges in {\em norm\/}
(to $L(c\overline{\beta_{p(x)}}(1))\psi(x)$),
and \eqref{eq:norm convergence} follows.
Hence
\[
L(c)\psi(x)L(c_i)\psi(y)^* \stackrel{\lVert\ \rVert}\longrightarrow L(c)\psi(x)\psi(y)^*,
\]
and since
\[
L(c)\psi(x)L(c_i)\psi(y)^*
= L(c)\psi(x)\psi(y)^*L(\beta_{p(y)}(c_i)) \in{\mathcal U},
\]
we deduce that $L(c)\psi(x)\psi(y)^*\in{\mathcal U}$.
Since $M({\mathcal U}) \subseteq {\mathcal U}''$, we have shown
in particular that the ranges of $L$ and $\psi$ are contained in ${\mathcal U}''$.
Consequently, any decomposition $1 = \sum Q_\lambda$ of the identity
as a sum of mutually orthogonal projections $Q_\lambda\in{\mathcal U}'$ gives
corresponding decompositions $L = \oplus Q_\lambda L$ and $\psi = \oplus Q_\lambda\psi$,
and by the first paragraph each pair $(Q_\lambda L,Q_\lambda\psi)$ is a covariant
representation of $(C,P,\beta,X)$. By the usual Zorn's Lemma argument we
can choose these projections such that ${\mathcal U}$ acts cyclically on $Q_\lambda{\mathcal H}$;
since $C^*(\{Q_\lambda L(c) Q_\lambda\psi(x): c\in C,\ x\in X\}) = Q_\lambda{\mathcal U}$
acts cyclically on $Q_\lambda{\mathcal H}$, this shows that every covariant representation of
$(C,P,\beta,X)$ decomposes as a direct sum of cyclic representations.
Let $S$ be a set of cyclic covariant representations with
the property that every cyclic covariant representation of $(C,P,\beta,X)$
is unitarily equivalent to an element in $S$.
It can be shown that such a set $S$ exists
by fixing a Hilbert space ${\mathcal H}$ of sufficiently
large cardinality (depending on the cardinalities of $C$ and $X$)
and considering only representations on ${\mathcal H}$.
Note that $S$ is nonempty because the system has a covariant representation,
which has a cyclic summand.
Define $i_C := \bigoplus_{(L,\psi)\in S} L$
and $i_X := \bigoplus_{(L,\psi)\in S} \psi$,
and let $C\textstyle{\rtimes_{\beta,X}} P$ be the $C^*$\nobreakdash- al\-ge\-bra generated by
$\{i_C(c)i_X(x): c\in C,\ x\in X\}$.
By the first paragraph, $(i_C,i_X)$ is a covariant representation of $(C,P,\beta,X)$,
and it is nondegenerate since each $(L,\psi)$ is.
We deduce that both $i_C$ and $i_X$ map into $M(C\textstyle{\rtimes_{\beta,X}} P)$,
and that condition (a) for a crossed product is satisfied by taking
$\sigma$ to be the identity representation. Condition (c) is trivial,
and (b) holds because every covariant representation decomposes as a direct sum
of cyclic representations.
We need to show that $i_C:C\to M(C\textstyle{\rtimes_{\beta,X}} P)$ and $i_X:X\to M(C\textstyle{\rtimes_{\beta,X}} P)$ are nondegenerate.
For this, let $c\in C$ and $x\in X$.
If $(a_i)$ is an approximate identity for $A = X_e$,
then by \eqref{eq:approximate identity} we have
$i_C(c)i_X(x)i_X(a_i) = i_C(c)i_X(x\cdot a_i) \to i_C(c)i_X(x)$
and
$i_X(a_i)i_C(c)i_X(x) = i_C(c)i_X(a_i)i_X(x) = i_C(c)i_X(\phi(a_i)x) \to i_C(c)i_X(x)$,
so $i_X$ is nondegenerate (Lemma~\ref{lemma:nondegenerate}).
If $(c_i)$ is an approximate identity for $C$,
then $i_C(c_i)i_C(c)i_X(x) = i_C(c_ic)i_X(x) \to i_C(c)i_X(x)$,
and since $i_C$ is nondegenerate as a representation on Hilbert space,
\eqref{eq:norm convergence} gives $i_C(c)i_X(x)i_C(c_i) \to i_C(c)i_X(x)$.
Thus $i_C$ is nondegenerate.
For the uniqueness assertion, suppose $(B, i_C', i_X')$ is another crossed product.
Condition (a) allows us to assume that $(i_C,i_X)$ and $(i_C', i_X')$
are covariant representations of $(C,P,\beta,X)$ on Hilbert spaces ${\mathcal H}$ and ${\mathcal H}'$.
Condition (b) then gives a representation $i_C'\times i_X':C\textstyle{\rtimes_{\beta,X}} P\to B({\mathcal H}')$
whose image is contained in $B$ since $i_C'\times i_X'(i_C(c)i_X(x)) = i_C'(c)i_X'(x)$.
Similarly one obtains a map $i_C\times i_X:B\toC\textstyle{\rtimes_{\beta,X}} P$ which is obviously an inverse
for $i_C'\times i_X':C\textstyle{\rtimes_{\beta,X}} P\to B$.
\end{proof}
If $P$ is a subsemigroup of a group $G$,
then there is a {\em dual coaction\/} of $G$ on $C\textstyle{\rtimes_{\beta,X}} P$:
\begin{prop}\label{prop:coaction}
Suppose $(C,P,\beta,X)$ is a twisted system which has a covariant
representation. If $P$ is a subsemigroup of a group $G$, then there is an injective
coaction
\[
\delta:C\textstyle{\rtimes_{\beta,X}} P \to (C\textstyle{\rtimes_{\beta,X}} P) \otimes_{\min} C^*(G)
\]
such that
\[
\delta(i_C(c)i_X(x)) = i_C(c)i_X(x) \otimes i_G(p(x)).
\]
If $G$ is abelian, there is a strongly continuous action
$\widehat\beta$ of $\widehat G$ on $C\textstyle{\rtimes_{\beta,X}} P$ such that
\[
\widehat\beta_\gamma(i_C(c)i_X(x)) = \gamma(p(x))i_C(c)i_X(x).
\]
\end{prop}
\begin{proof} We follow \cite[Proposition~2.7]{fowrae}.
Let $\sigma$ be a faithful nondegenerate representation $\sigma$ of $C\textstyle{\rtimes_{\beta,X}} P$
such that $(L,\psi) := (\overline\sigma\circ i_C, \overline\sigma\circ i_X)$
is a covariant representation of $(C, P, \beta, X)$,
and let $U$ be a unitary representation of $G$ whose integrated form $\pi_U$
is faithful on $C^*(G)$.
We claim that $(L\otimes 1, \psi\otimes (U\circ p))$
is a covariant representation of $(C, P, \beta, X)$.
Most of the verifications are routine, so we check only that
\begin{equation}\label{eq:coaction}
L(\beta_s(c))\otimes 1 = \alpha^{\psi\otimes(U\circ p)}_s(L(c)\otimes 1)
\qquad\text{for all $s\in P$ and $c\in C$.}
\end{equation}
For this, we show that $L(\beta_s(c))\otimes 1$ satisfies the properties
which characterize $\alpha^{\psi\otimes(U\circ p)}_s(L(c) \otimes 1)$
(Proposition~\ref{prop:alpha}).
First, let $x\in X_s$; we show that \eqref{eq:coaction}
holds on any vector in the range of
$(\psi\otimes(U\circ p))(x) = \psi_s(x)\otimes U_s$:
\begin{multline*}
(L(\beta_s(c))\otimes 1)(\psi_s(x)\otimes U_s)
= \alpha^\psi_s(L(c))\psi_s(x) \otimes U_s
= \psi_s(x)L(c)\otimes U_s \\
= (\psi_s(x)\otimes U_s)(L(c)\otimes 1)
= \alpha^{\psi\otimes(U\circ p)}_s(L(c)\otimes 1)(\psi_s(x)\otimes U_s).
\end{multline*}
Next, note that $\alpha^{\psi\otimes(U\circ p)}_s(1)$
is the projection onto
\begin{align*}
& \clsp\{(\psi\otimes(U\circ p))(x)\xi: x\in X_s, \xi\in{\mathcal H}_\sigma\otimes{\mathcal H}_U\} \\
& \qquad\qquad = \clsp\{\psi_s(x)h \otimes U_sk : x\in X_s, h\in{\mathcal H}_\sigma, k\in{\mathcal H}_U\} \\
& \qquad\qquad = \clsp\{\psi_s(x)h : x\in X_s, h\in{\mathcal H}_\sigma\} \otimes {\mathcal H}_U,
\end{align*}
which is precisely the range of $\alpha^\psi_s(1)\otimes 1$.
Since $L(\beta_s(c))\otimes 1 = \alpha^\psi_s(L(c)) \otimes 1$
vanishes on the range of $1 - \alpha^\psi_s(1)\otimes 1$,
\eqref{eq:coaction} follows from the uniqueness assertion
of Proposition~\ref{prop:alpha}.
Since $(L\otimes 1, \psi\otimes (U\circ p))$ is covariant,
there is a representation $\rho$ of $C\textstyle{\rtimes_{\beta,X}} P$ such that
\begin{align*}
\rho(i_C(c)i_X(x))
& = (L(c)\otimes 1)(\psi(x)\otimes U_{p(x)}) \\
& = (\sigma\otimes\pi_U)(i_C(c)i_X(x) \otimes i_G(p(x))).
\end{align*}
Since $\sigma$ and $\pi_U$ are faithful, $\sigma\otimes\pi_U$
is faithful on $(C\textstyle{\rtimes_{\beta,X}} P) \otimes_{\min} C^*(G)$,
and we can define $\delta := (\sigma\otimes\pi_U)^{-1}\circ\rho$.
By checking on generators it is easy to see that $\delta$
satisfies the coaction identity
$(\id\otimes\delta_G) \circ \delta = (\delta\otimes\id) \circ \delta$,
and $\delta$ is injective since $\sigma = (\sigma\otimes\epsilon)\circ\delta$,
with $\epsilon$ the augmentation representation of $C^*(G)$
(i.e., $\epsilon(i_G(s)) = 1$ for all $s\in G$).
When $G$ is abelian, $\widehat\beta$ is the action
canonically associated with $\delta$.
\end{proof}
\section{Nica covariance}
\label{section:covariance}
Now suppose $P$ is a subsemigroup of a group $G$ such that $P\cap P^{-1} = \{e\}$.
Then $s\le t$ iff $s^{-1}t\in P$ defines a partial order on $G$ which is left-invariant:
for any $r,s,t\in P$ we have $s\le t$ iff $rs \le rt$.
Following Nica \cite{nica}, we say that $(G,P)$ is a {\em quasi-lattice ordered group\/}
if every finite subset of $G$ which has an upper bound in $P$
has a least upper bound in $P$.
When $s,t\in P$ have a common upper bound, we denote their
least upper bound by $s\vee t$; when $s$ and $t$ have no common
upper bound we write $s\vee t = \infty$.
For a finite subset $C = \{t_1,\dots, t_n\}$ of $P$, we write
$\sigma C$ for $t_1 \vee \dots \vee t_n$.
\begin{definition}\label{defn:psicovariant}
Suppose $(G,P)$ is a quasi-lattice ordered group
and $X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules.
We call a Toeplitz representation $\psi:X\to B({\mathcal H})$ {\em Nica covariant\/} if
\[
\alpha^\psi_s(1)\alpha^\psi_t(1)
= \begin{cases} \alpha^\psi_{s\vee t}(1) & \text{if $s\vee t < \infty$} \\
0 & \text{otherwise.}
\end{cases}
\]
\end{definition}
\begin{remark} If $(G,P)$ is totally ordered,
then every Toeplitz representation of $X$ is Nica covariant.
\end{remark}
\begin{lemma}\label{lemma:fock covariant}
Let $l:X\to{\mathcal L}(F(X))$ be the Fock representation, and suppose $\pi$
is a representation of $A$ on a Hilbert space ${\mathcal H}$.
Then
\[
\Psi := F(X)\dashind_A^{{\mathcal L}(X)}\pi\circ l
\]
is a Nica-covariant Toeplitz representation of $X$.
If $\pi$ is faithful, then $\Psi$ is isometric.
\end{lemma}
\begin{proof} Since $l$ is a Toeplitz representation, so is $\Psi$.
Let $s\in P$. The range of $\alpha^\Psi_s(1)$ is
\begin{multline*}
\clsp\{\Psi(x)\xi: x\in X_s, \xi\in F(X)\otimes_A{\mathcal H}\} \\
= \clsp\{l(x)y\otimes_A h: x\in X_s, y\in F(X), h\in{\mathcal H}\}
= \bigoplus_{s \le r} X_r \otimes_A {\mathcal H}.
\end{multline*}
Hence for any $s,t\in P$, the range of $\alpha^\Psi_s(1)\alpha^\Psi_t(1)$ is
\[
\Bigl( \bigoplus_{s \le r} X_r \otimes_A {\mathcal H} \Bigr)
\cap
\Bigl( \bigoplus_{t \le r} X_r \otimes_A {\mathcal H} \Bigr),
\]
which is $\bigoplus_{s\vee t \le r} X_r \otimes_A {\mathcal H} = \ran\alpha^\Psi_{s\vee t}(1)$
if $s\vee t < \infty$, and is zero otherwise.
If $\pi$ is faithful then so is $F(X)\dashind_A^{{\mathcal L}(X)}\pi$;
since $l$ is isometric, this implies that $\Psi$ is isometric.
\end{proof}
\begin{prop}\label{prop:CP Nica}
Let $(G,P)$ be a quasi-lattice ordered group
such that every $s,t\in P$ have a common upper bound.
Let $X$ be a product system over $P$ of essential Hilbert $A$--$A$ bimodules
such that the left action of $A$ on each fiber $X_s$ is by compact operators.
Then every Toeplitz representation $\psi:X\to B({\mathcal H})$
which is Cuntz-Pimsner covariant is also Nica covariant.
\end{prop}
\begin{proof} Fix $s\in P$.
Since $(\psi_s,\psi_e)$ is Cuntz-Pimsner covariant and $\phi_s(A)\subseteq{\mathcal K}(X_s)$,
\cite[Lemma~1.6]{fmr} gives $\psi_e(A){\mathcal H}\subseteq\clsp\psi_s(X_s){\mathcal H}$.
But $X_s$ is essential, so the reverse inclusion holds as well,
and since $\clsp\psi_s(X_s){\mathcal H}$ is precisely the range of $\alpha^\psi_s(1)$,
we deduce that $\alpha^\psi_s(1)$ is constant in $s$.
Since $s\vee t<\infty$ for all $s,t\in P$, this implies that $\psi$
is Nica covariant.
\end{proof}
There are product systems for which Nica covariance is not a
$C^*$\nobreakdash- al\-ge\-braic condition;
that is, if $\psi:X\to B({\mathcal H})$ is Nica covariant
and $\sigma:C^*(\psi(X))\to B({\mathcal K})$ is a homomorphism,
the composition $\sigma\circ\psi$ need not be Nica covariant
\cite[Example~1.3]{fowler}.
We pause a moment to show how to adapt the methods
of \cite{fowler} to avoid this pathology.
The following Lemma collects some results we shall need
for both this and the sequel.
\begin{lemma}\label{lemma:rho}
Suppose $X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules,
$\psi:X\to B({\mathcal H})$ is a Toeplitz representation, and $s\in P$.
\textup{(1)}
There is a strict--strong continuous representation
$\rho^\psi_s:{\mathcal L}(X_s)\to B({\mathcal H})$
such that
\[
\rho^\psi_s(S)\psi_s(x)h = \psi_s(Sx)h
\qquad\text{for all $S\in{\mathcal L}(X_s)$, $x\in X_s$, and $h\in{\mathcal H}$,}
\]
and such that $\rho^\psi_s(S)$ vanishes on $(\psi_s(X_s){\mathcal H})^\perp$.
Moreover, $\rho^\psi_s(S) = \psi^{(s)}(S)$ for every $S\in{\mathcal K}(X_s)$.
\textup{(2)} $\rho^\psi_s(1) = \alpha^\psi_s(1)$
\textup{(3)} If $a\in A$ satisfies $\phi_s(a)\in{\mathcal K}(X_s)$, then
\begin{equation}\label{eq:absorb}
\psi_e(a)\rho^\psi_s(1) = \psi^{(s)}(\phi_s(a)) = \rho^\psi_s(1)\psi_e(a).
\end{equation}
\textup{(4)} If $Q\in\psi_e(A)'$,
then $\alpha^\psi_s(Q)\in\rho^\psi_s({\mathcal L}(X_s))'$.
Further, if $Q$ is a projection such that $\psi_e$ acts faithfully on $Q{\mathcal H}$,
then $\rho^\psi_s$ acts faithfully on $\alpha^\psi_s(Q){\mathcal H}$.
\textup{(5)} For all $S\in{\mathcal L}(X_s)$ and $t\in P$ we have
\[
\rho^\psi_{st}(S\otimes_A 1)
= \rho^\psi_s(S)\rho^\psi_{st}(1)
= \rho^\psi_{st}(1)\rho^\psi_s(S),
\]
where $S\otimes_A 1(xy) := (Sx)y$ for all $x\in X_s$ and $y\in X_t$.
\textup{(6)} If $t\in P$ and $z,w\in X_s$, then
$\rho^\psi_{st}(\Theta_{z,w} \otimes_A 1) = \psi(z)\alpha^\psi_t(1)\psi(w)^*$.
\end{lemma}
\begin{proof} (1) See \cite[Proposition~1.6(1)]{fowrae2}.
For the continuity assertion,
suppose $S_\lambda\to S$ strictly in ${\mathcal L}(X_s) = M{\mathcal K}(X_s)$,
$x\in X_s$, and $h\in{\mathcal H}$.
There exists $K\in{\mathcal K}(X_s)$ and $y\in X_s$ such that $x = Ky$,
and then
\[
\rho^\psi_s(S_\lambda)\psi_s(x)h
= \rho^\psi_s(S_\lambda K)\psi_s(y)h
\to \rho^\psi_s(SK)\psi_s(y)h
= \rho^\psi_s(S)\psi_s(x)h.
\]
(2) Both $\rho^\psi_s(1)$ and $\alpha^\psi_s(1)$ are the projection onto
$\clsp\{\psi_s(X_s){\mathcal H}\}$.
(3) If $x\in X_s$ and $h\in{\mathcal H}$, then
\[
\psi_e(a)\rho^\psi_s(1)\psi_s(z)h
= \psi_e(a)\psi_s(z)h
= \psi_s(\phi_s(a)z)h
= \psi^{(s)}(\phi_s(a))\psi_s(z)h;
\]
since both sides of \eqref{eq:absorb} are supported on $\clsp\psi_s(X_s){\mathcal H}$,
this implies that $\psi_e(a)\rho^\psi_s(1) = \psi^{(s)}(\phi_s(a))$.
By (2), $\rho^\psi_s(1)$ commutes with $\psi_e(a)$,
giving the other half of \eqref{eq:absorb}.
(4) When $Q$ is a projection, $\alpha^\psi_s(Q)$ is the projection
onto $\clsp\psi_s(X_s)Q{\mathcal H}$,
and the result follows from \cite[Proposition~1.6(2)]{fowrae2}.
(5) See \cite[Proposition~1.8(2)]{fowrae2}.
(6) $\rho^\psi_{st}(\Theta_{z,w} \otimes_A 1)
= \rho^\psi_{st}(1)\rho^\psi_s(\Theta_{z,w})
= \alpha^\psi_{st}(1)\psi(z)\psi(w)^*
= \psi(z)\alpha^\psi_t(1)\psi(w)^*$.
\end{proof}
\begin{prop}\label{prop:characterize covariance}
Suppose $(G,P)$ is a quasi-lattice ordered group
and $X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules.
A Toeplitz representation $\psi:X\to B({\mathcal H})$ is Nica covariant if and only if
\begin{equation}\label{eq:characterize covariance}
\rho^\psi_s(S)\rho^\psi_t(T)
= \begin{cases} \rho^\psi_{s\vee t}((S\otimes_A 1)(T\otimes_A 1))
& \text{if $s\vee t < \infty$} \\
0 & \text{otherwise.}
\end{cases}
\end{equation}
\end{prop}
\begin{proof} \cite[Proposition~1.4]{fowler}
If $\psi$ is Nica covariant, then
\begin{align*}
\rho^\psi_s(S)\rho^\psi_t(T)
& = \rho^\psi_s(S)\rho^\psi_s(1)\rho^\psi_t(1)\rho^\psi_t(T) \\
& = \rho^\psi_s(S)\rho^\psi_{s\vee t}(1)\rho^\psi_t(T)
= \rho^\psi_{s\vee t}((S\otimes_A 1)(T\otimes_A 1)),
\end{align*}
where the last equality uses Lemma~\ref{lemma:rho}(5).
Conversely, suppose \eqref{eq:characterize covariance} holds for all compact $S$ and $T$.
If $S\to 1$ strictly, then
\[
\rho^\psi_{s\vee t}((S\otimes_A 1))
= \rho^\psi_s(S)\rho^\psi_{s\vee t}(1)
\to \rho^\psi_s(1)\rho^\psi_{s\vee t}(1)
= \rho^\psi_{s\vee t}(1),
\]
where the convergence is in the strong operator topology.
Hence
\[
\rho^\psi_s(1)\rho^\psi_t(T)
= \begin{cases} \rho^\psi_{s\vee t}(T\otimes_A 1)
& \text{if $s\vee t < \infty$} \\
0 & \text{otherwise}
\end{cases}
\]
for every $T\in{\mathcal K}(X_t)$.
Letting $T\to 1$ strictly shows that $\psi$ is Nica covariant.
\end{proof}
When each product $(S\otimes_A 1)(T\otimes_A 1)$ is compact,
the previous Proposition allows us to give a $C^*$\nobreakdash- al\-ge\-braic
characterization of Nica covariance:
\begin{definition}\label{defn:CA}
Suppose $(G,P)$ is a quasi-lattice ordered group and
$X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules.
We say that $X$ is {\em compactly aligned\/}
if whenever $s,t\in P$ have a common upper
bound and $S$ and $T$ are compact operators on $X_s$ and $X_t$, respectively,
$(S\otimes_A 1)(T\otimes_A 1)$
is a compact operator on $X_{s \vee t}$.
If $X$ is compactly aligned and $\psi$ is a Toeplitz representation of $X$ in
a $C^*$\nobreakdash- al\-ge\-bra $B$, we say that $\psi$ is {\em Nica covariant\/} if
\[
\psi^{(s)}(S)\psi^{(t)}(T)
= \begin{cases} \psi^{(s\vee t)}((S\otimes_A 1)(T\otimes_A 1))
& \text{if $s\vee t < \infty$} \\
0 & \text{otherwise.}
\end{cases}
\]
whenever $s,t\in P$, $S\in{\mathcal K}(X_s)$ and $T\in{\mathcal K}(X_t)$.
\end{definition}
\begin{prop}\label{prop:CA conditions}
If $(G,P)$ is a total order, or if the left action of $A$ on each fiber $X_s$
is by compact operators, then $X$ is compactly aligned.
\end{prop}
\begin{proof} Suppose $s,t\in P$, $s\vee t < \infty$,
$S\in{\mathcal K}(X_s)$, and $T\in{\mathcal K}(X_t)$.
If $(G,P)$ is a total order then either $S\otimes_A 1 = S$
or $T\otimes_A 1 = T$; either way $(S\otimes_A 1)(T\otimes_A 1)$
is compact.
If the left action of $A$ on each fiber $X_s$
is by compact operators, then by \cite[Corollary~3.7]{pimsner},
both $S\otimes_A 1$
and $T\otimes_A 1$ are compact.
\end{proof}
\begin{prop}\label{prop:preserve covariance}
Suppose $X$ is compactly aligned.
Let $B$ and $C$ be $C^*$\nobreakdash- al\-ge\-bras,
let $\psi:X\to B$ be a Nica-covariant Toeplitz representation,
and let $\sigma:B\to C$ be a homomorphism.
Then $\sigma\circ\psi$ is Nica covariant.
\end{prop}
\begin{proof} By checking on an operator $\Theta_{x,y}\in{\mathcal K}(X_s)$,
one verifies that $(\sigma\circ\psi)^{(s)} = \sigma\circ\psi^{(s)}$,
and the result follows easily from this.
\end{proof}
\begin{prop}\label{prop:yz}
Suppose $X$ is a compactly-aligned product system,
$\psi$ is a Nica-covariant Toeplitz representation of $X$,
$s,t\in P$, $y\in X_s$, and $z\in X_t$.
If $s\vee t = \infty$, then $\psi(y)^*\psi(z) = 0$;
otherwise
\[
\psi(y)^*\psi(z)
\in\clsp\{\psi(f)\psi(g)^*: f\in X_{s^{-1}(s\vee t)}, g\in X_{t^{-1}(s\vee t)}\}.
\]
\end{prop}
\begin{proof} Express $y = Sy'$ with $S\in{\mathcal K}(X_s)$ and $y'\in X_s$;
similarly, express $z = Tz'$ with $T\in{\mathcal K}(X_t)$ and $z'\in X_t$.
Since $\psi$ is Nica covariant,
\[
\psi(y)^*\psi(z) = \psi(y')^*\rho^\psi_s(S^*)\rho^\psi_t(T)\psi(z')
\]
is zero if $s\vee t = \infty$, and otherwise
\[
\psi(y)^*\psi(z) = \psi(y')^*\rho^\psi_{s\vee t}(K)\psi(z'),
\]
where $K = (S^*\otimes_A 1)(T\otimes_A 1) \in{\mathcal K}(X_{s\vee t})$.
Since $K$ is compact it can be approximated in norm by a finite sum of operators
$\Theta_{u,v}$ with $u,v\in X_{s\vee t}$,
and hence $\rho^\psi_{s\vee t}(K)$ can be approximated by finite sums
of the form $\psi(u)\psi(v)^*$.
But any such $u$ can be approximated by finite sums
of products $u_1f'$ with $u_1\in X_s$ and $f'\in X_{s^{-1}(s\vee t)}$;
similarly, any such $v$ can be approximated by finite sums
of products $v_1g'$ with $v_1\in X_t$ and $g'\in X_{t^{-1}(s\vee t)}$.
Hence $\psi(y')^*\rho^\psi_{s\vee t}(K)\psi(z')$ can be approximated in norm
by finite sums of operators of the form
\[
\psi(y')^*\psi(u_1)\psi(f')\psi(g')^*\psi(v_1)^*\psi(z')
= \psi(\langle y',u_1 \rangle_A f')\psi(\langle z',v_1 \rangle_A g')^*.
\]
\end{proof}
The following Lemma is useful
when working with Nica-covariant Toeplitz representations.
\begin{lemma}\label{lemma:handy}
Suppose $(G,P)$ is a quasi-lattice ordered group,
$X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules,
$\psi$ is a Toeplitz representation of $X$ on ${\mathcal H}$,
$x\in X$, and $s\in P$.
\textup{(1)} If $p(x) \le s$, then
$\alpha^\psi_s(S)\psi(x) = \psi(x)\alpha^\psi_{p(x)^{-1}s}(S)$
for all $S\in\psi_e(A)'$.
\textup{(2)} If $\psi$ is Nica covariant, then
\[
\alpha^\psi_s(1)\psi(x) =
\begin{cases}
\psi(x)\alpha^\psi_{p(x)^{-1}(p(x) \vee s)}(1)
& \text{if $p(x) \vee s < \infty$,} \\
0 & \text{otherwise.}
\end{cases}
\]
\end{lemma}
\begin{proof} The proof is formally identical to that of \cite[Lemma~3.6]{fowrae}.
\end{proof}
\section{The system $(B_P,P,\tau,X)$}
\label{section:system}
For each $t\in P$, let $1_t\in\ell^\infty(P)$
be the characteristic function of $tP$.
Since the product $1_s1_t$ is either $1_{s\vee t}$ or $0$,
$B_P := \clsp\{1_t: t\in P\}$
is a $C^*$\nobreakdash- subalgebra of $\ell^\infty(P)$.
Left translation on $\ell^\infty(P)$ restricts to an action
$\tau$ of $P$ on $B_P$, determined by
$\tau_s(1_t) = 1_{st}$ for $s,t\in P$.
\begin{prop}\label{prop:Lpsi}
Suppose $(G,P)$ is a quasi-lattice ordered group and
$X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules.
\textup{(1)} If $(L, \psi)$ is a covariant representation of
$(B_P, P, \tau, X)$, then $\psi$ is a nondegenerate Nica-covariant Toeplitz representation of $X$ and
$L(1_s) = \alpha^\psi_s(1)$.
\textup{(2)} If $\psi$ is a
nondegenerate Nica-covariant Toeplitz representation of $X$ on a Hilbert space ${\mathcal H}$,
then there is a representation $L^\psi:B_P\to B({\mathcal H})$ such that
$L^\psi(1_s) = \alpha^\psi_s(1)$; moreover, $(L^\psi, \psi)$ is
then a covariant representation of $(B_P, P, \tau, X)$.
\end{prop}
\begin{proof}
The proof is formally identical to that of \cite[Proposition~4.1]{fowrae},
except that in (2) one must also note that $L^\psi(B_P) \subseteq\psi_e(A)'$
since $L^\psi(1_s) = \alpha^\psi_s(1) \in \psi_e(A)'$
and $\{1_s: s\in P\}$ generates $B_P$.
\end{proof}
\begin{cor}\label{cor:Lpsi}
The system $(B_P, P, \tau, X)$ has a covariant representation.
\end{cor}
\begin{proof}
Let $\pi$ be a nondegenerate representation of $A$ on a Hilbert space ${\mathcal H}$,
and let $l:X\to{\mathcal L}(F(X))$ be the Fock representation of $X$.
By Lemma~\ref{lemma:fock covariant},
$\Psi := F(X)\dashind_A^{{\mathcal L}(X)}\pi\circ l$
is a Nica-covariant Toeplitz representation of $X$.
Since $\pi$ is nondegenerate, so is $F(X)\dashind_A^{{\mathcal L}(X)}\pi$;
since $l$ is nondegenerate, $\Psi$ is as well.
The previous Proposition thus gives a covariant representation
$(L^\Psi,\Psi)$ of $(B_P,P,\tau,X)$.
\end{proof}
Let $i_X$ and $i_{B_P}$ be the canonical maps of $X$ and $B_P$ into $B_P \textstyle{\rtimes_{\tau, X}} P$.
Since $B_P$ is unital, $i_X(x) = i_{B_P}(1)i_X(x) \inB_P \textstyle{\rtimes_{\tau, X}} P$ for each $x\in X$.
We write ${\mathcal T}_{\cv}(X)$ for the $C^*$\nobreakdash- sub\-algebra of $B_P \textstyle{\rtimes_{\tau, X}} P$ generated
by $i_X(X)$; the following Theorem justifies this notation.
\begin{theorem}\label{theorem:subalgebra}
$({\mathcal T}_{\cv}(X),i_X)$ is universal for Nica-covariant Toeplitz representations of $X$,
in the sense that:
\textup{(a)} there is a faithful representation $\theta$ of ${\mathcal T}_{\cv}(X)$
on Hilbert space such that $\theta\circ i_X$ is a Nica-covariant
Toeplitz representation of $X$, and
\textup{(b)} for every Nica-covariant Toeplitz representation $\psi$ of $X$,
there is a representation $\psi_*$ of ${\mathcal T}_{\cv}(X)$ such that $\psi=\psi_*\circ i_X$.
\noindent Up to canonical isomorphism, $({\mathcal T}_{\cv}(X),i_X)$ is the unique
pair with this property.
If $X$ is compactly aligned, then $i_X$ is Nica covariant,
\begin{equation}\label{eq:span toeplitz}
{\mathcal T}_{\cv}(X) = \clsp\{i_X(x)i_X(y)^*: x,y\in X\},
\end{equation}
and
\begin{equation}\label{eq:span bpp}
B_P \textstyle{\rtimes_{\tau, X}} P = \clsp\{i_X(x)i_{B_P}(1_s)i_X(y)^*: x,y\in X, s\in P\}.
\end{equation}
\noindent If the left action of $A$ on each fiber $X_s$ is by compact operators,
then ${\mathcal T}_{\cv}(X)$ is all of $B_P \textstyle{\rtimes_{\tau, X}} P$;
if in addition every $s,t\in P$ have a common upper bound,
then the Cuntz-Pimsner algebra ${\mathcal O}_X$ is a quotient of ${\mathcal T}_{\cv}(X)$.
\end{theorem}
\begin{proof}[Proof of Theorem~\ref{theorem:subalgebra}]
Let $\sigma$ be a faithful representation of $B_P \textstyle{\rtimes_{\tau, X}} P$ on a Hilbert space ${\mathcal H}$
such that $(\overline\sigma \circ i_{B_P}, \sigma\circ i_X)$
is a covariant representation of $(B_P, P, \tau, X)$.
By Proposition~\ref{prop:Lpsi}(1), $\sigma\circ i_X$
is a Nica-covariant Toeplitz representation of $X$,
so we can take $\theta$ to be the restriction of $\sigma$ to ${\mathcal T}_{\cv}(X)$.
Suppose $\psi$ is a (nondegenerate) Nica-covariant Toeplitz representation of $X$.
Proposition~\ref{prop:Lpsi}(2) gives us a covariant representation
$(L^\psi,\psi)$ of $(B_P,P,\tau,X)$,
and hence a representation $L^\psi\times\psi$ of $B_P \textstyle{\rtimes_{\tau, X}} P$ such that
$(L^\psi\times\psi)\circ i_X=\psi$. Restricting
$L^\psi\times\psi$ to ${\mathcal T}_{\cv}(X)$ gives the required representation $\psi_*$.
Uniqueness of $({\mathcal T}_{\cv}(X),i_X)$ follows by the usual argument.
Suppose $X$ is compactly aligned. Since $i_X$ is the composition of the Nica-covariant
Toeplitz representation $\sigma\circ i_X$ and the homomorphism $\sigma^{-1}$
(restricted to $\sigma({\mathcal T}_{\cv}(X))$),
$i_X$ is Nica covariant by Proposition~\ref{prop:preserve covariance}.
Let $w\in X$, and express $w = z\cdot a$
for some $z\in X$ and $a\in A$.
Then $i_X(w) = i_X(z)i_X(a^*)^*$, so
${\mathcal A} := \clsp\{i_X(x)i_X(y)^*: x,y\in X\}$
contains $i_X(X)$.
Obviously ${\mathcal A}$ is a closed self-adjoint subspace of ${\mathcal T}_{\cv}(X)$,
and since $X$ is compactly aligned, Proposition~\ref{prop:yz} shows that
${\mathcal A}$ is closed under multiplication. This gives \eqref{eq:span toeplitz}.
Now let ${\mathcal B} := \clsp\{i_X(x)i_{B_P}(1_s)i_X(y)^*: x,y\in X, s\in P\}$.
Using Lemma~\ref{lemma:handy} with $\psi := \sigma\circ i_X$,
and then applying $\sigma^{-1}$, gives
\begin{equation}\label{eq:handy1}
i_X(x)i_{B_P}(1_s) = i_{B_P}(1_{p(x)s})i_X(x)
\end{equation}
and
\begin{equation}\label{eq:handy2}
i_{B_P}(1_s)i_X(x) =
\begin{cases} i_X(x)i_{B_P}(1_{p(x)^{-1}(p(x)\vee s)}) & \text{if $p(x)\vee s < \infty$} \\
0 & \text{otherwise.} \end{cases}
\end{equation}
Equation \eqref{eq:handy1} shows that
\[
i_X(x)i_{B_P}(1_s)i_X(y)^*
= i_{B_P}(1_{p(x)s})i_X(x)(i_{B_P}(1_{p(y)s})i_X(y))^*
\inB_P \textstyle{\rtimes_{\tau, X}} P,
\]
so ${\mathcal B}\subseteqB_P \textstyle{\rtimes_{\tau, X}} P$.
Since $B_P$ is generated by $\{1_s: s\in P\}$,
elements of the form $i_{B_P}(1_s)i_X(w)$ generate $B_P \textstyle{\rtimes_{\tau, X}} P$ as a $C^*$\nobreakdash- al\-ge\-bra;
with $w = z\cdot a$ as above, \eqref{eq:handy2} shows that
\begin{align*}
i_{B_P}(1_s)i_X(w)
& = i_{B_P}(1_s)i_X(z)i_X(a^*)^* \\
& = \begin{cases} i_X(z)i_{B_P}(1_{p(z)^{-1}(p(z)\vee s)})i_X(a^*)^*
& \text{if $p(z)\vee s < \infty$} \\
0 & \text{otherwise} \end{cases} \\
& \in{\mathcal B}.
\end{align*}
Hence to establish \eqref{eq:span bpp}, it remains only to show that ${\mathcal B}$ is closed
under multiplication. But Proposition~\ref{prop:yz} shows that the product
\[
i_X(x)i_{B_P}(1_s)i_X(y)^*i_X(z)i_{B_P}(1_t)i_X(w)^*
\]
of two typical generators of ${\mathcal B}$ is contained in the closed linear span of
elements of the form
\[
i_X(x)i_{B_P}(1_s)i_X(f)i_X(g)^*i_{B_P}(1_t)i_X(w)^*,
\]
which by \eqref{eq:handy2} simplifies to
\[
i_X(xf)i_{B_P}(1_{p(f)^{-1}(p(f)\vee s) \vee p(g)^{-1}(p(g)\vee t)})i_X(wg)^* \in{\mathcal B}.
\]
Suppose the left action of $A$ on each $X_s$ is by compact operators;
that is, $\phi_s(A) \subseteq{\mathcal K}(X_s)$ for all $s\in P$.
Let $x\in X$ and $s\in P$. Since $X_{p(x)}$ is essential,
we can express $x = \phi_{p(x)}(a)z$ for some $a\in A$ and $z\in X_{p(x)}$.
With $\psi := \sigma \circ i_X$, we then have
\begin{align*}
\sigma(i_{B_P}(1_s)i_X(x))
& = L^\psi(1_s)\psi(x)
= \rho^\psi_s(1)\psi_e(a)\psi(z) & \\
& = \psi^{(s)}(\phi_s(a))\psi(z) & \text{(Lemma~\ref{lemma:rho}(3))} \\
& = \sigma(i_X^{(s)}(\phi_s(a))i_X(z)), &
\end{align*}
so $i_{B_P}(1_s)i_X(x) = i_X^{(s)}(\phi_s(a))i_X(z) \in {\mathcal T}_{\cv}(X)$.
Since elements of the form $i_{B_P}(1_s)i_X(x)$ generate $B_P \textstyle{\rtimes_{\tau, X}} P$,
this gives $B_P \textstyle{\rtimes_{\tau, X}} P = {\mathcal T}_{\cv}(X)$.
If in addition every $s,t\in P$ have a common upper bound,
then by Proposition~\ref{prop:CP Nica} the universal map
$j_X:X\to{\mathcal O}_X$ is Nica covariant; the integrated form
$(j_X)_*:{\mathcal T}_{\cv}(X)\to{\mathcal O}_X$ is surjective
since it maps generators to generators.
\end{proof}
\section{Faithful representations}
\label{section:faithful}
Our strategy for characterising faithful representations of $B_P \textstyle{\rtimes_{\tau, X}} P$
follows \cite[Section~5]{fowrae}.
First we use the dual coaction $\delta$ of $G$
on $B_P \textstyle{\rtimes_{\tau, X}} P$ and the canonical trace $\rho$ on $C^*(G)$
to define a positive linear map $E_\delta := (\id\otimes\rho)\circ\delta$
of norm one of $B_P \textstyle{\rtimes_{\tau, X}} P$ onto the fixed-point algebra $(B_P \textstyle{\rtimes_{\tau, X}} P)^\delta$.
When $X$ is compactly aligned,
$(B_P,P,\tau,X)$ satisfies the spanning condition
\eqref{eq:span bpp}, and $E_\delta$ is determined by
\begin{equation}\label{eq:Edelta}
E_\delta(i_X(x)i_{B_P}(1_s)i_X(y)^*) = \begin{cases}
i_X(x)i_{B_P}(1_s)i_X(y)^* & \text{if $p(x) = p(y)$} \\
0 & \text{otherwise.}
\end{cases}
\end{equation}
\begin{definition}\label{defn:amenable}
The system $(B_P,P,\tau,X)$ is {\em amenable\/} if $E_\delta$
is faithful on positive elements.
\end{definition}
The argument of \cite[Lemma~6.5]{lacarae} shows that if
$G$ is an amenable group, then the system $(B_P,P,\tau,X)$ is amenable.
In Corollary~\ref{cor:amenability} we will show that
$(B_P,P,\tau,X)$ is also amenable when $X$ is compactly aligned
and $G$ is a free product $*(G^\lambda,P^\lambda)$ with each $G^\lambda$
an amenable group.
\begin{theorem}\label{theorem:faithfulness of representations}
Suppose $(G,P)$ is a quasi-lattice ordered group
and $X$ is a compactly-aligned product system over $P$
of essential Hilbert $A$--$A$ bimodules
such that the system $(B_P,P,\tau,X)$ is amenable.
Let $\psi$ be a Nica-covariant Toeplitz representation of $X$
on a Hilbert space ${\mathcal H}$.
Then $L^\psi\times\psi$ is a faithful representation of $B_P \textstyle{\rtimes_{\tau, X}} P$
if and only if
\begin{multline}\label{eq:psifaithful}
\text{for every $n\ge 1$ and $s_1, \dots, s_n\in P\setminus\{e\}$,
the subrepresentation} \\
a\in A\mapsto\psi_e(a)\prod_{k=1}^n \bigl(1 - L^\psi(1_{s_k})\bigr)
\text{ of $\psi_e$ is faithful.}
\end{multline}
\end{theorem}
\begin{proof}[Proof of necessity of \eqref{eq:psifaithful}]
Let $\pi:A\to B({\mathcal H})$ be a faithful nondegenerate representation of $A$
on a Hilbert space ${\mathcal H}$,
let $l:X\to{\mathcal L}(F(X))$ be the Fock representation of $X$,
and let $\Psi := F(X)\dashind_A^{{\mathcal L}(X)}\pi\circ l$;
by Lemma~\ref{lemma:fock covariant}, $\Psi$ is a Nica-covariant
Toeplitz representation of $X$ on $F(X)\otimes_A{\mathcal H}$.
We claim that
\[
a\in A \mapsto \Psi_e(a)\prod_{k=1}^n \bigl(1 - L^\Psi(1_{s_k})\bigr)
\]
is faithful.
Since $L^\Psi(1_{s_k}) = \alpha^\Psi_{s_k}(1)$ is the orthogonal
projection of $F(X)\otimes_A{\mathcal H}$ onto $\bigoplus_{t\in s_kP} X_t\otimes_A{\mathcal H}$
(see the proof of Lemma~\ref{lemma:fock covariant}),
each projection $1 - L^\Psi(1_{s_k})$ dominates the projection $Q_e$ onto
the $\Psi_e$-invariant subspace $X_e \otimes_A{\mathcal H}$.
To establish the claim it thus suffices to show that
the subrepresentation $Q_e\Psi_e$ of $\Psi_e$ is faithful.
But $\Psi_e = F(X)\dashind_A^A\pi$ decomposes as
$\bigoplus_{t\in P} X_t\dashind_A^A\pi$,
so $Q_e\Psi_e = A\dashind_A^A\pi$ is unitarily equivalent to $\pi$,
and hence faithful.
Now suppose that $L^\psi\times\psi$ is faithful and $a\in A$.
Let
\[
T := i_{B_P}\Bigl( \prod_{k=1}^n (1 - 1_{s_k}) \Bigr) i_X(a) \in B_P \textstyle{\rtimes_{\tau, X}} P.
\]
Then
\begin{multline*}
\norm a
= \norm{\Psi_e(a)\prod_{k=1}^n \bigl(1 - L^\Psi(1_{s_k})\bigr)}
= \norm{L^\Psi\times \Psi(T)}
\le \norm T \\
= \norm{L^\psi\times\psi(T)}
= \norm{\psi_e(a)\prod_{k=1}^n \bigl(1 - L^\psi(1_{s_k})\bigr)}
\le \norm a,
\end{multline*}
giving \eqref{eq:psifaithful}.
\end{proof}
Our proof that \eqref{eq:psifaithful} implies faithfulness
of $L^\psi\times\psi$ is based on the argument of \cite[Section~6]{fowrae}:
in Proposition~\ref{prop:fpa and EPsi}(1)
we prove that $L^\psi\times\psi$ is faithful on $(B_P \textstyle{\rtimes_{\tau, X}} P)^\delta$,
and in Proposition~\ref{prop:fpa and EPsi}(2)
we construct a spatial version $E_\psi$ of $E_\delta$ such that
$(L^\psi\times\psi)\circ E_\delta = E_\psi\circ(L^\psi\times\psi)$.
Faithfulness of $L^\psi\times\psi$ then follows easily:
if $L^\psi\times\psi(b) = 0$, then
\[
0 = E_\psi\circ(L^\psi\times\psi)(b^*b) = (L^\psi\times\psi)\circ E_\delta(b^*b),
\]
so by Proposition~\ref{prop:fpa and EPsi}(1), $E_\delta(b^*b) = 0$.
The amenability hypothesis then forces $b^*b = 0$, and hence $b=0$.
We begin by reviewing some notation and results from \cite[Remark~1.5]{lacarae}
and \cite[Remark~5.2]{fowrae}.
Let $F$ be a finite subset of $P$. A subset $C$ of $F$ is an
{\em initial segment\/} of $F$ if $c := \sigma C$ is finite
and $C = \{t\in F: t\le c\}$.
(Recall that $\sigma C$ is the least upper bound of $C$;
we use the convention that $\sigma\emptyset = e$.)
For each such $C$ there is a nonzero projection $Q_C$ in $B_P$ defined by
\[
Q_C := 1_c \prod_{\{t\in F: c < t\vee c < \infty\}} (1 - 1_t),
\]
and as $C$ ranges over the initial segements of $F$, these projections
form a decomposition of the identity in $B_P$.
\begin{lemma}\label{lemma:QC}
Suppose $(G,P)$ is a quasi-lattice ordered group,
$X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules,
$\psi$ is a Nica-covariant Toeplitz representation of $X$ on ${\mathcal H}$,
$F$ is a finite subset of $P$,
$C$ is an initial segment of $F$,
$x,y\in X$ and $s\in P$.
Let $c = \sigma C$, so that $C = \{t \in F: t \le c\}$.
\textup{(1)} If $p(x) = p(y)$,
then the operator $\psi(x)L^\psi(1_s)\psi(y)^*$
is in the commutant of $L^\psi(B_P)$.
In particular, it commutes with $L^\psi(Q_C)$.
\textup{(2)} If $p(x)s$, $p(y)s\in F$, then
\begin{multline*}
L^\psi(Q_C)\psi(x)L^\psi(1_s)\psi(y)^*L^\psi(Q_C) \\
= \left\{ \begin{array}{l}
L^\psi(Q_C)
\psi(x)L^\psi(1_{p(x)^{-1}c})L^\psi(1_{p(y)^{-1}c})\psi(y)^*
L^\psi(Q_C) \\
\phantom{0} \qquad\qquad\qquad\qquad \text{if $p(x)s\le c$ and $p(y)s\le c$} \\
0 \qquad \text{otherwise.}
\end{array}
\right.
\end{multline*}
\end{lemma}
\begin{proof} The proof, based on Lemma~\ref{lemma:handy},
is identical in form to the proof of \cite[Lemma~5.3]{fowrae}.
\end{proof}
\begin{lemma}\label{lemma:compute norm}
Suppose $(G,P)$ is a quasi-lattice ordered group,
$X$ is a product system over $P$ of essential Hilbert $A$--$A$ bimodules,
and $\psi$ is a Nica-covariant Toeplitz representation of $X$ which satisfies
\eqref{eq:psifaithful}.
Suppose further that $F$ be a finite subset of $P$
and $Z$ is a finite sum $\sum \psi(x_k)L^\psi(1_{s_k})\psi(y_k)^*$
such that $p(x_k)s_k = p(y_k)s_k \in F$ for each $k$.
Then
\begin{equation}\label{eq:normTC}
\norm Z = \max\{\norm{T_C}: \text{$C$ is an initial segment of $F$}\},
\end{equation}
where $T_C$ is the adjointable operator on $X_{\sigma C}$ defined by
\begin{equation}\label{eq:TC}
T_C := \sum_{p(x_k)s_k \le \sigma C}
\Theta_{x_k,y_k} \otimes_A 1^{p(x_k)^{-1}\sigma C}
\end{equation}
\end{lemma}
\begin{proof}
Since $\{Q_C: \text{$C$ is an initial segment of $F$}\}$
is a decomposition of the identity in $B_P$,
and since $L^\psi$ is a unital representation of $B_P$,
the projections $L^\psi(Q_C)$ decompose the identity operator.
By Lemma~\ref{lemma:QC}(1), $Z$ commutes with each $L^\psi(Q_C)$, and thus
\[
\norm{Z} = \max\{\norm{L^\psi(Q_C)Z}:
\text{$C$ is an initial segment of $F$.}\}
\]
Fix an initial segment $C$, and let $c:=\sigma C$.
By Lemma~\ref{lemma:QC}(2) and Lemma~\ref{lemma:rho}(6),
\begin{align*}
L^\psi(Q_C)Z
& = L^\psi(Q_C) \sum \psi(x_k)L^\psi(1_{s_k})\psi(y_k)^* \\
& = L^\psi(Q_C) \sum_{p(x_k)s_k\le c}
\psi(x_k)L^\psi(1_{p(x_k)^{-1}c})\psi(y_k)^* \\
& = L^\psi(Q_C) \sum_{p(x_k)s_k\le c}
\rho^\psi_c ( \Theta_{x_k,y_k} \otimes_A 1) \\
& = L^\psi(Q_C) \rho^\psi_c(T_C),
\end{align*}
so it suffices to show that
\begin{equation}\label{eq:compute norms2}
\norm{L^\psi(Q_C) \rho^\psi_c(T_C)} = \norm{T_C}.
\end{equation}
Let
\begin{equation}\label{eq:RC}
R_C := \prod_{\{t\in F: c < t \vee c < \infty\}}
( 1 - 1_{c^{-1}(t \vee c)}) \in B_P.
\end{equation}
Since $\psi$ satisfies \eqref{eq:psifaithful},
\[
a \mapsto \psi_e(a)\prod_{\{t\in F: c < t \vee c < \infty\}}
( 1 - L^\psi(1_{c^{-1}(t \vee c)}) )
= \psi_e(a)L^\psi(R_C)
\]
is a faithful representation of $A$.
By Lemma~\ref{lemma:rho}(4), the representation
$T\in{\mathcal L}(X_c) \mapsto \alpha^\psi_c(L^\psi(R_C))\rho^\psi_c(T)$
is thus also faithful.
But $\alpha^\psi_c(L^\psi(R_C)) = L^\psi(\tau_c(R_C)) = L^\psi(Q_C)$,
and hence \eqref{eq:compute norms2} is satisfied.
\end{proof}
\begin{prop}\label{prop:fpa and EPsi}
Suppose $(G,P)$ is a quasi-lattice ordered group,
$X$ is a compactly-aligned product system over $P$
of essential Hilbert $A$--$A$ bimodules,
and $\psi$ is a Nica-covariant Toeplitz representation of $X$ which satisfies
\eqref{eq:psifaithful}.
\textup{(1)} $L^\psi\times\psi$ is isometric on $(B_P \textstyle{\rtimes_{\tau, X}} P)^\delta$.
\textup{(2)} There is a linear map $E_\psi$ of norm one of
$L^\psi\times\psi(B_P \textstyle{\rtimes_{\tau, X}} P)$ onto $L^\psi\times\psi\bigl((B_P \textstyle{\rtimes_{\tau, X}} P)^\delta\bigr)$
such that $E_\psi \circ (L^\psi\times\psi) = (L^\psi\times\psi) \circ E_\delta$.
\end{prop}
\begin{proof}
(1) Since $X$ is compactly aligned, the spanning condition
\eqref{eq:span bpp} holds. Since $E_\delta$ is continuous
and maps onto $(B_P \textstyle{\rtimes_{\tau, X}} P)^\delta$,
we deduce that finite sums
\[
z := \sum i_X(x_k)i_{B_P}(1_{s_k})i_X(y_k)^*
\]
in which $p(x_k) = p(y_k)$ for all $k$ are dense in $(B_P \textstyle{\rtimes_{\tau, X}} P)^\delta$.
It therefore suffices to fix such a $z$
and show that $\norm{L^\psi\times\psi(z)} = \norm z$.
Let $\sigma$ be a faithful nondegenerate representation
of $B_P \textstyle{\rtimes_{\tau, X}} P$ such that $(\overline\sigma\circ i_{B_P},\sigma\circ i_X)$
is a covariant representation of $(B_P,P,\tau,X)$.
By Proposition~\ref{prop:Lpsi}, $i := \sigma\circ i_X$ is a covariant
representation of $X$ and $\overline\sigma\circ i_{B_P} = L^i$.
Since $L^i\times i = \sigma$ is faithful, $i$ satisfies \eqref{eq:psifaithful}.
Hence with $F := \{p(x_k)s_k\}$,
Lemma~\ref{lemma:compute norm} gives
\begin{align*}
\norm{L^\psi\times\psi(z)}
& = \norm{\sum \psi(x_k)L^\psi(1_{s_k})\psi(y_k)^*} \\
& = \max\{\norm{T_C}: \text{ $C$ is an initial segment of $F$}\} \\
& = \norm{\sum i(x_k)L^i(1_{s_k})i(y_k)^*}
= \norm{L^i\times i(z)}
= \norm z.
\end{align*}
(2) Since $X$ is compactly aligned,
finite sums of the form
\[
w := \sum i_X(x_k)i_{B_P}(1_{s_k})i_X(y_k)^*
\]
are dense in $B_P \textstyle{\rtimes_{\tau, X}} P$.
We will show that
$\norm{L^\psi\times\psi(E_\delta(w))} \le \norm{L^\psi\times\psi(w)}$;
it follows that $E_\psi$
is well-defined on operators of the form $L^\psi\times\psi(w)$
and extends to the desired linear contraction.
Let $F := \{p(x_k)s_k\} \cup \{p(y_k)s_k\}$,
and let $Z := L^\psi\times\psi(E_\delta(w))$;
by \eqref{eq:Edelta},
\[
Z = \sum_{p(x_k) = p(y_k)} \psi(x_k)L^\psi(1_{s_k})\psi(y_k)^*.
\]
By Lemma~\ref{lemma:compute norm}, there is an initial
segment $C$ of $F$ such that
$\norm Z = \norm{T_C}$.
Let $c := \sigma C$. We will construct a projection $R\in B_P$ such that
$a\in A \mapsto \psi_e(a)L^\psi(R)$ is faithful,
then define $Q := L^\psi(\tau_c(R)) = \alpha^\psi_c(L^\psi(R))$,
and show that $Q(L^\psi\times\psi(w))Q = Q\rho^\psi_c(T_C)$.
This will complete
the proof, since by Lemma~\ref{lemma:rho}(4) we then have
\[
\norm Z = \norm{T_C} = \norm{Q\rho^\psi_c(T_C)} =
\norm{Q(L^\psi\times\psi(w))Q} \le \norm{L^\psi\times\psi(w)}.
\]
For each $s,t\in C$ such that $s \ne t$ and
$s^{-1}c \vee t^{-1}c < \infty$,
define $d_{s,t}\in P$ as in \cite[Lemma~3.2]{lacarae}:
\[
d_{s,t} =
\begin{cases}
(s^{-1}c)^{-1}(s^{-1}c \vee t^{-1}c)
& \text{if $s^{-1}c < s^{-1}c \vee t^{-1}c$} \\
(t^{-1}c)^{-1}(s^{-1}c \vee t^{-1}c)
& \text{otherwise,} \\
\end{cases}
\]
noting in particular that $d_{s,t}$ is never the identity in $P$.
Let $R_C$ be as in \eqref{eq:RC}, and define
\[
R := R_C \prod_{\substack{s\ne t\in C\\ s^{-1}c \vee t^{-1}c < \infty}}
( 1 - 1_{d_{s,t}}).
\]
By condition \eqref{eq:psifaithful},
$a\in A \mapsto L^\psi(R)\psi_e(a)$ is faithful.
The proof that $Q(L^\psi\times\psi(w))Q = Q\rho^\psi_c(T_C)$ is exactly as in
\cite[Proposition~5.5]{fowrae}, so we omit it.
\end{proof}
\begin{prop}\label{prop:EPsi faithful}
Suppose $(G,P)$ is a quasi-lattice ordered group
and $X$ is a compactly-aligned product system over $P$
of essential Hilbert $A$--$A$ bimodules.
Let $\pi$ be a nondegenerate representation of $A$ on a Hilbert space ${\mathcal H}$,
and let $\Psi$ be the representation $F(X)\dashind_A^{{\mathcal L}(X)}\pi\circ l$,
where $l:X\to{\mathcal L}(F(X))$ is the Fock representation of $X$.
There is a projection $E_\Psi$ of norm one of $L^\Psi\times \Psi(B_P \textstyle{\rtimes_{\tau, X}} P)$ onto
$L^\Psi\times \Psi((B_P \textstyle{\rtimes_{\tau, X}} P)^\delta)$ such that
\begin{equation}\label{eq:EPsi}
E_\Psi \circ (L^\Psi\times\Psi) = (L^\Psi\times\Psi) \circ E_\delta;
\end{equation}
moreover, $E_\Psi$ is faithful on positive operators.
\end{prop}
\begin{proof} Denote by $Q_t$ the orthogonal projection
of $F(X)\otimes_A{\mathcal H}$ onto $X_t\otimes_A{\mathcal H}$.
Since the $Q_t$'s are mutually orthogonal, the formula
\[
E_\Psi(T) := \sum_{t\in P} Q_t T Q_t
\qquad\text{for $T\in L^\Psi\times \Psi(B_P \textstyle{\rtimes_{\tau, X}} P)$}
\]
defines a completely positive projection of norm one
which is faithful on positive operators.
We claim that
\begin{equation}\label{eq:EPsi2}
E_\Psi(\Psi(x)L^\Psi(1_s)\Psi(y)^*)
= \begin{cases}
\Psi(x)L^\Psi(1_s)\Psi(y)^* & \text{if $p(x) = p(y)$} \\
0 & \text{otherwise.}
\end{cases}
\end{equation}
Since $X$ is compactly aligned the spanning condition
\eqref{eq:span bpp} holds, and hence \eqref{eq:EPsi}
follows from \eqref{eq:EPsi2} and \eqref{eq:Edelta}.
Suppose $x,y\in X$ and $s\in P$.
For each $t \in P$, $\Psi(x)L^\Psi(1_s)\Psi(y)^*$ is zero on $X_t\otimes_A{\mathcal H}$
unless $p(y)s \le t$, in which case $\Psi(x)L^\Psi(1_s)\Psi(y)^*$ maps
$X_t\otimes_A{\mathcal H}$ into $X_{p(x)p(y)^{-1}t}\otimes_A{\mathcal H}$.
Thus if $p(x) \ne p(y)$, $Q_t\Psi(x)L^\Psi(1_s)\Psi(y)^*Q_t=0$ for every $t\in P$,
and $E_\Psi(\Psi(x)L^\Psi(1_s)\Psi(y)^*) = 0$.
If on the other hand $p(x) = p(y)$, then
$Q_t\Psi(x)L^\Psi(1_s)\Psi(y)^*Q_t = \Psi(x)L^\Psi(1_s)\Psi(y)^*Q_t$
for each $t\in P$, and thus
\begin{multline*}
E_\Psi(\Psi(x)L^\Psi(1_s)\Psi(y)^*)
= \sum_{t\in P} Q_t\Psi(x)L^\Psi(1_s)\Psi(y)^*Q_t \\
= \Psi(x)L^\Psi(1_s)\Psi(y)^*\sum_{t\in P} Q_t
= \Psi(x)L^\Psi(1_s)\Psi(y)^*.
\end{multline*}
\end{proof}
\begin{cor}\label{cor:EPsi faithful}
Suppose $\pi$ is faithful. Then the system $(B_P,P,\tau,X)$ is amenable
if and only if the representation $L^\Psi\times\Psi$ of $B_P \textstyle{\rtimes_{\tau, X}} P$ is faithful.
\end{cor}
\begin{proof} Suppose $L^\Psi\times\Psi$ is faithful.
By Proposition~\ref{prop:EPsi faithful},
$(L^\Psi\times\Psi) \circ E_\delta = E_\Psi \circ (L^\Psi\times\Psi)$ is faithful
on positive elements, hence so is $E_\delta$;
that is, $(B_P,P,\tau,X)$ is amenable.
Since $\Psi$ satisfies \eqref{eq:psifaithful}
(see the proof of necessity of \eqref{eq:psifaithful}),
the converse follows from Theorem~\ref{theorem:faithfulness of representations}.
\end{proof}
\section{Amenability}\label{section:amenability}
\begin{theorem}\label{theorem:amenability}
Suppose $\theta: (G, P) \to ({\mathcal G}, {\mathcal P})$ is a homomorphism of quasi-lattice
ordered groups such that, whenever $s\vee t<\infty$,
\begin{equation}\label{eq:theta}
\theta(s \vee t)= \theta(s) \vee \theta(t)\ \text{ and }\
\theta(s)= \theta(t)\Longrightarrow s = t,
\end{equation}
and suppose that ${\mathcal G}$ is amenable.
If $X$ is a compactly-aligned product system over $P$
of essential Hilbert $A$--$A$ bimodules,
then the system $(B_P,P,\tau,X)$ is amenable.
\end{theorem}
\begin{proof}
Our proof is essentially that of \cite[Theorem~6.1]{fowrae},
suitably modified to handle Hilbert bimodules.
The homomorphism $\theta:G \to {\mathcal G}$ induces a coaction
$\delta_\theta = (\id \otimes \theta) \circ \delta$ of ${\mathcal G}$ on $B_P \textstyle{\rtimes_{\tau, X}} P$,
and hence a conditional expectation $E_{\delta_\theta}$
of $B_P \textstyle{\rtimes_{\tau, X}} P$ onto the fixed-point algebra $(B_P \textstyle{\rtimes_{\tau, X}} P)^{\delta_\theta}$,
such that
\[
E_{\delta_\theta}(i_X(x)i_{B_P}(1_s)i_X(y)^*) =
\begin{cases}
i_X(x)i_{B_P}(1_s)i_X(y)^*& \text{if $\theta(p(x)) = \theta(p(y))$} \\
0 & \text{otherwise.}
\end{cases}
\]
Since ${\mathcal G}$ is amenable, $E_{\delta_\theta}$ is faithful on positive elements.
Let $l:X\to{\mathcal L}(F(X))$ be the Fock representation of $X$,
let $\pi$ be a faithful nondegenerate representation of $A$
on a Hilbert space ${\mathcal H}$, and let $\Psi := F(X)\dashind_A^{{\mathcal L}(X)}\pi\circ l$.
By Proposition~\ref{prop:EPsi faithful}, for every $b\inB_P \textstyle{\rtimes_{\tau, X}} P$ we have
\[
(L^\Psi\times\Psi)\circ E_\delta(b)
= E_\Psi(L^\Psi\times\Psi(E_{\delta_\theta}(b))).
\]
Since $E_{\delta_\theta}$ and $E_\Psi$ are faithful on positive elements,
to show that $(B_P,P,\tau,X)$ is amenable it suffices to show that
$L^\Psi\times\Psi$ is faithful on $(B_P \textstyle{\rtimes_{\tau, X}} P)^{\delta_\theta}$.
Let $\sigma$ be a faithful representation of
$B_P \textstyle{\rtimes_{\tau, X}} P$ such that $(\overline\sigma \circ i_{B_P}, \sigma \circ i_X)$
is a covariant representation of $(B_P, P, \tau, X)$.
By Proposition~\ref{prop:Lpsi},
$i = \sigma \circ i_X$ is a covariant representation of $X$
and $\overline\sigma \circ i_{B_P} = L^i$.
Observe that $i$ is isometric since, by Lemma~\ref{lemma:fock covariant},
\begin{align*}
\norm x
= \norm{\Psi(x)}
& = \norm{(L^\Psi\times \Psi)\circ i_X(x)} \\
& \le \norm{i_X(x)}
= \norm{\sigma\circ i_X(x)}
= \norm{i(x)}
\le \norm x.
\end{align*}
Let ${\mathcal F}$ be the set of all finite subsets $F$ of ${\mathcal P}$
which are closed under $\vee$ in the sense that $s\vee t\in F$
whenever $s,t\in F$ and $s\vee t < \infty$.
Exactly as in the proof of \cite[Theorem~6.1]{fowrae},
one can use Proposition~\ref{prop:yz} to show that,
for each $F\in{\mathcal F}$,
\[
{\mathcal U}_F := \clsp\{i_X(x)i_{B_P}(1_s)i_X(y)^*: \theta(p(x)s) = \theta(p(y)s) \in F\}
\]
is a $C^*$\nobreakdash- subalgebra of $B_P \textstyle{\rtimes_{\tau, X}} P$.
Applying $\Phi_{\delta_\theta}$ to both sides of \eqref{eq:span bpp} gives
\[
(B_P \textstyle{\rtimes_{\tau, X}} P)^{\delta_\theta}
= \clsp\{i_X(x)i_{B_P}(1_s)i_X(y)^*: \theta(p(x)) = \theta(p(y))\};
\]
since ${\mathcal F}$ is directed under set inclusion
(see the proof of \cite[Lemma~4.1]{lacarae}),
we deduce that
\[
(B_P \textstyle{\rtimes_{\tau, X}} P)^{\delta_\theta} = \overline{\bigcup_{F\in{\mathcal F}} {\mathcal U}_F}.
\]
By \cite[Lemma~1.3]{alnr}, to prove that $L^\Psi\times\Psi$ is faithful
on $(B_P \textstyle{\rtimes_{\tau, X}} P)^{\delta_\theta}$ it is enough to prove it is faithful
on each of the subalgebras ${\mathcal U}_F$.
We shall accomplish this by inducting on $\lvert F \rvert$.
First suppose $F = \{r\}$ for some $r\in{\mathcal P}$.
Let $W_r$ be the Hilbert $A$--$A$ bimodule
$\bigoplus_{t\in\theta^{-1}(r)} X_t$.
We claim that, for each Nica-covariant Toeplitz representation $\psi$ of
$X$ on a Hilbert space ${\mathcal K}$,
there is a linear map $\psi_r:W_r\to B({\mathcal K})$
which satisfies
$\psi_r(\oplus x_t) = \sum \psi_t(x_t)$,
and that $(\psi_r,\psi_e)$ is then a Toeplitz representation of $W_r$.
First observe that if $x,y\in X$ satisfy
$p(x) \ne p(y)$ and $\theta(p(x)) = \theta(p(y)) = r$,
then by \eqref{eq:theta} we have $p(x) \vee p(y) = \infty$,
and hence $\psi(x)^*\psi(y) = 0$.
Now suppose $\oplus x_t$ belongs to the algebraic direct sum
$\bigodot_{t\in\theta^{-1}(r)} X_t$; such vectors are dense in $W_r$.
Then
\begin{align*}
\norm{\sum_t \psi_t(x_t)}^2
& = \norm{\sum_{t,t'} \psi_t(x_t)^*\psi_{t'}(x_{t'})}
= \norm{\sum_t \psi_t(x_t)^*\psi_t(x_t)} \\
& = \norm{\sum_t \psi_e(\langle x_t,x_t \rangle_A)}
\le \norm{\sum_t \langle x_t, x_t \rangle_A} \\
& = \norm{\langle \oplus x_t, \oplus x_t \rangle_A}
= \norm{\oplus x_t}^2,
\end{align*}
ensuring the existence of $\psi_r$.
It is routine to check that
$(\psi_r,\psi_e)$ is a Toeplitz representation of $W_r$.
Write $\alpha^\psi_r$ for the endomorphism of $\psi_e(A)'$
which corresponds to $(\psi_r,\psi_e)$
(Proposition~\ref{prop:alpha}),
and write $\rho^\psi_r$ for the associated representation of ${\mathcal L}(W_r)$
(Lemma~\ref{lemma:rho}).
Suppose $Z$ is a finite sum
$\sum i_X(x_k)i_{B_P}(1_{s_k})i_X(y_k)^*$
such that
$\theta(p(x_k)s_k) = \theta(p(y_k)s_k) = r$
for every $k$;
to prove $L^\Psi\times\Psi$ faithful on ${\mathcal U}_{\{r\}}$
we will show that $\norm{L^\Psi\times\Psi(Z)} = \norm Z$.
For each $k$, let $\Theta_{x_k,y_k}\otimes_A 1^{s_k}$
denote the operator in ${\mathcal L}(W_r)$ which is the image of
\[
\Theta_{x_k,y_k}\in{\mathcal K}(X_{p(y_k)},X_{p(x_k)})
\mapsto \Theta_{x_k,y_k}\otimes_A 1^{s_k}
\in {\mathcal L}(X_{p(y_k)s_k},X_{p(x_k)s_k})
\subset {\mathcal L}(W_r).
\]
Define $T := \sum \Theta_{x_k,y_k}\otimes_A 1^{s_k} \in {\mathcal L}(W_r)$.
It is routine to check that
\[
\rho^\Psi_r(T)
= \sum \Psi(x_k)L^\Psi(1_{s_k})\Psi(y_k)^*
= L^\Psi\times\Psi(Z),
\]
and similarly $\rho^i_r(T) = L^i \times i(Z) = \sigma(Z)$.
Since $\Psi_e$ and $i_e$ are faithful representations of $A$,
the representations $\rho^\Psi_r$ and $\rho^i_r$ are isometric, and thus
\[
\norm{L^\Psi\times\Psi(Z)} = \norm{\rho^\Psi_r(T)} = \norm T
= \norm{\rho^i_r(T)} = \norm{\sigma(Z)} = \norm Z.
\]
For the inductive step,
suppose $F \in {\mathcal F}$ and $L^\Psi\times\Psi$ is faithful on ${\mathcal U}_{F'}$
whenever $F' \in {\mathcal F}$ and $\lvert F' \rvert < \lvert F \rvert$;
we aim to prove that $L^\Psi\times\Psi$ is faithful on ${\mathcal U}_F$.
Since $F$ is finite it has a minimal element;
that is, there exists $r_0 \in F$ such that
$r_0 < r_0 \vee r$ for each $r\in F \setminus\{r_0\}$.
As in the proof of \cite[Theorem~6.1]{fowrae}
we have $L^\Psi\times\Psi({\mathcal U}_{\{r\}})P_{r_0} = \{0\}$ for each
$r \in F \setminus\{r_0\}$,
where $P_{r_0}$ denotes the orthogonal projection of
$F(X) \otimes_A {\mathcal H}$
onto $\bigoplus_{t \in \theta^{-1}(r_0)} X_t\otimes_A{\mathcal H}$.
On the other hand, we have already demonstrated that
$L^\Psi\times\Psi$ maps ${\mathcal U}_{r_0}$ isometrically into
the range of $\rho^\Psi_{r_0}$, and an easy calculation
shows that $P_{r_0} = \alpha^\Psi_{r_0}(Q_e)$,
where $Q_e$ is the orthogonal projection onto $X_e\otimes_A{\mathcal H}$.
Since $a\mapsto\Psi_e(a)Q_e$ is faithful,
by Lemma~\ref{lemma:rho}(4) the representation
$S\in{\mathcal L}(W_{r_0}) \mapsto P_{r_0}\rho^\Psi_{r_0}(S)$ is also faithful.
Hence the map $Y \in {\mathcal U}_{r_0} \mapsto L^\Psi\times\Psi(Y)P_{r_0}$
is faithful.
Now suppose $Y \in {\mathcal U}_F$ and $L^\Psi\times\Psi(Y) = 0$.
We will show that $Y \in {\mathcal U}_{F \setminus \{r_0\}}$,
from which the inductive hypothesis implies that $Y = 0$.
Let $(Y_n)$ be a sequence in
\[
\Span\{i_X(x)i_{B_P}(1_s)i_X(y)^*: \theta(p(x)s) = \theta(p(y)s) \in F\}
\]
which converges in norm to $Y$, and express each $Y_n$ as a sum
$\sum_{r\in F} Y_{n,r}$, where $Y_{n,r} \in {\mathcal U}_{\{r\}}$.
For each $n$,
\[
\norm{L^\Psi\times\Psi(Y_n) P_{r_0}}
= \norm{L^\Psi\times\Psi(Y_{n,r_0}) P_{r_0}}
= \norm{Y_{n,r_0}},
\]
and consequently $Y_{n, r_0} \to 0$.
Thus $Y_n - Y_{n,r_0} \to Y$, which shows that
$Y \in {\mathcal U}_{F \setminus\{r_0\}}$, as claimed.
\end{proof}
\begin{cor}\label{cor:amenability}
Suppose $(G^\lambda, P^\lambda)$ is a quasi-lattice ordered
group with $G^\lambda$ am\-en\-able for each $\lambda$ belonging
to some index set $\Lambda$.
If $X$ is a compactly-aligned product system over $P := * P^\lambda$,
then the system $(B_P,P,\tau,X)$ is amenable.
\end{cor}
\begin{proof} The group $\bigoplus G^\lambda$ is amenable,
and by \cite[Proposition~4.3]{lacarae} the canonical map
$\theta: * G^\lambda \to \bigoplus G^\lambda$ satisfies \eqref{eq:theta}.
\end{proof}
\section{Applications}\label{section:applications}
In Section~\ref{section:crossed products},
we associated with each twisted semigroup dynamical system
$(A,P,\beta,\omega)$ a product system $X = X(A,P,\beta,\omega)$
of essential Hilbert $A$--$A$ bimodules over the opposite semigroup $P^o$
(Lemma~\ref{lemma:crossed product ps}),
and we showed that the Cuntz-Pimsner algebra ${\mathcal O}_X$ is canonically isomorphic
to the crossed product $A\rtimes_{\beta,\omega} P$;
we also showed that ${\mathcal T}_X$ has the structure of a certain
``Toeplitz'' crossed product ${\mathcal T}(A\rtimes_{\beta,\omega} P)$
(Proposition~\ref{prop:crossed product}).
Suppose now that $(G^o,P^o)$ is quasi-lattice ordered;
this is equivalent to $(G,P)$ being quasi-latticed ordered in its
{\em right\/}-invariant partial order
($s\le t \Leftrightarrow ts^{-1} \in P$).
Since the left action of $A$ on each fiber $X_s$ is by compact operators,
$X$ is compactly aligned (Lemma~\ref{prop:CA conditions})
and ${\mathcal T}_{\cv}(X) = B_P \textstyle{\rtimes_{\tau, X}} P$ (Theorem~\ref{theorem:subalgebra}).
Hence we can apply Theorem~\ref{theorem:faithfulness of representations}
to characterize the faithful representations of ${\mathcal T}_{\cv}(X)$.
This is particularly helpful when $(G^o,P^o)$ is a total order
since ${\mathcal T}_{\cv}(X) = {\mathcal T}_X$; more generally,
when every $s,t\in P^o$ have a common upper bound in $P^o$
(i.e. $Ps\cap Pt \ne \emptyset$),
the crossed product $A\rtimes_{\beta,\omega} P = {\mathcal O}_X$ is a quotient
of ${\mathcal T}_{\cv}(X)$ (Theorem~\ref{theorem:subalgebra}).
We begin by showing that ${\mathcal T}_{\cv}(X)$, too, has a crossed product structure:
\begin{definition}
Suppose $P$ is a subsemigroup of a group $G$
and $(G^o,P^o)$ is quasi-lattice ordered.
A {\em Nica-Toeplitz covariant representation\/} of $(A,P,\beta,\omega)$
is a Toeplitz covariant representation $(\pi,V)$
such that
\begin{equation}\label{eq:V covariant}
V_s^*V_sV_t^*V_t =
\begin{cases} V_{s\vee t}^*V_{s\vee t}
& \text{if $s\vee t < \infty$} \\
0 & \text{otherwise,} \end{cases}
\end{equation}
where $s\vee t$ denotes the least upper bound of $s$ and $t$
in the right-invariant partial order on $(G,P)$.
\end{definition}
The following Proposition establishes the existence
of a $C^*$\nobreakdash- al\-ge\-bra which is universal for such pairs $(\pi,V)$,
as in Definition~\ref{defn:omega crossed product}.
We call this algebra the {\em Nica-Toeplitz crossed product\/}
of $(A,P,\beta,\omega)$, and denote it
${\mathcal T}_{\cv}(A\rtimes_{\beta,\omega} P)$.
Let $i_X:X\to{\mathcal T}_{\cv}(X)$ be universal for Nica-covariant
Toeplitz representations of $X$.
Lemma~\ref{lemma:strict convergence} is easily adapted to this setting,
and allows us to define $i_P:P\to M{\mathcal T}_{\cv}(X)$ by
$i_P(s) = \lim i_X(s,\beta_s(a_i))^*$;
here $(a_i)$ is an approximate identity for $A$,
and the convergence is strict.
We also define $i_A:A\to{\mathcal T}_{\cv}(X)$ by $i_A(a) := i_X(e,a)$.
\begin{prop}\label{prop:crossed product2}
$({\mathcal T}_{\cv}(X),i_A, i_P)$
is a Nica-Toeplitz crossed product for $(A,P,\beta,\omega)$.
\end{prop}
\begin{proof}
As in the proof of Proposition~\ref{prop:crossed product},
$i_A$ is nondegenerate.
We verify the obvious analogues of conditions (a), (b), and (c)
in Definition~\ref{defn:omega crossed product}.
For (a), let $\sigma$ be a nondegenerate representation of ${\mathcal T}_{\cv}(X)$
on a Hilbert space ${\mathcal H}$,
let $\pi := \sigma\circ i_A$, and let $V := \overline\sigma\circ i_P$;
we must show that $(\pi,V)$ is a Nica-Toeplitz covariant representation
of $(A,P,\beta,\omega)$.
Exactly as in the proof of Proposition~\ref{prop:crossed product},
$(\pi,V)$ is a Toeplitz covariant representation of $(A,P,\beta,\omega)$,
so we need to establish \eqref{eq:V covariant}.
Fix $s\in P$.
For any $a\in A$ and $h\in{\mathcal H}$ we have
\begin{align*}
V_s^*\pi(a)h
& = \sigma(i_P(s)^*i_A(a))h
= \sigma(\lim i_X(s,\beta_s(a_i))i_X(e,a))h \\
& = \sigma(\lim i_X(s,\beta_s(a_i)a))h
= \sigma \circ i_X(s, \overline{\beta_s}(1)a)h,
\end{align*}
and since $\pi$ is nondegenerate this shows that
\[
V_s^*V_s{\mathcal H}
= \clsp\{\sigma\circ i_X(\xi)h: \xi\in X_s, h\in{\mathcal H}\}
= \alpha^{\sigma\circ i_X}_s(1).
\]
Since $X$ is compactly aligned,
$\sigma\circ i_X$ is Nica covariant
(Theorem~\ref{theorem:subalgebra} and Proposition~\ref{prop:preserve covariance}),
and \eqref{eq:V covariant} follows.
For condition (b), let $(\pi,V)$ be any Nica-Toeplitz covariant representation
on ${\mathcal H}$.
As in the proof of Proposition~\ref{prop:crossed product},
$\psi(s,x) := V_s^*\pi(x)$ defines a nondegenerate Toeplitz covariant representation
$\psi:X\to B({\mathcal H})$. To see that it is Nica-covariant,
let $s\in P$, and note that for any $a\in A$ we have
\begin{align*}
\psi(s,\overline{\beta_s}(1)a)
& = \lim \psi(s,\beta_s(a_i)a)
= \lim V_s^*\pi(\beta_s(a_i)a) \\
& = \lim V_s^*V_s\pi(a_i)V_s^*\pi(a)
= V_s^*\pi(a).
\end{align*}
Since $\pi$ is nondegenerate, this implies that
$\alpha^\psi_s(1) = V_s^*V_s$,
and hence $\psi$ is Nica covariant by \eqref{eq:V covariant}.
Defining $\pi\times V := \psi_*:{\mathcal T}_{\cv}(X)\to B({\mathcal H})$
gives the desired representation satisfying $(\pi\times V)\circ i_A = \pi$
and $\overline{\pi\times V}\circ i_P = V$.
Condition (c) is satisfied since $i_A(a)i_P(s) = i_X(s,\overline{\beta_s}(1)a^*)^*$,
and elements of this form generate ${\mathcal T}_{\cv}(X)$.
\end{proof}
Let $(G_i,P_i)$ be a collection of abelian lattice-ordered groups.
Since $(G_i,P_i)$ is quasi-lattice ordered
in both its left and its right-invariant partial order,
so is the free product $*(G_i, P_i)$.
\begin{theorem}\label{theorem:crossed products}
Suppose $(G,P) = *(G_i, P_i)$ is a free product of abelian lattice-ordered groups
and $(\pi,V)$ is a Nica-Toeplitz covariant representation
of the twisted semigroup dynamical system $(A,P,\beta,\omega)$
on a Hilbert space ${\mathcal H}$.
Then the integrated form $\pi\times V$ is a faithful representation
of ${\mathcal T}_{\cv}(A\rtimes_{\beta,\omega} P)$
if and only if
\begin{multline*}
\text{for every $n\ge 1$ and $s_1, \dots, s_n\in P\setminus\{e\}$,} \\
\text{$\pi$ acts faithfully on the range of
$\prod_{k=1}^n \bigl(1 - V_{s_k}^*V_{s_k})$.}
\end{multline*}
\end{theorem}
\begin{proof} Let $\theta$ be the canonical homomorphism of
$*(G_i,P_i)$ onto $\bigoplus (G_i,P_i)$.
By \cite[Proposition~4.3]{lacarae}, $\theta$ satisfies
the hypotheses of Theorem~\ref{theorem:amenability};
since $X = X(A,P,\beta,\omega)$ is compactly aligned,
the system $(B_P,P,\tau,X)$ is therefore amenable.
Identifying ${\mathcal T}_{\cv}(A\rtimes_{\beta,\omega} P)$
with ${\mathcal T}_{\cv}(X)$ as in the previous Proposition and defining
$\psi(s,x) := V_s^*\pi(x)$, the initial projection $V_s^*V_s$
is precisely $\alpha^\psi_s(1)$,
and the result follows
from Theorem~\ref{theorem:faithfulness of representations}.
\end{proof}
Nica covariance is automatic when $(G,P)$ is totally ordered:
\begin{cor}
Suppose $(G,P)$ is a totally ordered abelian group and $(\pi,V)$ is a Toeplitz
covariant representation of $(A,P,\beta,\omega)$ on a Hilbert space ${\mathcal H}$.
Then the integrated form $\pi\times V$ is a faithful representation
of ${\mathcal T}(A\rtimes_{\beta,\omega} P)$
if and only if
$\pi$ acts faithfully on $(V_s^*{\mathcal H})^\perp$ for every $s\in P\setminus\{e\}$.
\end{cor}
\begin{cor}
Suppose $\beta$ is an extendible endomorphism of $A$.
If $(\pi,V)$ is a Toeplitz representation of $(A,\field{N},\beta)$,
then $\pi\times V$ is a faithful representation of
${\mathcal T}(A\rtimes_\beta\field{N})$ if and only if $\pi$ acts faithfully on $(V^*{\mathcal H})^\perp$.
\end{cor}
\subsection*{Bicovariance}
Suppose $(G,P)$ is a quasi-lattice ordered group.
Following \cite{lacarae}, in \cite{fowrae} it was shown that
$B_P\rtimes_{\tau,\omega} P$ is universal for
isometric $\omega$\nobreakdash- representations of $P$
which are Nica covariant;
that is, which satisfy
\begin{equation}\label{eq:V covariant2}
V_sV_s^*V_tV_t^* =
\begin{cases} V_{s\vee t}V_{s\vee t}^*
& \text{if $s\vee t < \infty$} \\
0 & \text{otherwise.} \end{cases}
\end{equation}
Assuming that $(G^o,P^o)$ is also quasi-lattice ordered,
we now show that the Nica-Toeplitz crossed product
${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$ is universal for
partial isometric $\omega$\nobreakdash- representations of $P$
which are {\em bicovariant\/} in that they satisfy both
\eqref{eq:V covariant2} and \eqref{eq:V covariant}.
Note that bicovariance is automatic when $(G,P)$ is a totally ordered abelian group.
\begin{prop}\label{prop:universal bicovariant}
$i_P:P\to{\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$
is a bicovariant partial isometric $\omega$-represent\-ation of $P$
whose range generates ${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$ as a $C^*$\nobreakdash- al\-ge\-bra.
Moreover, for every bicovariant partial isometric
$\omega$-re\-pre\-senta\-tion $V$,
there is a representation $V_*$ of ${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$
such that $V_*\circ i_P = V$.
\end{prop}
\begin{proof}
Let $\sigma$ be a faithful nondegenerate representation of
${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$.
Then $V := \overline\sigma\circ i_P$ is a partial isometric
$\omega$\nobreakdash- representation of $P$ which satisfies \eqref{eq:V covariant},
and applying $\overline\sigma^{-1}$ we see that $i_P$ is as well.
Since $i_P(s)i_P(s)^* = i_{B_P}(1_s)$ for every $s\in P$,
$i_P$ also satisfies \eqref{eq:V covariant2}, and is hence bicovariant.
Since $\{1_s:s\in P\}$ generates $B_P$ linearly and
$\{i_{B_P}(a)i_P(t): a\in B_P, t\in P\}$ generates
${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$ as a $C^*$\nobreakdash- al\-ge\-bra,
elements of the form
$i_{B_P}(1_s)i_P(t) = i_P(s)i_P(s)^*i_P(t)$
are also generating.
If $V$ is any bicovariant partial isometric $\omega$-representation of $P$,
then by \cite[Proposition~1.3]{lacarae}
there is a representation $\pi_V$ of $B_P$
such that $\pi_V(1_s) = V_sV_s^*$ for every $s\in P$.
For any $s,t\in P$ the product $V_tV_s = \omega(t,s)V_{ts}$
is a partial isometry; hence by \cite[Lemma~2]{hw} the projections
$V_sV_s^*$ and $V_t^*V_t$ commute, and we deduce that
$\pi_V(a)V_t^*V_t = V_t^*V_t\pi_V(a)$ for every $a\in B_P$ and $t\in P$.
Further,
\begin{align*}
\pi_V(\tau_s(1_t))
& = \pi_V(1_{st})
= V_{st}V_{st}^* \\
& = (\overline{\omega(s,t)}V_sV_t)(\overline{\omega(s,t)}V_sV_t)^*
= V_sV_tV_t^*V_s^*
= V_s\pi_V(1_t)V_s^*,
\end{align*}
so $\pi_V(\tau_s(a)) = V_s\pi_V(a)V_s^*$ for every $s\in P$ and $a\in B_P$.
Thus $(\pi_V,V)$ is a Nica-Toeplitz covariant representation of
$(B_P,P,\tau,\omega)$.
The representation $V_* := \pi_V\times V$ satisfies $V_*\circ i_P = V$.
\end{proof}
We say that a bicovariant partial isometric $\omega$\nobreakdash- representation $V$
is {\em universal\/} if,
for every bicovariant partial isometric $\omega$\nobreakdash- representation $W$,
there is a homomorphism of $C^*\{V_s: s\in P\}$
which maps $V_s$ to $W_s$ for each $s\in P$.
\begin{theorem}\label{theorem:bicovariant}
Suppose $(G,P) = *(G_i, P_i)$ is a free product of abelian lattice-ordered groups
and $V$ is a bicovariant partial isometric $\omega$\nobreakdash- representation of $P$.
Then $V$ is universal if and only if
\[
\prod_{l=1}^m (V_rV_r^* - V_{rt_l}V_{rt_l}^*)\prod_{k=1}^n (1 - V_{s_k}^*V_{s_k}) \ne 0
\]
whenever $r\in P$, $m,n \ge 1$, and $s_1$, \dots, $s_n$,
$t_1$, \dots, $t_m\in P\setminus\{e\}$.
\end{theorem}
\begin{proof} $V$ is universal if and only if the representation
$V_* = \pi_V\times V$ of ${\mathcal T}_{\cv}(B_P\rtimes_{\tau,\omega}P)$
is faithful. By Theorem~\ref{theorem:crossed products}, this occurs if and only
if $\pi_V$ acts faithfully on the range of
$\prod_{k=1}^n (1 - V_{s_k}^*V_{s_k})$
whenever $s_1,\dots, s_n\in P\setminus\{e\}$,
and the result follows from \cite[Proposition~1.3]{lacarae}.
\end{proof}
Let $\field{F}_\infty$ be the free group on infinitely many generators
$z_1$, $z_2$, \dots, and let $\field{F}_\infty^+$ be the subsemigroup (with identity)
generated by the $z_i$;
the pair $(\field{F}_\infty, \field{F}_\infty^+)$ is quasi-lattice ordered.
In \cite{lacarae}, Laca and Raeburn realized the Cuntz
algebra ${\mathcal O}_\infty$ as the universal $C^*$\nobreakdash- al\-ge\-bra for covariant isometric
representations of $\field{F}_\infty^+$, and used their characterization
of the faithful representations of $B_P\rtimes_\tau P$
to derive Cuntz's simplicity result.
We finish by showing that
the universal $C^*$\nobreakdash- al\-ge\-bra for bicovariant partial isometric
representations of $\field{F}_\infty^+$ is reminiscent of ${\mathcal O}_\infty$,
and we derive a Cuntz-Krieger-type uniqueness theorem.
First some notation.
For a multi-index $\mu = (\mu_1,\dots, \mu_n)$ we write
$z_\mu := z_{\mu_1}\dotsm z_{\mu_n}$,
and we identify $\field{F}_\infty^+$ with the set of multi-indices
under concatenation via $z_\mu \leftrightarrow \mu$.
\begin{prop}
Suppose $S$ is a partial isometric representation of $\field{F}_\infty^+$
in a $C^*$\nobreakdash- al\-ge\-bra $B$; that is, $S$ is a semigroup homomorphism
and each $S_\mu$ is a partial isometry.
Then $C^*\{S_\mu: \mu\in\field{F}_\infty^+\}$ is generated by
$\{S_n: n \in \field{N}\}$,
and $S$ is bicovariant if and only if
\textup{(a)} the range projections $s_ks_k^*$ for $k\in\field{N}$
are pairwise orthogonal, and
\textup{(b)} the initial projections $s_k^*s_k$ for $k\in\field{N}$
are pairwise orthogonal.
\end{prop}
\begin{proof} The first statement is obvious.
In the left-invariant partial order on $\field{F}_\infty$,
two elements $\mu,\nu\in\field{F}_\infty^+$ have a common upper bound
if and only if one is an initial word of the other, and then
the least upper bound is the longer of the two words.
We will show that (a) holds if and only if
\begin{equation}\label{eq:S covariant}
S_\mu S_\mu^* S_\nu S_\nu^*
= \begin{cases}
S_\mu S_\mu^*
& \text{if $\nu^{-1}\mu \in \field{F}_\infty^+$,} \\
S_\nu S_\nu^*
& \text{if $\mu^{-1}\nu \in \field{F}_\infty^+$,} \\
0 & \text{otherwise;}
\end{cases}
\end{equation}
of course a similar statement holds for (b)
using the right-invariant partial order, and together these prove
the Proposition.
To begin with, \eqref{eq:S covariant} implies (a) since distict generators
of $\field{F}_\infty^+$ are not comparable.
For the converse, first suppose $\nu^{-1}\mu \in \field{F}_\infty^+$;
since $S_\nu$ is a partial isometry, we then have
\[
S_\mu S_\mu^* S_\nu S_\nu^*
= S_\mu S_{\nu^{-1}\mu}^* S_\nu^* S_\nu S_\nu^*
= S_\mu S_{\nu^{-1}\mu}^* S_\nu^*
= S_\mu S_\mu^*.
\]
The case $\mu^{-1}\nu \in \field{F}_\infty^+$ is similar.
Finally, suppose $\mu$ and $\nu$ are not comparable.
Then there exists $\sigma, \mu',\nu'\in\field{F}_\infty^+$
such that $\mu = \sigma\mu'$,
$\nu = \sigma\nu'$, and $\mu'_1 \ne \nu'_1$.
Condition (a) implies that $S_{\mu'}^* S_{\nu'} = 0$,
and by \cite[Lemma~2]{hw} the range projection of $S_{\nu'}$ commutes with the
initial projection of $S_\sigma$, so
\[
S_\mu^* S_\nu
= S_{\mu'}^* S_\sigma^* S_\sigma S_{\nu'}
= S_{\mu'}^* S_\sigma^* S_\sigma S_{\nu'} S_{\nu'}^* S_{\nu'}
= S_{\mu'}^* S_{\nu'} S_{\nu'}^* S_\sigma^* S_\sigma S_{\nu'}
= 0.
\]
\end{proof}
\begin{theorem}\label{theorem:free}
A bicovariant partial isometric representation $S$
of $\field{F}_\infty^+$ is universal if and only if each $S_\mu$ is nonzero.
\end{theorem}
\begin{proof}
Suppose each $S_\mu$ is nonzero.
To see that $S$ is universal, we apply Theorem~\ref{theorem:bicovariant}.
If $\nu\in\field{F}_\infty^+$, $m,n\ge 1$, and
$\sigma_1$, \dots, $\sigma_m$, $\tau_1$, \dots, $\tau_n\in \field{F}_\infty^+\setminus\{e\}$,
then we can choose $i,j\in\field{N}$ such that none of the multi-indices $\sigma_l$
begins with $i$, and none of the multi-indices $\tau_k$ ends with $j$.
Then
\[
\prod_{l=1}^m (S_\nu S_\nu^* - S_{\nu\sigma_l}S_{\nu\sigma_l}^*)
\prod_{k=1}^n (1 - S_{\tau_k}^*S_{\tau_k})
\ge S_\nu S_iS_i^*S_\nu^* S_j^*S_j
= S_j^*S_{j\nu i}S_{j\nu i}^*S_j
\]
is nonzero since
$S_j(S_j^*S_{j\nu i}S_{j\nu i}^*S_j)S_j^* = S_{j\nu i}S_{j\nu i}^* \ne 0$.
Hence $S$ is universal.
Now define
$T:\field{F}_\infty^+\to B(\ell^2(\field{F}_\infty^+)\otimes\ell^2(\field{F}_\infty^+))$
by
\[
T_\mu(\delta_\sigma \otimes \delta_\nu)
= \begin{cases} \delta_\sigma \otimes \delta_{\mu\nu}
& \text{if $\sigma$ ends in $\mu\nu$} \\
0 & \text{otherwise.} \end{cases}
\]
Then $T$ is a bicovariant partial isometric representation
of $\field{F}_\infty^+$ in which each $T_\mu$ is nonzero.
If $S$ is universal, then $S_\mu \mapsto T_\mu$ extends
to a homomorphism of $C^*\{S_\mu\}$, and hence each $S_\mu$
must be nonzero.
\end{proof}
|
\section{Introduction}
\label{introduction}
Spontaneous emission is a prime example of the action of vacuum
fluctuations on physically measurable processes. Since the early work of
Einstein \cite{Einstein} spontaneous emission has been a major
ingredient in the understanding of the effects of what one calls the
vacuum in quantum field theory. The radiation properties of an excited
atom located in free space have been a subject of many studies (for a
comprehensive list of original articles, see, e.g.,
\cite{Milonni94}).
In particular, the rate of spontaneous emission in free space
(half the Einstein coefficient) is given by
\begin{equation}
\label{1.1}
\Gamma_{\rm SE} = \Gamma_0
\equiv \frac{\omega_A^3 \mu^2}{3\pi \hbar \epsilon_0 c^3}\,,
\end{equation}
where $\omega_A$ is the transition frequency of the atom and $\mu$ is
the dipole matrix element of the transition. The question
has been arisen of how a surrounding medium modifies that decay.
Simple arguments based on the change of the mode density suggest
that the spontaneous emission rate inside a non-absorbing
dielectric should be modified according to \cite{Nienhuis76}
\begin{equation}
\label{1.2}
\Gamma_{\rm SE} = n \Gamma_0,
\end{equation}
where $n$ is the real refractive index of the medium.
In Eq.~(\ref{1.2}) it is assumed that the local field the
atom interacts with is identical with the
electromagnetic field in the continuous medium.
Since in reality the atom is in a small region of free space,
the local field felt by the atom is different from the field
in the continuous medium
\cite{Keller97}, and the decay rate may be expected to be modified to
\begin{equation}
\label{1.3}
\Gamma_{\rm SE} = n \xi \Gamma_0,
\end{equation}
where $\xi$ is the local-field correction factor.
Different models have been used to calculate it.
In the (Clausius-Mosotti) virtual cavity model
it is given by \cite{Knoester89}
\begin{equation}
\label{1.4}
\xi_{\rm CM} = \left(\frac{n^2+2}{3}\right)^2,
\end{equation}
whereas the (Glauber-Lewenstein) real cavity model
leads to \cite{Glauber91}
\begin{equation}
\label{1.5}
\xi_{\rm GL} = \left(\frac{3n^2}{2n^2+1}\right)^2.
\end{equation}
Recently, experiments have been reported
\cite{Rikken95,Schuurmans98} from which
the real-cavity model may be favored.
As already mentioned, in Eqs.~(\ref{1.2}) -- (\ref{1.5}) it is
assumed that the refractive index of the medium, which may
vary with frequency [i.e., $n$ $\!\to$ $\!n(\omega_A)$
in Eqs.~(1.2) -- (1.5)], is real.
However, in reality the refractive index must be a complex
function of frequency,
\begin{equation}
\label{1.6}
n(\omega) = \eta(\omega) + i\,\kappa(\omega).
\end{equation}
It is well known that causality requires
the permittivity of the medium, $\epsilon(\omega)$ $\!=$
$\!n^2(\omega)$, to be a complex function of frequency
whose real part (responsible for dispersion) and
imaginary part (responsible for absorption) are
related to each other by the Kramers-Kronig relation.
Only when the atomic transition frequency $\omega_A$ is sufficiently
far from a medium resonance, so that absorption may be
disregarded, the imaginary part of the refractive index
(at the atomic transition frequency) may
be neglected: $\!n(\omega_A)$ $\!\approx$ $\!\eta(\omega_A)$.
Describing the (undisturbed, continuous) medium in terms of a
complex permittivity,
in \cite{Barnett92,Barnett96} it is argued that
Eqs.~(\ref{1.3}) -- (\ref{1.5}) can be extended to the spontaneous
emission of an atom embedded in a lossy dielectric as
\begin{equation}
\label{1.7}
\Gamma_{\rm SE} = \eta(\omega_A) \xi(\omega_A) \Gamma_0 ,
\end{equation}
where the local-field correction factors
(\ref{1.4}) and (\ref{1.5}) are now regarded as being
squares of absolute values,
\begin{equation}
\label{1.8}
\xi_{\rm CM}(\omega_A) = \left|\frac{n^2(\omega_A)+2}{3}\right|^2,
\end{equation}
\begin{equation}
\label{1.9}
\xi_{\rm GL}(\omega_A)
= \left|\frac{3n^2(\omega_A)}{2n^2(\omega_A)+1}\right|^2 .
\end{equation}
Further, in \cite{Barnett96} the total
decay rate is decomposed as
\begin{equation}
\label{1.10}
\Gamma = \Gamma^\perp + \Gamma^\|,
\end{equation}
where the rates $\Gamma^\perp$ and $\Gamma^\|$, respectively,
are related to the transverse and longitudinal electromagnetic
fields in the medium. The rate $\Gamma^\perp$ is identified with
the cavity-radius-independent rate
$\Gamma_{\rm SE}$ given by Eq.~(\ref{1.7}),
and it is argued that the rate $\Gamma^\|$,
which depends on the cavity radius $R$ as $\Gamma^\|$ $\!\sim$ $\!R^{-3}$,
is responsible for nonradiative decay via energy transfer
between the atom and the surrounding (absorbing) dielectric.
>From the study of resonant energy transfer between two guest molecules
in a perfect lattice of absorbing molecules \cite{Juzeliunas94},
in \cite{Juzeliunas97} it is argued that
(within the approximations made) the rate of
spontaneous emission is given
by Eq.~(\ref{1.7}) together with Eq.~(\ref{1.8}), i.e., with the
local-field correction factor that corresponds to the virtual-cavity model.
However, the total decay rate is purely transverse; i.e., it results only
from the transverse part of the electromagnetic field in the medium,
\begin{equation}
\label{1.11}
\Gamma = \Gamma^\perp = \Gamma^{(1)} + \Gamma^{(2)} .
\end{equation}
It consists of an $R$-independent far-field term $\Gamma^{(1)}$,
which has the form of Eq.~(\ref{1.7}) [together
with Eq.~(\ref{1.8})] and is interpreted as the spontaneous emission
rate, and a $R$-dependent term $\Gamma^{(2)}$, which
in the near-field zone is proportional to $R^{-3}$ and describes
nonradiative energy transfer.
Recently it has been shown \cite{Scheel98b} that the decay rates
suggested in \cite{Barnett92,Barnett96} for the virtual-cavity model are
wrong in general, because the quantum vacuum in the presence of a
dispersive and absorbing dielectric is not introduced correctly.
The fluctuating part of the polarization field is not fully included
in the local field coupled to the atom and therefore effects such as
nonradiative energy transfer from the guest atom to the medium via virtual
photon exchange (i.e., transverse-field-assisted energy transfer)
are omitted. It is just the contribution to the local field of the
fluctuating part of the polarization which gives rise to
the relevant terms $\sim$ $\!R^{-3}$ and $\sim$ $\!R^{-1}$ in the transverse
decay rate of an excited atom surrounded by an absorbing medium
\cite{Scheel98b}. It is worth noting that the results have been
confirmed within a microscopic approach to the problem
more recently \cite{fleisch}.
In the virtual-cavity model, the electromagnetic field inside
the cavity, i.e., the local field, is modified by the presence
of the cavity, but the modification of the field outside the cavity
is disregarded. Hence the local field introduced in this way
is not exactly the field that couples to the atom in reality.
On the contrary, in the real-cavity model the
mutual modification of the fields outside and inside the
cavity are taken into account in a consistent way; i.e.,
the atom interacts with a field that exactly satisfies
both Maxwell's equations and the fundamental commutation rules of
quantum electrodynamics. It may be therefore expected that
the real cavity model is more suited for describing the
spontaneous decay than the virtual cavity model.
In particular, the Power-Zienau-Woolley transformation
(see, e.g., \cite{Craig84}) suggests
that (in dipole approximation) only the transverse
electromagnetic field contributes to the decay rate via
radiative decay {\em and} nonradiative decay
associated with virtual photon exchange,
the latter being typical for an absorbing medium.
In this article we consider, within the frame of rigorous
quantization of the electromagnetic field in an arbitrary
linear Kramers-Kronig consistent dielectric
\cite{Gruner96,Dung98,Scheel98a}, the spontaneous decay of
an excited atom embedded in an absorbing dielectric, applying
the real-cavity concept.
We find that the rate formulas suggested in \cite{Barnett96}
for the real-cavity model are essentially wrong. At first, only the
transverse electromagnetic field
contributes to the decay rate, i.e., $\Gamma^\|$ $\!\equiv$ $\!0$,
which contradicts \cite{Barnett96}.
At second, the (purely transverse)
rate not only contains an $R$-independent term but also
terms proportional to $R^{-1}$ and $R^{-3}$
which are closely related to nonradiative
decay -- a result which also contradicts \cite{Barnett96}.
As expected, nonradiative decay is only observed
for an absorbing medium.
It is worth noting that when the atomic transition frequency
is sufficiently far from an absorption band of the medium,
so that absorption may be neglected, our result exactly agrees
with that derived in \cite{Glauber91} for a non-absorbing medium.
The paper is organized as follows. After introducing the
quantization scheme, in Sect.~\ref{scheme} the problem of spontaneous
decay of an exited atom in an absorbing medium is considered.
In Sect.~\ref{CM} the results for decay rate with the virtual cavity model
are outlined, and Sect.~\ref{GL} presents a detailed analysis
of the decay rate with the real cavity model. The results are discussed in
Sect.~\ref{discussion}. Lengthy calculations are given in the Appendix.
\section{Basic equations}
\label{scheme}
Our analysis of the spontaneous decay of an excited atom
embedded in an absorbing medium is based
on the scheme for quantization of the
electromagnetic field in linear Kramers-Kronig dielectrics
developed in \cite{Gruner96,Dung98,Scheel98a}.
We start with the phenomenological Maxwell's equations in
the (temporal) Fourier space, without external sources,
\begin{equation}
\label{2.1}
{\bf \nabla} \cdot \underline{\hat{\bf B}}({\bf r},\omega) = 0,
\end{equation}
\begin{equation}
\label{2.1A}
{\bf \nabla} \cdot \left[ \epsilon_0 \epsilon({\bf r},\omega)
\underline{\hat{\bf E}}({\bf r},\omega)\right] =
\underline{\hat{\rho}}({\bf r},\omega) ,
\end{equation}
\begin{equation}
\label{2.1B}
{\bf \nabla} \times \underline{\hat{\bf E}}({\bf r},\omega) = i\omega
\underline{\hat{\bf B}}({\bf r},\omega) ,
\end{equation}
\begin{equation}
\label{2.1C}
{\bf \nabla} \times \underline{\hat{\bf B}}({\bf r},\omega) =
-i\frac{\omega}{c^2} \epsilon({\bf r},\omega) \underline{\hat{\bf
E}}({\bf r},\omega) +\mu_0 \underline{\hat{\bf j}}({\bf r},\omega),
\end{equation}
where $\epsilon({\bf r},\omega)$ $\!=$ $\!\epsilon_R({\bf r},\omega)$
$\!+$ $\!i\epsilon_I({\bf r},\omega)$ is the (spatially varying)
permittivity satisfying the
Kramers-Kronig relations. When there are no external charges
and currents, then $\underline{\hat{\rho}}({\bf r},\omega)$
and $\underline{\hat{\bf j}}({\bf r},\omega)$, respectively, are the
operator noise charge and current densities that are associated
with absorption according to the dissipation-fluctuation
theorem. They satisfy the equation of continuity,
\begin{eqnarray}
\label{2.1a}
{\bf \nabla} \cdot \underline{\hat{\bf j}}({\bf r},\omega)
= i \omega \underline{\hat{\rho}}({\bf r},\omega),
\end{eqnarray}
and they are related to the noise polarization
$\underline{\hat{\bf P}}^N({\bf r},\omega)$ as
\begin{eqnarray}
\label{2.2}
\underline{\hat{\bf j}}({\bf r},\omega) &=& -i\omega
\underline{\hat{\bf P}}^N({\bf r},\omega),
\\
\label{2.2a}
\underline{\hat{\rho}}({\bf r},\omega) &=& -{\bf \nabla} \cdot
\underline{\hat{\bf P}}^N({\bf r},\omega) .
\end{eqnarray}
Let $\hat{\bf f}({\bf r},\omega)$ be an infinite set
of bosonic field operators which may be viewed as being collective
excitations of the electromagnetic field, the medium polarization,
and the reservoir. All operators in the theory can then be
expressed in terms of these basic field operators using
the relation
\begin{equation}
\label{2.3}
\underline{\hat{\bf j}}({\bf r},\omega) = \omega \sqrt{\frac{\hbar
\epsilon_0}{\pi} \epsilon_I({\bf r},\omega)}
\,\hat{\bf f}({\bf r},\omega) .
\end{equation}
In particular, from Maxwell's equations the electric field
(in Fourier space)
is given by a convolution with the classical dyadic Green function,
\begin{equation}
\label{2.5}
\underline{\hat{E}}_k({\bf r},\omega) = i\mu_0 \int d^3{\bf r}'
\,\omega G_{kk'}({\bf r},{\bf r}',\omega)
\underline{\hat{j}}_{k'}({\bf r}',\omega) ,
\end{equation}
where $G_{kk'}({\bf r},{\bf r}',\omega)$ satisfies the partial
differential equation
\begin{equation}
\label{2.4}
\left[ \partial_i^r \partial_k^r - \delta_{ik} \left( \triangle^r
+ \frac{\omega^2}{c^2} \epsilon({\bf r},\omega) \right) \right]
G_{kk'}({\bf r},{\bf r'},\omega)
= \delta_{ik'} \delta({\bf r}-{\bf r'}) .
\end{equation}
Integration with respect to $\omega$
then yields the operator of the electric field as
\begin{equation}
\label{2.6}
\hat{\bf E}({\bf r}) = \hat{\bf E}^{(+)}({\bf r})
+ \hat{\bf E}^{(-)}({\bf r}),
\quad
\hat{\bf E}^{(-)}({\bf r})
= \left[\hat{\bf E}^{(+)}({\bf r})\right]^\dagger,
\end{equation}
\begin{equation}
\label{2.7}
\hat{E}_k^{(+)}({\bf r}) =
\int_{0}^{\infty} d\omega\,\underline{\hat{E}}_k({\bf r},\omega)
= i\mu_0 \int_{0}^{\infty} d\omega \int d^3{\bf r}'
\,\omega G_{kk'}({\bf r},{\bf r}',\omega)
\underline{\,\hat{\!j}}_{k'}({\bf r}',\omega) .
\end{equation}
Substituting in Eq.~(\ref{2.7}) for the
current density the expression given in Eq.~(\ref{2.3}) yields
the electric field in terms of the bosonic basic
fields. It can be proven \cite{Scheel98a} that the quantization scheme
is fully consistent with QED for arbitrary linear dielectrics, i.e.,
\begin{equation}
\label{2.7a}
\epsilon_0 \left[ \hat{E}_k({\bf r}) , \hat{B}_l({\bf r'}) \right]
= -i\hbar \epsilon_{klm} \partial_m^r \delta({\bf r}-{\bf r'}) ,
\end{equation}
\begin{equation}
\label{2.7b}
\left[ \hat{E}_k({\bf r}), \hat{E}_l({\bf r'})\right]
= \left[ \hat{B}_k({\bf r}), \hat{B}_l({\bf r'})
\right] = 0.
\end{equation}
The electric and magnetic fields can be of course expressed in terms
of vector ($\hat{\bf A}$) and scalar ($\hat{\varphi}$)
potentials. In what follows we will set the scalar potential
equal to zero. This gauge condition implies that both the transverse
and the longitudinal electric fields are obtained from the
vector potential
\begin{equation}
\label{2.8}
\hat{\bf A}({\bf r}) = \hat{\bf A}^{(+)}({\bf r})
+ \hat{\bf A}^{(-)}({\bf r}),
\end{equation}
\begin{equation}
\label{2.9}
\hat{A}_k^{(+)}({\bf r})
= \mu_0 \int_{0}^{\infty} d\omega \int d^3{\bf r}'
\, G_{kk'}({\bf r},{\bf r}',\omega)
\underline{\,\hat{\!j}}_{k'}({\bf r}',\omega) .
\end{equation}
Let us now consider the case when an external (two-level) atomic
system at position ${\bf r}_A$ is present.
Treating the interaction of such a guest atom with the
electromagnetic field in dipole and rotating wave approximations, the
Hamiltonian of the total system can be given by
\begin{equation}
\label{2.10}
\hat{H} = \int d^3{\bf r} \int_0^\infty d\omega \,\hbar\omega
\hat{\bf f}^\dagger({\bf r},\omega) \hat{\bf f}({\bf r},\omega)
+\sum_{\alpha=1}^{2}\hbar\omega_\alpha\hat{A}_{\alpha\alpha}
-\left[
i\omega_{21}\hat{A}_{21}\, \hat{\bf A}^{(+)}({\bf r}_A)\!\cdot\!{\bf d}_{21}
+{\rm H.c.}\right].
\end{equation}
Here the atomic operators $\hat{A}_{\alpha\alpha'}$
$\!=$ $\!|\alpha\rangle\langle \alpha'|$
are introduced, with $|\alpha\rangle$ being the energy eigenstates
of the guest atom
\mbox{($\alpha$ $\!=$ $\!1,2$)}. The energies of the two states are
$\hbar\omega_1$ and $\hbar\omega_2$ ($\hbar\omega_2$ $\!>$
$\!\hbar\omega_1$), and $\omega_{21}$ $\!=$
$\omega_2$ $\!-$ $\!\omega_1$ and ${\bf d}_{21}$, respectively,
are the atomic transition frequency and dipole moment.
Note that in the interaction term in Eq.~(\ref{2.10}) the
$\hat{\bf A}^2$ term and the counter-rotating terms have
been dropped.
In the Heisenberg picture the
equations of motion then read as, on recalling Eqs.~(\ref{2.3})
and (\ref{2.9}),
\begin{eqnarray}
\label{2.11}
\,\dot{\hat{\!A}}_{22} &=&
- \frac{\omega_{21}}{\hbar} \hat{A}_{21} \,
\hat{\bf A}^{(+)}({\bf r}_A)\!\cdot\!{\bf d}_{21}\, + {\rm H.c.},
\\
\label{2.11a}
\,\dot{\hat{\!A}}_{11} &=& -\,\dot{\hat{\!A}}_{22}\,,
\end{eqnarray}
\begin{equation}
\label{2.12}
\,\dot{\hat{\!A}}_{21} = i \omega_{21} \hat{A}_{21}
+ \frac{\omega_{21}}{\hbar} \hat{\bf A}^{(-)}({\bf r}_A)\!\cdot\!{\bf d}_{21}\,
\left(\hat{A}_{22}-\hat{A}_{11}\right),
\end{equation}
\begin{equation}
\label{2.13}
\,\dot{\hat{\!f}}_i({\bf r},\omega) = - i\omega \hat{f}_i({\bf r},\omega)
+\, \frac{\omega_{21}\omega}{c^2}
\sqrt{\frac{\epsilon_I({\bf r},\omega)}{\hbar\pi\epsilon_0}}
\, (d_{21})_k G_{ki}^\ast({\bf r}_A,{\bf r},\omega)\,\hat{A}_{12}\,.
\end{equation}
Substituting in the vector potential in Eqs.~(\ref{2.11}) --
(\ref{2.12}) for $\hat{f}_i({\bf r},\omega,t)$ the formal
solution of Eq.~(\ref{2.13}), i.e.,
\begin{equation}
\label{2.13a}
\hat{A}^{(+)}_i({\bf r},t) = \hat{A}^{(+)}_{{\rm free}\,i}({\bf r},t)
+\, \frac{\omega_{21}}{\pi \epsilon_0 c^2} (d_{21})_k \int_0^\infty
\!d\omega \,
\bigg[
{\rm Im}\,G_{ik}({\bf r},{\bf r}_A,\omega)
\int_{t'}^t \!d\tau\,
{\rm e}^{-i\omega (t-\tau)} \hat{A}_{12}(\tau)
\bigg],
\end{equation}
a system of integro-differential
equations for the atomic quantities is obtained.
[Note that Eq.~(\ref{B4}) has been used for
deriving Eq.~(\ref{2.13a}).]
At this stage a Markov approximation can be introduced,
and the integro-differential equations reduce to
Langevin-type differential equations
(Appendix \ref{markov})
\begin{eqnarray}
\label{2.15}
\,\dot{\hat{\!A}}_{22}
&=& - \Gamma \hat{A}_{22}
- \left[\hat{A}_{21}\, \frac{\omega_{21}}{\hbar}
\hat{\bf A}_{\rm free}^{(+)}({\bf r}_A,t)\!\cdot\!{\bf d}_{21}
+ {\rm H.c.} \right],
\\
\label{2.15a}
\,\dot{\hat{\!A}}_{11} &=& -\,\dot{\hat{\!A}}_{22}\,,
\end{eqnarray}
\begin{equation}
\label{2.16}
\,\dot{\hat{\!A}}_{21} =
\left[i(\omega_{21} - \delta\omega) - {\textstyle\frac{1}{2}\Gamma} \right]
\hat{A}_{21}
+ \,\frac{\omega_{21}}{\hbar}
\hat{\bf A}_{\rm free}^{(-)}({\bf r}_A,t)\!\cdot\!{\bf d}_{21}\,
\left(\hat{A}_{22}-\hat{A}_{11}\right),
\end{equation}
where $\Gamma$ is the rate of spontaneous decay of the excited state
of the guest atom,
\begin{eqnarray}
\label{2.17}
\Gamma = \frac{2\omega_A^2\mu_k\mu_{k'}}{\hbar\epsilon_0c^2}\,
{\rm Im}\,G_{kk'}({\bf r}_A,{\bf r}_A,\omega_A)
\end{eqnarray}
[$\mu_k$ $\!\equiv$ $\!(d_{21})_k$, $\omega_A$ $\!\equiv$ $\!\omega_{21}$],
and $\delta\omega$ is the (contribution of
the dielectric to the) Lamb shift [see Eq.~(\ref{B7})].
Note that $\hat{\bf A}_{{\rm free}}^{(\pm)}({\bf r},t)$ evolves
freely. From Eq.~(\ref{2.5}) together with Eqs.~(\ref{2.3}) and
(\ref{B4}) it can be proved that the quantization scheme
exactly yields, in agreement with the dissipation-fluctuation theorem,
the relation \cite{Abrikosov}
\begin{equation}
\label{2.18}
{\rm Im}\,G_{kk'}({\bf r},{\bf r}',\omega')\;\delta(\omega-\omega')
= \frac{\pi \epsilon_0 c^2}{\hbar \omega^2}
\,\langle 0 | \Big[\underline{\hat{E}}_k({\bf r},\omega),
\underline{\hat{E}}_{k'}^\dagger({\bf r}',\omega')\Big] | 0 \rangle .
\end{equation}
As long as the Markov approximation applies,
the spontaneous decay can be described in terms
of the rate (\ref{2.17}), the rate formula being
valid for arbitrary dielectrics and geometries.
Especially, for an atom in vacuum we have
\begin{equation}
\label{2.19}
{\rm Im} \,G_{kk'}({\bf r}_A,{\bf r}_A,\omega_A) =
\frac{\omega_A}{6\pi c} \, \delta_{kk'}
\end{equation}
[see Eqs.~(\ref{C1}) -- (\ref{C5}) for $\epsilon$ $\!=$ $\!1$],
which leads to the well-known result (\ref{1.1}),
\begin{equation}
\label{2.20}
\Gamma = \Gamma_0 = \frac{\omega_A^3 \mu^2}{3\pi \hbar \epsilon_0 c^3} \,.
\end{equation}
A guest atom in a dielectric is situated in a
small free-space region and is surrounded by medium atoms.
Frequently a cavity model is used for describing the
situation. An atom in an empty cavity in an otherwise continuous
medium is considered and it is assumed that the linear
dimensions of the cavity are much less then the atomic transition
wavelength. In particular, for isotropic systems a spherical cavity
of radius $R$ may be considered. With regard to Eq.~(\ref{2.17}), the
``only'' problem that remains is the calculation of (the imaginary part
of) the classical Green tensor for a dielectric medium
of given permittivity which is disturbed by a
small free-space inhomogeneity.
\section{Virtual cavity model}
\label{CM}
In the virtual cavity model it is assumed that the field outside
the sphere is not modified by the small region of free space
inside the sphere, and the (local) electric field
$\underline{\bf E}'({\bf r},\omega) $ inside the sphere is given by
\cite{bornwolf}
\begin{equation}
\label{3.1}
\underline{\hat{\bf E}}'({\bf r},\omega)
= \underline{\hat{\bf E}}({\bf r},\omega)
+ \frac{1}{3\epsilon_0} \underline{\hat{\bf P}}({\bf r},\omega) ,
\end{equation}
where $\underline{\bf E}({\bf r},\omega)$ and
$\underline{\bf P}({\bf r},\omega)$, respectively, are the
electric and polarization fields
in the unperturbed continuous medium. From
Maxwell's equations (\ref{2.1A}) and (\ref{2.1C})
together with Eqs.~(\ref{2.1a}) -- (\ref{2.3}) it is seen that
\begin{equation}
\label{3.2}
\underline{\hat{\bf P}}({\bf r},\omega)
= \epsilon_0 \left[\epsilon({\bf r},\omega) -1\right]
\underline{\hat{\bf E}}({\bf r},\omega) +
\underline{\hat{\bf P}}{^N}({\bf r},\omega) ,
\end{equation}
where
\begin{equation}
\label{3.2a}
\underline{\hat{\bf P}}{^N}({\bf r},\omega)
= i \sqrt{\frac{\hbar\epsilon_0}{\pi}\,\epsilon_I({\bf r},\omega)}
\,\hat{\bf f}({\bf r},\omega)
\end{equation}
is the noise polarization associated with absorption. For
classical optical fields at room temperatures the noise
polarization weakly contributes to the polarization and the
local field, and therefore it may be neglected. Obviously, for
quantum fields and especially for the quantum vacuum, whose
coupling to the guest atom gives rise to the spontaneous decay, the noise
polarization must not be omitted, because it is nothing other but a
part of the quantum vacuum. Combining Eqs.~(\ref{3.1}) and (\ref{3.2})
yields the local-field operator
\begin{equation}
\label{3.3}
\underline{\hat{\bf E}}'({\bf r},\omega)
= \frac{1}{3}
[\epsilon({\bf r},\omega) +2] \underline{\hat{\bf E}}({\bf r},\omega)
+\frac{1}{3\epsilon_0} \underline{\hat{\bf P}}{^N}({\bf r},\omega) .
\end{equation}
It can be shown that the local electromagnetic field satisfies
the equal-time commutation relations \cite{Scheel98b}
\begin{equation}
\label{3.4}
\epsilon_0 \left[ \hat{E}_k'({\bf r}) , \hat{B}_l'({\bf r'}) \right]
= -i\hbar \epsilon_{klm} \partial_m^r \delta({\bf r}-{\bf r'})
\left\{1+\textstyle{\frac{1}{9}} [\epsilon({\bf r},0) -1] \right\},
\end{equation}
\begin{equation}
\left[ \hat{E}_k'({\bf r}), \hat{E}_l'({\bf r'})\right]
= \left[ \hat{B}_k'({\bf r}), \hat{B}_l'({\bf r'})
\right] = 0,
\end{equation}
Comparing with the correct commutation relations,
we see that the virtual cavity model may be
regarded as being consistent with QED (over the whole
frequency domain), provided that
\begin{equation} \label{3.5}
\epsilon({\bf r},0) \ll 10;
\end{equation}
i.e., the value of the static permittivity must not be too large.
Now we can turn to the calculation of the spontaneous decay rate,
Eq.~(\ref{2.17}). Recalling Eq.~(\ref{2.18}), we may write
\begin{equation}
\label{3.6}
{\rm Im}\,G_{kk'}({\bf r},{\bf r}',\omega_A)\;\delta(\omega-\omega_A)
= \frac{\pi \epsilon_0 c^2}{\hbar \omega^2}
\,\langle 0 | \Big[\underline{\hat{E}}_k'({\bf r},\omega),
\underline{\hat{E}}_{k'}'^\dagger({\bf r}',\omega_A)\Big] | 0 \rangle
\end{equation}
with $|{\bf r}$ $\!-$ $\!{\bf r}_A|$, $\!|{\bf r}'$ $\!-$ $\!{\bf r}_A|$
$\!<$ $\!R$ and $\hat{\bf E}'$ from Eq.~(\ref{3.3}). Since
$\underline{\hat{\bf E}}$ in Eq.~(\ref{3.3}) is determined
by Eq.~(\ref{2.5})
with the Green tensor for the
field in the undisturbed continuous medium, knowledge of
the imaginary part of
that Green tensor is sufficient to calculate
the decay rate.
However, for \mbox{${\bf r},{\bf r}'$ $\!\to$ $\!{\bf r}_A$}
a singular contribution
to the rate is observed, which reflects the fact that
the description of the dielectric as a continuous medium
contradicts a precise determination of the position of the
guest atom.
The problem might be overcome by regularization, e.g.,
by averaging Eq.~(\ref{3.6})
over the sphere. Combining Eqs.~(\ref{2.17}) and (\ref{3.6})
and using Eq.~(\ref{3.3}) [together with Eqs.~(\ref{2.3}),
(\ref{2.5}), and (\ref{3.2a})] yields \cite{Scheel98b}
\begin{eqnarray}
\label{3.6A}
\lefteqn{
\Gamma_{\rm CM} = \frac{2\omega_A^2 \mu_k \mu_{k'}}{\hbar \epsilon_0 c^2}
\left| \frac{\epsilon(\omega_A)+2}{3}
\right|^2 {\rm Im}\,
\overline{G_{kk'}^{\rm M}({\bf r},{\bf r}',\omega)}
}
\nonumber \\&&\hspace{2ex}
+\,\frac{4\omega_A^2}{3\hbar \epsilon_0 c^2}
\epsilon_I(\omega_A) \mu_k \mu_{k'} \,
{\rm Re}\!\left[
\frac{\epsilon(\omega_A)+2}{3} \;
\overline{G_{kk'}^{\rm M}({\bf r},{\bf r}',\omega_A)}
\right]
+\,\frac{2}{9\hbar \epsilon_0} \epsilon_I(\omega_A) \mu_k \mu_{k'}
\overline{\delta_{kk'} \delta({\bf r}-{\bf r}')}
\end{eqnarray}
(the bar introduces averaging over the sphere), where
$G_{kk'}^{\rm M}({\bf r},{\bf r}',\omega)$ is the Green tensor
of the mean field in the undisturbed medium,
and $\epsilon(\omega_A)$ $\!\equiv$
$\!\epsilon({\bf r}_A,\omega_A)$. Note that the permittivity
can be assumed to be constant over the small sphere.
The first term in Eq.~(\ref{3.6A}) corresponds to the
result obtained in \cite{Barnett92,Barnett96}, without taking
account of the contribution of the noise polarization to the
quantum vacuum. The noise polarization gives rise to the second
term and the third term in Eq.~(\ref{3.6A}) -- terms that are
proportional to $\epsilon_I(\omega_A)$ and typically observed for
absorbing media.
When the position of the guest atom in the medium is sufficiently
far from inhomogeneities (such as the surface of the dielectric body)
the Green tensor $G_{kk'}^{\rm M}({\bf r},{\bf r}',\omega)$
in Eq.~(\ref{3.6A}) may be identified with that for bulk material
as given in Appendix \ref{greenbulk}.
Inserting for $G_{kk'}^{\rm M}({\bf r},{\bf r}',\omega_A)$ in
Eq.~(\ref{3.6A}) the result of
Eqs.~(\ref{C1}), (\ref{C2}), and (\ref{C5})
and averaging with respect to ${\bf r}$ and ${\bf r}'$ separately
over a sphere, on assuming equidistribution,
we derive \cite{note}
\begin{equation}
\label{3.7}
\Gamma_{\rm CM} = \Gamma_{\rm CM}^\| + \Gamma_{\rm CM}^\perp,
\end{equation}
where $\Gamma_{\rm CM}^\|$ and $\Gamma_{\rm CM}^\perp$,
respectively, are related to the longitudinal and transverse
parts of the Green tensor,
\begin{equation}
\label{3.7b}
\Gamma_{\rm CM}^\| =
\Gamma_0 \, \frac{4\epsilon_I(\omega_A)}{27|\epsilon(\omega_A)|^2}
\left( \frac{c}{\omega_A R} \right)^3,
\end{equation}
\begin{eqnarray}
\label{3.7a}
\lefteqn{
\Gamma_{\rm CM}^\perp = \Gamma_0 \Bigg\{ \eta(\omega_A)
\Bigg[ \Bigg|
\frac{\epsilon(\omega_A)+2}{3} \Bigg|^2 -
\frac{2\epsilon_I^2(\omega_A)}{9} \Bigg]
}
\nonumber \\ &&\hspace{4ex}
+ \,\epsilon_I(\omega_A) \left[\epsilon_R(\omega_A) +2 \right] \left[
\frac{8}{15} \left( \frac{c}{\omega_A R} \right)- \frac{2}{9}
\kappa(\omega_A) \right]
+ \,\frac{25\epsilon_I(\omega_A)}{54} \left( \frac{c}{\omega_A R}
\right)^3 \Bigg\} + {\cal O}(R)
\end{eqnarray}
($|R\sqrt{\epsilon(\omega_A)}\,\omega_A/c|$
$\!\ll$ $\!1$),
with $\Gamma_0$ being the free-space spontaneous emission rate defined
in Eq.~(\ref{1.1}). From inspection of
Eqs.~(\ref{3.7}) -- (\ref{3.7a})
it is seen that, when absorption can be disregarded, i.e.,
$\epsilon_I(\omega_A)$ $\!\approx$ $\!0$ and hence
$\epsilon(\omega_A)$ $\!\approx$ $\!\epsilon_R(\omega_A)$,
$n(\omega_A)$ $\!\approx$ $\!\sqrt{\epsilon_R(\omega_A)}$,
then $\Gamma_{\rm CM}$ $\!\approx$ $\!\Gamma_{\rm CM}^\perp$
reduces to $\Gamma_{\rm SE}$ given in Eq.~(\ref{1.3}) with
the local-field correction factor (\ref{1.4}). It is further
seen that for absorbing media the rate $\Gamma_{\rm CM}^\perp$
becomes quite different from that given in Eq.~(\ref{1.7}) with the
local-field correction factor (\ref{1.8}), because
of the effect of the noise polarization.
For more details, the reader is referred to \cite{Scheel98b}.
Most recently, a more microscopic derivation of the decay rate
has yielded, apart from regularization factors,
the same results \cite{fleisch}.
It is worth noting that the
$R$-dependent terms in Eq.~(\ref{3.7a}) solely result from the noise
polarization. In particular, the term $\sim R^{-3}$ may be regarded
as describing nonradiative decay via dipole-dipole energy
transfer from the guest atom to the surrounding medium. From
Eqs.~(\ref{3.7}) -- ~(\ref{3.7a}) it is seen
that the terms $\!\sim$ $\!R^{-3}$ can be combined to obtain
an overall rate for the nonradiative dipole-dipole energy
transfer. Obviously, the decomposition of $\Gamma_{\rm CM}$ in
$\Gamma_{\rm CM}^\perp$ and $\Gamma_{\rm CM}^\|$ has nothing
to do with a decomposition in radiative and nonradiative
decay channels in general.
It should be pointed out that the averages
in Eq.~(\ref{3.6A}), which correspond to
regularization at ${\bf r}$ $\!\to$ $\!{\bf r}'$,
can be taken in different ways. In other words,
the $R$-dependent terms in Eqs.~(\ref{3.7b})
and (\ref{3.7a}) are determined only up to some regularization
factors. Hence, not only the the cavity radius $R$ but also
the scaling factors of the absorption-assisted $\sim$ $\!R^{-1}$ and
$\sim$ $\!R^{-3}$ terms are undetermined in the model.
\section{Real cavity model}
\label{GL}
In the real cavity model the exact Green tensor for the system
disturbed by a small free-space inhomogeneity is inserted
in the rate formula (\ref{2.17}). In other words, the
electromagnetic field inside and outside the cavity exactly
solves Maxwell's equations (\ref{2.1}) -- (\ref{2.1C}) together
with the standard boundary conditions at the surface of the cavity.
In contrast to the virtual cavity approach, in the real cavity
approach the field inside the cavity exactly satisfies the
fundamental QED equal-time commutation relations (\ref{2.7a})
and (\ref{2.7b}), and the Green tensor does not lead to a
singular contribution to the decay rate.
The Green tensor for an inhomogeneous problem of that type can
always be written as a sum of the Green tensor for a homogeneous
problem and some tensor that obeys a source-free wave equation
and ensures the boundary conditions to be satisfied \cite{chew}.
Since the guest atom is situated in an empty cavity,
the relevant Green tensor reads as
\begin{equation}
\label{4.0a}
G_{kk'}({\bf r},{\bf r}_A,\omega_A)
= G_{kk'}^{\rm V}({\bf r},{\bf r}_A,\omega_A)
+ \tilde{G}_{kk'}({\bf r},{\bf r}_A,\omega_A)
\quad ({\bf r} \to {\bf r}_A)
\end{equation}
where $G_{kk'}^{\rm V}({\bf r},{\bf r}_A,\omega_A)$ is simply the
vacuum Green tensor, which is given by
Eqs.~(\ref{C1}) -- (\ref{C3})
with \mbox{$\epsilon(\omega)$ $\!=$ $\!1$}, and
$\tilde{G}_{kk'}({\bf r},{\bf r}_A,\omega_A)$ describes the
effect of reflection at the cavity surface. Obviously,
$G_{kk'}^{\rm V}({\bf r},{\bf r}_A,\omega_A)$ has no
longitudinal imaginary part,
\begin{equation}
\label{4.0b}
{\rm Im}\,G_{kk'}^{{\rm V}\|}({\bf r},{\bf r}_A,\omega_A) = 0
\qquad ({\bf r} \to {\bf r}_A)
\end{equation}
Since the tensor $\tilde{G}_{kk'}({\bf r},{\bf r}_A,\omega_A)$ is
related to a source-free problem, it is transverse, and hence
\begin{equation}
\label{4.0c}
\tilde{G}_{kk'}^\|({\bf r},{\bf r}_A,\omega_A) = 0
\qquad ({\bf r} \to {\bf r}_A).
\end{equation}
The imaginary part of $G_{kk'}({\bf r}_A,{\bf r}_A,\omega_A)$
is therefore equal to the imaginary part of the transverse part of the
Green tensor, so that the rate formula (\ref{2.17}) in the real
cavity model reads
\begin{eqnarray}
\label{4.0d}
\Gamma_{\rm GL} = \frac{2\omega_A^2\mu_k\mu_{k'}}{\hbar\epsilon_0c^2}\,
{\rm Im}\,G_{kk'}^\perp({\bf r}_A,{\bf r}_A,\omega_A).
\end{eqnarray}
In other words, in the real cavity model the longitudinal
field does not contribute to the decay rate. Thus, the
longitudinal decay rate $\Gamma_{\rm GL}^\|$ given in
\cite{Barnett96} is an artifact.
In order to calculate $\Gamma_{\rm GL}$ further, let us again consider
a spherical cavity of radius $R$ in bulk material, with the guest atom
being situated at the center of the sphere. The Green
tensor for a spherical two-layer system is given
in Appendix~\ref{green}. From Eqs.~(\ref{4.0a}) -- (\ref{4.0c})
together with
Eq.~(\ref{C5}) [for $\epsilon(\omega)$ $\!=$ $\!1$]
and Eq.~(\ref{A5}) it follows that
\begin{equation}
\label{4.a1}
{\rm Im}\, G_{kk'}^\perp({\bf r}_A,{\bf r}_A,\omega_A)
= \frac{\omega_A}{6\pi c} \left[ 1+ {\rm Re}\,C_1^N(\omega_A) \right]
\delta_{kk'},
\end{equation}
with the reflection coefficient $C_1^N(\omega_A)$ being given by
Eq.~(\ref{A6}). Hence, for a spherical cavity the spontaneous decay
rate (\ref{4.0d}) takes the form of
\begin{equation}
\label{4.a2}
\Gamma_{\rm GL}
= \Gamma_0 \left[ 1+ {\rm Re}\,C_1^N(\omega_A) \right],
\end{equation}
where $\Gamma_0$ is the free-space spontaneous emission
rate (\ref{1.1}). The reflection coefficient $C_1^N(\omega_A)$
in Eq.~(\ref{4.a2}) is a function of $R$
and given in Eq.~(\ref{A6}) explicitly. For $\omega_A R/c$
$\!=$ $\!2\pi R/\lambda_A$ $\!\ll$ $\!1$
we expand it in powers of $R$ to obtain
\begin{eqnarray}
\label{4.a5}
C_1^N(\omega_A)
&=& - \frac{3i[\epsilon(\omega_A)\!-\!1]}{2\epsilon(\omega_A)\!+\!1}
\left(\frac{c}{\omega_A R}\right)^3
-\, \frac{9i[4\epsilon^2(\omega_A)\!-\!3\epsilon(\omega_A)\!-\!1]}
{5[2\epsilon(\omega_A)+1]^2}
\left( \frac{c}{\omega_A R} \right)
+ \, \frac{9\epsilon^{5/2}(\omega_A)}{[2\epsilon(\omega_A)+1]^2}
\nonumber \\ && \hspace{5ex}
-1 +{\cal O}(R),
\end{eqnarray}
from which it follows that
\begin{eqnarray}
\label{4.a6}
\lefteqn{
\Gamma_{\rm GL}
= \Gamma_0\Bigg\{
\frac{9\epsilon_I(\omega_A)}{|2\epsilon(\omega_A)\!+\!1|^2}
\left( \frac{c}{\omega_A R} \right)^3
+\,\frac{9\epsilon_I(\omega_A)
[28|\epsilon(\omega_A)|^2\!+\!12\epsilon_R(\omega_A)\!+\!1]}
{5|2\epsilon(\omega_A)+1|^4 }
\left( \frac{c}{\omega_A R} \right)
}
\nonumber \\ &&\hspace{2ex}
+\,\frac{9\eta(\omega_A)}{|2\epsilon(\omega_A)+1|^4}
\Big[
4|\epsilon(\omega_A)|^4 +4\epsilon_R(\omega_A) |\epsilon(\omega_A)|^2
+\,\epsilon_R^2(\omega_A) -\epsilon_I^2(\omega_A)
\Big]
\nonumber \\ &&\hspace{2ex}
-\,\frac{9\kappa(\omega_A)\epsilon_I(\omega_A)}{|2\epsilon(\omega_A)+1|^4}
\left[ 4 |\epsilon(\omega_A)|^2 +2
\epsilon_R(\omega_A) \right]
\Bigg\}
+ {\cal O}(R) .
\end{eqnarray}
Needless to say that when setting $\epsilon(\omega)$ $\!=$ $\!1$,
then the free-space spontaneous emission rate is recovered.
When the atomic transition frequency is far from
an absorption band of the medium, then absorption may be disregarded,
i.e., $\epsilon_I(\omega_A)$ $\!\approx$ $\!0$ [and hence
$\epsilon(\omega_A)$ $\!\approx$ $\!\epsilon_R(\omega_A)$,
$n(\omega_A)$ $\!\approx$ $\!\sqrt{\epsilon_R(\omega_A)}$]. From
inspection of Eq.~(\ref{4.a6}) we see that for
$\epsilon_I(\omega_A)$ $\!\to$ $\!0$ the
term proportional to $R^0$ is the leading term, which exactly
gives rise to the rate formula (\ref{1.3}) together with the
correction factor (\ref{1.5}), i.e., we recover the familiar
result derived in \cite{Glauber91} for real refractive index.
We further see that for an absorbing medium the rate formula
cannot be given in the form of Eq.~(\ref{1.7}) together with
Eq.~(\ref{1.8}), as is suggested in \cite{Barnett96}.
Equation (\ref{4.a6}) reveals that for an absorbing medium
terms proportional to $R^{-3}$ and $R^{-1}$ are observed,
so that the decay rate sensitively depends on the radius of the sphere.
In particular, the near-field term proportional to $R^{-3}$ can
again be regarded as corresponding to nonradiative decay via dipole-dipole
energy transfer from the guest atom to the medium.
It should be pointed out that the
condition that \mbox{$\omega_AR/c$ $\!\ll$ $\!1$}; i.e.,
the (optical) wavelength $\lambda_A$ of the atomic transition
must be large compared with
the radius $R$ of the cavity, is in full agreement with
the Markov approximation used in order to introduce
a decay rate. From inspection of Eq.~(\ref{A6}) it is seen
that the (real part of the) reflection coefficient
$C_1^N(\omega)$ becomes a rapidly varying function
of frequency for $\omega R/c$ $\!\gtrsim$ $\!1$, and hence the
Markov approximation fails. In that case the
sphere acts like a micro-cavity resonator and
memory effects must be included in the temporal
evolution of the atom, which prevents the
excited state from decaying exponentially.
\section{Discussion}
\label{discussion}
To illustrate the results, we have computed the (virtual cavity model)
decay rate $\Gamma_{\rm CM}$, Eq.~(\ref{3.7}) -- (\ref{3.7a}),
and the (real cavity model) decay rate $\Gamma_{\rm GL}$, Eq.~(\ref{4.a6}),
of an atom in a spherical cavity of radius $R$ in a surrounding
medium with the single-resonance model permittivity
\begin{equation}
\label{4.1}
\epsilon(\omega) = 1+ \frac{\omega_P^2}
{\omega_T^2-\omega^2-i\gamma \omega_T}\,.
\end{equation}
Plots of the rates
as functions of the atomic transition frequency are
given in Figs.~\ref{fig1} -- \ref{fig6}. The figures reveal that
the two models can yield decay rates that are quite different
from each other. Far from the absorption band of the
medium the difference is rather quantitative than qualitative
[Figs.~\ref{fig2} and \ref{fig4}].
In the absorption band and in the vicinity of the absorption band,
i.e., in the region between the medium resonance $\omega_T$ and
the longitudinal frequency $\omega_L$ $\!=$ $\!\sqrt{\omega_T^2
+ \omega_P^2}$ (in the figures, $\omega_L$ $\!=$ $1.1\,\omega_T$),
a quantitatively and qualitatively different
behavior of the two rates can be observed
[Figs.~\ref{fig1}, \ref{fig3}, \ref{fig5}, and \ref{fig6}].
In particular, the rate obtained with the real cavity
model can substantially exceed the rate obtained with
the virtual cavity model. The differences between the
two rates are less pronounced for strong absorption;
i.e., when the value of the bandwidth parameter $\gamma$ in
Eq.~(\ref{4.1}) is sufficiently large (compare Fig.~\ref{fig1} with
Fig.~\ref{fig3}, and Fig.~\ref{fig5} with Fig.~\ref{fig6}).
In that region the rates sensitively respond to a change of the radius
of the cavity (compare Fig.~\ref{fig1}
with Fig.~\ref{fig5}, and Fig.~\ref{fig3} with Fig.~\ref{fig6}).
Obviously, an excited atom in an absorbing medium undergoes both
radiative and nonradiative damping, and in dense media
nonradiative decay can be much faster than radiative one.
In particular, for small cavity radius the $\sim R^{-3}$ dipole-dipole
energy transfer terms in the two rates can strongly enhance them.
Since the radiationless decay typically
happens at the longitudinal frequency $\omega_L$,
one observes, for sufficiently small values of $\gamma$, a shift
of the maximum of the decay rate from $\omega_T$ to $\omega_L$
with decreasing value of $R$
(compare Fig.~\ref{fig5} with Fig.~\ref{fig1}).
Even when the atomic transition frequency is relatively far from the
medium resonance, so that the imaginary part of the permittivity
becomes relatively small, the values of the two rates can
notably differ from those obtained from Eq.~(\ref{1.3})
together with either Eq.~(\ref{1.4}) or (\ref{1.5}), because
of the $\sim R^{-3}$ near field contributions to the rates. It should
be stressed that Eqs.~(\ref{1.3}) -- (\ref{1.5}) apply only when
nonradiative decay can be fully excluded from consideration.
Otherwise the near-field terms can give rise to observable effects,
as is illustrated in Figs.~\ref{fig2} and \ref{fig4}.
The rates $\Gamma_{\rm CM}$ and $\Gamma_{\rm GL}$
differ essentially in the way the cavity radius is introduced.
As already mentioned, in the virtual cavity
model the needed coincidence limit of the two spatial arguments
of the imaginary part of the Green tensor cannot be performed,
because of the singularity of the Green tensor of the (undisturbed) medium,
and regularization is required. In the paper,
a small fictitious distance \mbox{$|{\bf r}$ $-\!$ $\!{\bf r}'|$
$\!\neq$ $\!0$} between two neighboring atomic positions
inside a sphere of radius $R$
is kept in order to get a finite value, and the
result is then averaged with regard to ${\bf r}$ and
${\bf r}'$ separately over the sphere.
In contrast, in the real cavity
model the limit ${\bf r},{\bf r}'$ $\!\to$ ${\bf r}_A$ can
be performed exactly and a proper rate can be obtained,
$R$ being the radius of the real cavity.
>From the above it is suggested
that the value of the parameter $R$ may be different
in the two models in order to fit each other
(note that in \mbox{Figs.~\ref{fig1} -- \ref{fig6}}
the two rates are compared for equal values of $R$).
Another consequence of the via smoothing introduced
radius of the sphere in the virtual cavity model
is that there is a non-vanishing $\sim R^{-3}$ longitudinal-field
contribution to the decay rate.
Hence, the nonradiative
dipole-dipole energy transfer from the atom to the
surrounding medium is obtained from the interaction
of both transverse and longitudinal electromagnetic
field components with the atom,
$\Gamma_{\rm CM}$ $\!=$ $\Gamma_{\rm CM}^\|$ $\!+$ $\Gamma_{\rm CM}^\perp$.
On the contrary, the
real cavity model leads to a decay rate that solely
results from the interaction of the atom with
the transverse field, $\Gamma_{\rm GL}$ $\!=$ $\Gamma_{\rm
GL}^\perp$. Here, the dipole-dipole energy
transfer fully corresponds to a second-order process via
virtual photons. It is worth noting that for not too small
values of the radius of the virtual cavity (in our example,
$R$ $\!\gtrsim$ $\!0.1\,\lambda_A$) the contribution
of $\Gamma_{\rm CM}^\|$ to $\Gamma_{\rm CM}$ is small,
so that it may be disregarded and hence
$\Gamma_{\rm CM}$ $\!\approx$ $\Gamma_{\rm CM}^\perp$
(see Figs.~\ref{fig5} and \ref{fig6}).
Equation (\ref{2.17}) defines the total energy
relaxation rate of the (two-level) atom, which results from both
radiative and nonradiative decay, and the question arises
of what is the spontaneous emission rate. In \cite{Barnett96}
the transverse contribution to the decay rate is
associated with spontaneous emission, whereas the longitudinal
contribution is associated with nonradiative decay. However,
the exact result obtained with the real cavity model reveals that
there is no longitudinal contribution to the decay rate,
and hence the transverse contribution must be
associated with both spontaneous emission and nonradiative
decay. Similarly, the decay rate obtained from the study of
the resonant energy transfer between two guest molecules
surrounded by a perfect lattice of absorbing molecules
contains only transverse-field contributions and
describes both radiative and nonradiative relaxation
processes \cite{Juzeliunas94,Juzeliunas97}.
In \cite{Juzeliunas97}
it is suggested that the spontaneous emission rate be identified
with the $R$-independent (far-field) contribution to the decay
rate. Since the $\sim R^{-3}$ near field contribution
may be regarded as resulting from nonradiative decay via
dipole-dipole energy transfer, the question remains
of what is the meaning of the remaining terms.
Moreover, from our analysis of, e.g., the
real cavity model it is seen that $R$ must not substantially exceed
the atomic transition wavelength $\lambda_A$. Otherwise,
the Markov approximation does not apply and the calculated decay
rate becomes unphysical. In order to answer the question of what
is really spontaneous emission, the model should be extended
such that light detection at certain distances from the guest
atom is included.
Both in the virtual cavity model and the real cavity model
the dielectric is described in terms of a continuous polarization
field that does not resolve the positions of the microscopic
constituents of the medium. In reality an excited guest atom
does of course not interact with a continuous medium, but it
``sees'' the discrete distribution of the microscopic
constituents of the medium, at least the nearest-neighbor
grouping. Hence a refined treatment of the medium
should also allow for the presence in the cavity
of nearest-neighboring medium species whose interaction
with the guest atom is considered separately. The enlarged
cavity can then be chosen such that the guest atom
cannot ``resolve'' the discrete structure of the
medium outside the cavity and the continuous description
applies \cite{Krutitsky97,Krutitsky98}.
\acknowledgements
The authors thank M. Fleischhauer for helpful comments.
S. S. likes to thank J. Audretsch, F. Burgbacher, and K. Krutitsky for
helpful discussions during his stay at the University of Konstanz.
\begin{appendix}
\section{Markov approximation}
\label{markov}
Equation (\ref{2.13}) can be formally integrated to obtain
\begin{equation}
\label{2.14}
\hat{f}_i({\bf r},\omega,t) =
\hat{f}_{{\rm free}\,i}({\bf r},\omega,t)
+\,\frac{\omega_{21}\omega}{c^2}
\sqrt{\frac{\epsilon_I({\bf r},\omega)}{\hbar\pi\epsilon_0}}
\, (d_{21})_k G_{ki}^\ast({\bf r}_A,{\bf r},\omega)
\int_{t'}^t d\tau\, e^{-i\omega(t-\tau)}\hat{A}_{12}(\tau),
\end{equation}
where $\hat{\bf f}_{{\rm free}}({\bf r},\omega,t)$ evolves
freely. Substituting in the vector potential in Eqs.~(\ref{2.11})
-- (\ref{2.12})
for $\hat{f}_i({\bf r},\omega,t)$
the expression given in Eq.~(\ref{2.14}) yields
a system of integro-differential equations for the atomic
quantities, which cannot be solved analytically in general.
Usually the Markov approximation is introduced.
It is assumed that (after performing the $\omega$ integration)
the time integral effectively runs over
a small correlation time interval $\tau_{\rm c}$.
As long as we require that $t$ $\!-$ $\!t'$ $\!\gg$ $\!\tau_{\rm c}$,
we may extend the lower limit of the $\tau$ integral
in Eq.~(\ref{2.14}) to minus infinity with little error.
Further we require that $\tau_{\rm c}$ be small on a time scale
on which the atomic system is changed owing to the coupling
to the electromagnetic field. In this case in the $\tau$ integral
in Eq.~(\ref{2.14}) the slowly varying atomic quantity
$\hat{A}_{12}(\tau)e^{i\omega_{21}\tau}$ can be taken at time
$t$ and put in front of the integral,
\begin{eqnarray}
\label{B3}
\lefteqn{
\int_{t'}^t d\tau\, e^{-i\omega(t-\tau)}\hat{A}_{12}(\tau)
\approx
\int_{-\infty}^t d\tau\, e^{-i\omega(t-\tau)}\hat{A}_{12}(\tau)
}
\nonumber \\&&\hspace{2ex}
\approx \hat{A}_{12}(t)
\int_{-\infty}^t \!\! d\tau \,e^{-i(\omega-\omega_{21}) (t-\tau)}
= \hat{A}_{12}(t)\,\zeta(\omega_{21}-\omega) \nonumber \\
\end{eqnarray}
[$\zeta(x)$ $\!=$ $\!\pi\delta(x)$ $\!+$ $\!i{\cal P}x^{-1}$;
${\cal P}$ denotes the principal value]. Thus, the future of
the system is now determined by the present time only.
We substitute in Eq.~(\ref{2.14}) for the time integral
the expression given in Eq.~(\ref{B3}), calculate the vector
potential, Eqs.~(\ref{2.8}) and (\ref{2.9}).
With the help of the relation (see, e.g., \cite{Scheel98a})
\begin{equation}
\label{B4}
\int d^3{\bf s}\;
\frac{\omega^2}{c^2}\, \epsilon_I({\bf s},\omega)
G_{km}({\bf r},{\bf s},\omega) G^\ast_{lm}({\bf r'},{\bf s},\omega)
= {\rm Im}G_{kl}({\bf r},{\bf r'},\omega)
\end{equation}
we find after some calculation
\begin{equation}
\label{B5}
\hat{A}_i^{(+)}({\bf r}_A,t)
= \hat{A}_{{\rm free}\,i}^{(+)}({\bf r}_A,t)
+ \,\frac{\omega_{21}}{\pi\epsilon_0 c^2} (d_{21})_k
\int_0^\infty \!\!d\omega\,\zeta(\omega_{21}\!-\!\omega)
{\rm Im}\,G_{ik}({\bf r}_A,{\bf r}_A,\omega) \,\hat{A}_{12}(t) .
\end{equation}
In order to obtain Eqs.~(\ref{2.15}) -- (\ref{2.16}), we
eventually substitute in Eqs.~(\ref{2.11}) -- (\ref{2.12})
for the positive and negative frequency parts of the
vector potential the expressions according to Eq.~(\ref{B5}).
It can be easily seen that the real part of the $\zeta$ function
(i.e., the $\delta$ function) in Eq.~(\ref{B5}) leads to
$\Gamma$ given in Eq.~(\ref{2.17}).
The principal-value integral in Eq.~(\ref{B5})
which arises from the imaginary part of the $\zeta$ function
contributes to the Lamb shift and reads
\begin{eqnarray}
\label{B6}
\delta\omega=\frac{2\omega_{21}^2 (d_{21})_k (d_{21})_{k'}}{\hbar
\epsilon_0 c^2 \pi} \!\!\int_0^\infty\!\! d\omega \,
\frac{{\rm Im}\,G_{kk'}({\bf r}_A,{\bf r}_A,\omega)}{\omega -\omega_{21}}\,,
\end{eqnarray}
which can be rewritten as
\begin{equation}
\label{B7}
\delta\omega=\frac{2\omega_{21}^2 (d_{21})_k
(d_{21})_{k'}}{\hbar\epsilon_0 c^2}
\bigg[{\rm Re}\,G_{kk'}({\bf r}_A,{\bf r}_A,\omega_{21})
-\,\frac{1}{\pi} \int_0^\infty d\omega \,
\frac{{\rm Im}\,G_{ik}({\bf r}_A,{\bf r}_A,\omega)}{\omega
+\omega_{21}} \bigg].
\end{equation}
Equation (\ref{B7}) holds because of the Kramers-Kronig relation (or
Titchmarsh's theorem) for the Green function. Note that
the real part of the vacuum Green function is infinite
for ${\bf r}$ $\!=$ $\!{\bf r}'$ $\!=$ $\!{\bf r}_A$ and regularization
is required. The resulting
vacuum Lamb shift may be thought of as being included in the atomic
transition frequency, so that $\delta\omega$ in Eq.~(\ref{2.16})
may be regarded as being solely due to the surrounding dielectric.
\section{Green tensor for a homogeneous dielectric}
\label{greenbulk}
Following \cite{Barnett96,Abrikosov}, the
Green tensor for bulk material can be given by
\begin{equation}
\label{C1}
G_{kk'}({\bf r},{\bf r}',\omega)
= G_{kk'}^\|({\bf r},{\bf r}',\omega)
+ G_{kk'}^\perp({\bf r},{\bf r}',\omega),
\end{equation}
where ({\boldmath$\rho$} $\!=$ ${\bf r}$ $\!-$ $\!{\bf r}'$)
\begin{equation}
\label{C2}
G_{kk'}^\|({\bf r},{\bf r}',\omega)
= - \frac{c^2}{4\pi\omega^2\epsilon(\omega)} \left[
\frac{4\pi}{3}\delta(\mbox{\boldmath $\rho$})\,\delta_{kk'}
+ \left(\delta_{kk'}-\frac{3\rho_k\rho_{k'}}{\rho^2}\right)\frac{1}{\rho^3}
\right]
\end{equation}
and
\begin{eqnarray}
\label{C3}
G_{kk'}^\perp({\bf r},{\bf r}',\omega)
&=& \frac{c^2}{4\pi\omega^2\epsilon(\omega)}
\bigg\{
\left(\delta_{kk'}-\frac{3\rho_k\rho_{k'}}{\rho^2}\right)\frac{1}{\rho^3}
+\,k^3\bigg[
\left(\frac{1}{k\rho}+\frac{i}{(k\rho)^2}-\frac{1}{(k\rho)^3} \right)
\delta_{kk'}
\nonumber\\&&\hspace{2ex}
- \left(\frac{1}{k\rho}+\frac{3i}{(k\rho)^2}-\frac{3}{(k\rho)^3} \right)
\frac{\rho_k\rho_{k'}}{\rho^2}
\bigg] e^{ik\rho}
\bigg\},
\end{eqnarray}
are related to the longitudinal and transverse electric fields.
In Eq.~(\ref{C3}), the complex wave number
\begin{equation}
\label{C4}
k = \sqrt{\epsilon(\omega)}\,\frac{\omega}{c}
= \left[\eta(\omega)+i\kappa(\omega)\right]\frac{\omega}{c}
\end{equation}
has been introduced.
In particular for small values of $|k\rho|$,
$|k\rho|$ $\!\ll$ $\!1$, the exponential
$e^{ik\rho}$ in Eq.~(\ref{C3}) can be expanded to obtain
\begin{equation}
\label{C5}
G^\perp_{kk'}({\bf r},{\bf r}',\omega)
= \frac{1}{4\pi}
\left\{
\frac{\rho_k \rho_{k'}}{2\rho^3} +\frac{\delta_{kk'}}{2\rho}
+ \, \frac{2i\omega}{3c} \left[ \eta(\omega) +i\kappa(\omega) \right]
\delta_{kk'} \right\} +{\cal O}(\rho).
\end{equation}
\section{Green tensor for an empty sphere surrounded by
a homogeneous dielectric}
\label{green}
Following \cite{Li94},
the Green tensor of a system that consists of
an empty sphere surrounded by a homogeneous dielectric
can be given in terms of spherical Bessel functions and spherical
harmonics. When ${\bf r}$ and ${\bf r}'$ lie
in the sphere (with the center of the sphere being the origin of
the coordinate system), then the associated Green tensor
{\boldmath $G$}$({\bf r},{\bf r}',\omega)$
is given by
\begin{equation}
\label{A1}
\mbox{\boldmath $G$}({\bf r},{\bf r'},\omega)
= \mbox{\boldmath $G$}^{\rm V}({\bf r},{\bf r'},\omega)
+ \tilde{\mbox{\boldmath $G$}}({\bf r},{\bf r'},\omega),
\end{equation}
where \mbox{\boldmath $G$}$^{\rm V}({\bf r},{\bf r'},\omega)$
is the vacuum Green tensor, and
\begin{eqnarray}
\label{A1a}
\lefteqn{
\tilde{\mbox{\boldmath $G$}}({\bf r},{\bf r'},\omega)
= \frac{i\omega}{4\pi c} \sum\limits_{e,o} \sum\limits_{n=1}^\infty
\sum\limits_{m=0}^n
\bigg\{
\frac{2n\!+\!1}{n(n\!+\!1)} \frac{(n\!-\!m)!}{(n\!+\!m)!}
}
\nonumber \\ &&\hspace{2ex}\times \,
(2\!-\!\delta_{0m})\left[ C^M_n(\omega) {\bf M}_{{e \atop o}nm}
\left({\bf r},\frac{\omega}{c}\right)
{\bf M}_{{e \atop o}nm}\left({\bf r'},\frac{\omega}{c}\right)
+\, C^N_n(\omega) {\bf N}_{{e \atop o}nm}\left({\bf r},\frac{\omega}{c}\right)
{\bf N}_{{e \atop o}nm}\left({\bf r'},\frac{\omega}{c}\right) \right]
\bigg\}. \nonumber \\
\end{eqnarray}
Here ${\bf M}_{{e \atop o}nm}({\bf r},k)$ and ${\bf N}_{{e \atop
o}nm}({\bf r},k)$ are the ({\em e}ven and {\em o}dd)
vector Debye potentials, and the quantities $C^{M,N}_n(\omega)$
are the generalized reflection coefficients.
Introducing the abbreviating notations
\begin{eqnarray}
J_{ni} &=& j_n(k_i R),
\\
H_{ni} &=& h_n^{(1)}(k_i R),
\\
J'_{ni} &=& \frac{1}{\rho} \left. \frac{d[\rho j_n(\rho)]}{d\rho}
\right|_{\rho=k_i R},
\\
H'_{ni} &=& \frac{1}{\rho} \left. \frac{d[\rho
h_n^{(1)}(\rho)]}{d\rho}
\right|_{\rho=k_i R}
\end{eqnarray}
($k_1$ $\!=$ $\!\sqrt{\epsilon(\omega)}\,\omega/c$, $k_2$ $\!=$
$\!\omega/c$),
the reflection coefficients can be given by
\begin{equation}
\label{A2}
C^{M,N}_n(\omega) = \frac{T^{H,V}_{F,n}(\omega)
R^{H,V}_{P,n}(\omega)}{T^{H,V}_{P,n}(\omega)} \,,
\end{equation}
where
\begin{eqnarray}
R^H_{P,n}(\omega) &=& \frac{k_2 H'_{n2} H_{n1} - k_1 H'_{n1} H_{n2}}{k_2
J_{n1} H'_{n2} -k_1 J'_{n1} H_{n2}}\,, \\
R^V_{P,n}(\omega) &=& \frac{k_2 H_{n2} H'_{n1} - k_1 H_{n1} H'_{n2}}{k_2
J'_{n1} H_{n2} -k_1 J_{n1} H'_{n2}}\,, \\
T^H_{P,n}(\omega) &=& \frac{k_2 [ J_{n2} H'_{n2} - J'_{n2} H_{n2}]}{k_2
J_{n1} H'_{n2} - k_1 J'_{n1} H_{n2}}\,, \\
T^H_{F,n(\omega)} &=& \frac{k_2 [ J'_{n2} H_{n2} - J_{n2} H'_{n2}]}{k_2
J'_{n2} H_{n1} - k_1 J_{n2} H'_{n1}}\,, \\
T^V_{P,n}(\omega) &=& \frac{k_2 [ J'_{n2} H_{n2} - J_{n2} H'_{n2}]}{k_2
J'_{n1} H_{n2} - k_1 J_{n1} H'_{n2}}\,, \\
T^H_{F,n}(\omega) &=& \frac{k_2 [ J_{n2} H'_{n2} - J'_{n2} H_{n2}]}{k_2
J_{n2} H'_{n1} - k_1 J'_{n2} H_{n1}}\,.
\end{eqnarray}
The vector Debye potentials are defined by
\begin{eqnarray} \label{AA1}
{\bf M}_{{e \atop o}nm}({\bf r},k) &=& {\bf \nabla} \times \left[
\psi_{{e \atop o}nm}({\bf r},k) {\bf r}
\right] ,\\ \label{AA2}
{\bf N}_{{e \atop o}nm}({\bf r},k) &=& \frac{1}{k}{\bf \nabla} \times {\bf
\nabla} \times \left[ \psi_{{e \atop o}nm}({\bf r},k) {\bf r} \right]
\end{eqnarray}
with
\begin{equation}
\psi_{{e \atop o}nm}({\bf r},k) = j_n(kr) P_n^m(\cos\theta) {\cos
\choose \sin} m\phi ,
\end{equation}
and can be given by
\begin{equation}
\label{A3}
{\bf M}_{{e \atop o}nm}({\bf r},k) = \frac{im}{\sin\theta} \, j_n(kr)
P_n^m(\cos\theta) {\cos \choose \sin} m\phi \,{\bf e}_{\theta}
-\,j_n(kr)\, \frac{dP_n^m(\cos\theta)}{d\theta}
\,{\cos \choose \sin} m\phi \,{\bf e}_{\phi},
\end{equation}
\begin{eqnarray}
\label{A4}
\lefteqn{
{\bf N}_{{e \atop o}nm}({\bf r},k) = \frac{n(n+1)}{kr}\, j_n(kr)
P_n^m(\cos\theta) {\cos \choose \sin} m\phi \, {\bf e}_r
}
\nonumber \\ && \hspace{2ex}
+\,\frac{1}{kr} \frac{d[rj_n(kr)]}{dr} \left[
\frac{dP_n^m(\cos\theta)}{d\theta} {\cos \choose \sin} m\phi \, {\bf
e}_{\theta}
\mp \,\frac{im}{\sin\theta}\, P_n^m(\cos\theta) {\sin \choose \cos}
m\phi \, {\bf
e}_{\phi} \right] ,
\end{eqnarray}
$j_n(kr)$ is the spherical Bessel function
of the first kind and $P_n^m(\cos\theta)$ is the associated
Legendre polynomial.
Note that from Eqs.~(\ref{AA1}) and (\ref{AA2}) it follows that
$\mbox{\boldmath $G$}({\bf r},{\bf r}',\omega)$ is a (two-sided)
transverse tensor function.
Since for $kr$ $\!\to$ $\!0$ we have
\begin{equation}
j_n(kr) \stackrel{kr\to 0}{\longrightarrow} \frac{(kr)^n}{(2n+1)!!}
\left( 1-\frac{1}{2(2n+3)} +\ldots \right),
\end{equation}
from inspection of Eqs.~(\ref{A3}) and (\ref{A4}) we find that
\begin{eqnarray}
{\bf M}_{{e \atop o}nm}({\bf r},k) &\stackrel{kr\to
0}{\longrightarrow}& (kr)^n, \\
{\bf N}_{{e \atop o}nm}({\bf r},k) &\stackrel{kr\to
0}{\longrightarrow}& (kr)^{n-1}.
\end{eqnarray}
Hence, at the center of the sphere only the TM-wave
vector Debye potentials ${\bf N}_{{e \atop o}10}({\bf r},k)$
and ${\bf N}_{{e \atop o}11}({\bf r},k)$ contribute to
$\tilde{\mbox{\boldmath $G$}}({\bf r},{\bf r'},\omega)$
in Eq.~(\ref{A1a}),
\begin{equation} \label{A5}
\left.\tilde{G}_{kk'}({\bf r},{\bf r}',\omega)\right|_{{\bf r}={\bf r}'=0}
= \frac{i\omega}{6\pi c} C^N_1(\omega) \delta_{kk'} ,
\end{equation}
where $[n$ $\!\equiv$ $\!\sqrt{\epsilon(\omega)}]$
\begin{equation}
\label{A6}
C^N_1(\omega) =
\frac{\left[ i+z(n+1) - iz^2n - z^3 n^2/(n+1) \right] e^{iz}}
{\sin z - z(\cos z + in \sin z) + iz^2 n \cos z
- z^3 (\cos z -in \sin z) n^2/(n^2-1) }
\end{equation}
with
\begin{equation}
z = \frac{R\omega}{c} \,.
\end{equation}
\end{appendix}
|
\section{Introduction}
\lbl{sec.intro}
\subsection{A brief summary} In this paper we investigate relations
between three different phenomena in low-dimensional topology:
\begin{itemize}
\item[(a)] Massey products on the first cohomology $H^1 (M)$
with integer coefficients of $3$-manifolds $M$.
\item[(b)] the Johnson homomorphism on the mapping class group of an
orientable surface
\item[(c)]
the Goussarov-Habiro theory of finite-type invariants of 3-manifolds.
\end{itemize}
A key point of the connection between (a) and (b) is the notion of a
{\em homology cylinder}, i.e., a homology cobordism between
an orientable surface and itself. This notion generalizes the mapping
class group of a surface (in that case the cobordism is a product).
We will construct an extension of the Johnson homomorphism to homology
cylinders and use it to completely determine, in an
explicit fashion, the possible Massey products at the first
non-trivial level in a closed $3$-manifold (assuming that the first
homology $H_1$ is
torsion-free)---see Theorem \ref{thm.1}, Corollary \ref{cor.1} and
Theorem \ref{thm.2}.
This generalizes the known relationship
between the Johnson homomorphism and Massey products in the mapping
torus of a diffeomorphism of a surface to the more general situation
of homology cylinders---see Theorems \ref{thm.3}, \ref{th.alf} and
Remark \ref{rem.hcc}.
For historical reasons, we should mention early work of Sullivan \cite{Su}
on a relation between (a) and (b), and, for an alternative
point of view, work of Turaev \cite{Tu}.
With regards to the connection between (b) and (c), the main idea is
to consider Massey products as finite-type invariants of
$3$-manifolds, and to interpret them by a graphical calculus on
{\em trees}---see
Theorem \ref{thm.7}---in much the same way that
Vassiliev invariants of links have a graphical representation
and that Milnor's $\mu$-invariants are known to be
exactly the Vassiliev invariants of (concordance classes of)
string links which are represented by trees, see \cite{HM}.
A by-product of this investigation is a curious Lie algebra
structure on a vector space of the graphs which describe \fti s
of homology cylinders---see Proposition \ref{thm.4} and
Theorem \ref{thm.5}---that corresponds to the stacking of one homology
cylinder on top of another, and is closely related to
deformation quantization on a surface \cite{AMR1, AMR2,Ko3}.
\subsection{History}
\lbl{sub.his}
Years ago, Johnson introduced a homomorphism (the so-called Johnson
homomorphism) which he used to study the mapping class group,
\cite{Jo1,Mo3}. Morita \cite{Mo} discovered a close relation between
the Johnson homomorphism and the simplest \fti\ of 3-manifolds, namely
the Casson invariant; this relation was subsequently generalized
by the authors \cite{GL1, GL2} to
all \fti s of integral homology 3-sphere s (i.e., 3-manifolds $M$ with $H_1(M,\mathbb Z)=0$).
This generalization posed the question of
understanding the Johnson homomorphism (crucial to the structure
of the mapping class group) from the point
of view of \fti s. Unfortunately, this question is rather hard to answer
if we confine ourselves to invariants of integral homology 3-sphere s. This difficulty is overcome
by using a theory of \fti s based on the notion of surgery on $Y$-links, see
\cite{Gu1,Gu2,Ha,Oh,GGP}. Using this theory we will show that the Johnson
homomorphism is contained in its tree-level part, and we conjecture
that an extension of the Johnson homomorphism to homology cylinders
(i.e., 3-manifolds with boundary that homologically look like the product
of a surface with $[0,1]$), which
we define below, gives the full tree-level part; thus answering
questions raised by Hain and Morita \cite{Hain,Mo3}.
En route to answering the above question,
we were led to study this theory of invariants for
homology cylinders (studied also from a slightly different perspective
by Goussarov \cite{Gu1,Gu2} and Habiro \cite{Ha})
and discovered an apparently new Lie algebra of graphs colored by
$H_1(\Sigma)$
of a closed surface $\Sigma$, closely related to
deformation quantization on a surface \cite{AMR1, AMR2},
and to the curious graded group
$\mathsf{D}}%%%{\mathsf{L}^c(A)\overset{\text{def}}{=} \text{Ker}(A \otimes \mathsf{L}(A) \to \mathsf{L}(A))$,
(where $\mathsf{L}(A)$ denotes the free Lie ring of a torsion-free abelian
group
$A$) studied independently by several authors with a variety of
motivations
\cite{Jo2,Mo1,Ih,Dr,Ko1,Ko2,O1,O2,HM}.
It turns out that Massey products of 3-manifolds naturally take
values in $\mathsf{D}}%%%{\mathsf{L}^c(A)$,
and so does the Johnson homomorphism, which is also closely related to
Massey products--- a fact well-known to Johnson \cite{Jo2}, and later
proved by Kitano \cite{Ki}. However it is now known that the Johnson
homomorphism cannot
realize all elements of $\mathsf{D}}%%%{\mathsf{L}^c(A)$, but we will see that one can achieve
this realizability by replacing
surface diffeomorphisms by homology cylinders.
The generalized Johnson homomorphism actually provides
universally-defined invariants of homology cylinders, lifting the only
partially-defined Massey products. These are our explicit
candidates for the full
tree-level part of the Goussarov-Habiro theory (for homology cylinders).
This phenomenon was already observed when
one replaces 3-manifolds by string-links up to homotopy, see Bar-Natan
\cite{B-N} or by string-links up to concordance, see Habegger-Masbaum
\cite{HM}.
On the other hand, Massey products apply to more general manifolds and they
should provide partially-defined finite-type invariants.
\section{Statement of the results}
\lbl{sec.res}
\subsection{Conventions}
\lbl{sub.conv}
$F$ will always stand for a free group and $H$ for a torsion-free
abelian
group. The {\em lower central series} of
a group $G$ is inductively defined by $G_1=G$ and $G_{n+1}=[G,G_{n}]$.
A group homomorphism $p: K \to G$ is called an $n$-{\em equivalence}
if
it induces an isomorphism $K/K_n \cong G/G_n$. All manifolds will be
oriented, and all maps between them
will preserve orientation, unless otherwise mentioned.
The boundary of an oriented manifold is oriented with the ``outward
normal
first'' convention.
\subsection{Massey products}
\lbl{sub.massey}
By {\em Massey products} of length $n \geq 2$ in $H^2(\pi)$ we mean a Massey
product
$\langle a_1, \dots, a_n \rangle \in H^2(\pi)$ for $a_i \in H^1(\pi)$,
\cite{Ma,FS},
which are defined assuming that the ones of length $n-1$ are defined
and
vanish. We have the following theorem on universal Massey products:
\begin{theorem}
\lbl{thm.1}
(i) Given a connected topological space $X$ and $2$-equivalence $p:
F \to \pi\overset{\text{def}}{=}\pi_1(X)$, then $X$ has
vanishing Massey products of length less than $n$ if and only
if $p$ is an $n$-equivalence .
\newline
(ii) In that case, we have a short exact sequence\footnote{
Note that $\pi_n$ denotes the $n$th commutator subgroup of $\pi_1(X)$
and not the $n$th homotopy group of $X$.}
$$
H_2(X,\mathbb Z) \to \mathsf{L}_n(H_1(X,\mathbb Z)) \to \pi_{n}/\pi_{n+1} \to 0,
$$
where the first map determines and is determined by
all length $n$ Massey
products (for a precise expression, see Corollary \ref{cor.fi}) and
the second is induced by the Lie bracket. \newline
(iii) In addition, we have that
$$\alpha_1\smallsmile} %%\smallsmile \langle \alpha_2 ,\dots ,\alpha_{n+1}\rangle =\langle \alpha_1 ,\dots, a_n
\rangle \smallsmile} %%\smallsmile\alpha_{n+1} \in H^2(\pi), $$
for any $\alpha_1 ,\dots ,\alpha_{n+1}\in H^1 (\pi )$.
\end{theorem}
\begin{remark}
For the dependence of the short exact sequence in the above theorem
on the
map $p$, see Remark \ref{rem.dep}.
If $X$ satisfies the hypothesis of Theorem \ref{thm.1}
we have dually, over $\mathbb Q$:
$$
0 \to (\pi_{n}/\pi_{n+1})^\ast_{\mathbb Q} \to \mathsf{L}_n(H^1(X,\mathbb Q)) \to H^2(X,
\mathbb Q).
$$
Note that the first part of Theorem \ref{thm.1} appears in
\cite[Lemma 16]{O2}, and that the exact sequence above was first
suggested by Sullivan in \cite{Su} for $n=2$, and subsequently proven
by
Lambe in \cite{La} for $n=2$ using different techniques
involving minimal models.
\end{remark}
\begin{corollary}
\lbl{cor.1}
Given an $n$-equivalence\
$p :F \to \pi\overset{\text{def}}{=}\pi_1(M)$, where $M$ is a closed 3-manifold,
we have the exact sequence
$$
H^\ast \to \mathsf{L}_n(H) \to \pi_{n}/\pi_{n+1} \to 0$$
where $H = H_1(M,\mathbb Q)$.
If $\mu_n(M,p) \in H \otimes \mathsf{L}_n(H)$
denotes the first map (abbreviated by $\mu_n(M)$ if $p$ is clear),
then we have that $$ \mu_n(M,p) \in \mathsf{D}}%%%{\mathsf{L}^c_n(H).$$
In particular, $\mu_n(M,p)=0$ if and only if $p: F\to \pi$ is an
$(n+1)$-equivalence.
\end{corollary}
Given an integer $n$ and a torsion-free abelian group $H$,
it is natural to ask which elements
of $\mathsf{D}}%%%{\mathsf{L}^c_n(H)$ are realized by 3-manifolds as above. For this see
Theorem \ref{thm.2} below.
\subsection{The Johnson homomorphism}
\lbl{sub.johnson}
We now discuss the relation between Massey products and the Johnson
homomorphism.
Let $\Gamma_{g,1}$ denote the mapping class group\ of a surface $\Sigma_{g,1}$
of genus $g$ with one boundary component (i.e., the group of
surface diffeomorphisms that pointwise preserve the
boundary),
and let $\Gamma_{g,1}[n]$ denote its subgroup that consists of
surface diffeomorphisms that induce the identity on $\pi/\pi_{n+1}$,
where $\pi=\pi_1(\Sigma_{g,1})$. In \cite{Jo1}, Johnson defined a
homomorphism
$$ \tau_n: \Gamma_{g,1}[n] \to \mathsf{D}}%%%{\mathsf{L}^c_{n+1}(H),$$
where $H=H_1(\Sigma_{g,1},\mathbb Z)$, which he further extended to the case of
a closed surface.
Johnson was well-aware of the relation between his
homomorphism and Massey products on mapping torii, i.e., on twisted
surface
bundles over a circle; see \cite[p. 171]{Jo2}, further
elucidated by Kitano \cite{Ki}. In the present note, we extend this
relation to Massey products that come from an arbitrary pair
$(\Sigma,M)$ of an imbedding
$\iota: \Sigma \hookrightarrow M$ of a closed surface (not necessarily
separating)
in a 3-manifold.
Fix a closed 3-manifold $M$ and an $(n+1)$-equivalence
$F\to \pi\overset{\text{def}}{=}\pi_1(M)$. Given a pair $(\Sigma,M)$, and
$\phi \in \Gamma[n]$, let $M_\phi$ denote the result of cutting $M$ along
$\Sigma$, twisting by the element $\phi$ of its mapping class group\ and gluing back.
In this case, there exists a canonical cobordism $ N_\phi $ between $M$ and
$M_\phi$ such that the maps $\pi_1(M) \rightarrow \pi_1(N_\phi) \leftarrow
\pi_1(M_\phi)$ (induced by the inclusions $M,M_\phi \hookrightarrow
N_\phi$)
are $(n+1)$-equivalences; thus by Theorem \ref{thm.1}, $M_\phi$
has vanishing Massey products of length less than $n+1$. The ones of
length
$n+1$ on $M_{\phi}$ are determined in terms of those of $M$ and
the Johnson homomorphism as follows:
\begin{theorem}
\lbl{thm.3}
With the above assumptions, we have
$$ \mu_{n+1}(M_\phi)=\mu_{n+1}(M) + \iota_\ast \tau_n(\phi).$$
\end{theorem}
See also Remark \ref{rem.hcc}.
\subsection{Homology cylinders and realization}
\lbl{sub.homcyl}
It is well known \cite{Mo2,Hain} that the Johnson homomorphism
$\tau_n$
is not onto, in other words not every element of $\mathsf{D}}%%%{\mathsf{L}^c_{n+1}(H)$ can
be realized by surface diffeomorphisms. Generalizing
surface diffeomorphisms to a more general notion of homology
cylinders (defined below) allows
us to define an {\em ungraded} version
of the Johnson homomorphism, which then
induces, on the associated graded level, generalizations of the Johnson
homomorphisms. We will show that all of these are onto,
see Theorem \ref{th.alf}. As an application of this result, we will show
that we can realize
every element in $\mathsf{D}}%%%{\mathsf{L}^c_n(H)$ by 3-manifolds as in Corollary
\ref{cor.1}
and, in addition give a proof, free of spectral sequences, of
the isomorphism \eqref{eq.h3}, as mentioned above.
Let $\Sigma_{g,1}$ denote the compact orientable surface of genus $g$ with
one boundary component. A {\em homology cylinder} over $\Sigma_{g,1}$ is a
compact orientable $3$-manifold $M$ equipped with two
imbeddings $i^{-},i^{+}:\Sigma_{g,1}\to\partial M$ so that $i^{+}$ is
orientation-preserving and $i^{-}$ is orientation-reversing and if
we denote $\Sigma^{\pm}=\operatorname{Im} i^{\pm}(\Sigma_{g,1} )$, then $\partial M=\Sigma^+\cup\Sigma^-$ and
$\Sigma^{+}\cap\Sigma^{-}=\partial\Sigma^{+}=\partial\Sigma^{-}$. We also require that
$i^{\pm}$ be homology isomorphisms. We can multiply two homology
cylinders by identifying $\Sigma^{-}$ in the first with $\Sigma^{+}$ in
the second via the appropriate $i^{\pm}$. Thus $\mathcal H_{g,1}$, the set
of orientation-preserving diffeomorphism classes of homology
cylinders over $\Sigma_{g,1}$ is a semi-group with an obvious identity.
There is a canonical homomorphism $\Gamma_{g,1}\to\mathcal H_{g,1}$ that
sends $\phi$ to $(I\times\Sigma_{g,1}, 0\times\text{id}, 1\times \phi)$.
Nielsen showed that the natural map $\Gamma_{g,1}\to \mathrm{A}_0(F)$
is an isomorphism, where $F$ is
the free group on $2g$ generators $\{ x_i ,y_i\}$, identified with
the fundamental group of $\Sigma_{g,1}$ (with base-point on $\partial\Sigma_{g,1}$), and
$\mathrm{A}_0 (F)$ is the group of automorphisms of $F$ which fix the
element
$\omega_g =[x_1 ,y_1 ]\cdots [x_g ,y_g ]$, representing the boundary
of
$\Sigma_{g,1}$. It is natural to ask whether there exists an analogous isomorphism
for the semigroup $\mathcal H_{g,1}$. Below, we construct for every $n$
a homomorphism $\sigma_n :\mathcal H_{g,1}\to \mathrm{A}_0
(F/F_n )$, where $\mathrm{A}_0 (F/F_n )$ is the group of
automorphisms $\phi$
of $F/F_n$ such that a lift of $\phi$ to an endomorphism $\bar \phi$
of $F$
fixes $\omega_g\mod F_{n+1}$. It is easy to see that this condition
is independent of the lift. For example $A_0 (F/F_2 )=\text{Sp}(g, \mathbb Z )$.
Given $(M,i^{+},i^{-})\in\mathcal H_{g,1}$ consider
the homomorphisms $i^{\pm}_{*}:F\to\pi_{1} (M)$, where the
base-point is taken in $\partial\Sigma^{+}=\partial\Sigma^{-}$. In general, $i^{\pm}_{*}$
are not isomorphisms--- however, since $i^{\pm}$ are
homology isomorphisms, it follows from Stallings \cite{St} that they
induce isomorphisms
$i^{\pm}_{n}:F/F_{n}\to\pi_{1}(M)/\pi_{1}(M)_{n}$. We then
define $\sigma_{n}(M,i^{\pm})=(i^{-}_{n})^{-1} \circ i_{n}^{+}$. It is
easy to see that $\sigma_n(M,i^{\pm}) \in \mathrm{A}_0(F/F_n)$.
\begin{theorem}
\lbl{th.alf}
The map $\sigma_{n}: \mathcal H_{g,1}\to \mathrm{A}_0(F/F_n)$ is surjective.
\end{theorem}
\begin{remark}
\lbl{rem.hc}
We can convert $\mathcal H_{g,1}$ into a group $\mathcal H^c_{g,1}$ by considering
{\em homology cobordism classes } of homology cylinders. The inverse
of an
element is just the reflection in the $I$ coordinate. It is easy to
see
that the invariants $\sigma_n$ just depend on the homology bordism class
and so
define homomorphisms $\mathcal H^c_{g,1}\to\mathrm{A}_0 (F/F_n )$. The natural
homomorphism
$\Gamma_{g,1}\to\mathcal H^c_{g,1}$ is seen to be injective by the existence of
the $\sigma_n$
and the fact that the homomorphism $\Gamma_{g,1}\to\mathrm{A}_0 (F)$ is
an isomorphism.
In addition, we can combine the maps $\sigma_{n}$, for all $n$, to a single map
$\sigma^{\text{nil}}:\mathcal H_{g,1}\to \mathrm{A}_0(F^{\text{nil}})$,
where $F^{\text{nil}}$ is the nilpotent completion of $F$. Unlike
$\sigma$,
$\sigma^{\text{nil}}$ is not one-to-one, i.e.,
$\cap_{n}\ker\sigma_{n}\not=\{ 1\}$.
For example, if $P$ is any homology sphere, then the connected
sum $(I\times\Sigma_{g,1} )\sharp P$ defines an element in the kernel.
Also
$\sigma^{\text{nil}}$ is not onto, even though each $\sigma_{n}$ is. To
identify the image of $\sigma^{\text{nil}}$
we have to consider the {\em algebraic closure }
$\bar F\subseteq F^{\text{nil}}$, see \cite{Le2}. Using the arguments of
\cite{Le2}, we can show that any element of
$\operatorname{Im} (\sigma^{\text{nil}})$ restricts to an automorphism of $\bar F$
and, by
arguments similar to the proof of Theorem
\ref{th.alf}, it can be proved that $\operatorname{Im} (\sigma^{\text{nil}})$ consists
precisely of those $\phi\in A_0 (F^{\text{nil}})$ which restrict to an
automorphism of $\bar F$ and such that the element of $H_{2}(\bar F )$
associated to $\phi$ (see the
proof of Theorem \ref{th.alf}) is zero.
But since we do not know whether $H_{2} (\bar F
)=0$, this result does not seem very useful at this time.
\end{remark}
\begin{remark}\lbl{rem.mu}
The $\{\sigma_n\}$ can be described by numerical invariants if we consider the
coefficients of the Magnus expansion of $\sigma_n (M)(x_i ),\sigma_n (M) (y_i )$.
This is analogous to the definition of the $\mu$-invariants of a string
link. We can refer to these as {\em $\mu$-invariants of homology cylinders}.
\end{remark}
It will be useful for us to consider the filtration defined by the
maps $\sigma_n$, namely we define a decreasing {\em weight
filtration} on
$\mathcal H_{g,1}$ and on $\mathcal H^c_{g,1}$ by setting $\mathcal H_{g,1}[n]=\ker (\sigma_n)$.
\begin{proposition}
\lbl{prop.DA}
We have an exact sequence
$$ 1\to \mathsf{D}}%%%{\mathsf{L}^c_n(H)\to \mathrm{A}_0(F/F_{n+1})\to \mathrm{A}_0(F/F_n)
\to 1$$
and a commutative diagram
$$
\begin{diagram}
\node{\Gamma_{g,1}
[n]}\arrow[2]{e}\arrow{se,b}{\tau_n}\node[2]{\mathcal H_{g,1}[n]}
\arrow{sw,b}{\varsigma_n}\\
\node[2]{\mathsf{D}}%%%{\mathsf{L}^c_n(H)}
\end{diagram}
$$
where the map $\varsigma_n$, induced by $\sigma_n$, is onto. It follows that
$$ 0 \to \mathcal H^c_{g,1}[n+1] \to \mathcal H^c_{g,1}[n] \overset{\varsigma_n}\to \mathsf{D}}%%%{\mathsf{L}^c_n(H)
\to 0$$
is exact.
\end{proposition}
\noindent
\begin{remark}
A major problem in the study of the mapping class group is to
determine
the image of the Johnson homomorphism $\tau_n$, which largely
determines
the algebraic structure of the mapping class group since
$\cap_n \Gamma_{g,1}[n]=1$. In contrast, Theorem \ref{th.alf} largely
determines
the structure of $\mathcal H_{g,1}/\mathcal H_{g,1}[\infty]$, but in this case
$\mathcal H_{g,1}[\infty]=
\cap_n \mathcal H_{g,1}[n]$ is not trivial---see Question \ref{q.inf} at the end
of the paper.
\end{remark}
\begin{remark}\lbl{rem.sl}
It is instructive to consider the analogy between, on the one hand, the
mapping class group, homology cylinders and the invariant
$\sigma_n$ and the Johnson homomorphism, and, on the other hand, the pure braid
group, string links and the Milnor $\mu$-invariants.
There is an injection of the pure braid group on $g$ strands into the mapping
class group $\Gamma_{g,1}$, first defined by Oda and studied in \cite{Le2},
which preserves the weight filtrations and induces a monomorphism of the
associated graded Lie algebras.
This can, in fact, be generalized to an injection of the semi-group
$\mathcal S_g$ of string
links on $g$ strands into the semi-group $\mathcal H_{g,1}$ (and of the
string-link concordance group $ \mathcal S_g^c$ into $\mathcal H_{g,1}^c$), under
which $\sigma_n$ and
the $\mu$-invariants correspond. We will explain this in a future paper.
\end{remark}
\begin{theorem}
\lbl{thm.2}
Every element in $\mathsf{D}}%%%{\mathsf{L}^c_n(H)$ is realized by an $n$-equivalence\ $F\to
\pi_1(M^3)$, for some closed $3$-manifold $M$,
as in Corollary \ref{cor.1}. In addition, a $2$-equivalence $F \to H$
gives rise to a map $H_3(F/F_n)\to H \otimes \mathsf{L}_n(H)$ inducing
the isomorphism of Equation \eqref{eq.h3}.
\end{theorem}
We will give two different proofs of this theorem. One approach is to
apply results of Orr
\cite{O1} and Igusa-Orr \cite{IO} on $H_3 (F/F_n )$ and, in particular, the
isomorphism
\begin{equation}
\lbl{eq.h3}
\mathrm{cok}(H_3(F/F_{n+1}) \to H_3(F/F_n)) \cong \mathsf{D}}%%%{\mathsf{L}^c_n(H).
\end{equation}
A very similar argument appears in \cite{CGO}.
Alternatively we will see that this realizability is a consequence of
Theorem \ref{th.alf}.
This approach has the advantage of being ``spectral-sequence-free'' and
also gives another proof of \eqref{eq.h3}.
\subsection{Homology cylinders and \fti s of 3-manifolds}
\lbl{sub.fti}
Goussarov and Habiro \cite{Gu1,Gu2,Ha} have studied
two rather dual notions: an $n$-equivalence relation among
3-manifolds,
and a theory of invariants of 3-manifolds with values in an abelian
group.
Since their work is recent and not yet fully written, we will,
for the benefit of the reader, give a short introduction using
terminology and notation from \cite{GGP} (to which we refer the reader
for detailed proofs).
Both notions are intimately related to that of surgery $M_{\Gamma}$ along
a $Y$-link $\Gamma$ in a 3-manifold $M$, i.e., surgery along an imbedded link
associated to an imbedding of an
appropriately oriented, framed graph with trivalent and univalent vertices
so that the
univalent
ones end in ``leaves'' (explained below). Two manifolds are $n$-{\em
equivalent} if one can pass from
one to the other by surgery on a $Y$-link associated to a {\em
connected } Y-graph of degree (i.e., number of
trivalent
vertices) at least $n$. For example a theorem of Matveev \cite{M} says
that two closed manifolds are $1$-equivalent if and only if there is an
isomorphism between their first homology groups which preserves the
linking form on the torsion subgroups.
Similarly, a {\em finite type invariant} $\lambda$
ought to be the analog of a polynomial on the set of 3-manifolds,
in other words for some integer $n$ it satisfies a difference
equation
$$
\sum_{\Gamma' \subseteq \Gamma}(-1)^{|\Gamma'|} \lambda(M_{\Gamma})=0
$$
where $\Gamma$ is a $Y$-link in $M$ of more than $n$
components and the sum is over all $Y$-sublinks $\Gamma'$ of $\Gamma$.
In view of the above definition, it is natural to consider the free abelian
group $\mathcal M$ generated by homeomorphism classes of closed oriented
$3$-manifolds, and to define a decreasing filtration $\FY {}$
on $\mathcal M$ in such a way that $\lambda$ is an invariant of type $n$ if and
only if it vanishes on $\FY {n+1}$. Thus the question of how many
invariants
of degree $n$ there are translates into a question about the size of
the graded quotients $\GY {n}\overset{\text{def}}{=}\FY {n}/\FY {n+1}$. One
traditionally
approaches this problem by giving independently an upper bound and a
lower bound, which hopefully match.
In this theory, an upper bound has been obtained in terms of an
abelian group
of decorated graphs as follows. One observes
first that surgery along $Y$-links preserves the homology and linking
form of 3-manifolds, as well as the boundary. Define an equivalence
relation on compact $3$-manifolds: $M\sim N$ if there exists an isomorphism
$\rho :H_1 (M)\to H_1 (N)$ inducing an isometry of the linking forms, and a
homeomorphism $\partial M\to\partial N$ consistent with $\rho$. Thus if we let
$\mathcal M(M)$ denote the subgroup of
$\mathcal M$ generated by equivalent 3-manifolds, we have a direct sum
decomposition $\mathcal M=\oplus_{\sim}\ \mathcal M(M)$
(and also, $\FY {}=\oplus_{\sim}\ \FY {}(M)$), where the sum is over a
choice of one manifold $M$ from each equivalence class. In fact for closed
$3$-manifolds Matveev's theorem tells us that $ \GY {0}(M)=\mathbb Z$. After we fix
a 3-manifold $M$, and an oriented link
$\mathfrak{b}$
in $M$ that represents a basis of $H_1(M,\mathbb Z)/\mathrm{torsion}$, together with
a framing of $\mathfrak{b}$ (i.e., a choice of a trivialization of the normal
bundle of each component of $\mathfrak{b}$),
it turns out that there is a map\footnote{a more precise notation,
which
we will not use, would be $W^{M,\mathfrak{b}}_n$.}
\begin{equation}
\lbl{eq.graded}
W^{\mathfrak{b}}_n: \mathcal A_n(M) \to \GY {n}(M),
\end{equation}
which is onto over $\mathbb Q$ (actually, onto over $\mathbb Z [1/(2 |\mathrm{torsion}|)]$),
where $\mathcal A(M)$ is the group generated by graphs with univalent and
trivalent vertices, with a cyclic order along each trivalent
vertex, decorated by an element of $H_1(M,\mathbb Z)$ on each univalent
vertex,
modulo some relations, see \cite{GGP}.
Here $\mathcal A_n(M)$ is the subgroup generated by
graphs of degree $n$, i.e., with $n$ trivalent vertices; thus we have
$\mathcal A(M)=\oplus_n\mathcal A_n (M)=\mathcal A^t(M) \oplus \mathcal A^l(M)$,
where $\mathcal A^t(M)$ (resp. $\mathcal A^l(M)$) is the subgroup of $\mathcal A(M)$
generated by trees (resp. graphs with nontrivial first homology).
For a detailed discussion of the map $W^{\mathfrak{b}}$, see also Section \ref{sec.fti}.
We should point out that for $M=S^3$ (i.e., for integral homology 3-sphere s)
one can construct sufficiently many invariants of integral homology 3-sphere s to show that
$W^{\mathfrak{b}}$ is an isomorphism, over $\mathbb Q$, see \cite{LMO}. The same is true
for finite type (i.e., Vassiliev) invariants of links in $S^3$,
over $\mathbb Q$, see \cite{Ko2}.
However, it is at present unknown whether the map \eqref{eq.graded}
is one-to-one (and thus, an isomorphism), over $\mathbb Q$, for all 3-manifolds.
We now discuss a well-known isomorphism \cite{Ih,O2,Dr,HM}, over
$\mathbb Q$,
for a torsion-free abelian group $A$:
\begin{equation}
\Psi_n: \mathcal A^t_n(A) \cong_{\mathbb Q} \mathsf{D}}%%%{\mathsf{L}^c_{n+1}(A),
\end{equation}
which will help us relate the Johnson homomorphism to the tree-level
part of \fti s of 3-manifolds.
This map is defined as follows: Fix an oriented uni-trivalent tree
$T$
of degree $n$ (thus with $n+2$ legs, i.e., univalent vertices) and let
$c: \text{Leg}(T)\to A$ be a coloring of its legs. Given a leg $l$ of
$T$, $(T,l)$ is a rooted colored tree to which we can associate an
element
$(T,l)$ of $\mathsf{L}_{n+1}(A)$. Due to the $IHX$ relation (see Figure
\ref{lie}),
the function
$$
T \to \sum_{l \in \text{Leg}(T)} c(l) \otimes (T,l)
$$
descends to one $\mathcal A^t_n(A) \to A \otimes \mathsf{L}_{n+1}(A)$ so that its
composition
with $A\otimes \mathsf{L}_{n+1}(A)\to\mathsf{L}_{n+2}(A)$ vanishes,
thus defining the map $\Psi_n$.
There is a map $A \otimes \mathsf{L}_{n+1}(A) \to \mathcal A^t_n(A)$
(defined by sending $a \otimes b \in A \otimes \mathsf{L}_{n+1}(A)$ to the
rooted tree with one root colored by $a$ and $n$ additional legs
colored
by $c$), which shows that $\Psi_n$ is one-to-one; and by counting
ranks
it follows that it is in fact a vector space isomorphism. It is
unknown
to the authors whether $\Psi_n$ is an isomorphism over $\mathbb Z[1/6]$.
\begin{figure}[htpb]
$$ \eepic{lie1}{0.03} [c,[a,b]], \text{ } \eepic{lie2}{0.03}
a \otimes [c,b]+ c\otimes[b,a] +
b\otimes [a,c] $$
\caption{On the left, the map from rooted vertex-oriented trees to
the free
Lie algebra;
on the right the map $\Psi_1$.}\lbl{lie}
\end{figure}
It turns out that a skew-symmetric form $\mathfrak{c}:
A \otimes A \to \mathbb Q$ equips
$\mathcal A^t(A)$ with the structure of a graded Lie algebra, by defining the
Lie
bracket
\begin{equation}
\lbl{eq.bracket}
[\Gamma,\Gamma']^{\mathfrak{c}}= \sum_{a,b}
\mathfrak{c}( a, b )( \Gamma_a \mathrm{ glue }
\Gamma'_b ),
\end{equation}
where the sum is over each leg $a$ of $\Gamma$ and $b$ of $\Gamma'$ and
$\Gamma_a \mathrm{ glue }
\Gamma'_b$ is the graph obtained by gluing the legs $a$ and $b$ of $\Gamma$
and $\Gamma'$ respectively, with the
understanding that the sum over an empty set is zero.
In other words, $[\Gamma,\Gamma']^{\mathfrak{c}}$ is the sum of all
{\em contractions} of a leg of $\Gamma$ with a leg of $\Gamma'$.
This Lie bracket is not new, it has been
observed and used by Morita \cite{Mo1} and Kontsevich \cite{Ko1}
on a close relative of $\mathcal A^t(A)$, namely
$\mathsf{D}'}%%%\mathsf{L}^s(A)\overset{\text{def}}{=} A^\ast \otimes
\mathsf{L}(A)$ (which carries a bracket of degree $-1$).
We now explain the Lie bracket on $\mathcal A^t(A)$ from the point of view of
\fti s of 3-manifolds. Fixing a compact surface $\Sigma_{g,1}$ of genus
$g$ with one boundary component, it follows by definition that
$\mathcal M(\Sigma_{g,1}\times I)$ is generated by homology cylinders over $\Sigma_{g,1}$
and is a ring with multiplication
$M_1 \ast M_2$ defined by stacking $M_1$ below $M_2$.
Fix a framed oriented link $\mathfrak b$
in $\Sigma_{g,1}\times I$ that represents a basis of $H_1(\Sigma_{g,1}\times I;\mathbb Z)$ and consider
the associated onto map $W^{\mathfrak{b}} :\mathcal A(\Sigma _{g,1}\times I) \to \GY {}(\Sigma_{g,1}\times I)$ from
\eqref{eq.graded}, which is expected to be
an isomorphism. Thus, $\mathcal A(\Sigma_{g,1}\times I)$ should be equipped with a ring
structure. This is the content of the following
\begin{proposition}
\lbl{thm.4}
(i) $\mathfrak b$ induces a homomorphism
$\langle \cdot,\cdot \rangle^{\mathfrak b}: H_1(\Sigma_{g,1},\mathbb Z)\otimes
H_1(\Sigma_{g,1},\mathbb Z)\to
\mathbb Z$
satisfying\footnote{$ \langle \cdot, \cdot \rangle^{\mathfrak b} $
will often be denoted by $\langle \cdot , \cdot \rangle$ if
$\mathfrak b$ is clear from the context.}
$$ \langle a,b \rangle^{\mathfrak b} - \langle b,a \rangle^{\mathfrak b} = a\cdot
b,$$
where $\cdot$ is the natural symplectic form on $H_1(\Sigma_{g,1},\mathbb Z)$.
\newline
(ii) $\mathcal A(\Sigma_{g,1}\times I)$ is a ring with $\ast$-multiplication
(depending on $\mathfrak{b}$) defined as follows:
for $\Gamma,\Gamma' \in \mathcal A(\Sigma_{g,1}\times I)$,
$$
\Gamma \ast \Gamma'=\sum_{l=0}^\infty \langle \Gamma, \Gamma' \rangle_l ,$$
where
$$\langle \Gamma, \Gamma' \rangle_l=
(-1)^l \sum_{a,b}
\prod_{i=1}^l \langle a_i, b_i \rangle^{\mathfrak b}( \Gamma_a \mathrm{ glue }
\Gamma'_b )
$$
is the sum over all ordered subsets $a=(a_1, \dots
, a_l )$ and
$b=(b_1, \dots , b_l )$ of the set of legs of $\Gamma$ and $\Gamma'$
respectively,
$\Gamma_a \mathrm{ glue } \Gamma'_b$ is the graph obtained by gluing the
$a_i$-leg of $\Gamma$ to the $b_i$-leg of $\Gamma'$, for every $i$,
with the understanding that a sum over the empty set is zero
(thus the multiplication $\ast$ is a finite sum). \newline
(iii) $\mathcal A^c(\Sigma _{g,1}\times I)$ is a Lie subring of $\mathcal A(\Sigma_{g,1}\times I)$ with
bracket
defined by
$$
[\Gamma,\Gamma'] = \Gamma \ast \Gamma' - \Gamma' \ast \Gamma=
\sum_{l=1}^\infty \langle \Gamma ,\Gamma'\rangle_l - \langle \Gamma', \Gamma \rangle_l.
$$
(iv) Over $\mathbb Q$, there
is an algebra isomorphism $\mathsf{U}(\mathcal A^c(\Sigma_{g,1}\times I))\cong_{\mathbb Q} \mathcal A(\Sigma_{g,1}\times I)$,
where $\mathsf{U}$ is the universal enveloping algebra functor.
\end{proposition}
\begin{remark}
The leading term $\langle \cdot , \cdot \rangle_1$ of the $\ast$-multiplication
that involves contracting a single leg is independent of $\mathfrak{b}$
(see also part (iii) of Theorem \ref{thm.5}), whereas the subleading terms
$\langle \cdot , \cdot \rangle_l$ for $l \geq 2$ depend on $\mathfrak{b}$.
This is a common phenomenon in mathematical physics, analogous to the fact
that differential operators such as the Laplacian or the Dirac depend
on a Riemannian metric, but have symbols independent of it.
\end{remark}
\begin{remark}
It is interesting to compare
$\langle a,b \rangle^{\mathfrak b}$ to the Seifert matrix of a knot.
Both notions depend
on ``linking numbers'' of ``stacked'' curves, i.e. curves
pushed in a positive or negative direction and the relation of
$\langle \cdot , \cdot \rangle^{\mathfrak b}$ and the Seifert matrix to the symplectic
structure on $H_1 (\Sigma )$ is the same. The noncommutativity of
the stacking is reflected by the fact that the form $\langle \cdot , \cdot
\rangle^{\mathfrak b}$ is not symmetric. For a related appearance of this
noncommutativity, see also \cite{AMR1,AMR2}. Over $\mathbb Q$, the Lie algebra
$\mathcal A^c(\Sigma_{g,1}\times I)$
is closely related to a Lie algebra of chord diagrams on $\Sigma$
considered by
Andersen-Mattes-Reshetikhin in relation to deformation
quantization, loc. cit. We will postpone an explanation of this
relation to a subsequent publication.
\end{remark}
The ring structure on $\mathcal A(\Sigma_{g,1}\times I)$ would be of little interest were
it not
compatible with the one of $\GY {}(\Sigma_{g,1}\times I)$ and $\mathcal A^t(\Sigma_{g,1}\times I)$;
this is the content of the following
\begin{theorem}
\lbl{thm.5}
(i) The map $W^{\mathfrak{b}} : \mathcal A(\Sigma_{g,1}\times I)\to \GY {}(\Sigma_{g,1}\times I)$
preserves the ring structure. \newline
(ii) $\mathcal A^l(\Sigma_{g,1}\times I)$ is a Lie ideal of $\mathcal A^c(\Sigma_{g,1}\times I)$. \newline
(iii) The Lie bracket
of the quotient
$\mathcal A^c(\Sigma\times I)/\mathcal A^l(\Sigma_{g,1}\times I)\cong \mathcal A^t(\Sigma_{g,1}\times I)$
is equal to $(-1)^{\mathrm{deg}-1}$ times the Lie bracket of Equation
\eqref{eq.bracket}
using the symplectic form
on $H_1(\Sigma_{g,1}\times I)\cong H_1(\Sigma_{g,1})$. In particular, it is independent
of the basis $\mathfrak b$.
\end{theorem}
From now on, we will work over $\mathbb Q$.
We now show that the Johnson homomorphism
$\tau: \mathcal G\mathcal T_{g,1} \to \mathcal A^t(\Sigma_{g,1}\times I)$, or rather its signed version
$\overline{\tau}\overset{\text{def}}{=} (-1)^{\mathrm{deg}-1}\tau$,
can be recovered from the Lie algebra structure on $\mathcal A^c(\Sigma_{g,1}\times I)$,
where $\mathcal T_{g,1}(n) \subseteq \Gamma_{g,1}[n]$ is the subgroup of the Torelli
group generated by $n$-fold commutators, and $\mathcal G_n\mathcal T_{g,1}$ denotes the
quotient $\mathcal T_{g,1}(n)/\mathcal T_{g,1}(n+1)$.
Recall the map $\mathcal T_{g,1} \to \mathcal M(\Sigma_{g,1}\times I)$ defined by
changing the parametrization of the top part of homology cylinders,
its linear extension $(I \mathcal T_{g,1})^n \to \FY {n}(\Sigma_{g,1}\times I)$, where $I\mathcal T_{g,1}$
is the augmentation ideal of the group ring $\mathbb Q \Gamma_{g,1}$, and the induced
algebra map $\mathcal G\mathcal T_{g,1} \to \GY {}(\Sigma_{g,1}\times I)$.
The theorem below explains the statement that the Johnson
homomorphism is contained in the
tree-level part of a theory of invariants in $\Sigma_{g,1}\times I$.
\begin{theorem}
\lbl{thm.6} Given a surface $\Sigma_{g,1}$ as above of genus at least $6$,
there exists a map $\Phi: \mathcal G\mathcal T_{g,1}
\to \mathcal A^c(\Sigma_{g,1}\times I)$ and commutative diagrams of graded Lie
algebras:
$$
\divide\dgARROWLENGTH by2
\begin{diagram}
\node{\mathcal G\mathcal T_{g,1}}\arrow{se,r}{\overline{\tau}}\arrow{s,l}{\Phi} \\
\node{\mathcal A^c(\Sigma_{g,1}\times I)}\arrow{e}\node{\mathcal A^t(\Sigma_{g,1}\times I)}
\end{diagram}
\text{ and }
\begin{diagram}
\node{\mathcal G\mathcal T_{g,1}}\arrow{se}\arrow{s,l}{\Phi} \\
\node{\mathcal A^c(\Sigma_{g,1}\times I)}\arrow{e,t}{W^{\mathfrak{b}}}\node{\GY {}(\Sigma_{g,1}\times I)}
\end{diagram}
$$
where the left horizontal map is the natural projection on the tree
part. In other words, for $\phi \in \mathcal T_{g,1}(n)$ we have
$$ \Phi_n(\phi)=
(-1)^{n-1}\tau_n(\phi) + \mathrm{ loops } $$
in $ \mathcal A^c(\Sigma_{g,1}\times I)$.
\end{theorem}
\begin{remark}
\lbl{rem.closed}
For a closed surface $\Sigma_g$ of genus $g$, there is an identical version
of Proposition \ref{thm.4} and Theorem \ref{thm.5} above. As for
Theorem \ref{thm.6}, given
a closed surface $\Sigma_g$ of genus at least $6$, there exists
commutative diagrams
$$
\divide\dgARROWLENGTH by2
\begin{diagram}
\node{\mathcal G\mathcal T_{g}}\arrow{se,r}{\overline{\tau}}\arrow{s,l}{\Phi} \\
\node{\mathcal A^c(\Sigma_{g}\times I)/\Gamma_{\omega}}\arrow{e}\node{\mathcal A^t(\Sigma_{g}\times I)/\Gamma_{\omega}}
\end{diagram}
\text{ and }
\begin{diagram}
\node{\mathcal G\mathcal T_{g}}\arrow{se}\arrow{s,l}{\Phi} \\
\node{\mathcal A^c(\Sigma_{g}\times I)/\Gamma_{\omega}}\arrow{e,t}{W^{\mathfrak{b}}}\node{\GY {}(\Sigma_{g,1}\times I)}
\end{diagram}
$$
where $\Gamma_{\omega}$ is the ideal of $\mathcal A(\Sigma_{g}\times I)$ which is generated
by all elements of the form
$\sum_{i} \Gamma_{x_i,y_i,a}$ where $\{x_i,y_i\}$ is a standard symplectic
basis of $H_1(\Sigma_g)$, $a \in H_1(\Sigma_g)$ and $\Gamma_{b,c,d}$ denote the
degree $1$ graph
$$\underset{d}{\sideset{^b}{^c}\operatorname{\mathsf{Y}}}$$
with counterclockwise orientation.
\end{remark}
It is natural to ask for a statement of the above theorem
involving general (closed) 3-manifolds $M$.
How does one construct elements in $\FY {}(M)$? Given an imbedding
$\iota:\Sigma\hookrightarrow M$ of a closed surface $\Sigma$ and
$\phi \in \mathcal T(n)$, it was shown in \cite{GGP} that $M-M_\phi \in
\FY {n}(M)$.
The following theorem relates the Johnson homomorphism and the
tree-level part
of \fti s of 3-manifolds.
\begin{theorem}
\lbl{thm.7}
With the above assumptions, we can find $c^{\mathfrak b}_{\phi,n} \in
\mathcal A^l_{n}(M)$
so that we have in $\GY n (M)$:
$$ W^{\mathfrak{b}}_n(\Psi_n^{-1} \iota_\ast \overline{\tau}_n(\phi) + c^{\mathfrak b}_{\phi,n}) =
M-M_{\phi}.$$
\end{theorem}
The above theorem should be compared with \cite[Theorem 6.1]{HM},
where
they show that if a string link $L$ has vanishing $\mu$-invariants of
length less than $n$, then the degree $n$ tree-level part of the
Kontsevich
integral of $L$ is given by the degree $n$ $\mu$-invariants of
$L$. Note that these $\mu$-invariants are Massey products on the
closed
3-manifold obtained by $0$-surgery along the closure of the string
link.
\subsection{Plan of the proof}
\lbl{sub.plan}
The paper consists of two, largely independent sections; the reader
could easily skip one of them without any loss of understanding of the
results of the other. Two notions that jointly appear in Sections
\ref{sec.fti} and
\ref{sec.review} are the Johnson homomorphism and the notion of
homology cylinders.
In Section \ref{sec.fti}, we use combinatorial techniques
that are usually grouped under the name of \fti s (of knotted objects
such as braids, links, string links or 3-manifolds) or graph
cohomology. A key aspect is the introduction of a Lie
algebra $\mathcal A(\Sigma_{g,1}\times I)$ of graphs and its relation to the Johnson
homomorphism,
via Theorems \ref{thm.4}, \ref{thm.5}, \ref{thm.6} and \ref{thm.7}.
In Section
\ref{sec.review}, we use standard techniques from algebraic and
geometric
topology to prove Theorems \ref{thm.1} concerning
Massey products in general spaces and closed $3$-manifolds,in particular,
and Theorem \ref{thm.3} which relates the Johnson
homomorphism to Massey products. In addition, we use standard surgery
techniques adapted to homology cylinders to prove the two realization
Theorems
\ref{th.alf} and \ref{thm.2}.
Finally, in Section \ref{sec.que} we pose a set of questions that naturally
arise in our present study.
\section{Finite type invariants of 3-manifolds}
\lbl{sec.fti}
This section concentrates
on the proof of Theorems \ref{thm.4}, \ref{thm.5}, \ref{thm.6} and
\ref{thm.7}.
The techniques that we use are a combination of geometric and
combinatorial arguments.
\begin{proof}(of Proposition \ref{thm.4})
We only explain the first part. Statements (ii) and (iii) are
obvious and (iv)
follows by a theorem of Milnor-Moore \cite{MM} regarding the structure
of cocommutative graded connected Hopf algebras.
First we arrange that the components of $\mathfrak b$ project to immersions in
$\Sigma$ with transverse self-intersections and so that the framing has its
first componenet vector field pointing in the $I$ direction. We call such a
link {\em generic }. Given a two-component
sublink $\{b_1, b_2 \}$ of the framed oriented link
$\mathfrak b$ in $ \Sigma_{g,1}\times I$, let
$\{ p(b_1), p(b_2) \}$ denote its projection on $\Sigma\times 0$. Then
$p(b_1)$ and $p(b_2)$ intersect transversely at double points.
Define $\langle b_1, b_2 \rangle^{\mathfrak b}$ to be the sum with signs over
all
points in $p(b_1) \cap p(b_2)$
that $p(b_1)$ overcrosses $p(b_2)$, according to the convention
$$
\eepic{crossing}{0.03}.
$$
If $b_1 =b_2$, then we define $\langle b_1 ,b_1\rangle^{\mathfrak b}$ by counting
the self-intersections of $b_1$ in the above manner.
Since $\mathfrak b$ is a basis of $H_1(\Sigma_{g,1}\times I,\mathbb Z) \cong H_1(\Sigma,\mathbb Z)$,
this defines, by linearity, the desired map
$\langle \cdot, \cdot \rangle^{\mathfrak b}$.
Since $p(b_1) \cdot p(b_2)$ is the sum with signs over all points
$p(b_1) \cap p(b_2)$, it follows that $\langle a,b \rangle - \langle b,a \rangle= a
\cdot b$
for $a,b \in H_1(\Sigma,\mathbb Z)$.
\end{proof}
Before we proceed with the proof of Theorems \ref{thm.5} and
\ref{thm.7},
we need to recall the definition of the map $W^{\mathfrak{b}}: \mathcal A(M)\to\GY {}(M)$:
Given a colored graph $\Gamma$, we will construct an imbedding
of it in $M$ in two steps.
First, we imbed the leaves, as follows. Given the decoration
$x \in H_1(M,\mathbb Z)$, consider its projection $x^{tf} \in
H_1(M,\mathbb Z)/\mathrm{torsion}$
and write $x^{tf}=\sum_{b \in \mathfrak b} n_b [b]$, for integers
$n_b$.
Consider the oriented link $L_x$
obtained by the union (over $b$) of $n_b$ parallel
copies of $b$, where parallel copies of a component of $\mathfrak{b}$
are obtained by pushing off using the framing of $\mathfrak{b}$.
Choose a basing for
$\mathfrak b$, i.e., a set or meridians on each component of
$\mathfrak b$
together with a path to a base point. Join the components of $L_x$
using this basing to construct a knot $K_x$ in $M$.
Apply this construction to every leaf of $\Gamma$. Of course, the resulting
link depends on the above choices of basing and joining.
Second, imbed the edges of $\Gamma$ arbitrarily in $M$.
This defines an imbedding of $\Gamma$ in $M$ (which we denote by the
same name) that also depends on the above choices; however
the associated element $[M,\Gamma]$ in $ \GY {}(M)$ , where
$$[M,\Gamma]=\sum_{\Gamma' \subseteq \Gamma} (-1)^{|\Gamma'|} M_{\Gamma'}
$$
is the alternating sum over all $Y$-sublinks $\Gamma'$ of $\Gamma$, is
well-defined, depending only on the framed link $\mathfrak b$.
This follows
from the following equalities in $\GY {}(M)$ (for detailed proofs
see \cite{Gu2,Ha} and also \cite{GGP}):
\begin{equation}
\lbl{eq.r1}
[ \psdraw{STU1}{0.4in}]=[\psdraw{STU2}{0.4in}]
\end{equation}
\begin{equation}
\lbl{eq.r2}
[ \psdraw{Y4}{0.8in}]=[\psdraw{Y5}{0.8in}]+[\psdraw{Y6}{0.8in}]
\end{equation}
\begin{equation}
\lbl{eq.r3}
[ \psdraw{Y9}{0.8in}]=[\psdraw{Y10}{0.8in}]
\end{equation}
where in the above equalities $[M,\Gamma]$ is abbreviated by $[\Gamma]$.
Using further identities
in $\GY {}(M)$, one can show loc.cit.
that the map $\Gamma \to [M,\Gamma]$ factors through
further relations to define a map $W^{\mathfrak{b}}: \mathcal A(M)\to\GY {}(M)$.
\begin{proof}(of Theorem \ref{thm.5})
For the first part of the theorem, we begin by
choosing
$\mathfrak b$ in $\Sigma_{g,1}\times I$ to be a generic link. Note that $\mathfrak b$ can be
recovered
from its projection $p(\mathfrak b)$ together with a knowledge of the signs
at each overcrossing.
Given $\Gamma \in \mathcal A(\Sigma_{g,1}\times I)$, let $L^{\mathfrak b}(\Gamma)$ be an associated
$Y$-link in $\Sigma_{g,1}\times I$ such that $W^{\mathfrak b}(\Gamma)=
[\Sigma_{g,1}\times I, L^{\mathfrak b}(\Gamma)]$.
Without loss of generality, we can assume that the leaves $l^{\mathfrak
b}(\Gamma)$
of $L^{\mathfrak b}(\Gamma)$ form a generic link ,
and by abuse of notation, we can write that
$W^{\mathfrak b}(\Gamma)=
[\Sigma_{g,1}\times I, p(L^{\mathfrak b}(\Gamma))]$,
with the understanding that we have fixed the signs on the
overcrossings of
$p(L^{\mathfrak b}(\Gamma))$.
Now,
given $\Gamma,\Gamma' \in \mathcal A(\Sigma_{g,1}\times I)$,
let $L^{\mathfrak b}(\Gamma)$ and
$L^{\mathfrak b}(\Gamma')$ be the associated $Y$-links in $\Sigma\times [0,
1/2]$
and $\Sigma\times [1/2,1]$ respectively, and let $p_i: \Sigma_{g,1}\times I \to \Sigma\times
\{i\}$
denote the canonical projection.
Then we have that
\begin{eqnarray*}
W^{\mathfrak{b}}(\Gamma) \cdot W^{\mathfrak{b}}(\Gamma') & = & [\Sigma_{g,1}\times I, p_0(l^{\mathfrak b}(\Gamma))
\cup p_{1/2}(l^{\mathfrak b}(\Gamma'))] \\
& = & [(\Sigma_{g,1}\times I )_C, p_0(l^{\mathfrak b}(\Gamma))
\cup p_{0}(l^{\mathfrak b}(\Gamma'))] \\
& = & [(\Sigma_{g,1}\times I )_C, L^{\mathfrak b}(\Gamma)
\cup L^{\mathfrak b}(\Gamma')]
\end{eqnarray*}
where $C$ is a unit-framed trivial link in $\Sigma_{g,1}\times I$, whose
components encircle some crossings of $p_0(l^{\mathfrak b}(\Gamma))$
and $p_{0}(l^{\mathfrak b}(\Gamma'))$, so that surgery on $C$ brings
$p_0(l^{\mathfrak b}(\Gamma))$ below $p_0(l^{\mathfrak b}(\Gamma'))$.
The following lemma implies that the result of changing an
overcrossing
of $L^{\mathfrak b}(\Gamma)$ over $L^{\mathfrak b}(\Gamma')$ to an
undecrossing
can be achieved as a difference of the disjoint union of two graphs
minus the disjoint union of two graphs with a leg glued. Together
with Equation \eqref{eq.r2} and the definition of the multiplication
$\Gamma \ast \Gamma'$, it implies that
\begin{eqnarray*}
[(\Sigma_{g,1}\times I )_C, L^{\mathfrak b}(\Gamma)
\cup L^{\mathfrak b}(\Gamma')] & = & W^{\mathfrak{b}}(\Gamma \ast \Gamma')
\end{eqnarray*}
which finishes the proof of the first part of Theorem \ref{thm.5}.
\begin{lemma}
\lbl{lem.2things}
The following identity holds in $\GY {}(\Sigma_{g,1}\times I)$:
$$ [ \psdraw{Y1}{0.8in}]=[\psdraw{Y2}{0.8in}]-[\psdraw{Y3}{0.8in}] $$
where the framing of the unknot on the left hand side of the equation
is $+1$ and where we alternate with
respect to the $Y$-links of the figure. The vertical
arcs are arbitrary tubes.
\end{lemma}
\begin{proof}
This follows from the second equality above in $\GY {}(\Sigma_{g,1}\times I)$
and from
$$ [ \psdraw{Y7}{0.8in}]=-[\psdraw{Y3}{0.8in}] $$
$$ 2 [ \psdraw{Y11}{0.6in}]=0, $$
see \cite{GGP}.
\end{proof}
The second part of Theorem \ref{thm.5} is obvious from the definition
of the Lie bracket.
For the third part, notice that the Lie algebra structure on
the quotient $\mathcal A^t(\Sigma_{g,1}\times I)\cong \mathcal A^c(\Sigma_{g,1}\times I)/\mathcal A^l(\Sigma_{g,1}\times I)
$
is given by
$$
[\Gamma,\Gamma']=
\langle \Gamma , \Gamma'\rangle_1 -\langle \Gamma , \Gamma' \rangle_1
=-[\Gamma , \Gamma']^{\omega},
$$
where the last equality follows from the first part of
Proposition \ref{thm.4} and where $\omega$ is the symplectic form.
\end{proof}
\begin{proof}(of Theorem \ref{thm.6} and Remark \ref{rem.closed})
For a surface $\Sigma_{g,1}$ of genus $g \geq 3$ with one boundary component,
Johnson \cite{Jo1} introduced a homomorphism $\tau_1: \mathcal G_1\mathcal T_{g,1} \to
\Lambda^3(H)$, where, following the conventions of
\cite[Chapter 4]{Jo1}, $H=H_1(\Sigma_{g,1},\mathbb Z)$ and
$\Lambda^m(H)$ is identified
with a submodule of the $m$-th tensor power $\mathsf{T}^k(H)$
by defining
$$a_1 \wedge \cdots \wedge a_m =\sum_{\pi \in \text{Sym}_m} a_{\pi(1)} \otimes
\cdots\otimes a_{\pi(m)}.$$
In subsequent work, Johnson showed that modulo $2$-torsion his
homomorphism coincides with the abelianization of $\mathcal T_{g,1}$, thus
one gets, over $\mathbb Q$,
an onto map of Lie algebras $\mathsf{L}(\Lambda^3(H))\to \mathcal G\mathcal T_{g,1}$.
For the rest of the proof we will work over $\mathbb Q$. In \cite[Section 11]{Hain}
Hain proved that for genus $g \geq 6$,
the above map of Lie algebras has kernel generated by quadratic
relations $R_{g,1}$ (Hain's notation for $\mathcal G_n\mathcal T_{g,1}$ is
$\mathfrak{t}_{g,1}(n)$). Combining the proof of \cite{Hain}
Proposition 10.3 with Theorem 11.1 and Proposition 11.4], it follows that
the relation set $R_{g,1}$ is the symplectic submodule
of $\mathsf{L}_2(\Lambda^3(H))=\Lambda^2(\Lambda^3(H))$
generated by
$$[ x_1 \wedge x_2 \wedge y_2, x_3 \wedge x_4 \wedge y_4]=0$$
in terms of a standard symplectic basis $\{x_i,y_i\}$ of $H$.
Using the isomorphism $\Lambda^3(H)\cong \mathcal A^t_1(\Sigma_{g,1}\times I)$
given by mapping $a \wedge b \wedge c \in \Lambda^3(H)$
to the degree $1$ graph $\Gamma_{a,b,c}$ as in remark \ref{rem.closed},
we obtain a map of Lie algebras $\mathsf{L}(\Lambda^3(H))\to \mathcal A^c(\Sigma_{g,1}\times I)$.
Since for every
choice of $a \in \{x_1,x_2,y_2 \}$ and
$b \in \{ x_3,x_4,y_4 \}$ we have $a \cdot b =0$, the
first part of Proposition \ref{thm.4} implies
for every basis $\mathfrak b$ we have
$\langle a,b \rangle^{\mathfrak b} = \langle b,a \rangle^{\mathfrak b}$. This implies,
by definition, that $[\Gamma_{x_1,x_2,y_2}, \Gamma_{x_3,x_4,y_4}]=0 \in
\mathcal A^c_2(\Sigma_{g,1}\times I)$, thus obtaining the desired map $\Phi: \mathcal G\mathcal T_{g,1}
\cong \mathsf{L}(\Lambda^3(H))/(R_{g,1}) \to \mathcal A^c(\Sigma_{g,1}\times I)$.
Since $\mathcal G\mathcal T_{g,1}$ is generated by its elements of degree $1$, the
commutativity of the two diagrams follows by their commutativity in
degree $1$; the later follows by definition
for the first diagram, and by the fact that surgery on a $Y$-link
of degree $1$ with counterclockwise orientation and leaves decorated
by $a,b,c$ is equivalent to cutting, twisting and gluing by an element
of the Torelli group (of a surface of genus $3$ with one boundary component,
imbedded in $\Sigma_{g,1}$)
whose image under the Johnson homomorphism is equal to $a \wedge b \wedge c$,
see \cite{GGP}.
This concludes the proof of Theorem \ref{thm.6}.
We now prove the statements in Remark \ref{rem.closed}.
For a closed surface $\Sigma_g$ of genus $g \geq 3$,
Johnson \cite{Jo1} gave a version of his homomorphism $\tau_1:
\mathcal G\mathcal T_{g}\to \Lambda^3_0(H)$ where $\Lambda^3_0(H)$ is defined to be
the cokernel of the homomorphism $H\to\Lambda^3(H)$
that sends $x$ to $\omega \wedge x$, where
$\omega=\sum_{i} x_i \wedge y_i$ is the symplectic form of
$\Sigma_g$, for a choice of symplectic basis.
Working, from now on, over $\mathbb Q$,
Johnson \cite[Chapter 4]{Jo1} gave an identification of $\Lambda^3_0(H)$
with $\ker(C)$, where $C:\Lambda^3(H)\to H$ is given by
$$
C(x \wedge y \wedge z)= 2((x\cdot y)z + (y \cdot z)x + (z \cdot x)y)
$$
and $\cdot$ denotes the symplectic form.
Explicitly, we will think of $\Lambda^3_0(H)$ as the submodule of $\Lambda^3(H)$
which is generated by elements of the form $2(g-1)a -\omega \wedge
C(a)$ for $a \in \Lambda^3(H)$.
In \cite[Theorems 1.1 and 10.1]{Hain} Hain proved that for genus $g \geq 6$,
there is an isomorphism of graded Lie algebras $\mathsf{L}(\Lambda^3_0 (H))/(R_g)
\to \mathcal G\mathcal T_g$ which, in degree $1$, is the inverse of the Johnson
homomorphism, where
$R_g$ is the symplectic submodule
of $\mathsf{L}_2(\Lambda^3_0(H))=\Lambda^2(\Lambda^3_0(H))$
generated by the relations
$$[ 2(g-1)x_1 \wedge x_2 \wedge y_2 -x_1 \wedge \omega,
2(g-1)x_3 \wedge x_4 \wedge y_4 -x_3 \wedge \omega ]=0.$$
The slight difference of $2$ in the relations that Hain gave and the ones
mentioned above are due to the difference in the normalization of the
$\wedge$-product between Hain and Johnson. The restriction of the map
$\mathsf{L}(\Lambda^3(H))\to \mathcal A^c(\Sigma_{g,1}\times I)\cong \mathcal A^c(\Sigma_{g}\times I) \to
\mathcal A^c(\Sigma_{g}\times I)/(\Gamma_\omega)$ to $\mathsf{L}(\Lambda^3_0(H))$ gives a map
$\mathsf{L}(\Lambda^3_0(H)) \to \mathcal A^c(\Sigma_{g}\times I)/(\Gamma_\omega)$ which sends
the relations $R_{g}$ to zero (this really follows from the calculation
of the surface with one boundary component together with the fact that
$a \wedge \omega \in \Lambda^3(H)$ is sent into the ideal $\Gamma_\omega$
of $A^t_1(\Sigma_{g}\times I)$), thus inducing the desired map $\Phi:
\mathcal G\mathcal T_{g}\to \mathcal A^c(\Sigma_{g}\times I)/(\Gamma_\omega)$.
We claim that $W^{\mathfrak b}(M): \mathcal A(\Sigma_{g}\times I) \to \GY {}(\Sigma_{g}\times I)$
maps $\Gamma_\omega$ to zero. This follows from the identity
$$ (\tau_\partial)^{2g-2}=\prod_{i} [ \tau_{x_i}, \tau_{y_i}]$$
of Dehn twists on the mapping class group of $\Sigma_{g,1}$
\cite[Theorem 5.3]{Mo}, where $x_i ,y_i$ refer to the standard
meridian, longitude pairs associated with a symplectic basis of $H_1
(\Sigma_{g,1})$ and $\partial$ is the boundary curve of $\Sigma_{g,1}$. Thus we
have the relation
$$ 1=\prod_i \tau_a [ \tau_{x_i}, \tau_{y_i}] \tau_{a'}^{-1}$$
on the mapping class group of $\Sigma_g$ (where $a,a'$ are simple closed curves
in $\Sigma_{g,1}$ with isotopic images in $\Sigma_g$), together with the fact
surgery along the $Y$-link $\Gamma_{x_i,y_i,a}$ corresponds to the Dehn
twist $\tau_a [ \tau_{x_i}, \tau_{y_i}] \tau_{a'}^{-1}$ in $\mathcal T_g$,
\cite{GGP}.
Since $\mathcal G\mathcal T_{g}$ is generated by its elements of degree $1$, the
commutativity of the two diagrams follows by their commutativity in
degree $1$; this is shown in the same way as for a surface with one
boundary component.
This concludes the proof of Remark \ref{rem.closed}.
\end{proof}
\begin{proof}(of Theorem \ref{thm.7})
For a closed surface $\Sigma$ of genus at least $6$, Theorem \ref{thm.7} follows
from Remark \ref{rem.closed}
and the following Lemma \ref{lem.nm}, perhaps of independent interest.
For a closed surface $\Sigma$ of genus less than $6$, fix a disk and consider an
imbedding of its complement to a surface $\Sigma'$ in $M$ of genus at
least $6$.
Choose a lifting of $\phi$ to a diffeomorphism of the punctured
surface that preserves the boundary and extend it trivially to a
diffeomorphism
of $\phi'$ of $\Sigma'$. Since the Johnson homomorphism is stable with
respect
to increase in genus, and since $M_{\phi}=M_{\phi'}$, the result
follows from the previous case.
\end{proof}
\begin{lemma}
\lbl{lem.surface}
If $\Gamma \subseteq M$ is an imbedded graph in $M$ with a
distinguished leaf that bounds a surface disjoint from the other leaves
of $\Gamma$, then $[M,\Gamma]=0 \in \GY {}(M)$.
\end{lemma}
\begin{proof}
First of all, recall that $[M,\Gamma ]=0$ if any leaf bounds a disk
disjoint from the other leaves of $\Gamma$. As explained in \cite{GGP}, an
alternative way of writing Equation
\eqref{eq.r2} is as follows:
\begin{equation}
\lbl{eq.r4}
[ \psdraw{leavesab}{0.4in}]=[\psdraw{leavesa}{0.4in}] +
[\psdraw{leavesb}{0.4in}]
\end{equation}
for arbitrary disjoint imbeddings of two based oriented knots in $M$.
Given a based knot $\alpha$ in $M$, let $\overline{\alpha}$ denote the based knot
obtained by a push-off of $\alpha$ in its normal direction (any will do)
followed by reversing the orientation.
The above identity implies that $[M,\Gamma_{\overline{\alpha}}]=-[M,\Gamma_{\alpha}]$
in $\GY {}(M)$, where $\Gamma_{\kappa}$ is any imbedded graph in $M$
with a distinguished leaf the based oriented knot $\kappa$ in $M$.
Given $\Gamma$ as in the statement of the lemma, it follows that its
distinguished leaf is the connected sum of disjoint based knots of the form
$\alpha_i \sharp \beta_i \sharp \overline{\alpha_i} \sharp \overline{\beta_i}$; thus
it follows from the above discussion that $[M,\Gamma]=0$ in $\GY {}(M)$.
\end{proof}
\begin{lemma}
\lbl{lem.nm}
Given an imbedding $\iota: N\to M$ of (not-necessarily closed)
3-manifolds,
and links $\mathfrak{b}(N)$ (resp. $\mathfrak{b}(M$)) in $N$ (resp. $M$)
representing a basis of $H_1(N)$ (resp. $H_1(M)$),
there is an induced map $ \iota_\ast :\mathcal A(N) \to \mathcal A(M)$
induced by $\iota_*: H_1(N)\to H_1(M)$ on the colorings of
the legs of the graphs and a diagram
$$\begin{diagram}
\node{\mathcal A(N)}\arrow{s,l}{W^{\mathfrak{b}(N)}}\arrow{e,t}{\iota_\ast}\node{\mathcal A(M)}
\arrow{s,r}{W^{\mathfrak{b}(M)}}\\
\node{\GY {}(N)}\arrow{e,t}{\iota_\ast}\node{\GY {}(M)}
\end{diagram}
$$
that commutes up to $W^{\mathfrak{b}(M)}(\mathcal A^l(M))$.
In particular, for $\Gamma \in \mathcal A^t(N)$ we have
$$
\iota_\ast W^{\mathfrak{b}(N)}(\Gamma)=W^{\mathfrak{b}(M)}(\iota_\ast(\Gamma))
+ \mathrm{loops}.$$
\end{lemma}
\begin{proof}
Fix a graph $\Gamma \in \mathcal A^t(N)$. Let $\Gamma_L$ (resp. $\Gamma_{L'}$) denote
the $Y$-links in $M$ (with leaves $L$ (resp. $L'$)) such that
$$
\iota_\ast W^{\mathfrak{b}(N)}(\Gamma)=[M,\Gamma_L] \text{ and }
W^{\mathfrak{b}(M)}(\iota_\ast \Gamma)=[M,\Gamma_{L'}].
$$
It follows by definition that $L$ and $L'$ are homologous
links in $M$. After choosing a common base point for each pair $(L_i,L'_i)$
of components of $L$ and $L'$, it follows that the connected
sum $L_i \sharp \overline{L'_i}$ is nullhomologous in $M$ and thus bounds
a surface $\Sigma_i$ in $M$. The surface $\Sigma_i$ might intersect the other
components of $L$ or $L'$ at finitely many points; however by deleting disks
around the points of intersection of $\Sigma_i$ with $L \cup L'$, we can find
a nullhomotopic based link $L''_i$ and a surface $\Sigma_i'$ disjoint from
$L \cup L'$ with based boundary
such that $L_i \sharp \overline{L'_i}= L_i'' \sharp \partial \Sigma_i'$.
Equation \eqref{eq.r4} and Lemma \ref{lem.surface} imply that
$[M,\Gamma_L]-[M,\Gamma_{L'}]$ is a sum of terms over $Y$-links $\Gamma_\kappa$
in $M$ which are trees, with at least one
component $\kappa$ being nullhomotopic. By choosing a sequence of
crossing changes (represented by a unit-framed trivial link
$C(\kappa)$) that trivialize $\kappa$ and using Lemma \ref{lem.2things},
it follows that $[M,\Gamma_\kappa]=[M,\Gamma_{\text{trivial}}]=0$
modulo terms that involve joining
some legs of $\Gamma$ (thus modulo terms that involve graphs with
loops), which concludes the proof.
\end{proof}
\section{Massey products and the Johnson homomorphism}
\lbl{sec.review}
\subsection{Universal Massey products}
\lbl{sub.universal}
In this section, homology will be with integer coefficients, unless
otherwise
stated. A useful tool in the proof of Theorem \ref{thm.1} is a
{\em five-term} exact sequence of Stallings \cite{St}:
given a short exact sequence of groups $1 \to H \to G \to K \to 1$,
there is an associated five-term exact sequence
$$
H_2(G) \to H_2(K) \to H/[G,H] \to H_1(G) \to H_1(K) \to 1.
$$
Applying the five-term sequence to the exact sequence
$1 \to R \to F \to G \to 1$, (where $F$ is a free group) we get Hopf's
theorem \cite{Ho}
$$H_2(G)\cong (R \cap [F,F])/[R,F], \text{ and in particular, }
H_2(F/F_{n}) \cong F_{n}/F_{n+1}.$$
In the rest of this section, we will give a proof of Theorem
\ref{thm.1}
and its corollaries.
We will follow a rather traditional notation involving local
coordinates,
\cite{Ma,FS,Dw}.
Let $F$ denote the free group with basis $( x_1 ,\dots, x_m )$; we
will
denote by the same name the corresponding basis of $H=H_1 (F)$.
Let $(u_1 ,\dots, u_m)$ be the dual basis of $H^1 (F)\cong H^1 (F/F_n
)$. The
graded vector space $\bigoplus_n F_n /F_{n+1} $ has the structure of
a Lie
algebra induced by commutator, and is naturally identified with the
free
Lie algebra $\mathsf{L}(H)=\bigoplus_n \mathsf{L}_n (H)$.
We consider the {\em Magnus} expansion, \cite{MKS}.
Let $\mathbb Z[\![t_1 ,\dots,t_m ]\!]$ denote the power series ring in
non-commuting variable $\{t_1 ,\dots, t_m\}$. Define $\delta
:F\to\mathbb Z[\![t_1 ,\dots,t_m ]\!]$ to be the
multiplicative map defined by $\delta (x_i )=1+t_i$. This is an
imbedding and
induces imbeddings $\delta_n
:\mathsf{L}_n (H)=F_n /F_{n+1} \to\mathbb Z[\![t_1 ,\dots,t_m ]\!]_n$, where $\mathbb Z[\![t_1 ,\dots,t_m ]\!]_n$ is
the subspace of homogeneous polynomials of degree $n$.
We also recall the isomorphism $H_2 (F/F_n )\cong F_n /F_{n+1}
=\mathsf{L}_n (H)$
from above. We
now describe the Massey product structure on $H^ 1 (F/F_n )$.
For a sequence $I=(i_1,\dots,i_r)$ of numbers $i_j \in \{1,\dots,m\}$
(of length $|I| \overset{\text{def}}{=} r$), we will let $u_I$ denote the sequence
$(u_{i_1},\dots,u_{i_r})$ and, if each $u_i\in H^1 (F/F_n )$ we let
$\langle u_I \rangle$ denote the length $r$
Massey
product; we also let $t_I$ denote the element $\prod_{j=1}^r t_{i_j}$.
\begin{proposition}\label{th.um}
Any Massey product $\langle u_I \rangle$ of $F/F_n$ vanishes if
$|I|<n$. The action of any Massey product $\langle u_I \rangle$ on
$H_2 (F/F_n )\cong F_n /F_{n+1} =\mathsf{L}_n (H)\subseteq\mathbb Z[\![t_1 ,\dots,t_m ]\!]$ is
determined by the formula
\begin{equation}
\lbl{eq.mp}
\langle u_I\rangle \cdot\
t_J =
\begin{cases} 1 &\text{ if }
I=J \text{ and } |I|=n \\ 0 & \text{ otherwise }
\end{cases}
\end{equation}
In other words, the set $\{ \langle u_I\rangle | |I|=n \}$ defines the
basis of
$\mathbb Z[\![t_1 ,\dots,t_m ]\!]^{\ast}_n$ dual to $\{ t_I | |I|=n \}$.
\end{proposition}
See \cite{O2} for a slightly less explicit version of this theorem.
\begin{proof}
This follows easily from \cite{FS}. Suppose $w \in F_n$ is some
$n$-fold commutator. Then consider the one-relator group $G=F/\langle w
\rangle $
and the
projection $p:F/F_n\to G$. Consider $I$ of length $r \leq n$.
By induction we can assume that
$\langle u_I \rangle $ is uniquely defined in $F/F_n$ and by \cite{FS},
it is well-defined in $G$. Moreover, by naturality under $p^{*}$,
they take
the same value on $w$. If $r<n$ this is zero by \cite{FS}. Since this
holds
for all $w$ it follows that $\langle u_I \rangle =0$. If $r=n$
Equation \eqref{eq.mp} follows directly from the formula in \cite{FS}.
\end{proof}
\begin{corollary}
\lbl{cor.j1}
Let $p :F\to\pi$ be a $2$-equivalence.
Then $p$ is an $n$-equivalence\ if and only if all Massey
products in $H_1 (\pi )$ of length less than $n$ vanish.
\end{corollary}
See also \cite[Proposition 6.8]{CGO}.
\begin{proof}
The ``if'' part follows directly from the above proposition.
To prove the
``only if'' part we proceed by induction on $n$.
The inductive step presents us with a map $\pi \to \pi/\pi_{n-1}
\cong F/F_{n-1} $; consider the diagram
$$
\divide\dgARROWLENGTH by2
\begin{diagram}
\node{}\node{F/F_n}\arrow{s}\\
\node{\pi}\arrow{e}\arrow{ne,..}\node{F/F_{n-1}}
\end{diagram}$$
The obstruction to lifting this map is the pullback of the
characteristic class
in $H^2 (F/F_{n-1} ; F_{n-1} /F_n )$ of the central extension
$F_{n-1} /F_n \to F/F_{n-1} \to F/F_n $. But
Proposition~\ref{th.um}
implies that $H^2 (F/F_{n-1} )$ is
generated by Massey products of length $n-1$ and so the pullback is
zero if
and only if all Massey products of length $n-1$ vanish in $H^2 (\pi
)$.
Thus, we can inductively lift the map to $\pi \to F/F_n$, thus to a
map
$\pi/\pi_n\to F/F_n$, which is still a $2$-equivalence.
On the other hand, since $p$ is a $2$-equivalence, it induces an
onto map $F/F_n\to \pi/\pi_n$, which is also a $2$-equivalence.
Composing
with the map $\pi/\pi_n\to F/F_n$, we get an endomorphism of $F/F_n$
which is a $2$-equivalence. Stalling's theorem \cite{St}
implies that this endomorphism of $F/F_n$ is an isomorphism which
implies that the map $F/F_n\to \pi/\pi_n$ is one-to-one and thus
an $n$-equivalence .
\end{proof}
This proves the first assertion of Theorem \ref{thm.1}.
\begin{corollary}\lbl{cor.lam}
\label{cor.fi}
Suppose $p :F\to\pi$ is an $n$-equivalence .
Then there is an exact sequence:
$$ H_2 (\pi )\overset{\hat{p}}\to F_n /F_{n+1}
\overset{p_{\ast}}\to\pi_n
/\pi_{n+1} \to 0, $$
where $\hat{p}$ is defined by the formula
$$\hat{p} (\alpha )=\sum_{I}(\langle u_I \rangle
\cdot\alpha )\ t_I, $$
where $\alpha \in H_2(\pi)$, the summation is over $I$ of length $n$,
$\cdot : H^\ast \otimes H_\ast \to \mathbb Z$ is the
evaluation map, and where
the right hand side is asserted to lie in $\mathsf{L}_n (H)=F_n /F_{n+1}
\subseteq\mathbb Z[\![t_1 ,\dots,t_m ]\!]$.
\end{corollary}
\begin{proof}
Apply Stallings five-term exact sequence to the short exact
sequence of groups $1\to\pi_n\to\pi\to\pi /\pi_n\to 1$ to obtain
$$ H_2 (\pi )\to H_2 (\pi /\pi_n )\to \pi_n /\pi_{n+1}\to 1. $$
Combining this with the map $p$ gives the commutative diagram
$$\begin{diagram}
\divide\dgARROWLENGTH by2
\node{H_2 (\pi )}\arrow{e}\node{H_2 (\pi /\pi_n )}\arrow{e}\node{\pi_n
/\pi_{n+1}}\arrow{e}\node{1} \\
\node[2]{H_2 (F/F_n )}\arrow{n,l}{\cong}\arrow{e,t}{\cong}\node{F_n
/F_{n+1}}\arrow{n,r}{p_{*}}
\end{diagram}$$
This diagram yields the exact sequence of the corollary, where
$\hat{p}$ is
defined as the composition
$$H_2 (\pi )\to H_2 (\pi /\pi_n )\cong H_2 (F/F_n )\cong
F_n /F_{n+1} $$
To prove the formula for $\hat{p}$ first note that, for any $\alpha\in H_2
(F/F_n )\cong F_n /F_{n+1}=\mathsf{L}_n (H)\subseteq\mathbb Z[\![t_1 ,\dots,t_m ]\!]_n$ we have
$$\alpha =\sum_{I}(\langle u_I \rangle \cdot\alpha )\ t_I, $$
as follows directly from Proposition ~\ref{th.um}.
But now the corollary follows
from the definition of $\hat{p}$ and naturality.
\end{proof}
This proves the second assertion of Theorem \ref{thm.1} for $K(\pi,1)$
spaces.
\begin{remark}
\lbl{rem.dep}
Given two choices $p, p'$ of maps as in Corollary \ref{cor.fi},
we get a commutative diagram:
$$
\divide\dgARROWLENGTH by2
\begin{diagram}
\node{H_2(\pi)}\arrow{e}\arrow{s,=}\node{F_n/F_{n+1}}\arrow{e}\arrow{s}
\node{\pi_n/\pi_{n+1}}
\arrow{e}\arrow{s,=}\node{0}\\
\node{H_2(\pi)}\arrow{e}\node{F_n/F_{n+1}}\arrow{e}\node{\pi_n/\pi_{n+1}}
\arrow{e}\node{0}
\end{diagram}
$$
where the middle map is the automorphism of $F_n /F_{n+1}\cong H_2
(F/F_n )$ defined by the composition of isomorphisms
$$H_2 (F/F_n )\overset{p_*}\to H_2 (\pi /\pi_n )\overset{p'_*}\gets H_2
(F/F_n ). $$ In particular, if $p'\equiv p\ \mod F_2$ then the two exact
rows are identical.
\end{remark}
The third assertion of Theorem \ref{thm.1} for $K(\pi,1)$ spaces
follows from the following
\begin{proposition}
\label{lem.cyc}
Let $p: F \to \pi$ be an $n$-equivalence\ and
$\alpha_1 ,\dots ,\alpha_{n+1}\in H^1 (\pi )$. Then we have:
$$\alpha_1\smallsmile} %%\smallsmile \langle \alpha_2 ,\dots ,\alpha_{n+1}\rangle =\langle \alpha_1 ,\dots, a_n
\rangle \smallsmile} %%\smallsmile\alpha_{n+1}.$$
\end{proposition}
\noindent
See also \cite{Kr} for a related result.
\begin{proof}
We will use Dwyer's formulation \cite{Dw} of the Massey products.
Choose
cocycles $a_i$ representing $\alpha_i$, for $1\le i\le n+1$. Since we are
assuming all Massey products of length less than $n$ are defined and
vanish,
we can choose cochains $a_{ij}$, for $1\le i<j\le n+2$, with the
exception
of the three cases
$$i=1,j=n+1\qquad i=2, j=n+2 \qquad i=1, j=n+2 $$
so that $a_{i,i+1}=a_i$ and
$\delta a_{rs}=\sum_{r<i<s}a_{ri}\smallsmile} %%\smallsmile a_{is}$. For two of the three
exceptional cases the cochains
$$b_{1,n+1}=\sum_{1<i<n+1}a_{1i}\smallsmile} %%\smallsmile a_{i,n+1}\quad\text{ and }\quad
b_{2,n+2}=\sum_{2<i<n+2}a_{2i}\smallsmile} %%\smallsmile a_{i,n+2}$$
the $b_{ij}$ are cocycles but not necessarily coboundaries. In fact
they
represent the Massey products $\langle \alpha_1 ,\dots ,\alpha_n \rangle $ and $\langle
\alpha_2,\dots
,\alpha_{n+1}\rangle $ respectively.
Thus $\langle \alpha_1 ,\dots ,\alpha_n \rangle \smallsmile} %%\smallsmile\alpha_{n+1}$ is represented by the
cocycle $b_{1,n+1}\smallsmile} %%\smallsmile a_{n+1,n+2}$
and $\alpha_1\smallsmile} %%\smallsmile \langle \alpha_2 ,\dots ,\alpha_{n+1}\rangle $ is represented by the
cocycle $a_{12}\smallsmile} %%\smallsmile b_{2,n+2}$.
Now consider the cochain
$$ c=\sum_{1<i<r<n+2}a_{1i}\smallsmile} %%\smallsmile a_{ir}\smallsmile} %%\smallsmile a_{r,n+2} $$
By grouping the terms in one way we see that
\begin{align*}
c &=a_{12}\smallsmile} %%\smallsmile (\sum_{2<r<n+2}a_{2r}\smallsmile} %%\smallsmile
a_{r,n+2})+\sum_{2<i<r<n+2}(a_{1i}\smallsmile} %%\smallsmile a_{ir}\smallsmile} %%\smallsmile a_{r,n+2})\\
&=a_{12}\smallsmile} %%\smallsmile b_{2,n+2}+\sum_{2<i<r<n+2}(a_{1i}\smallsmile} %%\smallsmile\delta a_{i,n+2})
\end{align*}
Grouping the terms in another way gives
\begin{align*}
c&=\sum_{1<i<r<n+1}(a_{1i}\smallsmile} %%\smallsmile a_{ir}\smallsmile} %%\smallsmile
a_{r,n+2})+(\sum_{1<i<n+1}a_{1i}\smallsmile} %%\smallsmile a_{i,n+1})\smallsmile} %%\smallsmile a_{n+1,n+2}\\
&=\sum_{1<r<n+1}(\delta a_{1r}\smallsmile} %%\smallsmile a_{r,n+2})+b_{1,n+1}\smallsmile} %%\smallsmile a_{n+1,n+2}
\end{align*}
Now subtracting these two formulae for $c$ gives
\begin{align*}
a_{12}\smallsmile} %%\smallsmile b_{2,n+2}-b_{1,n+1}\smallsmile} %%\smallsmile a_{n+1,n+2}&=\sum_{1<r<n+1}\delta
a_{1r}\smallsmile} %%\smallsmile a_{r,n+2}-\sum_{2<i<n+2}a_{1i}\smallsmile} %%\smallsmile\delta a_{i,n+2}\\
&=\delta (\sum_{2<i<n+1}a_{1i}\smallsmile} %%\smallsmile a_{i,n+2})
\end{align*}
since $a_{12}$ and $a_{n+1,n+2}$ are cocycles. Since the left side
of this
equation represents
$$\alpha_1\smallsmile} %%\smallsmile \langle \alpha_2 ,\dots ,\alpha_{n+1}\rangle -\langle \alpha_1 ,\dots ,\alpha_n
\rangle \smallsmile} %%\smallsmile\alpha_{n+1}$$
and the right side is a coboundary the proof is complete.
\end{proof}
This concludes the proof of Theorem \ref{thm.1} for $K(\pi,1)$ spaces.
The general case follows from the fact that the canonical map
$X \to K(\pi,1)$ induces an onto map $H_2(X)\to H_2(\pi)$.
\qed
\begin{proof}(of Corollary \ref{cor.1})
The first part is immediate from Theorem \ref{thm.1},
using the Poincar\'e duality isomorphism $H^\ast=
H_1(M,\mathbb Q)=H^2(M,\mathbb Q)$.
In local coordinates, it implies (see Corollary \ref{cor.fi})
that the Massey product
$$ \mu_n(M)
\in H \otimes \mathsf{L}_n(H) \cong \text{Hom}(H^1(M), \mathsf{L}_n(M))
$$
is given by
\begin{equation}
\lbl{eq.lam}
\mu_n(M) ( u) = \sum_{I} [M]\smallfrown} %%\smallfrown
(u \smallsmile} %%\smallsmile \langle u_I\rangle ) \ t_I,
\end{equation}
where $u \in H^1(M)$, the summation is over $I$ of length $n$,
$\smallsmile} %%\smallsmile$ indicates cup product and $[M]\smallfrown} %%\smallfrown$ indicates cap
product with the fundamental homology class of $M$.
Let $[ \cdot ]: H \otimes \mathsf{L}(H)\to\mathsf{L}(H)$ be the Lie algebra
bracket,
defined by $[a\otimes b]
=[a,b]$, for $a \in H, b \in \mathsf{L}(H)$.
If we regard $\mathsf{L}(H)\subseteq\mathbb Z[\![t_1 ,\dots,t_m ]\!]$ then $[ \cdot ]$ can be
expressed by the formula
$$[x_i\otimes c]=(1+t_i )c-c(1+t_i )=t_i c-ct_i.$$
Equation \eqref{eq.lam} implies that
$$\mu_n(M)=\sum_{i,I}x_i\otimes ([M]\smallfrown} %%\smallfrown
(u_i\smallsmile} %%\smallsmile \langle u_I\rangle ) )\ t_I $$
and so
$$[\mu_n(M)]=\sum_{i,I}
[M]\smallfrown} %%\smallfrown (u_i\smallsmile} %%\smallsmile \langle u_I\rangle ) \ (t_i t_I-t_I t_i ). $$
Thus Corollary \ref{cor.1} follows from the third assertion of
Theorem \ref{thm.1} (or its coordinate version, Proposition
\ref{lem.cyc}).
\end{proof}
\subsection{Realization results}
\lbl{sub.real}
\begin{proof}(of Theorem \ref{th.alf})
Given an element $\phi\in \mathrm{A}_0 (F/F_{n} )$ we construct maps
$f^{\pm}:\Sigma_{g,1}\to K(F/F_{n},1)$,
where $f^{+}_{*}:\pi_{1}(\Sigma_{g,1} )\to F/F_{n}$ corresponds to the
canonical projection $p :F\to F/F_{n}$ under the identification
of $\pi_{1}(\Sigma_{g,1} )$ with $F$, and $f_{*}^{-}=\phi\circ p$. Since
$\phi_{*}(\omega_g )=\omega_g $, we have
$f^{+}|\partial\Sigma_{g,1}\simeq f^{-}|\partial\Sigma_{g,1}$ and so
we can
combine the two maps to define a map
$f:\hat{\Sigma}_{g,1}\to K(F/F_{n},1)$, where
$\hat{\Sigma}_{g,1}$ is the double of $\Sigma_{g,1}$. We would like to
extend
$f$ to a map $\Phi:M\toK(F/F_{n},1)$, for some compact orientable
$3$-manifold with $\partial M=\hat{\Sigma}_{g,1}$ the obstruction to the
existence of $\Phi$ is the element $\theta\in\Omega_{2}(F/F_{n})\cong
H_{2}(F/F_{n})$ represented by $f$, where $\Omega_{*}(F/F_{n})$
are
the oriented bordism groups of $F/F_{n}$. Since
$H_{2}(F/F_{n})\not= 0$ we must be careful in our choices to
assure that $\theta =0$. Redo the construction of $f^{+},f^{-}$
and $f$ but using $K(F/F_{n+1},1)$ instead of $K(F/F_{n},1)$ and using
a lift of $\phi$ to an automorphism $\bar \phi$ of $F/F_{n+1}$
instead of $\phi$. Our restriction on $\phi$ assures that $\bar
\phi(\omega_g )=\omega_g$ and so we obtain
$\bar f:\hat{\Sigma}_{g,1}\to K(F/F_{n+1},1)$ and
an obstruction element $\bar\theta\in\Omega_{2}(F/F_{n+1})$. Now
this element may not be zero, but since the projection map
$H_{2}(F/F_{n+1})\to H_{2}(F/F_{n})$ is zero, and clearly
$\bar\theta$ maps to $\theta$, we conclude that $\theta =0$.
Thus $f$ extends to the desired $\Phi:M\toK(F/F_{n},1)$.
Let $i^{\pm}:\Sigma_{g,1}\to\partial M$ be the obvious diffeomorphisms onto
the
domains of $f^{\pm}$. It is clear that if $M$ were a homology
cylinder over $\Sigma^{+}$, then $\sigma_{n}(M)=\phi$. But this is not
necessarily true and so we will perform surgery on the map $\Phi$,
adapting the arguments in \cite{KM} to our situation. See also
\cite[Theorem 1]{Tu} for similar surgery arguments which are used to show
that any finite $3$-dimensional Poincar\'e complex is homology equivalent
to a closed $3$-manifold.
\begin{lemma}
Suppose $\alpha\in\ker \Phi_{*}:H_{1}(M)\to H$. Then there exists
$\bar\alpha\in\pi_{1}(M)$ such that $\bar\alpha\in\ker \Phi_{*}:\pi_{1}(M)\to
F/F_{n}$ and $\bar\alpha$ represents $\alpha$.
\end{lemma}
\begin{proof}
If $\bar\alpha\in\pi_{1}(M)$ is any representative of $\alpha$, then
$\Phi_{*}(\bar\alpha )\in F_{2}/F_{n}$. Choose an element
$\beta\in\pi_{1}(\Sigma^{+})_{2}$ so that $\Phi_{*}(\beta )=f^{+}_{*}(\beta
)=\bar\alpha$. Then $\bar\alpha\beta^{-1}\in\ker \Phi_{*}$ and $\bar\alpha\beta^{-1}$
represents $\alpha$.
\end{proof}
Thus for any $\alpha\in\ker \Phi_{*}:H_{1}(M)\to H$ we can do surgery on
a curve representing $\alpha$ and extend $F$ over the trace of the
surgery.
The first step in killing $\ker \Phi_{*}$ will be to kill the
torsion-free part. Note that $H_{1}(M)\cong H_{1}(\Sigma^{+})\oplus\ker
\Phi_{*}$, since $\Phi_{*}\circ i^{+}_{*}$ is an isomorphism, and
so,
under the canonical map $H_{1}(M)\to H_{1}(M,\partial M),\ \ker \Phi_{*}$
maps onto $H_{1}(M,\partial M)$. Choose an element $\alpha\in\ker \Phi_{*}$
which maps to a primitive element of $H_{1}(M,\partial M)$. Now surgery
on a simple closed curve $C$ representing $\alpha$ will produce a new
manifold $M'$ so that, if $\beta\in H_{1}(M')$ is the element
represented by the meridian of $C$, then
\begin{equation}\lbl{eq.surg}
H_{1}(M)/\langle \alpha \rangle \cong H_{1}(M')/\langle \beta \rangle
\end{equation}
(see \cite{KM}). Since $\alpha$ is primitive in $H_{1}(M,\partial M)$,
there is a $2$-cycle $z$ in $M$ whose intersection number
with $C$ is $+1$. Thus the intersection of $z$ with $M'$ is
a $2$-chain whose boundary is $\beta$. So, by Equation
\eqref{eq.surg}, $H_{1}(M')\cong H_{1}(M)/\langle \alpha \rangle $.
A sequence of such surgeries will kill the torsion-free part of
$H_{1}(M,\partial M)$. But this implies that $\ker \Phi_{*}$ is torsion
by the following simple homology argument. Consider the exact
sequence:
$$0\to H_{2}(M)\to H_{2}(M,\partial M)\to H_{1}(\partial M)\to H_{1}(M)\to
H_{1}(M,\partial M)\to 0
$$
Since $\operatorname{rank} H_{2}(M)=\operatorname{rank} H_{1}(M,\partial M)=0$ and $H_{1}(\Sigma^{+})$
imbeds into $H_{1}(M), \operatorname{rank} H_{2}(M,\partial M)=\operatorname{rank} H_{1}(M)\ge 2g$.
But, since $\operatorname{rank} H_{1}(\partial M)=4g$, we conclude that $\operatorname{rank}
H_{1}(M)=2g$. Therefore
$$ 2g=\operatorname{rank} H_{1}(M)=\operatorname{rank} H_{1}(\Sigma^{+})+\operatorname{rank}\ker \Phi_{*} $$
and so $\operatorname{rank}\ker \Phi_{*}=0$.
We now follow the argument in \cite{KM} to kill the torsion group
$T=\ker \Phi_*$. The linking pairing $l:T\otimes T\to\mathbb Q /\mathbb Z$ is
non-singular
since $T=\operatorname{tors} H_1 (M)$ maps isomorphically to
$\operatorname{tors} H_1 (M,\partial M)=H_1 (M,\partial M)$. According to \cite[Lemma 6.3]{KM}
if,
for $\alpha\in T$, $l(\alpha ,\alpha )\not=0$, then we can choose the normal
framing to
any closed curve $C$ representing $\alpha$ so that the element $\beta\in H_1
(M')$
is of finite order smaller than the order of $\alpha$. Thus the torsion
subgroup
of $H_1 (M')$ is smaller than $T$. Continuing in this way we reach
the point
where all the self-linking numbers are $0$. According to \cite[Lemma
6.5]{KM}
this implies that $T$ is a direct sum of copies of $\mathbb Z /2$. Now
choose any
non-zero element $\alpha\in T$. We will show that surgery on $\alpha$ reduces
the
rank of $H_1 (M,\partial M;\mathbb Z /2)$. Denote by $V$ the trace of the
surgery and $M'$
the result of the surgery. Then we have a diagram of homology groups
(coefficients in $\mathbb Z /2$) with exact row:
$$
\divide\dgARROWLENGTH by2
\begin{diagram}
\node{H_2 (V,\partial M')}\arrow{e}\node{H_2 (V,M')}\arrow{e}\node{H_1
(M',\partial M')}\arrow{e}\node{H_1 (V,\partial M')}\arrow{e}\node{0}\\
\node{H_2 (M,\partial M)}\arrow{n}\arrow{ne,..}
\end{diagram}
$$
Now $H_1 (V,\partial M')\cong H_1 (M,\partial M)/\langle \alpha \rangle $ and so has rank
one less than $H_1 (M,\partial M)$. Since $H_2 (V,M')$ is generated by the
transverse disk bounded by the meridian curve representing $\beta$, the
dotted
arrow can be interpreted as the ($\mathbb Z /2$) intersection number with
$\alpha$. By
Poincar\' e duality this map is non-zero and so
$H_1 (M',\partial M')\cong H_1 (V,\partial M')$, proving the claim. As in
\cite{KM}
we can assume the normal framing chosen so that $\beta$ has order $2$ or
$\infty$. Thus the possibilities for $H_1 (M',\partial M';\mathbb Z)$ are either
$\mathbb Z\oplus (s-2)\mathbb Z /2$ or $\mathbb Z /4\oplus (s-2)\mathbb Z /2$, where
$s=\operatorname{rank} H_1 (M,\partial M)$. We can then do a surgery to kill the $\mathbb Z$
factor,
in the first case, or reduce the order of $H_1 (M',\partial M';\mathbb Z )$,
in the second case. Continuing this way we eventually kill $\ker
\Phi_*$,
producing the desired $(M,\Phi)$.
\end{proof}
\begin{proof}(of Proposition \ref{prop.DA})
Let $\mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n(H)$ denote the kernel of the natural
projection
$\mathrm{A}_0(F/F_{n+1})\to\mathrm{A}_0(F/F_{n})$.
We first construct a map $D_n:\mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n(H)\to \mathsf{D}}%%%{\mathsf{L}^c_n(H)$ as
follows.
If $h \in \mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n(H)$ we can write $h(a)=a \psi(a)$,
where $\psi(a) \in
F_n/F_{n+1}\cong \mathsf{L}_n(H)$.
Then, we define
$D_n(h)([a])=\psi(a)$,
where $[a] \in H$ and $a \in F/F_n$ is a lift of $[a]$.
Using the isomorphism $\text{Hom}(H, \mathsf{L}_n(H)) \cong H \otimes
\mathsf{L}_n(H)$
this defines a map (denoted by the same name)
$$\mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n(H)\to H \otimes \mathsf{L}_n(H)$$
with corresponding description in local coordinates given by
$$D_n(h)=\sum_{i} x_i \otimes \psi(y_i)- y_i \otimes \psi(x_i) \in
H \otimes \mathsf{L}_n(H).$$
If $h\in \mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n (H)$, as above, then $\prod_i [h(x_i
),h(y_i )]\equiv\prod_i
[x_i ,y_i ]\mod F_{n+2}$ and so
$$\prod_i [x_i\psi (x_i ),y_i\psi (y_i )]\equiv\prod_i [x_i ,y_i ][\psi
(x_i
),y_i ][x_i ,\psi (y_i )]\mod F_{n+2} $$
Therefore $\prod_i [\psi (x_i ),y_i ][x_i ,\psi (y_i )]\in F_{n+2}$,
which implies that $D_n(h) \in \mathsf{D}}%%%{\mathsf{L}^c_n(H)$.
It is clear that $D_n$ is one-to-one.
We now show that it is onto.
Suppose we have an element $\theta =\sum_i (x_i\otimes\alpha_i
-y_i\otimes\beta_i )\in\mathsf{D}}%%%{\mathsf{L}^c_n(H) $. Lift $\alpha_i ,\beta_i$ into $F_n$
(denoted by
the same symbols) and define an endomorphism $h$ of $F$ by
$$h(x_i )=x_i\alpha_i ,\ h(y_i )=y_i\beta_i .$$
It follows by Stalling's theorem \cite{St} that
$h$ induces an automorphism of $F/F_{n+1}$ which restricts to the
identity automorphism of $F/F_n$. We note that
$$h(\prod_i [x_i ,y_i ])=\prod_i [x_i\alpha_i ,y_i\beta_i ]\equiv\prod_i [x_i
,y_i
][x_i ,\alpha_i ][\beta_i ,y_i ]\mod F_{n+2}$$
But $\prod_i [x_i ,\alpha_i ][\beta_i ,y_i ]$ represents
the image of $theta$ under the Lie bracket
$H \otimes \mathsf{L}_n(H) \to \mathsf{L}_{n+1}(H)$, which vanishes since
$\theta \in \mathsf{D}}%%%{\mathsf{L}^c_n(H)$; thus $\prod_i [x_i ,\alpha_i ][\beta_i ,y_i ] \in
F_{n+2}$.
This shows that $h\in\mathrm{A}_0 (F/F_{n+1})$ and clearly
$D_n (h)=\theta$.
The fact that the projection
$\mathrm{A}_0(F/F_{n+1})\to\mathrm{A}_0(F/F_{n})$ is onto follows
immediately from Theorem \ref{th.alf}. It is not hard, however, to give a
direct argument; we leave this as an exercise for the reader.
Finally, it is clear by the definitions that the diagram in
Proposition \ref{prop.DA} commutes, and that the sequence below it is exact.
\end{proof}
\begin{remark}
The action of $\mathrm{A}_0(F/F_{n})$ on $\mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n(H)\cong
\mathsf{D}}%%%{\mathsf{L}^c_n(H)$ induced by conjugation by elements of $ \mathrm{A}_0(F/F_{n+1})$
coincides with the natural action of
$\mathrm{A}_0(F/F_{2})\cong\text{Sp}(2g,\mathbb Z )$ on $ \mathsf{D}}%%%{\mathsf{L}^c_n(H)$, via the
projection $ \mathrm{A}_0(F/F_{n})\to \mathrm{A}_0(F/F_{2})$.
\end{remark}
\begin{proof}(of Theorem \ref{thm.2})
Let $M\in\mathcal H_{g,1}$ and define $S(M)^o=T_+\cup M\cup T_-$, where
$T_{\pm}$
are two copies of the solid handlebody $T$ of genus $g$, which are
attached
to $\partial M$ via the diffeomorphisms $i^{\pm}$ so that, referring to a
basis
$\{ x_i ,y_i\}$ of $F$ corresponding to a symplectic basis of $H$,
the $\{
x_i\}$ are represented by the boundaries of meridian disks in $T$.
Thus
$\pi_1 (T)=F'$, the free group generated by $\{ y_i\}$ (or, more
precisely,
their images in $\pi_1 (T)$). $S(M)^o$ is a 3-manifold with boundary
$S^2$,
which we can fill-in to obtain a closed 3-manifold $S(M)$.
If $M\in\mathcal H_{g,1}[n]$, then the inclusion
$T_+\subseteq S(M)$ induces an isomorphism $p :F'/F'_n\cong \pi_1
(S(M))/\pi_1
(S(M))_n$ and we can consider $\mu_n(S(M),p) \in \mathsf{D}}%%%{\mathsf{L}^c_n(H')$,
where $H'=H_1 (F')$.
Suppose that $\sigma_{n+1}(M)=h\in\mathrm{A}_0 (F/F_{n+1})$.
Set $a_i =\rho (h(x_i ))\in
F'_n$, where $\rho :F\to F'$is the projection defined by $\rho (x_i
)=1$.
Then $ \mu_n(S(M),p)=\sum_i [y_i ]\otimes [a_i ]$, where $[y_i ]
\in H' ,[a_i ] \in \mathsf{L}_n(H')$
are the classes represented by $y_i ,a_i$. This assertion is just the
obvious generalization of Corollary \ref{cor.1} and the proof is the
same.
Now let $\sum_i [y_i ]\otimes [\lambda_i ]$ be an arbitrary element in
$\mathsf{D}}%%%{\mathsf{L}^c_n(H')$, where $\lambda_i\in F'_n$, i.e. $\prod_i
[y_i ,\lambda_i ]\in F_{n+2}$. We want to construct $(N,p )$ such that
$\mu_n(N,p)=\sum_i [y_i ]\otimes [\lambda_i ]$. Consider the
endomorphism $h$ of $F$ defined by
\begin{equation}\lbl{eq.O}
\begin{split}
h(x_i )&=x_i\lambda_i \\
h(y_i )&=\lambda_i^{-1} y_i\lambda_i.
\end{split}
\end{equation}
Denote also by $h$ the induced
automorphism of $F/F_{n+1}$.
To see that $h\in\mathrm{A}_0 (F/F_{n+1})$, we compute
$$\prod_i [h(x_i ),h(y_i )]=\prod_i (x_i y_i x_i^{-1}\lambda_i^{-1}
y_i^{-1}\lambda_i
)=\prod_i [x_i ,y_i ][y_i ,\lambda_i^{-1} ]\equiv\prod_i [x_i ,y_i ]\mod
F_{n+2}.$$
Therefore, by Theorem \ref{th.alf}, $h=\alpha_n (M)$ for some
$M\in\mathcal H_{g,1}$. Since $\lambda_i\in F'_n$ we have $h\in \mathsf{D}}%%%{\mathsf{L}^c^{\mathrm{a}}_n(H)$
and so $M\in\mathcal H_{g,1}[n]$.
By the discussion above, $\mu_n(N,p)=\sum_i [y_i ]\otimes
[\lambda_i ]$.
\end{proof}
For completeness, we close this section by a sketch of a more direct
proof of Theorem \ref{thm.2} using the results of \cite{O1,IO}.
Similar arguments can be found in \cite{CGO}.
\begin{lemma}
\lbl{lem.real}
For every $\alpha \in H_3(F/F_n)$ there is a closed 3-manifold $M$
and an $n$-equivalence\ $p:F\to\pi\overset{\text{def}}{=} \pi_1(M)$ such that $p_\ast[M]=\alpha$.
\end{lemma}
\begin{remark}
Since $\Omega_3 (F/F_n )\cong H_3 (F/F_n )$, it follows
that every element $\alpha \in \Omega_3 (F/F_n )$ is represented by
some closed 3-manifold $M$ and map $p :\pi\overset{\text{def}}{=} \pi_1(M)\to F/F_n$
so that
$p_* [M]=\alpha$. The point is to arrange that
$p_* :H_1 (M)\cong H_1
(F/F_n )$, which would imply that $p$ is an $n$-equivalence .
\end{remark}
\begin{proof}
We apply the constructions and results of \cite{O1}. Consider the
mapping cone $K_n$ of the natural
map $K(F,1)\to K(F/F_n,1)$ of Eilenberg-MacLane spaces. $K_n$ is
constructed from $K(F/F_n,1)$ by adjoining
$2$-cells $e^2_i$ along the generators $x_i\in F\twoheadrightarrow
F/F_n$.
Then $K_n$ is simply-connected and $H_i (K_n )\cong H_i (F/F_n )$ for
$i\ge
2$. So we have the Hurewicz epimorphism $\rho :\pi_3 (K_n
)\twoheadrightarrow
H_3 (K_n )\cong H_3 (F/F_n )$, where $\pi_3$
denotes the third homotopy group. Suppose $\rho (\theta )=\alpha$. Then
the
Pontrjagin-Thom construction gives us a map $f:S^3 \to K_n$
representing
$\theta$, such that, if $x_i\in e^2_i$ is some interior point, then $f^{-1}
(x_i )=L_i$, a zero-framed imbedded circle in $S^3$, see \cite{O1}.
Now let
$M^3$ be the result of framed surgery on $S^3$ along the $\{ L_i\}$.
Then
$f$ induces a map $p :M\to K(F/F_n )$ and it is clear that $p_*
[M]=\alpha$. Finally we note that $p_* :H_1 (M)\cong H_1 (F/F_n )$.
\end{proof}
From another viewpoint, we have
defined
a homomorphism $\mu_n ' :H_3 (F/F_n )\to \mathsf{D}}%%%{\mathsf{L}^c_n(H)$ by the formula
\begin{equation}\lbl{eq.hom}
\mu_n '(\alpha )=\sum_{i,I} x_i\otimes (\alpha\smallfrown (u_i\smallsmile} %%\smallsmile
\langle u_I \rangle ))t_I
\end{equation}
and so, by Corollaries \ref{cor.1} and \ref{cor.lam} $\mu_n (M,p )=\mu_n '
(p_* [M])$. It follows from Theorem \ref{thm.1} that the kernel of $\mu_n
'$ is precisely
the image of
$H_3 (F/F_{n+1})\to H_3 (F/F_n )$, inducing, therefore, an injection
$\mathrm{cok} ( H_3 (F/F_{n+1})\to H_3 (F/F_n ))
\rightarrowtail \mathsf{D}}%%%{\mathsf{L}^c_n(H)$. But it is shown in
\cite{O1,IO} that both sides are free
finitely generated abelian groups of the same rank and so they are
isomorphic. In fact K. Orr has pointed out to us that a
straightforward
examination of the spectral sequence of the group extension
$F_{n}/F_{n+1}\to F/F_{n+1}\to F/F_{n}$ shows that the rank of
$\mathrm{cok} ( H_3 (F/F_{n+1})\to H_3 (F/F_n ))$ is at least that of $\mathsf{D}}%%%{\mathsf{L}^c_n(H)$.
\qed
\subsection{Massey products and the Johnson homomorphism}
In this section we will give a proof of Theorem \ref{thm.3}.
We first recall the definition of the Johnson homomorphism. Let
$\mathrm{A}_0 (F/F_n)$ be as in Section \ref{sub.homcyl}.
We define the Johnson homomorphism
$$\tau_n :\mathrm{A}_0 (F/F_n)\to
\hom (H,\mathsf{L}_n (H))$$
where $H=H_1 (F)$,
as follows. If $\phi\in \mathrm{A}_0 (F/F_n)$,
then $\tau_n (\phi )\cdot [\alpha ]=\alpha
\phi(\alpha^{-1} )\in F_n /F_{n+1}\cong\mathsf{L}_n (H)$, where $[\alpha ]\in H$ is the
homology
class of $\alpha\in F$. Let $K(\phi )\subseteq F$ be the normal closure of all
elements of
the form $\alpha\phi(\alpha^{-1} )$ for $\alpha\in F$; note that $K(\phi )\subseteq F_n$ if
$ \phi\in \mathrm{A}_0 (F/F_n)$.
Let $\pi =F/K( \phi )$. Then the projection $F\to\pi$
induces an isomorphism $p_{ \phi}:F/F_n\overset\cong\to \pi /\pi_n$ and
the
homomorphism
$\hat p_{ \phi }$ defined (as $\hat p$) in Corollary~\ref{cor.fi} is given
by the
composition
$$H_2 (\pi )\cong K( \phi ) /[F,K(\phi ) ]\to F_n /F_{n+1}\cong\mathsf{L}_n (H) $$
The first isomorphism is given by Hopf's theorem. Now consider the
natural
homomorphism $i:H\to H_2 (\pi )$ defined by $i([\alpha ])=[\alpha\phi(\alpha )^{-1}
]$, then
\begin{equation}\lbl{eq.J}
\tau_n (\phi )=\hat p_{\phi}\circ i.
\end{equation}
Now let $\Sigma$ be a compact orientable surface of genus $g$ with one
boundary
component. Then $\pi_1 (\Sigma )\cong F$, the free group on $2g$
generators.
By Nielsen's theorem $\Gamma_{g,1}\subseteq \mathrm{A}(F)$ and so $\Gamma_{g,1}
[n]=\Gamma_{g,1}\cap \mathrm{A}_0 (F/F_n)$.
For $ \phi\in\Gamma_g$ define an associated closed $3$-manifold $T_{\phi}$ as
follows. Let $T'_{\phi}$ be the mapping torus of $ \phi$, i.e. $I\times\Sigma$
with
$1\times\Sigma$ identified with $0\times\Sigma$ by the homeomorphism $(1,x)\to
(0, \phi(x))$ for every $x\in\Sigma$. Since
$ \phi |\partial\Sigma =\text{id}$ there is a canonical identification of $\partial
T'_{\phi}$ with
the torus
$S^1\times S^1$. $T_{\phi}$ is defined by pasting in $D^2\times S^1$. Note
that $\pi_1 (T_{\phi})\cong F/K(\phi )$, where $K(\phi )$ was defined above.
A theorem of the following sort was first suggested by Johnson
\cite{Jo1} and a
proof (which we sketch, for completeness) was given in \cite{Ki}.
\begin{proposition}\lbl{cor.2}
$\tau_n (\phi )=\mu_n (T_{\phi })$, where $H_1 (\Sigma )$ and $H^1 (T_{\phi})$ are
identified by
the string of isomorphisms
$$H^1 (T_{\phi})\overset\cong\to H^1 (\Sigma )\cong H_1 (\Sigma )$$
induced by the inclusion $\Sigma\subseteq T_{\phi}$ and Poincar\'e duality for
$\Sigma$.
\end{proposition}
\begin{proof}
It follows from the definitions that we need to establish the
commutativity of
the following diagram
$$\begin{diagram}
\node{H_1 (\Sigma )=H}\arrow[2]{e,t}{i}\arrow{s,l,<>}{\cong}\node[2]{H_2
(\pi )}\\
\node{H^1 (\Sigma )}\node{H^1
(T_{\phi})}\arrow{w,t}{j^*}\arrow{e,t,<>}{\cong}\node{H_2
(T_{\phi})}\arrow{n,r}{\theta}\arrow{wnw,t}{j_*}
\end{diagram}$$
where $i$ is defined in equation~\eqref{eq.J}, $j:\Sigma\subseteq I \times 0 \subseteq
I\times\Sigma\to
T_{\phi}$
is the inclusion map inducing $j^*$ and the Gysin homomorphism $j_*$,
and
$\theta$ is the Hopf map $H_2 (X)\to H_2 (\pi_1 (X))$ when $X=T_{\phi}$.
Now suppose $\alpha\in\pi_1 (\Sigma )$ is represented by a $1$-cycle $z$ in
$\Sigma$. since $\phi(z)$ is homologous to $z$, there exists a $2$-chain $c$
in $\Sigma$ such that $\partial c=\phi(z)-z$. Consider the chain $I\times z$ in
$I\times \Sigma\to T_{\phi}$ with boundary $1\times z-0\times z$. Since
$0\times
z$ is identified with $1\times \phi(z)$ in $T_{\phi}$, the chain $\xi
=I\times
z+1\times c$ is a $2$-cycle in $T_{\phi}$---let $\beta$ denote its homology
class in $H_2 (T_{\phi})$.
It follows from the definition of $\theta$ that $\theta (\beta )=i([\alpha
])$.
On the other hand $j_* (\beta )=[\alpha ]$ since the $2$-cycle $\xi$
intersects
$\epsilon\times\Sigma$ transversely in the $1$-cycle $\epsilon\times z$, if $0<\epsilon <1$.
This establishes the desired commutativity.
\end{proof}
Combining Proposition~\ref{cor.2} and Corollary~\ref{cor.lam} we have
\begin{corollary}\lbl{cor.mas}
If $\{ x_i\}$ is a basis of $H=H_1 (\Sigma )\cong H^1 (T_{\phi})$ and $\{
u_i\}$
is the dual basis of $H_1 (T_{\phi})$, then
$$ \tau_n (h)=\sum_{i,I}x_i\otimes
([T_{\phi}]\smallfrown
(u_i\smallsmile \langle u_{I}\rangle )) t_{I} $$
using the
inclusion $\mathsf{L}_n (H)\subseteq\mathbb Z[\![t_1 ,\dots,t_{2g}]\!]_n$.
\end{corollary}
We now turn to the proof of Theorem \ref{thm.3}. Fix an $n$-equivalence\
$p: F \to \pi_1(M)$, an imbedding
$\iota: \Sigma \to M$ of a (closed) surface in a (closed) 3-manifold $M$,
and
an element $\phi \in \Gamma_g[n]$.
The homomorphisms $f: \pi_1(M)\to F/F_{n+1}$
and $f_{\phi}: \pi_1(M_{\phi})\to F/F_{n+1}$ induced by the $n$-equivalence\
$p$
(in the discussion before the statement of the theorem) determine
maps $f: M \to K(F/F_{n+1},1)$ and $f_{\phi}: M_{\phi} \to
K(F/F_{n+1},1)$
in the Eilenberg-MacLane space $K(F/F_n,1)$. We will construct a
cobordism
$F: V^4 \to K(F/F_{n+1})$ between $f$ and $f_{\phi}$.
First we push $\iota(\Sigma)$ in the positive and negative normal
directions in
$M$ to obtain two copies of $\iota(\Sigma)$, $\Sigma_p$ and $\Sigma_n$,
respectively.
Then we attach $I\times\iota(\Sigma)$ to $M$ along its boundary by
identifying
$0\times\iota(\Sigma)$ to $\Sigma_n$ by the homeomorphism $(0,x)\to x$, for
$x\in\iota(\Sigma)$, and $1\times\iota(\Sigma)$ to $\Sigma_p$ by $(1,x)\to
\phi(x)$.
We can
now thicken up $I\times\iota(\Sigma)$ and $M$ to obtain a manifold $W$,
see
Figure \ref{W}.
\begin{figure}[htpb]
$$ \eepic{W}{0.04} $$
\caption{The manifold $W$. The handle shown represents a thickening
of $\iota(\Sigma)$, shown as the core.}\lbl{W}
\end{figure}
Note that the boundary of $W$ consists of three components:
$M$,$M_\phi$ and the mapping torus $T_{\phi}$ of $\phi$ on
$\iota(\Sigma)$. We make one small
modification to obtain $V$. Remove a disk $D$ from $\iota(\Sigma)$ to
obtain
a surface $\Sigma^o$ with boundary and lift $\phi$ to a diffeomorphism
$\phi^o$
of $\Sigma^o$.
Then $S^1\times D$ is naturally
imbedded in $T_\phi$, and so in the boundary of $W$. We attach
$D^2\times D$ to $W$ along $S^1\times D$ to obtain $V$.
The boundary of $V$ is now given by $\partial V=M_\phi
-T_{\phi^o}-M$. It is
not difficult to check that the inclusions $M\to V\gets M_\phi$
are $(n+1)$-equivalences since
since $\phi\in\Gamma_g [n+1]$.
Then $f,f_{\phi}$ extend to a map $F:V\to K(F/F_{n+1} ,1)$.
In addition the inclusions
$\Sigma^o\subseteq T_{\phi^o}\subseteq V$ induces isomorphisms
$$F/F_{n+1}\cong\pi_1 (\Sigma^o )/\pi_1 (\Sigma^o)_{n+1} \cong
\pi_1 (T_{\phi^o})/\pi_1 (T_{\phi^o})_{n+1} \cong \pi_1 (V)/\pi_1
(V)_{n+1}$$
So the bordism $(F,V)$ implies that $(f_{\phi})_* [M_\phi ]=f_* [M]
+(p_{\phi})_*
[T_{\phi^o}] \in \Omega_3(F/F_{n+1}) \cong H_3(F/F_{n+1})$.
Applying the map of Equation \eqref{eq.h3} concludes the proof of
Theorem
\ref{thm.3}.
\qed
\begin{remark}
\lbl{rem.hcc}
Theorem \ref{thm.3} and Proposition \ref{cor.2} can be
generalized easily to the context of homology cylinders, by essentially the
same arguments. For any homology cylinder $N$ there is an obvious notion
of mapping torus $T_N$---then Proposition \ref{cor.2} can be rephrased,
replacing $T_{\phi}$ by $T_N$, $\phi\in\Gamma_{g,1}[n]$ by $N\in\mathcal H_{g,1}[n]$
and $\tau_n$ by $\varsigma_n$. For Theorem \ref{thm.3} we consider
$N\in\mathcal H_{g,1}$ and an imbedding $\Sigma\subseteq M$, where $\Sigma$ is a closed
orientable surface of genus $g$. We can cut $M$ open along $\Sigma$ and paste
in $\bar N$---where $\bar N$ is obtained from $N$ by filling it in, in the
obvious way, to get a homology cylinder over $\Sigma$---and so obtain a new
manifold $M_N$. Then Theorem \ref{thm.3} can be rephrased, replacing
$M_{\phi}$ by $M_N$.
\end{remark}
\begin{remark}\lbl{rem.fil}
We can consider another filtration of $\mathcal H_{g,1}$. Let $\mathcal H_{g,1}(n)$ denote
the set of all homology cylinders $n$-equivalent to $\Sigma_{g,1}\times I$ (see Section
\ref{sub.fti}). Thus if $M\in\mathcal H_{g,1}(n)$ then $M-(\Sigma_{g,1}\times I )\in\mathcal F_n^Y (\Sigma_{g,1}\times I
)$. It is easy to see that $\mathcal H_{g,1}(n)$ is a subsemigroup of $\mathcal H_{g,1}$.
It is a natural conjecture that the quotient $\mathcal H_{g,1}/ \mathcal H_{g,1}(n)$ is a
group.
According to \cite{GGP} we get the same filtration if we ask that $M$ be
obtained from $\Sigma_{g,1}\times I$ by cutting open along some imbedded closed orientable
surface $\Sigma '\subseteq\Sigma_{g,1}\times I$ and reattaching by some element of $\mathcal T_n $, the
$n$-th lower central series subgroup of the Torelli group $\mathcal T$ of $\Sigma '$.
It is clear that $\mathcal H_{g,1}(n)\subseteq \mathcal H_{g,1}[n]$ since the effect of cutting
and reattaching in $M$ by an element of $\mathcal T_n$ does not change $\pi_1
(M)/\pi_1 (M)_n$. If $\mathcal G\mathcal H_{g,1}(*)$ and $\mathcal G\mathcal H_{g,1}[*]$ denote the
associated graded groups of these filtrations then we have a natural map $
\mathcal G\mathcal H_{g,1}(*)\to \mathcal G\mathcal H_{g,1}[*]$. Note also that the natural homomorphism
$\Gamma_{g,1}\to\mathcal H_{g,1}$ induces maps $(\mathcal T_{g,1})_n\to\mathcal H_{g,1}(n)$ and so
$\mathcal G (\mathcal T_{g,1})_*\to\mathcal G\mathcal H_{g,1}(*)$.
Putting this, and some of the other maps constructed in this paper, all
together, we have a commutative diagram:
$$
\divide\dgARROWLENGTH by2
\begin{diagram}
\divide\dgARROWLENGTH by2
\node[3]{\mathcal G (\mathcal T_{g,1})_*}\arrow[2]{sw}\arrow[2]{s}\arrow[2]{e}\node[2]{\mathsf{D}}%%%{\mathsf{L}^c_*
(H)}\arrow[2]{s,r}{\cong}\\ \\
\node{\mathcal A^c (\Sigma_{g,1}\times I
)}\arrow[2]{ese}\arrow[2]{e,..}\arrow[2]{se}\node[2]{\mathcal G\mathcal H_{g,1}(*)}
\arrow{s,-}\arrow[2]{e} \node[2]{\mathcal G\mathcal H_{g,1}[*]}\arrow[2]{s,r}{\cong}\\
\node[3]{}\arrow{s}\\
\node[3]{\mathcal G^Y\mathcal M (\Sigma_{g,1}\times I )}\node[2]{\mathcal A^t (\Sigma_{g,1}\times I )}
\end{diagram}
$$
The dotted arrow denotes a conjectured lifting.
\end{remark}
\section{Questions}
\lbl{sec.que}
It is well known that there is a set of moves that generates (string) link
concordance, \cite{Tr}. These moves, together with
the existence of the Kontsevich integral, were the key to the proof
that the tree-level part of the Kontsevich integral of string-links
is given by Milnor's invariants, or equivalently, by Massey products,
see \cite{HM}.
\begin{question}
Is there a set of local moves that generates homology cobordism of homology
cylinders?
\end{question}
\begin{question}
Does Theorem \ref{thm.6} generalize to homology cylinders, using
our extension of the Johnson homomorphism? See Remark \ref{rem.fil}.
\end{question}
A positive answer to the above question would imply that the full tree-level
part of the theory of \fti s on 3-manifolds is given by our extension
of the Johnson homomorphism to homology cylinders.
\begin{question}
If $(\Sigma_{g,1}\times I )-M\in\mathcal F^Y_n (\Sigma_{g,1}\times I )$, is $M\in\mathcal H_{g,1}(n)\otimes\mathbb Q$? See Remark
\ref{rem.fil}. When $g=0$ the answer is yes---see \cite{GL2}. Are the
$\mu$-invariants of homology cylinders (see Remark \ref{rem.mu})
finite-type in the Goussarov-Habiro sense?
\end{question}
We now consider the group $\mathcal H^c_{g,1}$ of homology cobordism classes of
homology cylinders defined in Remark \ref{rem.hc}. The subgroup
$\mathcal H^c_{g,1}[2]$ (see Proposition \ref{prop.DA}) is the analogue of the
Torelli group, which we denote $\mathcal T\mathcal H^c_{g,1}$. We can consider the lower
central series filtration $(\mathcal T\mathcal H^c_{g,1})_n$ and, just as in the case of
the Torelli group (see \cite{Mo2}), we have
$(\mathcal T\mathcal H^c_{g,1})_n\subseteq\mathcal H^c_{g,1}[n+1]$. Recall also that Hain \cite{Hain}
proved that, over $\mathbb Q$, the associated graded Lie algebra of the lower
central
series filtration of the Torelli group maps onto the associated graded Lie
algebra of the weight filtration
of the Torelli group, and that these two filtrations are known to differ
in degree $2$ by a factor of $\mathbb Q$, \cite{Hain,Mo3}. Whether they differ
in degrees other than $2$ is an interesting question.
\begin{question}
What is the relation between the filtrations $ (\mathcal T\mathcal H^c_{g,1})_n$ and $
H^c_{g,1}[n]$?
\end{question}
Notice that the answer to above question is not known for the group
of concordance classes of string-links, see \cite{HM}, but it is
known (in the positive) for the group of homotopy classes of string-links,
see \cite{HL}, and for the pure braid group, see \cite{Kh}.
\begin{question}
We have $(\mathcal T_{g,1})_n\subseteq\mathcal H^c_{g,1}(n)$ (see Remark \ref{rem.fil}) and
obviously $(\mathcal T_{g,1})_n\subseteq (\mathcal T\mathcal H^c_{g,1})_n$. What is the relation between
the filtrations $ \mathcal H^c_{g,1}(n)$ and $ (\mathcal T\mathcal H^c_{g,1})_n$?
\end{question}
We now consider the center $\mathcal Z (\mathcal H^c_{g,1})$ of $\mathcal H^c_{g,1}$. This
contains, at least, the group $\theta^H$ of homology cobordism classes of
homology $3$-spheres (see Remark \ref{rem.hc}). Furthermore $\mathcal Z
(\mathcal H^c_{g,1})$ contains also the element $\tau_{\partial}$ defined by a Dehn
twist about the boundary of $\Sigma_{g,1}$. For the mapping class group
$\Gamma_g$ it seems to be true (according to J. Birman and C. McMullen)
that the center is trivial, at least for large $g$.
\begin{question} Determine $\mathcal Z (\mathcal H^c_{g,1})$. Is it generated by
$\theta^H$ and $\tau_{\partial}$?
\end{question}
We now consider the subgroup $ \mathcal H^c_{g,1}[\infty]\subseteq \mathcal H^c_{g,1}$. This
contains $\theta^H\times \mathcal S_g^c [\infty ]$, where $ \mathcal S_g^c
[\infty ]$ denotes the subgroup of the string-link concordance group $
\mathcal S_g^c$ consisting of string links with vanishing $\mu$-invariants
(see Remarks \ref{rem.hc} and \ref{rem.sl}). $\mathcal S_g^c [\infty ]$
contains, for example, all boundary string links and, in particular, the
knot concordance group, which is an abelian group of infinite rank
(see\cite{Le1}).
\begin{question}\lbl{q.inf}
Determine $ \mathcal H^c_{g,1}[\infty]$. Is it equal to $\theta^H\times\mathcal
S_g^c [\infty ]$?
\end{question}
\begin{question}
Is $ \mathcal Z (\mathcal H^c_{g,1})/
\mathcal Z (\mathcal H^c_{g,1})\cap \mathcal H^c_{g,1}[\infty]$ the infinite cyclic group
generated by $\tau_\partial$?
\end{question}
In \cite{Mo3}, Morita calculated the
symplectic invariant part $\mathsf{D}}%%%{\mathsf{L}^c(H)^{\mathfrak s\mathfrak p}$ of $\mathsf{D}}%%%{\mathsf{L}^c(H)$
in terms of a beautiful space of chord diagrams.
The group $ \mathcal Z (\mathcal H^c_{g,1})/(
\mathcal Z (\mathcal H^c_{g,1})\cap \mathcal H^c_{g,1}[\infty])$ is closely related
to $\mathsf{D}}%%%{\mathsf{L}^c(H)^{\mathfrak s\mathfrak p}$, since the image under the map
$\sigma_n$ of Theorem \ref{th.alf} of an element in $
\mathcal Z (\mathcal H^c_{g,1})\cap \mathcal H^c_{g,1}[n]$ lies in
$\mathsf{D}}%%%{\mathsf{L}^c_n(H)^{\mathfrak s\mathfrak p}$.
Thus any element of $ \mathcal Z (\mathcal H^c_{g,1})/
\mathcal Z (\mathcal H^c_{g,1})\cap \mathcal H^c_{g,1}[\infty]$ provides a geometric
construction of an element of $\mathsf{D}}%%%{\mathsf{L}^c(H)^{\mathfrak s\mathfrak p}$.
The following question is important to the philosophical notion of
finite type.
\begin{question}
Is $\mathcal H^c_{g,1}$ finitely-generated?
Is its abelianization finitely-generated?
Note that both the mapping
class group $\Gamma_{g,1}$ and the Torelli group $\mathcal T_{g,1}$ are
finitely-generated.
Note also that $\mathcal S_g^c$ and $ \theta^H$ are infinitely-generated
abelian (see
\cite{F}) and, as for the analogous question for string-links, since the knot
concordance group has infinite rank, the
abelianization of the string-link concordance group (on any number of strings)
has infinite rank.
\end{question}
\begin{question} Let $\mathcal H_g$ denote the semigroup of homology cylinders over
the {\em closed } surface $\Sigma_g$ of genus $g$. The kernel of the
obvious epimorphism $\mathcal H_{g,1}\to\mathcal H_g$ is related to concordance classes of
framed proper arcs in $I\times \Sigma_g$. Describe this more explicitly
and consider also the homology cobordism groups $\mathcal H_g^c$.
\end{question}
\subsection{Note}
The present paper was completed in 1999, and its follow-up by the second
author appeared in \cite{Le3}.
\ifx\undefined\bysame
\newcommand{\bysame}{\leavevmode\hbox
to3em{\hrulefill}\,}
\fi
|
\section*{\rm Introduction}
Let $X$ be a smooth projective threefold with big $K_X$. Many
authors have studied the pluricanonical systems or the
pluricanonical maps of $X$(see \cite{Ben}, \cite{Ch1}, \cite{E-L},
\cite{Ka1}, \cite{Kol}, \cite{L1}, \cite{L2}, \cite{Ma},
\cite{Sho}, \cite{Wi} etc). Suppose $h^0(X, kK_X)\ge 2$. J.
Koll\'ar(\cite{Kol}, Corollary 4.8) first proved that the
$(11k+5)$-canonical map is birational. Then, in \cite{Ch2}, it has
been proved that either the $(7k+3)$-canonical map or the
$(7k+5)$-canonical map is birational. We denote by $\fei{m}$ the
m-canonical map of $X$. Since it's still unclear whether the
birationality of $\fei{m}$ does imply the birationality of
$\fei{m+1}$, we hope to find certain "stable property" of
$\fei{m}$. In this paper, we modify the ${\Bbb Q}$-divisor method
and then present much better results than in \cite{Kol} and
\cite{Ch2,Ch3}. The ${\Bbb Q}$-divisor method was originally
developed by Kawamata, Reid, Shokorov and others in connection
with the minimal model program initiated by Mori. It was also
exploited much effectively, for various purpose, by many authors
such as Ein, Koll\'ar, Lazarsfeld, Viehweg and so on. As far as
our method can tell here, the results are as follows.
\begin{thm}\label{T:0.1} Let $X$ be a smooth projective 3-fold of general type.
Suppose the $k$-th plurigenus $P_k(X)\ge 2$. Then the $m$-canonical map is
birational onto its image for all $m\ge 5k+6$.
\end{thm}
\begin{thm}\label{T:0.2} Let $X$ be a smooth projective 3-fold of general type.
Suppose $p_g(X)\ge 2$. Then $\fei{8}$ is birational onto its image.
\end{thm}
The base field is always supposed to be algebraically closed of characteristic 0.
For readers' convenience, let us recall the definition of the so-called {\it relative canonical stability} as follows.
Let $X$ be a nonsingular projective variety of general type of dimension
$n$. We define
$k_0(X):=min\{\ k|\ P_k(X)\ge 2\};$
$k_s(X):=min\{\ k|\ \phi_m$ is birational for $m\ge k\}$, which is called {\it
the canonical stability of} $X$;
$\mu_s(X):=\frac{k_s(X)}{k_0(X)}$, which is called {\it the relative
canonical stability} of $X$.
$\mu_s(n):=\text{sup}\{\mu_s(X)| X$ is a smooth projective n-fold of general type$\}$,
which is referred to as {\it the n-th relative canonical stability}.
It's well-known that $\mu_s(1)=3$ and $\mu_s(2)=5$. A conjecture
concerning $\mu_s$ is that $\mu_s(n)=2n+1$. Theorem \ref{T:0.1}
and Theorem \ref{T:0.2} imply the following
\begin{cor} $\mu_s(3)\le 8$.
\end{cor}
\begin{remark} $\mu_s(n)$ is very interesting for $n\ge 3$. For example, if $\mu_s(4)<+\infty$,
then the following is true:
there is a constant $m_0$ such that $\fei{m_0}$ is birational for all smooth projective 3-fold $X$ of general type with
$p_g(X)>0$.
\noindent In fact, taking the product of $X$ with a curve of genus $\ge 2$, one can realize the above statement very easily.
Unfortunately, few information found on $\mu_s(4)$ yet.
\end{remark}
\section{\rm Key lemmas}\label{S:1}
Let $S$ be a smooth projective surface and $D$ a divisor on $S$. We would
like to know when the system $|K_S+D|$ gives a birational rational map onto
its image. It is well-known that Reider (\cite{Rdr}) gave much effective
results when $D$ is nef and big. Because, in our cases, $D$ may be not nef, we
hope to find an effective sufficient condition in order to treat our situation.
Suppose $M$ is a divisor on $S$ with $h^0(S, M)\ge 2$ and $|M|$ is
movable. Taking necessary blow-ups $\pi:S'\longrightarrow S$, along the
indeterminency of the system, such that the movable part of
$|\pi^*(M)|$ is basepoint free, we can get a morphism
$g:=\Fi{M}\circ \pi: S'\longrightarrow W$ where $W$ is the image of $S'$ in
${\Bbb P}^{h^0(S, M)-1}$. If $\dim(W)=1$, we can take the Stein
factorization $g: S'\overset{f}\longrightarrow B\longrightarrow W$, where $B$ is a
smooth projective curve and $f$ is a fibration. Denote $b:=g(B)$.
\begin{defn}\label{D:1.1} If $b=0$, we say that $|M|$ is {\it composed of a
rational pencil of curves}. Otherwise, $|M|$ is {\it composed of an
irrational pencil of curves.}
\end{defn}
We can write $\pi^*(M)\sim_{\text{lin}} M_0+Z_0$ where $M_0$ is the movable part and
$Z_0$ the fixed one. According to Bertini's theorem, a general member
$C\in |M_0|$ is a smooth irreducible curve if $|M_0|$ is not composed of
pencil of curves. Otherwise, we can write $M_0\sim_{\text{lin}} \sum_{i=1}^mC_i$ where
the $C_i$'s are fibers of the fibration $f$ for all $i$. Denote by $C$ a
general fiber of $f$ in the latter case. Then $C$ is still an irreducible
smooth curve.
\begin{defn}\label{D:1.2} In the above setting, $C$ is called
{\it a generic irreducible element of} $|M_0|$. Meanwhile, $\pi(C)$ is
called {\it a generic irreducible element of} $|M|$. Note that, however,
$\pi(C)$ may be a singular curve.
\end{defn}
Now we can set up the following lemma.
\begin{lemma}\label{L:1.3} Let $S$ be a smooth projective surface and $D$ a divisor on $S$. The rational map $\Fi{K_S+D}$ is birational onto its image if the following conditions hold.
(i) There is a divisor $M$ on $S$ with $h^0(S, M)\ge 2$ and $|M|$ movable such that
$K_S+D\ge M$. If $|M|$ is composed of an irrational pencil of curves, denoting by $C$ its generic irreducible element, $D-2C\ge \roundup{A_0}$ holds
for certain nef and big ${\Bbb Q}$-divisor $A_0$ on $S$ (with $A_0\cdot C>1$ in the case
$\kappa(S)=-\infty$).
(ii) $D-C\ge\roundup{A}$ holds for certain nef and big ${\Bbb Q}$-divisor $A$ on $S$ with $A\cdot C>2$, where $C$ is a generic irreducible element of $|M|$.
\end{lemma}
\begin{proof}
It's clear that one may suppose that $|M|$ is base point free. Taking the
Stein-factorization of $\Fi{M}$, we can get
$$\Fi{M}: X\overset{f}{\longrightarrow} W\overset{s}{\longrightarrow} {\Bbb P}^{h^0(M)-1},$$
where $f$ has connected fibers. A generic irreducible element of $|M|$ is a smooth
projective curve.
If $|M|$ is not composed of a pencil of curves, then it's sufficient to verify
the birationality of $\Fi{K_S+D}|_{C}$ by virtue of Tankeev's principle (\cite{Ta}, Lemma 2).
If $|M|$ is composed of a pencil of curves, then $f$ is a fibration onto the
smooth curve $W$. In this case, $C$ is a general fiber of $f$.
When $W$ is a rational curve, $|K_S+D|$ can distinguish different
fibers of $f$ since $K_S+D\ge C$. When $W$ is irrational, suppose $C_1$ and $C_2$
are two general fibers of $f$. By assumption, we have
$$D-C_1-C_2=\roundup{A_0}+F,$$
where $F$ is an effective divisor on $S$ and $A_0$ is a nef and big
${\Bbb Q}$-divisor. We consider the system $|K_S+D-F|$. According to
Kawamata-Viehweg vanishing theorem, we have the surjective map
$$H^0(S, K_S+D-F)\longrightarrow H^0(C_1, K_{C_1}+G_1)\oplus H^0(C_2, K_{C_2}+G_2).$$
By (i), we have $h^0(C_1, K_{C_1}+G_1)>0$ and $h^0(C_2, K_{C_2}+G_2)>0$. Thus
$|K_S+D-F|$ can distinguish $C_1$ and $C_2$, so can $|K_S+D|$. Therefore
it's also sufficient to verify the birationality of $\Fi{K_S+D}|_{C}$.
By (ii), we can write $D-C=\roundup{A}+F_1$ where $F_1$ is an effective divisor and
$A$ is a nef and big ${\Bbb Q}$-divisor. We consider the system $|K_S+D-F_1|$.
By vanishing theorem, we have the surjective map
$$H^0(S, K_S+D-F_1)\longrightarrow H^0(C, K_C+G)$$
where $G$ is a divisor on $C$ with $\deg (G)\ge 3$. Thus
$\Fi{K_S+D-F_1}|_{C}$ is an embedding, so is $\Fi{K_S+D}|_{C}$.
\end{proof}
\begin{lemma}\label{L:1.4}
Let $S$ be a smooth projective surface of general type. Then
$K_S+\roundup{A}+D$ is effective if $A$ is a nef and big ${\Bbb Q}$-divisor
and if $h^0(S, D)\ge 2$.
\end{lemma}
\begin{proof}
We may suppose that
$|D|$ is basepoint free. Denote by $C$ a generic irreducible element of
$|D|$. Then the vanishing theorem gives the exact sequence
$$H^0(S, K_S+\roundup{A}+C)\longrightarrow H^0(C, K_C+H)\longrightarrow 0,$$
where $H:=\roundup{A}|_C$ is a divisor of positive degree. It is obvious that
$h^0(C, K_C+H)\ge 2$ since $C$ is a curve of genus $\ge 2$. The proof is completed.
\end{proof}
\begin{lemma}\label{L:1.5} (\cite{Ch3}, Lemma 2.7)
Let $X$ be a smooth projective variety of dimension $\ge 2$. Let $D$ be a divisor
on $X$, $h^0(X, \Co{X}(D))\ge 2$ and $S$ be a smooth irreducible divisor on $X$ such that $S$ is not a fixed component of $|D|$. Denote by $M$ the movable part of $|D|$ and by $N$ the movable part of $|D|_S|$ on $S$. Suppose the natural restriction map
$$H^0(X, \Co{X}(D))\overset\theta\longrightarrow H^0(S, \Co{S}(D|_S))$$
is surjective. Then $M|_S\ge N$ and thus
$$h^0(S,\Co{S}(M|_S))=h^0(S,\Co{S}(N))=h^0(S,\Co{S}(D|_S)).$$
\end{lemma}
\section{\rm The case with $k_0\ge 2$}\label{S:2}
Sometimes, for simplicity, we denote $k_0(X)$ and $k_s(X)$ by $k_0$ and
$k_s$ respectively.
\begin{prop}\label{P:2.1} Let $X$ be a minimal projective 3-fold of general
type with only ${\Bbb Q}$-factorial terminal singularities. If
$\dim\fei{k_0}(X)\ge 2$, then $P_m(X)\ge 2$ for all $m\ge 2k_0$.
\end{prop}
\begin{proof}
First we take a birational modification $\pi: X'\longrightarrow X$, according to
Hironaka, such that
(1) $X'$ is smooth;
(2) the movable part of $|k_0K_{X'}|$ defines a morphism;
(3) the fractional part of $\pi^*(K_X)$ has supports with only normal
crossings.
Denote by $S_0:=S_{k_0}$ a generic irreducible element of the movable part
of $|k_0K_{X'}|$. Then $S_0$ is a smooth projective surface of general type
by Bertini's theorem. By the vanishing theorem, we have the exact sequence
\begin{align*}
&H^0(X', K_{X'}+\roundup{(t+k_0)\pi^*(K_X)}+S_0)\\
\longrightarrow &H^0(S_0, K_{S_0}+
\roundup{(t+k_0)\pi^*(K_X)}|_{S_0})\longrightarrow 0,
\end{align*}
where $t\ge 0$ is a given integer and
$$\roundup{(t+k_0)\pi^*(K_X)}|_{S_0}\ge \roundup{t\pi^*(K_X)|_{S_0}}+D_0,$$
$D_0:=S_0|_{S_0}$ has the property
$h^0(S_0,D_0)\ge 2$ according to the assumption.
If $t=0$, then
$$P_{2k_0+1}(X)\ge h^0(S_0, K_{S_0}+D_0)\ge 2$$
by Lemma 1.2 of \cite{Ch2}.
If $t>0$, we still have the following exact sequence
$$H^0(S_0, K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+C)\longrightarrow
H^0(K_C+G)\longrightarrow 0,$$
where $C$ is a generic irreducible element of the movable part of $|D_0|$ and
$G:=\roundup{t\pi^*(K_X)|_{S_0}}|_C$
is a divisor of positive degree on $C$.
Since $C$ is a curve of genus $\ge 2$, we have
$h^0(C,K_C+G)\ge 2.$
Thus we can see that
$P_{2k_0+t+1}\ge 2$. The proof is completed.
\end{proof}
\begin{cor}\label{C:2.2} Let $X$ be a minimal projective 3-fold of general
type with only ${\Bbb Q}$-factorial terminal singularities. If
$\fei{k_0}$ is birational, then $k_s\le 3k_0$.
\end{cor}
\begin{proof} This is obvious according to Proposition 2.1.
\end{proof}
\begin{thm}\label{T:2.3} Let $X$ be a minimal projective 3-fold of general
type with only ${\Bbb Q}$-factorial terminal singularities.
If $\dim\fei{k_0}(X)=3$ and $\fei{k_0}$ is not birational, then
$k_s\le 3k_0+2$.
\end{thm}
\begin{proof}
Taking the same modification $\pi:X'\longrightarrow X$ as in the proof of Proposition 2.1,
we still denote by $S_0$ the general member of the movable part of
$|k_0K_{X'}|$. Note that both $|k_0K_{X'}|$ and $|\roundup{k_0\pi^*(K_X)}|$
have the same movable part. For a given integer $t>0$, we have
$$K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}+S_0\le (t+3k_0+1)K_{X'}.$$
It is sufficient to prove the birationality of the rational map defined by
$$|K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}+S_0|.$$
Because
$K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}$
is effective by the proof of Proposition 2.1, we have
$$K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}+S_0\ge S_0.$$
Thus we only need to prove the birationality of
$$\Fi{K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}+S_0}\bigm|_{S_0}.$$
We have the following exact sequence by the vanishing theorem
\begin{align*}
&H^0(X, K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}+S_0)\\
\longrightarrow
&H^0(S_0, K_{S_0}+\roundup{(t+2k_0)\pi^*(K_X)}|_{S_0})\longrightarrow 0,
\end{align*}
which means
$$|K_{X'}+\roundup{(t+2k_0)\pi^*(K_X)}+S_0|\bigm|_{S_0}=
|K_{S_0}+\roundup{(t+2k_0)\pi^*(K_X)}|_{S_0}|.$$
Noting that
$$K_{S_0}+\roundup{(t+2k_0)\pi^*(K_X)}|_{S_0}\ge
K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+2L_0,$$
where $L_0:=S_0|_{S_0}$, we want to show that
$\Fi{K_{S_0}+\roundup{t\pi^(K_X)|_{S_0}}+2L_0}$
is birational. Because $|L_0|$ gives a generically finite map, we see from
Lemma 1.4 that
$K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+L_0$
is effective. On the other hand,
let $C$ be a generic irreducible element of $|L_0|$, then $\dim\Fi{L_0}(C)=1$.
So $L_0\cdot C\ge 2$ and thus $(t\pi^*(K_X)|_{S_0}+L_0)\cdot C>2$. By Lemma 1.3,
$|K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+2L_0|$
gives a birational map. The proof is completed.
\end{proof}
\begin{thm}\label{T:2.4} Let $X$ be a minimal projective 3-fold of general
type with only ${\Bbb Q}$-factorial terminal singularities.
If $\dim\fei{k_0}(X)=2$, then $k_s\le 4k_0+4$.
\end{thm}
\begin{proof}
First we take the same modification $\pi:X'\longrightarrow X$ as in the proof of
Proposition 2.1. We also suppose that $S_0$ is the movable part of $|k_0K_{X'}|$.
For a given integer $t>0$, we obviously have
$$K_{X'}+\roundup{(t+2k_0+2)\pi^*(K_X)}+2S_0\le (t+4k_0+3)K_{X'}.$$
Thus it is sufficient to verify the birationality of the rational map defined
by
$$|K_{X'}+\roundup{(t+2k_0+2)\pi^*(K_X)}+2S_0|.$$
By Proposition 2.1,
$$K_{X'}+\roundup{(t+2k_0+2)\pi^*(K_X)}+S_0$$
is effective.
Thus we only have to prove the birationality of the restriction
$$\Fi{K_{X'}+\roundup{(t+2k_0+2)\pi^*(K_X)}+2S_0}\bigm|_{S_0}$$
for the general $S_0$. The vanishing theorem gives the exact sequence
\begin{align*}
&H^0(X',K_{X'}+\roundup{(t+2k_0+2)\pi^*(K_X)}+2S_0)\\
\longrightarrow
&H^0(S_0, K_{S_0}+\roundup{(t+2k_0+2)\pi^*(K_X)}\bigm|_{S_0}+S_0|_{S_0})\longrightarrow
0.
\end{align*}
This means
$$\Fi{K_{X'}+\roundup{(t+2k_0+2)\pi^*(K_X)}+2S_0}\bigm|_{S_0}
=\Fi{K_{S_0}+\roundup{(t+2k_0+2)\pi^*(K_X)}|_{S_0}+S_0|_{S_0}}.$$
Suppose $M_{2k_0+2}$ is the movable part of $|(2k_0+2)K_{X'}|$. We
have to study some property of $\bigm|M_{2k_0+2}|_{S_0}\bigm|$.
Note that $M_{2k_0+2}$ is also the movable part of
$$|\roundup{(2k_0+2)\pi^*(K_X)}|.$$
We have
$K_{X'}+\roundup{\pi^*(K_X)}+2S_0\le (2k_0+2)K_{X'}.$
The vanishing theorem gives the exact sequence
$$H^0(X',K_{X'}+\roundup{\pi^*(K_X)}+2S_0)\overset{\alpha}\longrightarrow
H^0(S_0, K_{S_0}+\roundup{\pi^*(K_X)}|_{S_0}+L_0)\longrightarrow 0,$$
where $L_0:=S_0|_{S_0}$. Denote by $M_{2k_0+2}'$ the movable part of
$|K_{X'}+\roundup{\pi^*(K_X)}+2S_0|$
and by $G$ the movable part of
$|K_{S_0}+\roundup{\pi^*(K_X)}|_{S_0}+L_0|.$
By Lemma 2.3, we have
$G\le M_{2k_0+2}'|_{S_0}\le M_{2k_0+2}|_{S_0}.$
Noting that $|L_0|$ is a free pencil, we can suppose $C$ is a generic irreducible
element of $|L_0|$. Now the key step is to show that $\dim\Fi{G}(C)=1$.
In fact, the vanishing theorem gives
$$|K_{S_0}+\roundup{\pi^*(K_X)|_{S_0}}+L_0|\bigm|_C=|K_C+D|,$$
where $D:=\roundup{\pi^*(K_X)|_{S_0}}\bigm|_C$ is a divisor of positive degree.
Because $C$ is a curve of genus $\ge 2$, $|K_C+D|$ gives a finite map.
This shows
$$\dim\Fi{K_{S_0}+\roundup{\pi^*(K_X)|_{S_0}}+L_0}(C)=1,$$
thus
$\dim\Fi{G}(C)=1$. Therefore
$\dim\Fi{M_{2k_0+2}|_{S_0}}(C)=1.$
and so $M_{2k_0+2}|_{S_0}\cdot C\ge 2$.
Noting that
$h^0(S_0, M_{2k_0+2}|_{S_0})\ge h^0(S_0,G)\ge 2,$
we get from Lemma 1.4 that
$K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+M_{2k_0+2}|_{S_0}$
is effective.
Now Lemma 1.3 implies the birationality of the rational map defined by
$|K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+M_{2k_0+2}|_{S_0}+L_0|.$
Because
\begin{align*}
&|K_{S_0}+\roundup{t\pi^*(K_X)|_{S_0}}+M_{2k_0+2}|_{S_0}+L_0|\\
\subset &|K_{S_0}+\roundup{(t+2k_0+2)\pi^*(K_X)}|_{S_0}+L_0|,
\end{align*}
$\Fi{K_{S_0}+\roundup{(t+2k_0+2)\pi^*(K_X)}|_{S_0}+S_0|_{S_0}}$
is birational.
We have proved the theorem.
\end{proof}
From now on, we suppose that $\dim\fei{k_0}(X)=1$. We can take the
same modification $\pi:X'\longrightarrow X$ as in the proof of Proposition
2.1. Let $g:=\fei{k_0}\circ\pi$ be the morphism from $X'$ onto
$W\subset {\Bbb P}^{P_{k_0}-1},$ where $W$ is the Zariski closure
of the image of $X$ through $\fei{k_0}$. Let $g:X'\overset{f}\longrightarrow
Q\longrightarrow W$ be the Stein-factorization. Then $Q$ is a smooth
projective curve. Denote $b:=g(Q)$, the genus of $Q$.
\begin{setup}\label{Set:2.5}
If $b>0$, we have already known from \cite{Ch2} that $k_s\le 2k_0+4$.
\end{setup}
In the rest of this section, we mainly study the case when $Q$ is the
rational curve ${\Bbb P}^1$. We have a derived fibration $f:X'\longrightarrow{\Bbb P}^1$. Let
$S$ be a general fiber of the fibration. Then $S$ is a smooth projective
surface of general type. Note that $S$ is also the generic irreducible
element of the movable part of the system $|k_0K_{X'}|$.
Let $\sigma: S\longrightarrow S_0$ be the contraction onto the minimal model.
\begin{thm}\label{T:2.6} Let $X$ be a minimal projective 3-fold of general
type with only ${\Bbb Q}$-factorial terminal singularities.
If $\dim\fei{k_0}(X)=1$ and $b=0$, then $k_s\le 5k_0+6$.
\end{thm}
\begin{proof}
For all $i>0$, denote by $M_i$ the movable part of $|iK_{X'}|$. By
Koll\'ar's method (\cite{Kol} or see \cite{Ch2}, 2.2), we have
$M_{9k_0+4}|_S\ge 4\sigma^*(K_{S_0})$.
By the vanishing theorem, one has
\begin{align*}
&|K_{X'}+\roundup{(9k_0+4)\pi^*(K_X)}+S||_S=
|K_S+\roundup{(9k_0+4)\pi^*(K_X)}_S|\\
&\supset |K_S+M_{9k_0+4}|_S|\supset
|5\sigma^*(K_{S_0})|.
\end{align*}
By Lemma 1.5, we see that $M_{10k_0+5}|_S\ge
5\sigma^*(K_{S_0}).$ Repeatedly performing this process, one has
$$ M_{9k_0+4+m(k_0+1)}|_S\ge (m+4)\sigma^*(K_{S_0})$$
for all integer $m>0$.
This means that we can write
$$(5k_0+(m+4)(k_0+1))\pi^*(K_X)|_S\sim_{\Bbb Q} (m+4)\sigma^*(K_{S_0})+
E_{\Bbb Q}^{(m)},$$
where $E_{\Bbb Q}^{(m)}$ is an effective ${\Bbb Q}$-divisor only relating to
$m$. Thus
$$(\frac{5k_0}{m+4}+(k_0+1))\pi^*(K_X)|_S\sim_{\text{num}} \sigma^*(K_{S_0})+
\frac{1}{m+4}E_{\Bbb Q}^{(m)}. \eqno(2.6.1)$$
We can write
$2\sigma^*(K_{S_0})\sim_{\text{lin}} C+Z,$
where $C$ is the movable part and $Z$ the fixed one. According to
Xiao(\cite{X}), $|C|$ is composed of a pencil of curves if and only if
$K_{S_0}^2=1$ and $p_g(S)=0$.
We prove this theorem step by step.
\smallskip
Step 1. $tK_{X'}$ is effective for all $t\ge 3k_0+4$.
\smallskip
Denote by $\alpha\ge 2k_0+3$ a positive integer. We have
$$|(\alpha+k_0+1)K_{X'}|\supset |K_{X'}+\roundup{\alpha\pi^*(K_X)}+S|.$$
By the vanishing theorem, one has
$$|K_{X'}+\roundup{\alpha\pi^*(K_X)}+S||_S=
|K_S+\roundup{\alpha\pi^*(K_X)}|_S|\supset |K_S+\roundup{\alpha\pi^*(K_X)|_S}|.$$
Now we consider a sub-system
$$|K_{S}+\roundup{\alpha\pi^*(K_X)|_S-Z-\frac{2}{m+4}E_{\Bbb Q}^{(m)}}|.$$
Because
$$\alpha\pi^*(K_X)|_S-Z-\frac{2}{m+4}E_{\Bbb Q}^{(m)}-C\sim_{\text{num}}
t_0\pi^*(K_X)|_S$$
where $t_0:=\alpha-\frac{10k_0}{m+4}-2k_0-2>0$ for big $m$. Thus we have
$$|K_{S}+\roundup{\alpha\pi^*(K_X)|_S-Z-\frac{2}{m+4}E_{\Bbb Q}^{(m)}}||_S=
|K_C+D|,$$
where $D$ is divisor on $C$ with $\deg(D)>0$. Therefore
$P_{\alpha+k_0+1}(X)\ge h^0(C, K_C+D)\ge 2$ since $g(C)\ge 2$.
\smallskip
Step 2. The birationality.
\smallskip
Denote by $\beta\ge 4k_0+5$ a positive integer. Considering the system
$|K_{X'}+\roundup{\beta\pi^*(K_X)}+S|$, we have
$$ |K_{X'}+\roundup{\beta\pi^*(K_X)}+S||_S\supset
|K_S+\roundup{\beta\pi^*(K_X)|_S}|.$$
By Step 1, it's sufficient to verify the birationality of
$\Phi_{K_S+\roundup{\beta\pi^*(K_X)|_S}}$.
It's obvious that
$$K_S+\roundup{\beta\pi^*(K_X)|_S}\ge K_S+
\roundup{\beta\pi^*(K_X)|_S
-Z-\frac{4}{m+4}E_{\Bbb Q}^{(m)}}.$$
Denote by $A:=
\roundup{\beta\pi^*(K_X)|_S
-Z-\frac{4}{m+4}E_{\Bbb Q}^{(m)}}-2\sigma^*(K_{S_0})$. We have
$$A\sim_{\text{num}} (\beta-4k_0-4-\frac{20k_0}{m+4})\pi^*(K_X)|_S.$$
So $A$ is also nef and big for big $m$. Now we have
$$|K_S+\roundup{\beta\pi^*(K_X)|_S-Z-\frac{4}{m+4}E_{\Bbb Q}^{(m)}}|=
|K_S+\roundup{A}+2\sigma^*(K_{S_0})+C|.$$
By Lemmas 1.3 and 1.4, one can easily see that the above system defines a birational
map onto its image. The theorem follows.
\end{proof}
\section{\rm The case with $k_0=1$}\label{S:3}
{}From now on, we only consider the case with $p_g(X)\ge 2$. We always suppose $X$ is a minimal
projective 3-fold of general type with ${\Bbb Q}$-factorial terminal singularities.
The first effective result was obtained by Koll\'ar who proved that $\fei{16}$
is birational. In \cite{Ch3}, it has been proved that $\fei{9}$ is birational. Here
we would like to prove the birationality of $\fei{8}$.
By virtue of Theorem 2.3, Theorem 2.4 and (2.5), we only have to consider the situation
with $\dim\phi_1(X)=1$
and $b=0$. We have a derived fibration $f:X'\longrightarrow C$. A general fiber $S$ is a
projective smooth surface of general type. We note that $p_g(S)>0$ in this case.
In order to formulate our proof, we classify $S$ into the following types:
(I) $K_{S_0}^2=1$, $p_g(S)=2$;
(II) $K_{S_0}^2=2$, $p_g(S)=3$;
(III) $K_{S_0}^2=2$, $p_g(S)=2$;
(IV) $K_{S_0}^2=1$, $p_g(S)=1$;
(V) $K_{S_0}^2\ge 3$;
(VI) $K_{S_0}^2=2$, $p_g(S)=1$.
\begin{claim}\label{Claim:3.1} If $S$ is of type (I), then $\Fi{8K_X}$ is
birational.
\end{claim}
\begin{proof}
Denote by $C$ a generic irreducible element of the movable part of $|K_S|$. Then it's well-known
that $C$ is a smooth curve of genus 2. According to Claim in Proposition 5.3 of \cite{Ch3}, we have
$\xi:=\pi^*(K_X)\cdot C\ge \frac{3}{5}$. For a positive integer $t$, we have
$$|K_{X'}+\roundup{t\pi^*(K_X)}+S||_S=|K_S+\roundup{t\pi^*(K_X)}|_S|\supset
|K_S+\roundup{t\pi^*(K_X)|_S}|.$$
Since $p_g(X)>0$, taking $t=1$ and applying Lemma 1.5, one has $M_3|_S\ge C$. Taking $t=3$
and applying Lemma
1.5 once more, one has $M_5|_S\ge 2C$. This means
$$5\pi^*(K_X)|_S\sim_{\Bbb Q}2C+E_{\Bbb Q}^{(5)},$$
where $E_{\Bbb Q}^{(5)}$ is an effective ${\Bbb Q}$-divisor. Thus we have
$$\frac{5}{2}\pi^*(K_X)|_S\sim_{\text{num}} C+\frac{1}{2}E_{\Bbb Q}^{(5)}.$$
Now $t\pi^*(K_X)|_S-C-\frac{1}{2}E_{\Bbb Q}^{(5)}\sim_{\text{num}} (t-\frac{5}{2})\pi^*(K_X)|_S$. It is easy to see that, for $t\ge 6$,
$(t-\frac{5}{2})\pi^*(K_X)|_S\cdot C>2$. Lemma 1.3 implies that $\Fi{K_S+
\roundup{6\pi^*(K_X)|_S}}$ is birational and so is
$\Fi{8K_X}$.
\end{proof}
\begin{claim}\label{Claim:3.2} If $S$ is of type (II), then $\Fi{7K_X}$ is
birational.
\end{claim}
\begin{proof}
We still denote by $C$ a generic irreducible element of the movable part of $|K_S|$. It's well-known that
$|C|$ defines a generically finite map and $C$ is a smooth curve of genus 3. By a parallel argument as in
the proof of Claim 3.1, we have $M_{2m+1}|_S\ge mC$ for any positive integer $m$. This means
$$(2m+1)\pi^*(K_X)|_S\sim_{\Bbb Q}mC+E_{\Bbb Q}^{(m)},$$
where $E_{\Bbb Q}^{(m)}$ is an effective ${\Bbb Q}$-divisor depending on $m$. Thus we have
$$\frac{2m+1}{m}\pi^*(K_X)|_S\sim_{\text{num}} C+\frac{1}{m}E_{\Bbb Q}^{(m)}.$$
Therefore $\eta:=\pi^*(K_X)|_S\cdot C\ge \frac{m}{2m+1}C^2\ge \frac{2m}{2m+1}$ for all $m>0$.
So $\eta\ge 1$.
We want to verify the birationality of $|K_S+\roundup{t\pi^*(K_X)|_S}|$ for certain $t$. Because
$$t\pi^*(K_X)|_S-C-\frac{1}{m}E_{\Bbb Q}^{(m)}\sim_{\text{num}} (t-\frac{2m+1}{m})\pi^*(K_X)|_S=:A,$$
Fix a big $m$, one can see that $A\cdot C>2$ for $t\ge 5$. Thus, by Lemma 1.3,
$\Fi{K_S+\roundup{5\pi^*(K_X)|_S}}$ is birational and so is $\Fi{7K_X}$.
\end{proof}
\begin{claim}\label{Claim:3.3} If $S$ is of type (III), then $\Fi{7K_X}$ is
birational.
\end{claim}
\begin{proof}
Denote by $C$ a generic irreducible element of the movable part of $|K_S|$.
Recall that $\sigma:S\longrightarrow S_0$ is the contraction onto the minimal model. $C_1:=\sigma_{*}(C)$ is the movable part of
$|K_{S_0}|$. It's easy to see that $C_1$ has two types:
(3.3.1) $|K_{S_0}|=|C_1|+Z$, where $C_1^2=0$ and $C_1$ is smooth curve of genus 2.
(3.3.2) $|K_{S_0}|=|C_1|$ , where $C_1$ is a smooth curve of genus 3.
In either cases, we always have $\sigma^*(K_{S_0})\cdot C=K_{S_0}\cdot C_1=2$. By
Theorem 3.1 of \cite{Ci},
$|mK_{S_0}|$ is basepoint free for $m\ge 2$. Now Koll\'ar's technique gives
$7\pi^*(K_X)|_S\ge 2\sigma^*(K_{S_0})$ and so $M_7|_S\ge 2\sigma^*(K_{S_0})$. By a parallel
argument as in the proof of Claim 3.1, we get
$$(2m+3)\pi^*(K_X)|_S\ge M_{2m+3}|_S\ge m\sigma^*(K_{S_0})$$
for all positive integer $m$. Thus
$$\pi^*(K_X)|_S\cdot C\ge \frac{m}{2m+3}\sigma^*(K_{S_0})\cdot C\ge \frac{2m}{2m+3}.$$
So $\pi^*(K_{S_0})\cdot C\ge 1$. Now by the same argument as in the proof of Claim 4.2, one
can easily obtain the birationality of $\Fi{7K_X}$.
\end{proof}
\begin{claim}\label{Claim:3.4} If $S$ is of type (IV), then $\Fi{8K_X}$ is
birational.
\end{claim}
\begin{proof}
We consider the natural map
$$H^0(X',M_2)\overset{\alpha_2}{\longrightarrow}\Lambda_2\subset H^0(S,M_2|_S)\subset
H^0(S,2K_S),$$
where $\Lambda_2$ is the image of $\alpha_2$. Since $P_2(S)=3$,
$0<\dim_{\Bbb C}\Lambda_2\le 3$.
(3.4.1) $\dim_{\Bbb C}\Lambda_2=3$. In this case, $\Lambda_2$ defines the bicanonical map of
$S$. By \cite{Ci}, $|2K_{S_0}|$ is base point free. Thus we see that $M_2|_S\ge 2\sigma^*(K_{S_0})$.
We can write $2\pi^*(K_X)|_S\sim 2\sigma^*(K_{S_0})+E_{\Bbb Q}$, where $E_{\Bbb Q}$ is an effective
${\Bbb Q}$-divisor. Denote by $C$ a general member of $|2\sigma^*(K_{S_0})|$.
Now we have $4\pi^*(K_X)|_S-C-E_{\Bbb Q}\sim_{\text{num}} 2\pi^*(K_X)|_S$ and
$2\pi^*(K_X)|_S\cdot C\ge 4$. By Lemma 1.3, $\Fi{K_S+\roundup{4\pi^*(K_X)|_S}}$ is birational and so is
$\Fi{6K_X}$.
(3.4.2) $\dim_{\Bbb C}\Lambda_2=2$. Because $\Lambda_2$ defines a morphism,
the movable part of $\Lambda_2$ forms a complete linear pencil. The pencil is rational because
$q(S)=0$. Denote by $C_1$ a generic irreducible element of the movable part of
$\Lambda_2$. By Lemma 3.7 below, $\sigma^*(K_{S_0})\cdot C_1\ge 2$.
Because $M_2|_S\ge C_1$, we can write $2\pi^*(K_X)|_S\sim_{\text{num}} C_1+E_{\Bbb Q}'$, where
$E_{\Bbb Q}'$ is an effective ${\Bbb Q}$-divisor.
Now $5\pi^*(K_X)|_S-C_1-E_{\Bbb Q}'\sim_{\text{num}} 3\pi^*(K_X)|_S$ and, by
(3.6.1), $\pi^*(K_X)|_S\cdot C_1\ge \frac{1}{2}\sigma^*(K_{S_0})\cdot C_1\ge 1$.
According to Lemma 1.3, $\Fi{K_S+\roundup{5\pi^*(K_X)|_S}}$ is birational and so is
$\Fi{7K_X}$.
(3.4.3) $\dim_{\Bbb C}\Lambda_2=1$. In this case, $|2K_X|$ is composed of
a pencil of surfaces. One can see that $q(X)=h^2(\Co{X})=0$, whence
$\chi(\Co{X})\le -1$. Applying Reid's R-R formula(\cite{R2}), one has
$P_2(X)\ge 4$.
We can write $2\pi^*(K_X)\sim_{\text{num}} aS+E'$,where $a\ge 3$ and $E'$ is an effective divisor.
We hope to prove the birationality of $\Fi{8K_X}$. Because $p_g(X)>0$,
it's sufficient to prove the birationality of $\Fi{8K_X}|_S$. By virtue of
Koll\'ar's method, one can see that $|7K_{X'}||_S\supset |2\sigma^*(K_{S_0})|$.
Therefore we are reduced to prove the birationality of $(\Fi{8K_{X'}}|_S)|_C$, where
$C$ is a general member of $|2\sigma^*(K_{S_0})|$.
By the vanishing theorem, one has
$|K_{X'}+\roundup{7\pi^*(K_X)-\frac{1}{a}E'}||_S\supset
|K_S+\roundup{L}|$ where $L\sim_{\text{num}} (7-\frac{2}{a})\pi^*(K_X)|_S$.
(3.5.1) below is still true when $S$ is of type (IV), i.e.
$$\frac{2(2m+3)}{m}\pi^*(K_X)\sim_{\text{num}} C+E_{\Bbb Q}^{(m)},$$
where $E_{\Bbb Q}^{(m)}$ is an effective ${\Bbb Q}$-divisor.
Thus $|K_S+\roundup{L-E_{\Bbb Q}^{(m)}}||_C=|K_C+D|$, where
$D:=\roundup{L-E_{\Bbb Q}^{(m)}}|_C$. Because
$L-C-E_{\Bbb Q}^{(m)}\sim_{\text{num}} (3-\frac{2}{3}-\frac{6}{m})\pi^*(K_X)|_S$ for all $m>0$,
we see that $\deg(D)>2$ whenever $m$ is large. Thus $\Fi{K_S+\roundup{L}}|_C$ is birational and so is
$\Fi{8K_X}$.
\end{proof}
\begin{claim}\label{Claim:3.5} If $S$ is of type (V), then $\Fi{7K_X}$ is
birational.
\end{claim}
\begin{proof}
By \cite{Ci}, $|mK_{S_0}|$ is basepoint free for all $m\ge 2$.
Denote by $C$ the movable part of $|2K_S|$. Then $C\sim_{\text{lin}}
2\sigma^*(K_{S_0})$. It's sufficient to prove the birationality of
$\Fi{7K_X}|_S$. By Koll\'ar's method, $|7K_{X'}||_S\supset
|2\sigma^*(K_{S_0})|=|C|$. We only need to verify the
birationality of $(\Fi{7K_{X'}}|_S)|_C $. The vanishing theorem
gives
$|K_{X'}+\roundup{7\pi^*(K_X)}+S||_S\supset|K_S+\roundup{7\pi^*(K_X)|_S}|.$
Applying Lemma 1.5, we get $M_9|_S\ge 3\sigma^*(K_{S_0})$.
Repeatedly proceeding the above process while replacing "7" by "9,
11, $\cdots$", one can obtain $M_{2m+3}|_S\ge m\sigma^*(K_{S_0})$
and so
$$(2m+3)\pi^*(K_X)|_S\sim_{\text{num}} m\sigma^*(K_{S_0})+E_{\Bbb Q}^{(m)},$$
where $m$ is a positive integer. Thus we have
$$\frac{2(2m+3)}{m}\pi^*(K_X)|_S\sim_{\text{num}} 2\sigma^*(K_{S_0})+
\frac{2}{m}E_{\Bbb Q}^{(m)}.\eqno(3.5.1)$$
It's easy to see that $\pi^*(K_X)|_S\cdot C\ge 3$.
Now taking a very big $m$, one has
$$|K_S+\roundup{5\pi^*(K_X)|_S-\frac{2}{m}E_{\Bbb Q}^{(m)}}||_C=|K_C+D|,$$
where $D:=\roundup{5\pi^*(K_X)|_S-\frac{2}{m}E_{\Bbb Q}^{(m)}}|_C$ and
$\deg(D)\ge (1-\frac{6}{m})\pi^*(K_X)|_S\cdot C>2.$
Therefore we have proved that $(\Fi{7K_{X'}}|_S)|_C$ is birational. So $\Fi{7K_X}$ is birational.
\end{proof}
\begin{claim}\label{Claim:3.6} If $S$ is of type (VI), then $\Fi{8K_X}$ is
birational.
\end{claim}
\begin{proof}
We keep the same notations as in the proof of Claim 3.5. The proof is almost the same except
that we have here $\pi^*(K_X)|_S\cdot C\ge 2$. Thus we can prove that
$|K_S+\roundup{6\pi^*(K_X)|_S-\frac{2}{m}E_{\Bbb Q}^{(m)}}||_C$ is birational by the same
argument. This, in turn, proves the birationality of $\Fi{8K_X}$. We conclude the claim.
\end{proof}
Claims 3.1 through 3.6 imply Theorem 0.2.
\begin{lemma}\label{L:3.7}
Let $S$ be a smooth projective surface of general type. Let $\sigma: S\longrightarrow S_0$ be the
contraction onto the minimal model. Suppose we have an effective irreducible curve $C$
on $S$ such that $C\le \sigma^*(2K_{S_0})$ and $h^0(S, C)=2$. If $K_{S_0}^2=p_g(S)=1$,
then $C\cdot \sigma^*(K_{S_0})\ge 2$.
\end{lemma}
\begin{proof}
We can suppose $|C|$ is a free pencil. Otherwise, we can blow-up $S$ at base points of
$|C|$. Denote $C_1:=\sigma(C)$. Then $h^0(S_0, C_1)\ge 2$. Suppose
$C\cdot \sigma^*(K_{S_0})=1$. Then $C_1\cdot K_{S_0}=1$. Because $p_a(C_1)\ge 2$,
we can see that $C_1^2>0$. {}From $K_{S_0}(K_{S_0}-C_1)=0$, we get
$(K_{S_0}-C_1)^2\le 0$, i.e. $C_1^2\le 1$. Thus $C_1^2=1$ and $K_{S_0}\sim_{\text{num}} C_1$.
This means $K_{S_0}\sim_{\text{lin}} C_1$ by virtue of \cite{Ca}, which is impossible because $p_g(S)=1$.
So $C\cdot\sigma^*(K_{S_0})\ge 2$.
\end{proof}
\begin{remark}\label{R:3.8} Slightly modifying our method, one can even prove the following
statements:
Let $X$ be a minimal projective 3-fold of general type with ${\Bbb Q}$-factorial terminal singularities.
Suppose $p_g(X)\ge 2$. Then
(1) $\fei{7}$ is birational whenever $p_g(X)\ne 2$.
(2) $\fei{6}$ is birational whenever $p_g(X)\ne 2,\ 3$.
(3) $\fei{5}$ is birational whenever $p_g(X)\ne 2,\ 3,\ 4$.
We don't give the explicit proof since we think that the calculation is more complicated. However
it's not difficult for a reader to verify these statements once he understands our method.
\end{remark}
\centerline{\bf Acknowledgement}
This paper was begun while I was doing a post-doc
research at the Mathematical Institute of the University of Goettingen, Germany between
1999 and 2000. It was finally
revised when I visited the IMS, CUHK in 2003. I appreciate very
much for financial supports from both the institutes.
Thanks are also due to both Professor Fabrizio Catanese for
effective discussions and Professor Kang Zuo for many helps and
hospitality.
|
\section*{Abstract}
The potential energy surface of an off-lattice model protein is characterized
in detail by constructing a disconnectivity graph and by examining the
organisation of pathways on the surface. The results clearly
reveal the frustration exhibited by this system and explain why it
does not fold efficiently to the global potential energy minimum.
In contrast, when the frustration is removed by constructing a `G\=o\xspace-type'
model, the resulting graph exhibits the characteristics expected for
a folding funnel.
\section{Introduction}
The potential energy surface (PES) of an interacting system determines its
structural, dynamic, and thermodynamic properties. Formally, the links between
the PES and these properties are fully defined by the stationary points on the
PES, its gradient (which gives the forces on the particles), and the partition
function. However, it is only relatively recently that explicit connections
have been sought between the overall structure of the PES,
or potential energy `landscape', and the behaviour of the system it describes.
This approach promises to provide insight into a number of fields,
including protein folding, global optimization and glass formation.
\par
In the present contribution we provide a global characterization of the
PES for a model heteropolymer, and show how this picture explains the
dynamical properties observed in previous simulations. In the original
model `frustration' prevents efficient relaxation to the global
potential energy minimum. However, when the frustration is removed by
constructing the corresponding `G\=o\xspace-like' model, the deep traps disappear and
the resulting surface resembles a funnel. The term frustration was first used
in the context of spin glasses,\cite{Toulouse77a} where it is impossible to
satisfy all favourable interactions simultaneously. Analogous effects exist
in proteins:\cite{Bryngelson95a} a three-dimensional structure that brings
together two mutually attractive residues may involve generating unfavourable
contacts elsewhere (`energetic frustration'), and the interconversion of
two similar structures may require the disruption of existing favourable
interactions (`geometric frustration').
\par
The major difficulty in providing a fundamental explanation of structure,
dynamics and thermodynamics in terms of the underlying potential energy
surface is that the number of stationary points grows very rapidly with
the size of the system.\cite{Stillinger99a} This growth is, in fact, the
basis of Levinthal's `paradox',\cite{Levinthal69a} which points out the
apparent impossibility of a protein finding its biologically active state
in a random search amongst the astronomical number of available structures.
Some attempts to resolve the paradox proposed a reduction in the search
space from the full configuration space\cite{Levinthal68a,Camacho93a,Sali94a,Socci94a}.
Although it seems unlikely that this reduction is the solution to the
paradox, there is an implicit realization in such approaches that, in some
way, the search is not random. In terms of the energy landscape there are
two reasons for this. Firstly, conformations have different statistical
weights in the thermodynamic ensemble, and secondly, they are not arranged
at random in configuration space. Levinthal's analysis assumes that the
energy landscape is flat, like a golf course with a single hole corresponding
to the native state.\cite{Bryngelson95a} By constructing a simple model that
includes an energetic bias towards the native structure, it can be shown that
the search time on the full conformational space is dramatically reduced to
physically meaningful scales.\cite{Zwanzig92a,Zwanzig95a}
\par
One of the first studies
to consider more explicitly the organization of the energy landscape was
that of Leopold, Montal, and Onuchic.\cite{Leopold92a} These authors proposed
that the landscape of a natural protein consists of a collection of convergent
kinetic pathways that lead to a unique native state which is thermodynamically
the most stable. Such a landscape structure was termed a `folding funnel' because
it focuses the manifold misfolded states towards the correct target. This approach
highlights the fundamental fallacy of the random search in Levinthal's `paradox'.
\par
Funnel theory has gained widespread acceptance through its development by
Wolynes and coworkers in terms of a {\em free} energy
landscape.\cite{Bryngelson95a,Wolynes95a}
The funnel can be described in terms of the free energy gradient towards
the native structure, and the roughness---a measure of the barrier heights between
local free energy minima, which can act as kinetic traps. Folding is encouraged when
the roughness is not large compared with the energy gradient. Simulations have shown
that the folding ability can be measured by the ratio of the folding temperature,
$T_{\rm f}$, where the native state becomes thermodynamically the most stable,
to the glass transition temperature, $T_{\rm g}$, where the kinetics slow down
dramatically because of the free energy barriers.\cite{Socci94a,Socci96a} $T_{\rm g}$
is usually defined as the temperature at which the folding time passes through a certain
threshold. Folding is easiest for large $T_{\rm f}/T_{\rm g}$, since the native
state is then statistically populated at temperatures where it is kinetically accessible.
The effect of frustration is to increase the roughness of the energy landscape
relative to its gradient towards the native structure, thereby hindering relaxation
to the latter.
\par
We have recently shown\cite{Wales98c,Doye99b} how a new visualization
of the potential energy surface using disconnectivity graphs\cite{Becker97a}
reveals the features which determine relaxation of clusters
to their global potential energy minimum.
This approach has already been used by others
to examine the energy landscape of a tetrapeptide\cite{Becker97a,Czerminski90a}
and to study the effects of conformational constraints in hexapeptides\cite{Levy98a}
employing an all-atom model. In the present contribution we analyse a
coarse-grained representation of a larger polypeptide
with 46 residues. Connected sequences of minima have been
reported before for this system\cite{Berry97a}, and we will show how the disconnectivity
graph approach provides a clearer picture of the relation between the energy landscape
and dynamics.
\section{The Model Potential}
Intermediate in detail between lattice and all-atom models of proteins are
continuum bead models, in which each monomer is represented by a single bead on a chain.
These off-lattice systems have received relatively little attention in terms
of landscape analysis, but provide a useful medium for such an approach, since
atomistic representations of proteins are computationally demanding.
\par
Here we examine the effects of frustration in a model heteropolymer
introduced by Honeycutt and Thirumalai.\cite{Honeycutt90a,Honeycutt92a} These
authors proposed a `metastability hypothesis' that a polypeptide
may adopt a variety of metastable folded conformations with similar structural
characteristics but different energies. The particular state reached in the
folding process depends on the initial conditions. We shall see that this scenario
arises from frustration effects intrinsic to the model, which are not expected
for a `good folder'.
\par
The heteropolymer has $N=46$ beads linked by stiff bonds. There are three types
of bead: hydrophobic (B), hydrophilic (L), and neutral (N), and the sequence is
\begin{displaymath}
{\rm B}_9{\rm N}_3({\rm LB})_4{\rm N}_3{\rm B}_9{\rm N}_3({\rm LB})_5{\rm L}.
\end{displaymath}
The potential energy is given by\cite{Honeycutt92a}
\begin{multline}
V={\textstyle\frac{1}{2}} K_r\sum_{i=1}^{N-1}(r_{i,i+1}-r_{\rm e})^2
+{\textstyle\frac{1}{2}} K_\theta\sum_i^{N-2}(\theta_i-\theta_{\rm e})^2 \\
+\epsilon\sum_i^{N-3}\left[A_i(1+\cos\varphi_i)+B_i(1+\cos3\varphi_i)\right] \\
+4\epsilon\sum_{i=1}^{N-2}\sum_{j=i+2}^N C_{ij}\left[\left(\frac{\sigma}{r_{ij}}\right)^{12}
-D_{ij}\left(\frac{\sigma}{r_{ij}}\right)^6\right],
\label{eq:barrelpot}
\end{multline}
where $r_{ij}$ is the separation of beads $i$ and $j$.
The first term represents the bonds linking successive beads. The bond lengths
were constrained at $r_{\rm e}$ in \paper{Honeycutt92a}, but here we follow Berry
et al.\cite{Berry97a} by replacing these constraints with stiff springs:
$K_r=231.2\,\epsilon\sigma^{-2}$, where $\sigma$ and $\epsilon$ are the units of length
and energy defined by the last term in \eq{barrelpot}. To put the energy parameter
in a physical context, the value of $\epsilon$ suggested by Berry et al.\cite{Berry97a}
is $121\,{\rm K}$, such as might be used for the van der Waals interactions between argon atoms.
The second term in \eq{barrelpot}
is a sum over the bond angles, $\theta_i$, defined by the triplets
of atomic positions ${\bf r}_i$ to ${\bf r}_{i+2}$, with
$K_\theta=20\,\epsilon\,{\rm rad}^{-2}$ and $\theta_{\rm e}=105^\circ$. The third term
is a sum over the dihedral angles, $\varphi_i$, defined by the quartets ${\bf r}_i$ to
${\bf r}_{i+3}$. If the quartet involves no more than one N monomer then $A_i=B_i=1.2$, generating
a preference for the trans conformation ($\varphi_i=180^\circ$), whereas if two or three
N monomers are involved then $A_i=0$ and $B_i=0.2$. This choice makes the three neutral
segments of the chain flexible and likely to accommodate turns. The last term in \eq{barrelpot}
represents the non-bonded interactions, and $\sigma$ is set equal to $r_{\rm e}$.
The coefficients for the various combinations of monomer types are as follows.
\begin{alignat*}{3}
i,j\in{\rm B}& \qquad &C_{ij}=1 \quad &D_{ij}=1 \\
i\in{\rm L},\ j\in{\rm L,B}& \qquad &C_{ij}={\textstyle\frac{2}{3}} \quad &D_{ij}=-1 \\
i\in{\rm N},\ j\in{\rm L,B,N}& \qquad &C_{ij}=1 \quad &D_{ij}=0,
\end{alignat*}
with $C_{ij}=C_{ji}$ and $D_{ij}=D_{ji}$. Hence, hydrophobic monomers experience a mutual
van der Waals attraction, and all other combinations are purely repulsive, with interactions
involving a hydrophilic monomer being of longer range.
\par
The global minimum of this system, which we call the BLN model, is a four-stranded
$\beta$-barrel,\cite{Honeycutt92a} illustrated in \fig{p46}. The hydrophobic segments
congregate at the core, and there are turns at the neutral segments. By cutting the
sequence at these regions, Vekhter and Berry have also used this model to study the
self-assembly of the $\beta$-barrel from the separated strands.\cite{Vekhter99a}
\section{Characterizing the Energy Landscape}
The most important points on a PES are the minima and the transition states that
connect them. A transition state is a stationary point at which the Hessian matrix
has exactly one negative eigenvalue whose eigenvector corresponds to the reaction
coordinate. Minima linked by higher-index saddles (the index being the number of
negative Hessian eigenvalues) must also be linked by one or more true transition
states of lower energy.\cite{Murrell68a} The pattern of stationary points and their
connectivities define the topology of the PES.
\subsection{Exploring the Landscape\label{sect:explore}}
All the transition states in the present work were located by
eigenvector-following\cite{Pancir74a,Cerjan81a,Simons83a,Banerjee85a,Baker86a,Wales94a},
where the energy is maximized along one direction and simultaneously minimized in all
the others. Details of our implementation have been given before\cite{Wales96c}.
The minima connected to a given transition state are defined by the end points
of the two steepest-descent paths commencing parallel and antiparallel to the
transition vector (i.e., the Hessian eigenvector whose eigenvalue is negative)
at the transition state. Rather than steepest-descent minimization, we have
employed a conjugate-gradient method (using only first derivatives of the potential)
to calculate the pathways. This technique gives
similar results, and has the advantage of being much faster. However, it is possible
for conjugate-gradient minimization to converge to a stationary point of higher index
than a minimum. To guard against this problem, each optimization was followed
by reoptimization with eigenvector-following to a local minimum. In the majority
of cases, the reoptimisation converged in a few steps, indicating that the
conjugate-gradient method had indeed found a true minimum.
\par
A number of similar approaches have been developed for
systematically exploring a PES by hopping between potential
wells,\cite{Tsai93b,Doye97a,Barkema96a,Mousseau98a} and these can be
adapted to obtain a topographical database in several ways.
Here we want to explore the energy landscape thoroughly, working from the
global minimum upwards. In our scheme, we commenced at the lowest-energy
known minimum, and performed an eigenvector-following search for a transition
state along the eigenvector with the smallest non-zero eigenvalue. Having
located a transition state, the connected minima were found by evaluating the
path as described above. The process was then repeated,
always starting at the lowest-energy minimum found so far, and searching
along eigenvectors in both directions in order of increasing eigenvalue.
To enable the search to explore away from the starting minimum, an
upper limit, $n_{\rm ev}$, was imposed on the number of eigenvectors to
be searched from each minimum. When $n_{\rm ev}$ eigenvectors had been exhausted,
the search moved onto the next-lowest energy minimum. We note that, even if
$n_{\rm ev}$ is set to its maximum value of $3N-6$, there is no guarantee of
finding all the transition states connected to a given minimum.
\par
The low-energy regions of the BLN model energy landscape were explored using
$n_{\rm ev}=10$ until 250 minima had been found. Because of the harmonic
bond potential, following normal modes uphill in energy did not always lead to
a transition state in a reasonable number of iterations in this system. To
compensate for this problem, the value of $n_{\rm ev}$ was raised to 20 and the
search continued until a total of 500 minima had been found. The final number of
transition states was 636.
\subsection{Visualization\label{sect:visualisation}}
A useful visual representation of a PES is provided by the disconnectivity
graph of Becker and Karplus.\cite{Becker97a} This technique was first introduced to
interpret a structural database of the tetrapeptide isobutyryl-(ala)$_3$-NH-methyl,
produced by Czerminski and Elber,\cite{Czerminski90a} and was subsequently applied
to study the effects of conformational constraints in hexapeptides.\cite{Levy98a}
The method is formally expressed\cite{Becker97a} in the language of graph theory,
but can easily be summarized as follows.
\par
At a given total energy, $E$, the minima can be grouped into disjoint sets,
called `super-basins', whose members are mutually accessible at that energy.
In other words, each pair of minima in a super-basin are connected directly or
through other minima by a path whose energy never exceeds $E$,
but would require more energy to reach a minimum in another super-basin.
At low energy there is just one super-basin---that containing the global minimum.
At successively higher energies, more super-basins come into play as new
minima are reached. At still higher energies, the super-basins coalesce as higher
barriers are overcome, until finally there is just one containing
all the minima (provided there are no infinite barriers).
\par
A disconnectivity graph is constructed by performing the super-basin
analysis at a series of energies, plotted on a vertical scale. At
each energy, a super-basin is represented by a node, with lines joining
nodes in one level to their daughter nodes in the level below. The
choice of the energy levels is important; too wide a spacing and
no topological information is left, whilst too close a spacing
produces a vertex for every transition state and hides the overall
structure of the landscape. The horizontal position of the
nodes is arbitrary, and can be chosen for clarity.
In the resulting graph, all branches terminate at local minima, while
all minima connected directly or indirectly to a node are mutually
accessible at the energy of that node.
\par
Visualization of the PES in terms of connectivity patterns between minima represents
a mapping from the full configuration space onto the underlying `inherent
structures'\cite{Stillinger82a}. Although this approach discards information about
the volume of phase space associated with each minimum, the density of minima with
energy can provide a qualitative impression of the volumes associated with the various
regions of the energy landscape.
\par
Some example schematic potential energy surfaces and the corresponding disconnectivity
graphs are illustrated in \fig{demo}. The first two examples demonstrate that
a funnel-shaped landscape produces a disconnectivity graph with a single stem
leading to the global minimum, from which branches sprout corresponding to
local minima that are progressively cut off as the energy descends below the
barriers. The contrasting nature of the funnels in \figs{demo}(a) and (b)
is immediately discernible from the corresponding graphs, where we see that
the higher barriers and lower potential energy gradient towards the global minimum in
(a) produce long dangling branches in the disconnectivity graph. \fig{demo}(c)
is qualitatively different. The PES possesses a hierarchical arrangement of
barriers, giving rise to multiple sub-branching in the graph. The strength of
the disconnectivity graph in representing the topology of the PES is that it
is independent of the dimensionality of the system, whereas schematic
plots of the potential energy itself are restricted to one or two dimensions.
\par
The disconnectivity graph for the low-energy regions of the BLN model
landscape is shown in \fig{treep46}, using the sample of 500 minima and 636
transition states obtained in \sect{explore}.
It is immediately apparent that the PES is not a single funnel.
In fact, it is a good example of a rough energy landscape, with repeated
splitting at successive nodes and long descending branches. A number of low-energy
structures exist which are separated by high barriers. Even if the barriers
were not so high, there would be little thermodynamic driving force towards the
global minimum. The fact that the attractive forces are of relatively long range
and non-specific character means that it is possible to construct many significantly
different structures from common motifs such as the four strands in the global minimum.
For example, some of the low-energy minima differ only by the relative positions of the
two purely hydrophobic strands. These can register with each other in a number of
positions, related visually by a parallel slide. However, such a slide would be
an unlikely mechanism because all the non-bonded interactions would be disrupted
at once. Instead, the shortest path between such structures typically proceeds
through over ten separate rearrangements.
\par
Other ways in which low-energy structures are related involve a reorientation
of the hydrophobic strands, so that the beads which are outermost and those
that come into contact in the core in \fig{p46} are interchanged. Again, such
a process involves many steps and a high barrier. The neutral turn regions
can also adopt a number of configurations. The barriers between structures
related in this way tend to be somewhat smaller, since the torsion potential
is weaker in these regions.
\par
The same structural database that is used to construct the disconnectivity graph
can also be analysed in terms of `monotonic sequences' of connected
minima in which the potential energy decreases with every step.\cite{Berry95a,Kunz95a}
The collection of sequences leading to a
particular minimum define what we will call a monotonic sequence basin (MSB).
Whilst the super-basin of the disconnectivity graph is defined at a specified energy,
a monotonic sequence basin is a fixed feature of the landscape.
\par
Berry et al.\cite{Berry97a} have characterized some monotonic sequences leading
to the global minimum of the BLN model. The sequences are connected by barriers
that are relatively low compared with the energy gradient along the sequence,
leading these authors to place the BLN model into the category of `structure seekers'.
We note, however, that only 67 of our sample of 500 minima lie on monotonic
sequences to the global minimum, so that such sequences are not representative
of paths to the global minimum. Furthermore, other low-energy minima also lie
at the bottom of separate monotonic sequences of comparable or even larger sets
of minima. Hence, this system `seeks' only a general $\beta$-barrel structure;
consideration of the arrangement of the monotonic sequences shows that significantly
different low-energy minima will be reached from different starting configurations,
and interconversion of these minima will be relatively slow with little preference
for any given one.
\subsection{The Effects of Frustration}
The folding characteristics of the BLN model have recently been questioned in other
studies. Guo and Brooks\cite{Guo97a} used MD simulations and a histogram
method to study the thermodynamics of folding. They identified a collapse transition
to compact states with a peak in the specific heat, and a folding transition in
terms of a similarity parameter with the global minimum. The free energy surface as
a function of this parameter and the compactness showed that collapse occurs before
any appreciable native structure is attained, rather than the cooperative collapse
and structuring expected for a good folder. Nymeyer et al.\cite{Nymeyer98a} inferred
the roughness of the energy landscape from the model's thermodynamic and dynamic
behaviour.\cite{potentialnote}
To demonstrate the effects of frustration, they compared their simulations of the BLN
model with a modified version in which the frustration is largely eliminated. We now
characterize the energy landscape of this modified model.
\par
To remove the effects of frustration in the BLN model, all attractive interactions
between pairs of monomers that are not in contact in the native state (global minimum)
are removed. This transformation is equivalent to setting $D_{ij}=0$ in
\eq{barrelpot} for non-bonded pairs of hydrophobic monomers which are separated by
more than $1.167\,\sigma$ in the global minimum. This change increases the heterogeneity of
the interactions, since it makes the attractive forces more specific. The modified
potential was termed `G\=o\xspace-like', following G\=o\xspace and collaborators, who constructed
model lattice proteins by defining attractive interactions between neighbouring
non-bonded monomers in an assumed ground state structure.\cite{Ueda78a}
\par
Performing a survey of the energy landscape of the G\=o\xspace-like model as for the BLN
model above produced 805 transition states linking the 500 low-lying minima. The
disconnectivity graph is shown in \fig{treeGo}. The appearance is much more
funnel-like, with no low-energy minima separated from the global minimum by
large barriers. Relaxing the BLN global minimum with the G\=o\xspace-like potential
actually produces the second-lowest
energy structure; a similar structure differing in the
orientation of one of the turns lies slightly lower. The energy range of the
disconnectivity graph is a much larger proportion of the global minimum well depth
than in the analogous graph for the BLN model (\fig{treep46}). This range reflects the
lower density of minima per unit energy in the G\=o\xspace-like system that results from
the specificity of the attractive forces. The highest-energy minima in the BLN
sample were still relatively compact, whereas those for the G\=o\xspace-like model showed
considerable unfolding of the $\beta$-barrel.
\par
The plots of energy versus shortest integrated path length to the global minimum
in \fig{sgminp46} display the difference between the BLN and G\=o\xspace-like energy
landscapes clearly. For the BLN model there is little correlation between distance and
energy, whereas for the G\=o\xspace-like model the energy rises with distance, as one would
expect in a funnel-like landscape.\cite{LJpaper} The number of
individual rearrangements along the shortest paths to the global minimum is shown
for both models in \fig{ngmbarrel}. The distribution for the BLN model is broader,
with some minima lying as far as 24 steps from the global minimum, in contrast with
a maximum of 15 for the G\=o\xspace-like model. This reveals the greater organization of
the G\=o\xspace-like energy landscape into pathways converging at the global minimum.
\par
A funnel-like interpretation for the G\=o\xspace-like model is also encouraged by the changes
in the average properties of the individual paths between minima, as demonstrated
in \tab{p46paths}. Uphill barriers are, on average, higher and downhill barriers
lower for the G\=o\xspace-like model, producing a steeper downhill gradient between minima.
However, the funnel of the G\=o\xspace-like model is far from ideal. A monotonic sequence
analysis shows that only 124 of the 500 minima lie in the primary MSB, so that the
relaxation from an arbitrary structure to the global minimum is likely to involve
a number of uphill steps.
\par
In simulations, Nymeyer et al.\cite{Nymeyer98a} found that the collapse from
unfolded states and the formation of native structure occurred cooperatively for
the G\=o\xspace-like model, producing a single narrow peak in the heat capacity. They
also showed that glassy dynamics, as measured by non-exponential relaxation
from unfolded states, starts at temperatures just below the collapse for the
BLN model, hindering the search for the native structure. In the G\=o\xspace-like
model, in contrast, glassy dynamics only set in below the folding temperature,
where the global minimum still has a large equilibrium probability. These
results are entirely in accord with those expected from the direct characterization
of the energy landscape presented here.
\section{Conclusions}
The disconnectivity graph analysis of the 3-colour, 46-bead model polypeptide
reveals a frustrated energy landscape with a number of low-lying $\beta$-barrel
structures in competition with the global potential energy minimum. Although
relaxation to one of these $\beta$-barrel minima may be quite efficient, much longer
time scales are needed for the system to reliably locate the global minimum,
in agreement with previous simulations.
\par
In contrast, when the frustration is removed by changing the potential to
a G\=o\xspace-type model, the landscape is transformed to one where the global minimum
should be located easily. The competitive low-lying minima disappear
following the transformation, and the metastable minima are organised with an
energy gradient towards the global minimum. Our results illustrate the utility of the
disconnectivity graph approach as a tool to rationalize and predict structural,
dynamic and thermodynamic behaviour from the potential energy surface.
\section*{Acknowledgements}
We are grateful to the Royal Society and the EPSRC for supporting this research,
and to Dr John Rose for providing his Fortran routines for the model potential.
\newpage
\bibliographystyle{thesis}
\input{jnames}
|
\part{{\Large Pion-Kaon Scattering near the Threshold in Chiral SU(2)
\title{Pion-Kaon Scattering near the Threshold in Chiral SU(2) Perturbation
Theory}
\author{A. Roessl \\ {\small Institut de physique th\'{e}orique, Universit\'{e} de
Lausanne, Switzerland}}
\date{}
\maketitle
\begin{abstract}
In the context of chiral $SU(2)$ perturbation theory, pion-kaon scattering is
analysed near the threshold to fourth chiral order. The scattering amplitude
is calculated both in the relativistic framework and by using an approach
similar to heavy baryon chiral perturbation theory. Both methods lead to
equivalent results.
We obtain relations between
threshold parameters, valid to fourth chiral order,
where all those combinations of low-energy constants which are not associated
with chiral-symmetry breaking terms drop out. The remaining low-energy constants
can be estimated using chiral $SU(3)$ symmetry. Unfortunately, the experimental
information is not precise enough to test our low-energy theorems.
\end{abstract}
\pagebreak
\pagebreak \section{Introduction}
By the end of the fifties it was clear that meson-meson scattering
plays an important part in the understanding of the forces between baryons.
For energies below the two-kaon threshold it is sufficient to take into
account the interactions between pions. At higher energies, heavier
degrees of freedom (kaon, eta, rho\dots) become relevant. Pion-kaon
scattering -- with a threshold at about $650$ MeV -- constitutes the simplest
mesonic system with unequal mass kinematics and carrying strangeness. For
reviews of theoretical and experimental knowledge in the meson sector up to
the late seventies see \cite{Petersen}, \cite{Martin}, \cite{Lang}.
In the late seventies and early eighties, it was discovered that the low-energy
structure of Green's functions of QCD can be determined systematically
\cite{Weinberg, GaLe84, GaLe85}. The method -- called chiral
perturbation theory ($\chi{\rm PT}$) -- is based on a simultaneous expansion of
Green's functions in powers of external momenta and light quark masses, the
latter taking into account the breaking of the $SU(N)_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} \times SU(N)_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}$ (N
flavour)
symmetry of the strong interactions. In the context of $\chi SU(3)$ the
expansion parameters $M_{\pi}^2 /(4\pi F_\pi)^2$ and $M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2/(4\pi F_\pi)^2$ are
of the same chiral order despite the fact that the pion mass $M_\pi$ is
about three times smaller than the kaon mass $M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}$. Chiral $SU(3)$ perturbation
theory was applied to pion-kaon scattering in \cite{BKMthresh, MBK}. In these
references the results for scattering lengths and other threshold parameters
are given only numerically.
In 1987 $\chi{\rm PT}$ was extended to include nucleons, particles whose masses
are non-zero in the chiral limit (of vanishing quark masses) \cite{GSS}.
Interactions of these heavy degrees of freedom with pseudo-Goldstone bosons
are not constrained by the full chiral group $G=SU(2)_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} \times SU(2)_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}$ but
only by the diagonal subgroup $H=SU(2)_V$. There are therefore more unknown
low-energy constants than there would be in the case of `Goldstones' only.
In recent years, several attempts have been made to extend the range of
application of the $\chi SU(3)$ scattering amplitude to higher energies. In
\cite{Hannah}, \cite{Beldjoudi} and \cite{Borges} this is achieved by using
unitarisation methods and dispersion relations. Another possibility of
partially including higher-order corrections is to take resonances
into account explicitly \cite{BKMres}.
An S-matrix parametrisation that conforms with unitarity \cite{Ishida}
was applied to the data of two experiments on kaon-nucleon scattering \cite{Aston,
Estabrooks} performed at SLAC in the seventies and eighties respectively. The authors
find evidence for the existence of an $s$-wave resonance $\kappa(900)$, which
constitutes an improvement in the understanding of $\pi$--$K$ partial waves.
The light scalar mesons and their possible classifications are of much
current interest (see for example \cite{Beveren}).
The purpose of the present article is to calculate the pion-kaon scattering
amplitude in the context of
$\chi SU(2)$ perturbation theory, where the strange quark mass is fixed by
definition, to fouth chiral order and to find relations between threshold
parameters where as much low-energy constants as possible cancel. An advantage
of this approach is that the biggest expansion parameter (near the threshold)
is now $M_{\pi}^2/M_{K}^2 <M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2/(4\pi F_\pi)^2$ so that the chiral
expansion is expected to converge more rapidly than in the $\chi SU(3)$ theory.
However, analogously to baryon $\chi{\rm PT}$, the number of low-energy
constants increases.
In section \ref{symm.sec} the chiral symmetry of the strong
interactions is briefly reviewed. The role of closed kaon loops is discribed
in section \ref{kta}. The transformation laws of the meson fields
are given in section \ref{effF} and the general effective
Lagrangian is constructed in section \ref{efflag}.
Section \ref{masses.sec} is devoted to the chiral expansion of
observables which can be extracted from the two point functions
(pion and kaon masses, pion decay constant). In
sections \ref{Methods.sec} and \ref{scatter},
we construct the isospin $3/2$ amplitude to fourth
chiral order as a function of the low-energy constants. Section \ref{counting}
concerns the matching of the $\chi SU(2)$ with the $\chi SU(3)$ amplitude and
provides an expansion of the $\chi SU(2)$ coupling
constants in powers of the strange quark mass. Low-energy theorems are given in
section \ref{LET} and section \ref{summary} contains the conclusions.
\section{Chiral symmetry of the strong interactions}\label{symm.sec}
In the limit of vanishing light-quark masses ($m_u=m_d=0$) the Lagrangian of
QCD,
\begin{eqnarray}\label{QCD1}
{\cal L}^{\rm QCD} \hspace{-1mm}&=&\hspace{-1mm}-{1\over 2g^2} tr_c G_{\mu\nu}G^{\mu\nu} + {\rm i} \bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$R$}}}
\gamma^\mu(\partial_\mu -{\rm i}G_\mu)q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} + {\rm i} \bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}
\gamma^\mu(\partial_\mu -{\rm i}G_\mu)q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} \nonumber \\
&& \qquad -\bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} {\cal M} q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} - \bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} {\cal M} q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}},
\end{eqnarray}
is invariant under global chiral transformations \mbox{$(v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}}, v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}})
\in G \hspace{-0.5mm}=\hspace{-0.5mm}SU(2)_{{\mbox{\tiny\hspace{-0.4mm}$R$}}}
\hspace{-0.5mm}\times\hspace{-0.5mm} SU(2)_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}$}. The
$3 \times 3$ colour matrix $G_\mu$ denotes the gluon field with associated field
strength $G_{\mu \nu}$, and ${\cal M}={\rm diag}(m_u,m_d \dots)$ is the quark-mass
matrix in flavour space. The action on the
chiral up- and down-quarks,
\[q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} = {1\over 2}(1 + \gamma_5)q,\qquad q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} = {1\over
2}(1 - \gamma_5)q, \] is defined by
\[q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} \rightarrow v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}},\qquad q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} \rightarrow v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}
q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}. \]
The gluon field and the heavy quarks $q_s$, $q_c$, $q_b$ and $q_t$ are not transformed
under the chiral group. Therefore, these degrees of freedom are suppressed in
the following.
The vacuum of QCD is only assumed to be invariant under the isospin subgroup
$H=SU(2)_V$ of $G$. According to Goldstone's
theorem \cite{Goldstone} this implies the ocurrence of an internal $(N_G
- N_H)$-dimensional subspace degenerate with the vacuum, the Goldstone bosons.
Here $N_G$ and $N_H$ denote the number of parameters in the Lie groups $G$ and
$H$ respectively (in our case $N_G=6$ and $N_H=3$).
The Green's functions of the conserved currents as well as the scalar and
pseudoscalar densities are generated by the vacuum-to-vacuum amplitude
\[
{\rm e}^{{\rm i}Z(r,l,s,p)} = _{\rm out}\hspace{-1mm}\bra 0 \vert 0\ket_{\rm in}
\]
based on the Lagrangian
\begin{equation}\label{QCD2}
{\cal L}={\cal L}^{\rm QCD}_{{\cal M}=0} + \bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} r_\mu\gamma^\mu q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} +
\bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} l_\mu\gamma^\mu q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} - \bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} (s-{\rm i} p \gamma_5)
q_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} - \bar{q}_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} (s-{\rm i} p \gamma_5) q_{{\mbox{\tiny\hspace{-0.4mm}$R$}}}.
\end{equation}
The external fields $r_\mu$, $l_\mu$, $s$ and $p$ are hermitian colour-neutral
matrices in flavour space, $r_\mu$ and $l_\mu$ being traceless.
The Lagrangian (\ref{QCD2}) is now invariant under {\it local} chiral
transformations provided the sources transform as
\begin{eqnarray}
r_\mu&\rightarrow& v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} r_\mu v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} + {\rm i} v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} \partial_\mu
v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}}^+, \nonumber\\
l_\mu&\rightarrow& v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} l_\mu v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}} + {\rm i} v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}\partial_\mu
v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}^+, \nonumber\\
s+{\rm i}p&\rightarrow& v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} (s + {\rm i}p) v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}. \nonumber
\end{eqnarray}
Matrix elements of vector and axial vector currents
\[V^\mu=\bar{q}\gamma^\mu\sigma^{i}q, \hspace{2cm} \qquad
A^\mu=\bar{q}\gamma^\mu\gamma^5\sigma^{i}q \]
can be obtained by deriving the generating functional $Z$ with respect to
\[\begin{array}{cc}
v_\mu^{i} =
{1\over 2}(r_\mu^{i} + l_\mu^{i}) & \mbox{and } a_\mu^{i} =
{1\over 2}(r_\mu^{i} - l_\mu^{i})
\end{array}\]
where $r_\mu^{i}$ and $l_\mu^{i}$ are defined by $r_\mu=r_\mu^{i}\sigma^{i}$
and $l_\mu=l_\mu^{i}\sigma^{i}$.
\section{The role of closed kaon loops}\label{kta}
The role of closed nucleon loops is analysed in the context of baryon $\chi
{\rm PT}$ in \cite{GSS}. The situation is
analogue in the case of kaons instead of nucleons. The general effective Lagrangian describing the process $\pi K \rightarrow
\pi K$ is given by a chirally symmetric string of terms
\begin{eqnarray}
{\cal L}\hspace{-1mm}&=&\hspace{-1mm}{\cal L}_{\pi \pi} + {\cal L}_{\pi K}
\nonumber \\
{\cal L}_{\pi K} \hspace{-1mm}&=&\hspace{-1mm}{\cal L}_{K K} +
{\cal L}_{K K K K} + \dots \nonumber
\end{eqnarray}
The term ${\cal L}_{\pi \pi}$ does not contain the kaon field $K$ whereas
${\cal L}_{K K}$, ${\cal L}_{K K K K}$ \dots stand for local Lagrangians which
are bilinear, quadrilinear, \dots in the kaon field. The bilinear term can be
written in the form
\[
{\cal L}_{K K} = K^+ D K
\]
where $D$ denotes a $2 \times 2$ matrix-differential operator which contains
only the pion fields $U$ and the external sources $a_\mu$, $v_\mu$, $s$ and $p$. The
explicit form of $D$ is given in section \ref{Lagdens4.sub}. For brevity
the fields $s$, $p$ and $v_\mu$ are dropped in the following. With an
additional source $\rho$ for the kaon field,
\[
S(U,K,a,\rho)=\int {\rm d}^4 x \left( {\cal L}_{\pi \pi} + {\cal L}_{\pi K} +
\rho^+ K + K^+ \rho \right)
\]
and the vacuum transition amplitude is given by
\begin{eqnarray}\label{pfadint}
_{\rm out}\hspace{-0.3 mm}\bra 0 \vert 0\ket_{\rm in} &=& {\rm e}^{{\rm i}
Z(a,\rho,\rho^+)} \nonumber \\
&=& \int{\cal D}U {\cal D} K^+ {\cal D} K {\rm e}^{{\rm i} S(U, K, a, \rho)}.
\end{eqnarray}
In the effective theory with pions only n-loop
contributions are suppressed by a factor of $O(q^{2n})$ \cite{Weinberg}. In the
presence of heavy particles the situation is more complicated. First,
there are closed kaon loops which are not suppressed. These loop contributions
are real below the two-kaon threshold and, in a regularisation scheme which
respects chiral symmetry, can be taken into account by a redefinition
of fields and low-energy constants. This is always possible because --
by construction -- the general chiral Lagrangian
${\cal L}$ contains all chirally invariant structures. In particular,
as we are only interested in processes with one incoming and one outgoing kaon,
vertices with more than two kaon legs can be dropped. In other words, we only consider Feynman
diagrams with a single uninterrupted kaon line going through the diagram.
The integration over the kaon fields can
therefore be performed symbolically,
\begin{eqnarray}
{\rm e}^{{\rm i} Z(a,\rho,\rho^+)}
&=& \int{\cal D}U {\cal D} K^+ {\cal D} K {\rm e}^{{\rm i} \int{{\rm d}^4 x
\left( {\cal L}_{\pi \pi} \hspace{0.5mm}+\hspace{0.5mm} K^+ D K \hspace{0.5mm}
+ \hspace{0.5mm}
\rho^+ K + K^+ \rho \right) }} \nonumber \\
&=& {1 \over {\rm det}(D)} \int{\cal D}U {\rm e}^{{\rm i} \int{{\rm d}^4 x
({\cal L}_{\pi \pi} \hspace{0.5mm}-\hspace{0.5mm} \rho^+ S_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}(U, a) \rho})},\nonumber
\end{eqnarray}
if the low-energy constants are adjusted accordingly.
Here $S_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}$ denotes the kaon propagator in the presence of
pion and external fields,
\[
D S_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}(x,y)= \delta^4(x-y).
\]
The above argument can be used again to justify the replacement
of the determinant by $1$. It's only after this step that the low-energy
constants contained in ${\cal L}_{\pi \pi}$ are rigorously the same as the
ones in the original article by Gasser and Leutwyler \cite{GaLe84}.
Secondly, there are still unsuppressed loop contributions, arising from an
internal kaon line, if the loop integrals are evaluated in the standard
relativistic way. In the pion-nucleon case there are at least
three different methods for evaluating such loop integrals in order to obtain
the scattering amplitude to third chiral order. These methods, whose relationship has
only been clarified recently \cite{Tang, Becher}, will
be described in section \ref{Methods.sec}.
\section{Effective fields and transformation laws}\label{effF}
To model the dynamics of the pion-kaon system by an effective field theory,
the action of the chiral group $G$ has to be chosen in such a way as to
represent correctly the low-energy structure of QCD. The fact that the
isospin subgroup leaves the vacuum invariant (in the chiral limit) means that
$H$ must be the stability group (isotropy group) of each orbit of the
carrier space $X$ containing the pions. This fact already implies that each
orbit of $X$ must be homeomorphic to the coset space $G/H$ which is itself
isomorphic to $S^3$ (the three-dimensional sphere) and also to
$SU(2)$. It is convenient to select one representative
element out of each equivalence class of $G/H$ defined by
$(u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} =u,u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}=u^+)$. This choice is unambiguous near the neutral element of
$SU(2)$. The action of $(v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} ,v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}})\in G$ on $(u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} ,u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}})\in X$ is then:
\[
(u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} ,u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}) \rightarrow (v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} h^+,v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}h^+)
\]
where the compensatory field $h\in SU(2)$ has to be chosen in such a way that
$(v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} h^+,v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}h^+)$ is the representative element of the class containing
$(v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} ,v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}})$, i.e. so that $v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} u h^+ =
(v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}u^+ h^+)^+$. Because kaons have isospin ${1\over 2}$, this compensator
can also be used to define the action of $G$ on the kaon fields
$ K(x) \in \C^2 $:
\begin{equation}\label{chitrans}
\left[(u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} ,u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}});K\right] \rightarrow \left[(v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} u_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} h^+,v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}u_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}h^+);hK\right],
\end{equation}
so that in our case
\begin{equation}\label{chitrans}
\left( u, K \right) \rightarrow \left( v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} u h^+, hK \right).
\end{equation}
The compensator $h$ does not depend on $u$ if the
action is restricted to $H$ i.e. if $v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} =v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}$. It was shown in a more general
context by Coleman et al. \cite{CCWZ} that any action on the pair
$(u,K)$ respecting the chiral symmetries given above is
equivalent\footnote{Equivalence must be understood here to mean that the
action (\ref{chitrans}) can be obtained by a field redefinition leaving the
S-matrix unchanged.} to (\ref{chitrans}). In this work $u$ and $K$ are
parameterised in terms of the physical particle fields as
\begin{equation}\label{mesonpar}
\begin{array}{ll}
u(x) = {\rm e}^{{\rm i}{\Phi \over 2F}}, & \Phi(x)=\left( \begin{array}{cc}
\pi^0 (x) & -\sqrt{2} \pi^+ (x) \\ \sqrt{2} \pi^- (x) & -\pi^0 (x)
\end{array}\right), \\
K(x)= \left( \begin{array}{c} K^+ (x) \\ K^0 (x) \end{array} \right), &
K^+(x) = \left( {\bar K}^0 (x), K^-(x) \right).
\end{array}
\end{equation}
\subsection{Basic building blocks}\label{BBB}
The fields $U= u^2$ and $\chi=2 B (s+{\rm i}p)$
of standard chiral perturbation theory transform as $U \rightarrow v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} U v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}^{-1}$
under independent right and left chiral rotations $(v_{{\mbox{\tiny\hspace{-0.4mm}$R$}}} ,v_{{\mbox{\tiny\hspace{-0.4mm}$L$}}}) \in SU(2) \times
SU(2)$.
In order to construct systematically all independent terms respecting the
symmetries of the underlying theory, it is convenient to introduce fields $A$
which all transform in the same way under the chiral group. For the present
application we choose
\begin{equation}\label{hah}
A \rightarrow hAh^+
\end{equation}
where $h$ is the compensator defined in section \ref{effF}. In this case the
pions are contained in the matrix field $\Delta_{\mu}$ defined in table
\ref{blocks}, which collects a set of basic building blocks sufficient for the
construction of the fourth-order Lagrangian.
\begin{table}
\caption{{\small Definition and transformation of the basic building blocks under parity and charge
conjugation. The space-time dependence of the fields is not shown. It is
understood that the space arguments of parity-transformed fields undergo a
change of sign. All the fields transform under chiral rotations as indicated in
(\ref{hah}).}}
\begin{center} \begin{tabular}{|c|c|c|c|}
\hline
building block & definition & parity & charge conj.\\
\hline \hline
$\Delta_{\mu}$ & ${1\over2} u^+ D_\mu U u^+$ & $-\Delta^{\mu}$ &
$(\Delta_{\mu})^T$ \\
\hline
$\Delta_{\mu\nu}$ & ${1\over 2}(D_\mu\Delta_\nu + D_\nu\Delta_\mu)$ & $-\Delta^{\mu\nu}$ & $(\Delta_{\mu\nu})^T$ \\
\hline
$KK^+$& & $KK^+$ & $(KK^+)^T$\\
\hline
$K_{\pm}^{\mu}$ & $D^{\mu}KK^+ \pm KD^{\mu}K^+$ & $K_{\mu\pm}$ & $\pm (K_{\pm}^{\mu})^T$ \\
\hline
$K_{\pm}^{\mu\nu}$ & $D^{\mu}KD^{\nu}K^+ \pm D^{\nu}KD^{\mu}K^+$ & $K_{\mu\nu\pm }$ & $\pm (K_{\pm}^{\mu\nu})^T$\\
\hline
$K^{*\mu\nu}_{\pm}$ & $D^{\mu\nu}KK^+ \pm KD^{\mu\nu}K^+$ & $K^{*}_{\mu\nu\pm}$ & $\pm (K^{*\mu\nu}_{\pm})^T$\\
\hline
$K_{\pm}^{\mu\nu\rho}$ & $D^{\mu\nu}KD^{\rho}K^+ \pm D^{\rho}KD^{\mu\nu}K^+$ & $K_{\mu\nu\rho\pm}$ & $\pm (K_{\pm}^{\mu\nu\rho})^T$\\
\hline
$K^{\mu\nu\rho\sigma}_{\pm}$ & $D^{\mu\nu}KD^{\rho\sigma}K^+ \pm D^{\rho\sigma}KD^{\mu\nu}K^+$ & $K_{\mu\nu\rho\sigma\pm}$ & $\pm (K^{\mu\nu\rho\sigma}_{\pm})^T$\\
\hline
$\chi_{\pm}$ & $u^+ \chi u^+ \pm u \chi^+u$ &$\pm \chi_{\pm}$ & $\chi_{\pm}^T$\\
\hline
$\chi^{\mu}_{\pm}$ & $u^+ D^{\mu}\chi u^+ \pm u D^{\mu}\chi^+u$ & $\pm \chi_{\mu\pm}$ & $(\chi^{\mu}_{\pm})^T$\\
\hline
$F^{\mu\nu}_{\pm}$ & $u^+ r^{\mu\nu}u \pm u l^{\mu\nu}u^+$ & $\pm F_{\mu\nu\pm}$ & $\mp (F^{\mu\nu}_{\pm})^T$ \\
\hline
$F^{\mu\nu\rho}_{\pm}$ & $u^+ D^{\mu}r^{\nu\rho}u \pm u D^{\mu}l^{\nu\rho}u^+$ & $\pm F_{\mu\nu\rho\pm}$ & $ \mp (F^{\mu\nu\rho}_{\pm})^T$\\
\hline
\end{tabular} \end{center}
\label{blocks}
\end{table}
The symbol $D^{\mu\nu}$ is defined by $D^{\mu\nu} = D^{\mu}D^{\nu} +
D^{\nu}D^{\mu}$ where $D_{\mu}$ is the covariant derivative, transforming like the
field on which it acts. For example
\begin{eqnarray}\label{Gamma}
D_{\mu} U &=& \partial_{\mu} U - {\rm i} r_{\mu} U + {\rm i} U l_{\mu} \nonumber \\
D_{\mu} K &=& \partial_{\mu} K + \Gamma_{\mu} K, \nonumber \\
\Gamma_{\mu} &=& {1 \over 2} \left( u^+ (\partial_{\mu} -{\rm i} r_{\mu}) u +
u (\partial_{\mu} -{\rm i} l_{\mu} ) u^+ \right).
\end{eqnarray}
Kaon matrices like $KK^+$, which are defined by
\[
(KK^+)_{ij} = K_i K_j^+,
\]
are introduced in order to have the same transformation laws for all fields
under the chiral group and so that trace relations can be used to eliminate
redundant terms. In fact, for any set $A_1,A_2,A_3,A_4$ of complex $2\times 2$
matrices,
{\footnotesize \begin{eqnarray}
\sum_{2p} \bra A_1A_2A_3 \ket \hspace{-2mm}&=&\hspace{-2mm} \sum_{3p} \bra A_1A_2 \ket \bra A_3 \ket -
\bra A_1\ket \bra A_2\ket \bra A_3\ket, \nonumber \\
\sum_{6p} \bra A_1A_2A_3A_4 \ket \hspace{-2mm}&=&\hspace{-2mm} \sum_{3p} \bra A_1A_2 \ket \bra A_3A_4
\ket + \sum_{6p} \bra A_1A_2\ket \bra A_3\ket \bra A_4\ket - 3 \bra A_1 \ket \bra A_2\ket
\bra A_3\ket \bra A_4\ket. \nonumber
\end{eqnarray}}
\unskip The sum $\d\sum_{np}$ is taken over all $n$ permutations giving different
expressions. For example
\[ \sum_{3p} \bra A_1A_2 \ket \bra A_3 \ket = \bra A_1A_2 \ket \bra A_3 \ket +
\bra A_2A_3 \ket \bra A_1 \ket + \bra A_3A_1 \ket \bra A_2 \ket .\]
The differences $D_\mu D_\nu - D_\nu D_\mu$ and
$D_\mu\Delta_\nu - D_\nu\Delta_\mu$ do not appear in the
list of basic building blocks: due to the relations
\begin{eqnarray}
(D^{\mu}D^{\nu}-D^{\nu}D^{\mu})A &=& (-[\Delta^{\mu},\Delta^{\nu}] -{{\rm i}
\over{2}}F^{\mu\nu}_+ )A, \nonumber \\
D^\mu\Delta^\nu - D^\nu\Delta^\mu &=& -{{\rm i}\over 2} F_-^{\mu\nu}, \nonumber
\end{eqnarray}
such terms are contained in the set involving the fields
$\Delta^{\mu}$ and $F^{\mu\nu}_{\pm}$.
Terms with three or more covariant derivatives on the field $K$ can be
transformed by the total derivative argument described for example in
\cite{Scherer} to give terms with two covariant derivatives only.
Table \ref{blocks} also shows the transformation properties under parity and
charge conjugation.
\section{Construction of the effective Lagrangian to fourth chiral
order}\label{efflag}
Our aim is to construct an effective field theory describing pion-kaon
scattering in the energy region where spatial momenta of the mesons in the
$CM$ frame are small compared to the heavy mass $M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}$:
\[
q < M_\pi \sim 140 \hspace{1mm}{\rm MeV} < M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \sim 490 \hspace{1mm}{\rm MeV}.
\]
The theory should be invariant under the Lorentz group, parity and charge
conjugation and should represent correctly the chiral-symmetry breaking
structure. We consider the spatial
momenta and the pion mass to be of $O(p^1$) whereas the kaon mass is
considered to be of $O(p^0)$.
Chiral-power counting is non-trivial in chiral perturbation theory if there
are particles whose masses do not vanish in the chiral limit \cite{GSS}. Loops
of these {\it heavy} particles have unsuppressed contributions if they are
evaluated using the standard relativistic regularisation \cite{GaLe85}.
If the heavy particle is a kaon, the kinetic term giving rise to the propagator
$(M^2-p^2)^{-1}$ is
\begin{equation}\label{ackin}
S_{\rm kin}=\int {\rm d}^4 x (\partial_\mu K^+ \partial^\mu K - M^2 K^+ K ).
\end{equation}
In ${\rm HB}\chi{\rm PT}$ a consistent chiral-power counting procedure is obtained by a field
redefinition \cite{JM}, which in the kaon case takes the form
\begin{equation}\label{nonreltrans}
K(x)={\rm e}^{{\rm i}Mvx} k(x),
\end{equation}
where $v$ is an arbitrary light-like four-component vector with $v^2=1$.
In this way the mass term in equation (\ref{ackin}) is eliminated,
\[
S_{\rm kin}=\int {\rm d}^4 x (-2{\rm i}M v^\mu k^+ \partial_\mu k + \partial_\mu
k^+ \partial^\mu k),
\]
and the propagator becomes $(-2Mvp)^{-1}$. The heavy-particle mass appears
only as a global factor and can be taken in front of the loop integral,
whereas the term $\partial_\mu k^+ \partial^\mu k$ must be treated as a
vertex of second order.
Equivalently, the heavy-particle expansion can be discussed on the level of
Feynman diagrams. In this language, the relativistic propagator $(M^2-p^2)^{-
1}$ is developed in powers of the light momentum $l_\mu$ defined by
$p_\mu=Mv_\mu+l_\mu$ :
\[
{1\over M^2-(Mv+l)^2}
= -{1\over 2Mvl} +{l^2\over 4M^2(vl)^2} + O(l^1).
\]
The heavy mass again factorises out and integrals with a single uninterupted
kaon line are powers -- given by dimensional considerations -- of the light
scales only. It is shown explicitly in appendix \ref{rel.ap} that the
development of the kaon propagator under the integral sign does not lead to
inconsistencies in the application to elastic $\pi$--$K$ scattering.
The above discussion illustrates that chiral-power counting at the Lagrange level is
easier in the `non-relativistic' formalism: terms with one $\partial_\mu$ just
count as $O(p^1)$ and $\chi_{\pm}$ as $O(p^2)$. To
obtain the Lagrange density to fourth chiral order, we therefore write down in
a first step all invariant non-relativistic structures. These terms are given
in appendix \ref{nonrelstr}. In a second step, we determine a set of
relativistic terms, which generate -- via the relation (\ref{nonreltrans}) --
those non-relativistic ones.
\subsection{Lagrange density to fourth order}\label{Lagdens4.sub}
The effective $\chi SU(2)$ Lagrangian contributing to the pion-kaon scattering
amplitude up to fourth chiral order is given by
\begin{eqnarray}\label{Lagrangian}
{\cal L} &=& {\cal L}^{(2)}_{\pi\pi} + {\cal L}_{\pi\pi}^{(4)} +
{\cal L}_{\pi K}^{(1)} + {\cal L}_{\pi K}^{(2)} + {\cal L}_{\pi K}^{(3)} +
{\cal L}_{\pi K}^{(4)}, \nonumber \\
\nonumber \\
{\cal L}_{\pi\pi}^{(2)} &=& F^2 \left( -\bra\Delta_\mu\Delta^\mu\ket +
{1\over 4} \bra \chi_+ \ket \right), \nonumber \\
\nonumber \\
{\cal L}_{\pi\pi}^{(4)} &=& 4 l_1 \bra\Delta_\mu\Delta^\mu\ket^2 +
4 l_2 \bra\Delta_\mu\Delta_\nu\ket\bra\Delta^\mu\Delta^\nu\ket \nonumber \\
&&+ {1\over 16} (l_3 + l_4) \bra \chi_+ \ket^2
- {1\over 2} l_4 \bra\Delta_\mu\Delta^\mu\ket\bra\bar{\chi}_+\ket, \nonumber \\
\nonumber \\
{\cal L}_{\pi K}^{(1)} &=& D_\mu K^+D^\mu K -M^2 K^+K, \nonumber \\
\nonumber \\
{\cal L}_{\pi K}^{(2)} &=&
A_1\;\bra\Delta_\mu\Delta^\mu\ket K^+K \nonumber \\
&&+ A_2\;\bra\Delta^\mu\Delta^\nu\ket D_\mu K^+D_\nu K \nonumber \\
&&+ A_3\;K^+\chi_+K \nonumber \\
&&+ A_4\;\bra \chi_+\ket K^+K, \nonumber \\
\nonumber \\
{\cal L}_{\pi K}^{(3)} &=& B_1 \left( K^+\left[\Delta^{\nu\mu},\Delta_\nu\right]D_\mu K-D_\mu
K^+\left[\Delta^{\nu\mu},\Delta_\nu\right]K \right) \nonumber \\
&&+ B_2\;\bra\Delta^{\mu\nu} \Delta^\rho\ket
\left(D_{\mu\nu} K^+D_\rho K + D_\rho K^+ D_{\mu\nu}K \right) \nonumber \\
&&+ B_3\; \left( K^+\left[\Delta_\mu,\chi_-\right]D^\mu K-D_\mu
K^+\left[\Delta^\mu,\chi_-\right]K \right),
\end{eqnarray}
\begin{eqnarray}
{\cal L}_{\pi K}^{(4)} &=&
C_1\;\bra\Delta_\nu \Delta^{\mu\nu}\ket \left(K^+D_\mu K+D_\mu
K^+K\right) \nonumber \\
&&+ C_2\;\bra\Delta^{\mu\rho} \Delta^\nu\ket
\left(D_{\mu\nu} K^+D_\rho K + D_\rho K^+ D_{\mu\nu}K\right) \nonumber \\
&&+ C_3\;\left( \bra\Delta^{\mu\nu} \Delta^\rho\ket
\left(D_{\mu\nu} K^+D_\rho K + D_\rho K^+ D_{\mu\nu}K \right) \right.
\nonumber \\
&& \left. -2 \left( D^{\mu\nu} K^+ \Delta_\mu\Delta_{\nu\rho} D^\rho K +
D^\rho K^+\Delta_{\nu\rho}\Delta_\mu D^{\mu\nu} K \right) \right) \nonumber \\
&&+ C_4\;\bra\Delta^{\mu\nu} \Delta^{\rho\sigma}\ket
\left(D_{\mu\nu} K^+D_{\rho\sigma} K + D_{\rho\sigma} K^+ D_{\mu\nu}K\right)
\nonumber \\
&&+ C_5\; \left( D_{\mu}K^+\chi_+D^{\mu}K - M^2 K^+\chi_+K \right) \nonumber \\
&&+ C_6\; \left( \bra \chi_+\ket D_\mu K^+ D^\mu K-M^2 \bra \chi_+\ket K^+K
\right) \nonumber \\
&&+ C_7\; \bra\Delta_\mu\chi_-\ket\left(K^+ D_\mu K+D_\mu
K^+K\right) \nonumber \\
&&+ C_8\; \bra\Delta_\mu\Delta^\mu\ket K^+ \chi_+K \nonumber \\
&&+ C_9\; \bra\Delta_\mu\Delta^\mu\ket\bra\chi_+\ket K^+K
\nonumber \\
&&+ C_{10}\; \bra\Delta^\mu\Delta^\nu\ket \left( D_\mu K^+ \chi_+
D_\nu K + D_\nu K^+ \chi_+ D_\mu K \right) \nonumber \\
&&+ C_{11}\; \bra\Delta^\mu\Delta^\nu\ket \bra \chi_+ \ket
\left( D_\mu K^+ D_\nu K + D_\nu K^+ D_\mu K \right) \nonumber \\
&&+ C_{12}\; D_\mu K^+ \{ \{\Delta^\mu,\Delta^\nu \}, \chi_+ \} D_\nu K
\nonumber \\
&&+ C_{13}\; \bra \chi_+\ket K^+\chi_+K \nonumber \\
&&+ C_{14}\; \bra \chi_+^2\ket K^+K \nonumber \\
&&+ C_{15}\; \bra \chi_+\ket^2 K^+K \nonumber \\
&&+ C_{16}\; \bra \chi_-^2\ket K^+K. \nonumber
\end{eqnarray}
The chiral order associated with each class of terms corresponds to the
dominating non-relativistic structures generated from the relativistic terms,
or equivalently, to the chiral order of the leading tree-contributions.
The constants $l_i$ in the pionic part of the Lagrangian of fourth order are
chosen in such a way that they coincide with the original ones first
introduced in \cite{GaLe84}, where a different parameterisation is used for the
unitary matrix $U$. Here, by convention, all the terms involving
$K^{*,\mu\nu}_{\pm}, K^{\mu\nu\rho}_{\pm}, K^{\mu\nu\rho\sigma}_{\pm},
\chi_\mu$ and $F_{\pm}^{\mu\nu\rho}$ are eliminated by shifting the
covariant derivatives to other fields, thus creating equation-of-motion terms,
total derivatives or other basic building blocks. The elimination of the field
$\chi_\mu$, which is used in \cite{GaLe84}, explains the appearence of the
combination $l_3 + l_4$ in the third term of ${\cal L}^{(4)}_{\pi \pi}$.
Some chiral-symmetry breaking terms are proportional to each other in the isospin
limit ($m_u=m_d$). They are, however, distinguished for reasons of generality.
\subsection{The classical equations of motion}
The classical equations of motion associated with the fourth order Lagrangian
are given by
\begin{eqnarray}
4F^2 D_\mu\Delta^\mu &=& F^2\chi_-
-{1\over2} F^2 \bra \chi_- \ket + O(K), \nonumber \\
D_\mu D^\mu K + M^2 K &=& O(p^2).\nonumber
\end{eqnarray}
The explicit form of the higher-order terms is not needed in the present approach.
Here the only use of these equations is to eliminate interaction terms
proportional to $D_{\mu} \Delta^{\mu}$ or $D_\mu D^\mu K$ by appropriate field
redefinitions \cite{Scherer}.
\section{Masses and the pion decay constant}\label{masses.sec}
Using the definitions for the effective fields $\Delta_\mu$, $\Gamma_\mu$,
$\chi_+$ and $\chi_-$ given in section \ref{effF}, one can turn off all
external sources and develop the general Lagrange density in powers of the
pion and kaon fields:
{\small \begin{eqnarray}
{\cal L}&=&{1 \over 4} (1+2 {m_0^2\over F^2} l_4) \bra \partial_\mu \Phi
\partial^\mu \Phi \ket + {m_0^2 \over 4} \left[1+ 2 {m_0^2\over F^2} (l_3+l_4)
\right] \bra \Phi
\Phi \ket \nonumber \\
&&\hspace{2mm}+ \left[ 1+ 2 m_0^2(C_5 +2 C_6)\right] \partial_\mu K^+ \partial^\mu K
\nonumber \\
&&\hspace{4mm}+ \left[ M^2-2 m_0^2(A_3+2 A_4-M^2 (C_5 + 2 C_6))+O(m_0^4)\right] K^+ K \nonumber \\
&&\hspace{6mm}+ O(\Phi^4,\Phi^2 K^2),\nonumber
\end{eqnarray}}
where $m_0^2 = 2B{\hat m}$ and ${\hat m}=m_u=m_d$.
The field redefinitions
\begin{eqnarray}
\Phi '&=& \left( 1 + {m_0^2\over F^2} l_4 \right) \Phi, \nonumber\\
K '&=& \left( 1+ m_0^2(C_5 +2 C_6) \right) K, \nonumber
\end{eqnarray}
give the standard form for the kinetic terms in the Lagrangian,
\begin{eqnarray}
{\cal L}&=&{1\over 4} \left( \bra \partial_\mu \Phi ' \partial^\mu \Phi ' \ket +
M_{\pi,{\rm t}}^2 \bra \Phi ' \Phi ' \ket \right) \nonumber \\
&&\hspace{2mm}+ \left( \partial_\mu K'^+ \partial^\mu K' + M_{K,{\rm t}}^2 K'^+ K' \right)
+ O(\Phi^4,\Phi^2 K^2),\nonumber
\end{eqnarray}
where $M_{\pi,{\rm t}}^2=m_0^2(1+2 l_3 m_0^2/F^2)$ and
$M_{K,{\rm t}}^2=M^2-2 m_0^2 (A_3 + 2 A_4) + O(m_0^4)$ are the masses at tree level. Including
the pion one-loop contributions shown in figure \ref{twopoint.fig}, the
physical pion and kaon masses to fourth chiral order are
\begin{figure}
\begin{center}
\epsfig{file=diagr4.eps,width=.75\textwidth}
\end{center}
\vspace{-1cm}
\caption{{\small Loop contributions to the kaon (a) and pion (b) two-point functions.
The continuous and dashed lines stand for kaon and pion propagators
respectively. The relevant interactions are ${\cal L}^{(2)}_{\pi K}$ for the
kaon and ${\cal L}^{(2)}_{\pi \pi}$ for the pion. Vertices from
${\cal L}^{(1)}_{\pi K}$ do not contribute. As explained in section \ref{kta},
closed kaon-loops must not be taken into account.}}\label{twopoint.fig}
\end{figure}
\begin{eqnarray}\label{mpimk}
M_\pi^2 &=& m_0^2 \left( 1 +2{m_0^2\over F^2}l_3^r + {m_0^2\over F^2}{\bar
\mu}_\pi \right), \nonumber \\
M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 &=& M^2 + M^{(2)} m_0^2 + M^{(4)} m_0^4,
\end{eqnarray}
where
\begin{eqnarray}
M^{(2)}&=&-2 \sigma_0, \nonumber \\
M^{(4)}&=& 4 \sigma_0 (C_5^r +2 C_6^r)-8(C_{13}^r+C_{14}^r+2C_{15}^r)-{3 M^2
A_2 \over 256 \pi^2 F_{\pi}^2} \nonumber \\
&&+{\bar \mu}_{\pi}\left( {3
A_1 \over F_{\pi}^2 }+{6 \sigma_0 \over F_{\pi}^2 }+{3 M^2 A_2 \over
4 F_{\pi}^2 } \right), \nonumber \\
\sigma_0&=&A_3+2 A_4, \nonumber \\
{\bar \mu}_\pi &=& { 1 \over 32 \pi^2} \ln {m_0^2 \over \mu^2}.\nonumber
\end{eqnarray}
Since the first two coefficients in the kaon-mass expansion are finite, the
constants $M$ and $\sigma_0$ are not renormalised.
The pion decay constant has been calculated in \cite{GaLe84} to fourth chiral
order as
\begin{equation}\label{pidec.const}
F_{\pi} = F \left(1 + {m_0^2\over F^2} l_4^r -2 {m_0^2 \over F^2} {\bar
\mu}_\pi \right).
\end{equation}
This result is not modified because the effect of closed kaon loops is
already included in the $\chi SU(2)$ coupling constants.
The {\it kaon} decay constant cannot be obtained in the present framework,
because strangeness-changing, axial currents have not been coupled to
external sources in the general $\chi SU(2)$ Lagrangian.
\subsection{Matching with chiral $SU(3)$}
In the $\chi SU(3)$ theory the strange-quark mass is treated as an expansion
pa\-ra\-me\-ter, whereas in the $\chi SU(2)$ case it is fixed and contained
in the low-energy constants, just like all the other heavy-quark
masses. Comparing the expressions for $M_{\pi}$, $M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}$ and $F_{\pi}$, given by
(\ref{mpimk}) and (\ref{pidec.const}) respectively, with
the corresponding $\chi SU(3)$ results \cite{GaLe85} leads to expansions of the
low-energy constants in powers of the strange-quark mass (${\bar M}_K^2=B_0
m_s$):
{\small \begin{eqnarray}
M^2 \hspace{-2mm}&=&\hspace{-2mm} {\bar M}_K^2 +{ {\bar M}_K^4 \over F_0^2} \left[ { 1 \over 36 \pi^2}
\ln({4 {\bar M}_K^2 \over 3 \mu^2}) - 8(2L_4^r +L_5^r - 4L_6^r-2L_8^r) \right]
\hspace{-1mm}+ O({\bar M}_K^6), \nonumber \\
\nonumber\\
\sigma_0 \hspace{-2mm}&=&\hspace{-2mm} -{1\over 4} - { {\bar M}_K^2
\over F_0^2} \left[ {1\over 288 \pi^2} + { 1 \over 72 \pi^2}\ln({4 {\bar M}_K^2
\over 3 \mu^2}) - 4(2L_4^r +L_5^r - 4L_6^r-2L_8^r) \right] \hspace{-1mm} + O({\bar M}_K^4),
\nonumber \\
&& \hspace{-1.3cm}{F^2\over
8}(C_5^r+2C_6^r)+F^2(C_{13}^r+C_{14}^r+C_{15}^r) + {3 M^2 A_2 \over 2048 \pi^2} \nonumber
\\
&=&\hspace{-1mm} -{5 \over 9216 \pi^2}-{1 \over 2304 \pi^2} \ln({4 {\bar M}_K^2 \over 3 \mu^2})
+{1\over 4}(4L_4^r +L_5^r - 8L_6^r-2L_8^r) + O({\bar M}_K^2). \nonumber
\end{eqnarray}}
The combination of low-energy constants associated with the second order Lagrangian
${\cal L}_{\pi K}^{(2)}$,
\[
A_1 +{M^2 A_2 \over 4} = {1\over 2}+O({\bar M}_K^2),
\]
is evaluated to next-to-leading order in section (\ref{counting}).
The results for the expansions of pionic low-energy constants coincide with
those given in \cite{GaLe85} and are not repeated here.
\section{Relativistic versus non-relativistic loops}\label{Methods.sec}
Chiral-power counting is a non-trivial problem whenever particles are present
whose masses do not vanish in the chiral limit. In baryon $\chi{\rm PT}$ there
are three main approaches to the calculation of the scattering amplitude to
third chiral order \cite{GSS, Tang, MartinM}. The relativistic
approach \cite{GSS} maintains relativistic invariance during the whole calculation
and uses the same techniques of dimensional regularisation that are used in
standard $\chi{\rm PT}$ with (pseudo-) Goldstone bosons only. The chiral power $D$ of
the non-polynomial part of a given loop diagram without closed nucleon loops
is given by
\begin{equation}\label{chidim}
D=2 L + 1+\sum_d (d-2) N^M_d +\sum_d (d-1) N^{M N}_d
\end{equation}
where $L$ is the number of loops and $N^M_d$($N^{M N}_d$) the number of pure
meson (meson-nucleon) vertices, whose tree contributions are of chiral order
$d$. However, loop diagrams usually contain also terms of lower order than
$D$.
Although these terms can be absorbed into the low-energy
constants, most applications to pion-nucleon scattering are performed
in the `non-relativistic' approach (${\rm HB}\chi{\rm PT}$).
In ${\rm HB}\chi{\rm PT}$ \cite{MartinM} the Lorentz-invariant Lagrange
density is written as a sum of terms, which break in general Lorentz
invariance, and loop calculations are performed using these non-invariant
vertices. In this approach the spin structure is simplified and the power-counting
formula (\ref{chidim}) can easily be proved.
Lorentz invariance can be recovered at a later stage but there
are many more vertices than in the relativistic case. Furthermore
${\rm HB}\chi{\rm PT}$ suffers from a deficiency: The perturbation series
corresponding to the non-relativistic expansion fails to converge in part of
the low-energy region \cite{Becher}.
An alternative to the above method is used in \cite{Tang},
where the heavy-particle expansion is used only under the loop integral and
not -- as in ${\rm HB}\chi{\rm PT}$ -- on the Lagrangian level. This method is
relativistic at every stage of the calculation and the power-counting formula
(\ref{chidim}) is also valid for the divergent part. Although this method
suffers in general from the same deficiency as ${\rm HB}\chi{\rm PT}$ it
seems to be the most efficient one for pion-kaon scattering to one loop
because in that case -- due to the absence of cubic
interaction terms -- there are no such low-energy divergencies as those
mentioned above in the case of ${\rm HB}\chi{\rm PT}$. This is shown
explicitely in appendix \ref{rel.ap}.
\section{The scattering amplitude}\label{scatter}
The isospin $3/2$ amplitude is defined by
\[
_{\rm out}\bra\pi^+(\vec{p}_3)K^+(\vec{p}_4)|\pi^+(\vec{p}_1)K^+(\vec{p}_2)\ket_{\rm
in}={\rm i}(2 \pi)^4\delta^{(4)}(p_1+p_2-p_3-p_4)T^{3\over
2}(s,t,u). \]
where $s,u$ and $t$ are the conventional Mandelstam variables,
\begin{eqnarray}
s &=& (p_1 + p_2)^2, \nonumber\\
u &=& (p_1 - p_4 )^2, \nonumber \\
t &=& (p_1-p_3)^2,\nonumber
\end{eqnarray}
subject to the constraint $s+u+t=2(M_{\pi}^2 + M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2)$. The isospin ${1\over
2}$ amplitude can be obtained from $T^{3/2}$ by $s \leftrightarrow u$ crossing:
\[
T^{1/2}(s,t,u)={3\over 2} T^{3/2}(u,t,s)-{1\over 2}T^{3/2}(s,t,u).
\]
It is useful to define the amplitudes
\begin{eqnarray}
T^+(s,t,u)&=&{1\over 3}\left( T^{1/2}(s,t,u) + 2 T^{3/2}(s,t,u) \right), \label{nueven}
\\
T^-(s,t,u)&=&{1\over 3}\left( T^{1/2}(s,t,u) - T^{3/2}(s,t,u) \right), \label{nuodd}
\end{eqnarray}
where $T^+$($T^-$) is even (odd) under $s \leftrightarrow u$ crossing. In the
$s$-channel the decomposition into partial waves is given by
\begin{equation}\label{partialwaves}
T^I(s(q),t(q,z))=16 \pi \sum_{l=0}^{\infty} (2l+1) t_l^I(q) P_l(z).
\end{equation}
In the expansion of the amplitude to fourth chiral order, only the first three Legendre
polynomials $P_l$ are needed:
\begin{eqnarray}
P_0(z)&=& 1, \nonumber\\
P_1(z)&=& z, \nonumber \\
P_2(z)&=& {1\over 2}(3 z^2-1).\nonumber
\end{eqnarray}
The dependence of $s$ and $t$ on the momentum $q$ of center of mass and the
scattering angle $\theta$ is given by
\begin{eqnarray}
s(q)&=& \left( \sqrt{M_{\pi}^2+q^2}+\sqrt{M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 +q^2} \right)^2, \nonumber\\
t(q,z)&=& -2q^2 (1-z), \nonumber \\
z&=&\cos \theta.\nonumber
\end{eqnarray}
In the elastic region, the partial-wave amplitudes satisfy the unitarity
constraint
\begin{equation}\label{unitarity.const}
{\rm Im}\, t_l^I={2q\over \sqrt{s}} \arrowvert t_l^I \arrowvert,
\end{equation}
and close to the threshold the Taylor coefficients of their real parts
define the scattering lengths $a$, effective ranges $b$,\dots:
\begin{equation}\label{scatlen.def}
{\rm Re}\, t_l^I(q)= {\sqrt{s(q)} \over 2} q^{2l}\left( a_l^I+b_l^I q^2+c_l^I
q^4 + O(q^6) \right).
\end{equation}
\subsection{Evaluation of Feynman diagrams}
For the calculation of the off-shell scattering amplitude it would be necessary
to evaluate the general form of the matrix elements of two axial currents (or
pseudoscalar densities) between two kaon states. Here we are only interested in
the on-shell amplitude and, as explained in \cite{GaLe86},
the standard techniques of Feynman diagrams can be used: all external sources are
turned off in the general Lagrangian (\ref{Lagrangian}) and the traces of the
matrix fields (\ref{mesonpar}) are developed up to the sixth power in the physical
meson fields.
As in the $\chi SU(3)$ theory there are only three different types of Feynman
diagrams at the one-loop level. These are shown in figure \ref{tretafi.fig}.
Tree diagrams must be evaluated using all the vertices of the complete
fourth order Lagrangian. The tadpole contributions are
proportional to the pion propagator at $x=0$:
\begin{eqnarray}
{1 \over {\rm i}} \Delta_{\pi}(0) &=& 2 M_{\pi}^2 \lambda + 2 M_{\pi}^2
{\bar \mu_{\pi}}, \nonumber \\
\lambda &=& {\mu^{d-4} \over 16 \pi^2} \left( {1 \over d-4} - {1\over 2}
\left( \ln 4\pi + \Gamma'(1) + 1 \right) \right), \nonumber \\
{\bar \mu_{\pi}} &=& {1 \over 32 \pi^2} \ln { M_{\pi}^2 \over \mu^2}. \nonumber
\end{eqnarray}
and their chiral order $D$ is given by $D=D_{\rm tree}+2$ where
$D_{\rm tree}$ denotes the chiral order of the tree contribution of the same
vertex. In order to obtain the scattering amplitude to fourth chiral order,
vertices from ${\cal L}_{\pi K}^{(1)}$ and ${\cal L}_{\pi K}^{(2)}$ must be
taken into account. Remember that closed kaon
loops are not calculated in this approach, because such contributions are considered
to be already included in the coupling constants (see section \ref{kta}).
\begin{figure}
\begin{center}
\epsfig{file=file1.eps,width=.75\textwidth}
\end{center}
\vspace{-5mm}
\caption{{\small {\bf Tree, tadpole and fish diagram.} The continuous and
dashed lines stand for kaon and pion propagators
respectively. Crossed fish-diagrams are not shown.}}\label{tretafi.fig}
\end{figure}
There are five different fish diagrams contributing to the amplitude according
to the five different combinations of particles running in the loop. These
combinations are:
$\pi^0$--$\pi^0$ ($t$-channel), $\pi^{\pm}$--$\pi^{\mp}$ ($t$-channel),
$K^0$--$\pi^0$ ($u$-channel), $K^+$--$\pi^+$ ($s$-channel) and
$K^{\pm}$--$\pi^{\mp}$ ($u$-channel). Figure \ref{tretafi.fig}
shows the $s$-channel $K^+$--$\pi^+$ diagram which is proportional to
\[
J_{K \pi}(s)={1\over {\rm i}} \int {{\rm d}^d l \over (2 \pi)^d} {1\over M_{\pi}^2 -l^2}{1\over M_{K}^2 -
(p-l)^2},
\]
where $p=p_1+p_2$ is the total momentum.
The function $J_{K \pi}(s)$ is Lorentz-invariant and can therefore be evaluated
at any $p'=\Lambda p$ where $\Lambda$ is an element of the Lorentz group. It is
expedient to choose the Lorentz transformation such that the initial
kaon momentum $p_2$ is just $p_2=M_{K} v$, with $v=(1,0,0,0)$, and to develop the
kaon propagator under the integral sign. The $M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2$ terms
in the denominator then cancel and in this
{\it heavy kaon HK formalism} the function $J_{K \pi}$ becomes
\begin{eqnarray}\label{hubhub}
J_{K \pi}^{HK}(s) \hspace{-2mm}&=&\hspace{-2mm} -{1\over 2 {\rm i} M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}} \int {{\rm d}^d l \over (2 \pi)^d}
{1\over M_{\pi}^2 -l^2}{1\over v(p_1-l)} \left( 1 - {(p_1-l)^2 \over 2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} v
(p_1 - l)} + O(l^2) \right). \nonumber \\
&&
\end{eqnarray}
This integral starts contributing at $O(p^1)$ and the two vertices in the fish
diagram must therefore be taken from ${\cal L}_{\pi K}^{(m)}$ and
${\cal L}_{\pi K}^{(n)}$ with $m+n \leq 3$. This confirms the general
power-counting formula (\ref{chidim}).
It is shown explicitly in appendix \ref{rel.ap} that the power series resulting
from the development of the kaon propagator converges to the correct result.
The expression (\ref{hubhub}) can be written as a sum of standard integrals of
heavy-baryon $\chi{\rm PT}$:
\[
J_{K \pi}^{HK}(s) = {1\over 2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}} J_0(\omega) + {1\over 4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2} (M_{\pi}^2 G_0(\omega)
-2 \omega G_1(\omega) + d G_2(\omega) + G_3(\omega))
\]
where
\[
\omega = {s-M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2-M_{\pi}^2 \over 2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}}={p_1 p_2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}}.
\]
The functions $J_i$ and $G_i$ for $(i = 0, 1, 2)$ are taken from
\cite{MartinM} and \cite{Meissnerczech} and are given in appendix
\ref{heavy.app}.
Posing $\nu = s-u $, which differs from the conventional definition, the full
fourth order amplitude is given by
{\small
\begin{eqnarray}\label{t32.eqn}
T^{3 \over 2} \hspace{-1mm}&=&\hspace{-1mm} T^{3 \over 2}_1+T^{3 \over 2}_2+T^{3 \over 2}_3+T^{3 \over
2}_4, \nonumber \\
\nonumber \\
T^{ 3\over 2}_1 \hspace{-1mm}&=&\hspace{-1mm} -{\nu \over 4 F_\pi^2}, \nonumber \\
\nonumber \\
T^{3 \over 2}_2 \hspace{-1mm}&=&\hspace{-1mm} t{A_1 \over 2 F_\pi^2}-\nu^2{{A_2} \over
16 F_\pi^2} - M_\pi^2 \left( {A_1 \over F_\pi^2} +{{2 \rho }\over
F_\pi^2} \right), \nonumber \\
\nonumber \\
T^{3 \over 2}_3 \hspace{-1mm}&=&\hspace{-1mm} {\nu^2 \over {16 F_\pi^4}}
\left[ f_s(\nu,t) + 3 f_u(\nu,t) \right] + {\bar J}_{\pi\pi}(t) \left( {{M_\pi^2 \nu}\over 6 F_\pi^2}-{{t \nu}\over
24 F_\pi^2} \right) + {\bar \mu}_\pi \left( {{t \nu}\over {12 F_\pi^4}} + {\nu^3 \over {16
F_\pi^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 }}\right) \nonumber \\
&&+ \nu^3 \left( {B_2^r\over 32 F_\pi^2} - {1\over {512 \pi^2 F_\pi^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}}
\right) + t \nu \left( -{B_1^r \over 4 F_\pi^2} + {1\over {384 \pi^2 F_\pi^4
}}\right) \nonumber \\
&&+ M_\pi^2 \nu \left( {B_1^r - 4B_3^r \over 2 F_\pi^2} -{1\over 96 \pi^2
F_\pi^4 }\right), \nonumber \\
\nonumber \\
T^{3 \over 2}_4 \hspace{-1mm}&=&\hspace{-1mm} f_s(\nu,t) \left[ -{{t \nu}\over {8 F_\pi^4}} + M_\pi^2 \nu
\left( {1 \over {4 F_\pi^4}} + {\rho \over F_\pi^4} \right) + \nu^3 \left(
-{1 \over {64 F_\pi^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}} + {A_1 \over {32 F_\pi^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}}+ {A_2 \over {32
F_\pi^4}} \right) \right] \nonumber \\
&&+ f_u(\nu,t) \left[ {{3 t \nu}\over {8 F_\pi^4}} + M_\pi^2 \nu
\left( -{3 \over {4 F_\pi^4}} + {\rho \over F_\pi^4} \right) + \nu^3 \left(
{3 \over {64 F_\pi^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}} + {A_1 \over {32 F_\pi^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}}+ {A_2 \over {32
F_\pi^4}} \right) \right] \nonumber \\
&&+ {\bar J}_{\pi\pi}(t) \left[ t^2 \left( { A_1 \over {2 F_\pi^4}} + {
{M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 A_2} \over {12 F_\pi^4}}\right) + M_\pi t \left( -{{5
A_1} \over {4 F_\pi^4}} - {
{3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 A_2} \over {8 F_\pi^4}} - {{2 \rho } \over F_\pi^4}
\right)
\right. \nonumber \\
&&+ \left. M_\pi^4 \left( {A_1 \over {2 F_\pi^4}} + {{ M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 A_2} \over
{6 F_\pi^4}} + {\rho \over F_\pi^4}\right) \right] \nonumber \\
&&+ \lambda_{(\nu^4)} \nu^4 + \lambda_{(\nu^2 t)} \nu^2 t + \lambda_{(t^2)} t^2
+ \lambda_{(\nu^2 m^2)} \nu^2 M_{\pi}^2 + \lambda_{(t m^2)} t M_{\pi}^2 +
\lambda_{(m^4)} M_{\pi}^4,
\end{eqnarray}}
where
\[
\rho = \sigma_0 \left[ 1-2m_0^2 (C_5^r+2 C_6^r) \right] + O(m_0^4).
\]
The loop functions are given in terms of the variables $\nu$ and $t$ by
{\small \begin{eqnarray}
f_s(\nu,t) \hspace{-1mm}&=&\hspace{-1mm} \left({1 \over 32 \pi^2} -{\nu \over 64 \pi^2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}\right) \sqrt{{
(t - \nu )^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^4} -16 {M_{\pi}^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}} \left[ {\rm i}\pi-
{\rm arccosh}\left( {\nu - t \over 4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} M_{\pi} }\right)\right], \nonumber \\
\nonumber \\
f_u(\nu,t) \hspace{-1mm}&=&\hspace{-1mm} \left({1 \over 32 \pi^2} +{\nu \over 64 \pi^2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}\right) \sqrt{{
(t + \nu )^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2}-16 {M_{\pi}^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2} } {\rm arccosh}\left( {
\nu + t \over 4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} M_{\pi} }\right), \nonumber \\
\nonumber \\
{\bar J}_{\pi \pi}(t) \hspace{-1mm}&=&\hspace{-1mm} {1 \over 16 \pi^2} \left( \sigma(t) \ln {\sigma(t) -1
\over \sigma(t) +1} + 2 \right), \hspace{7mm} (t \leq 0), \nonumber \\
\sigma(t) \hspace{-1mm}&=&\hspace{-1mm} \sqrt{1-{4 M_{\pi}^2 \over t}}.\nonumber
\end{eqnarray}}
In the physical energy region $\nu + t \geq 4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} M_{\pi}$ and $t \leq 0$.
The constants $A_1$, $A_2$, and $\sigma_0$ are scale-independent as are the
combinations $\lambda_{(pol)}$ of low-energy constants which are
defined in appendix \ref{leccomb}. It is easy to see whether a given term in
the scattering amplitude (\ref{t32.eqn}), arises from a tree or a loop diagram:
loop contributions carry a factor $\pi^2$ in the denominator. As a consistency
check we have verified that the partial-wave amplitudes $t^I_l$, defined by
equation (\ref{partialwaves}), satisfy the unitarity constraint
(\ref{unitarity.const}).
\section{Comparison with the chiral $SU(3)$ amplitude}\label{counting}
A numerical estimate of the low-energy constants can be obtained by
comparing the above amplitude with the corresponding amplitude of the $\chi
SU(3)$ theory which is given in \cite{MBK} and whose result is based on the
generating functional given in \cite{GaLe85}. The standard method of Feynman
diagrams leads to the same result\footnote{There is a misprint in equation (3.16)
of \cite{MBK} in the term proportional to $J_{K \eta}^r$ where $3/2$ must be
replaced by $2/3$ (see also \cite{Borges} page 53).}. The amplitude involves
loop functions of heavy particles ($P, Q \in \{K, \eta \}$),
{\small \begin{eqnarray}
M_{P Q}^r (s) \hspace{-2mm}&=&\hspace{-2mm} {1 \over 12 s}(s-2 \Sigma) {\bar J}_{P Q} (s)+{\Delta^2 \over 3 s^2}
\; \bar{\hspace*{-1.4mm}\bar{J}}\!_{P Q} (s)-{1 \over 6}k_{P Q} + {1 \over 288 \pi^2}, \nonumber \\
\nonumber \\
L_{P Q} \hspace{-2mm}&=&\hspace{-2mm} { \Delta^2 \over 4 s} {\bar J}_{P Q}, \qquad
K_{P Q} = { \Delta \over 2 s} {\bar J}_{P Q}, \nonumber \\
\nonumber \\
k_{P Q} \hspace{-2mm}&=&\hspace{-2mm} {1 \over 32 \pi^2} { M_P^2 \ln (M_P^2/ \mu^2) - M_Q^2 \ln (M_Q^2/ \mu^2)
\over M_P^2 - M_Q^2 }, \nonumber \\
\nonumber \\
\; \bar{\hspace*{-1.4mm}\bar{J}}\!_{P Q} (s) \hspace{-2mm}&=&\hspace{-2mm}
{\bar J}(s) - s {\bar J}'(0), \qquad
{\bar J}'(0) = {1 \over 32 \pi^2} \left( {\Sigma \over \Delta
^2} + 2 { M_P^2 M_Q^2 \over \Delta^3} \ln {M_Q^2 \over M_P^2} \right), \nonumber
\\
\nonumber \\
{\bar J}_{P Q}(s) \hspace{-2mm}&=&\hspace{-2mm} {1 \over 32 \pi^2} \left( 2 + { \Delta \over s}\ln {M_Q^2
\over M_P^2 } - {\Sigma \over \Delta} \ln {M_Q^2 \over M_P^2}
- { \nu \over s} \ln {(s+\nu)^2 - \Delta^2 \over (s-\nu)^2 - \Delta^2}
\right), \nonumber \\
\nonumber \\
\Sigma \hspace{-2mm}&=&\hspace{-2mm} \Sigma_{P Q} = M_P^2 + M_Q^2,\qquad
\Delta = \Delta_{P Q} = M_P^2 - M_Q^2, \nonumber
\end{eqnarray}}
\unskip which, close to the threshold, give real
contributions to the scattering amplitude and can be developed in powers of
$M_{\pi}$, $t$ and $\nu=s-u$. For example, the function ${\bar J}_{\eta,K}$ is
needed to third $SU(2)$ order:
{\footnotesize \begin{eqnarray}\label{jnk}
&&\hspace{-0.8cm}{\bar J}_{\eta K}(u) = \nonumber \\
&&\hspace{-0.6cm}{ 1 \over \pi^2} \left( {1 \over 16} +
{\nu \over 32 M_0^2} - {m_0^2 \over 8 M_0^2} + {t \over 32 M_0^2} + {13 \nu^2 \over 1024 M_0^4} - {27 m_0^2 \nu \over 512 M_0^4} +
{13 t \nu \over 512 M_0^4} + {83 \nu^3 \over 16384 M_0^6} \right) \nonumber \\
&&\hspace{-0.6cm}- {\arctan (\sqrt{2}) \over \sqrt{2} \pi^2} \left(\hspace{-0.3mm} {1 \over 6}
\hspace{-0.3mm}+\hspace{-0.3mm} {5 \nu \over 96 M_0^2}\hspace{-0.3mm}-\hspace{-0.3mm}
{11 m_0^2 \over 96 M_0^2} \hspace{-0.3mm}+\hspace{-0.3mm} {5 t \over 96 M_0^2}
\hspace{-0.3mm}+\hspace{-0.3mm} {53 \nu^2 \over 3072 M_0^4} \hspace{-0.3mm}-\hspace{-0.3mm}
{35 m_0^2 \nu \over 512 M_0^4} \hspace{-0.3mm}+\hspace{-0.3mm}
{53 t \nu \over 1536 M_0^4} \hspace{-0.3mm}+\hspace{-0.3mm}
{343 \nu^3 \over 49152 M_0^6 } \hspace{-0.3mm}\right) \nonumber \\
&&\hspace{-0.6cm}+ {\ln (4/3) \over \pi^2} \left( {5 \over 24} - {\nu \over 192 M_0^2}
+ {5 m_0^2 \over 24 M_0^2} - {t \over 192 M_0^2} - {\nu^2 \over 384 M_0^4}+{m_0^2 \nu \over 64 M_0^4} -
{t \nu \over 192 M_0^4} - {\nu^3 \over 768 M_0^6} \right), \nonumber \\
&&
\end{eqnarray}}
\vspace{-1cm}
\begin{eqnarray}
\hspace{-2cm}\mbox{where} \hspace{2cm} M_0^2 &=& B_0 (m_s + {\hat m}), \nonumber \\
m_0^2 &=& 2 B_0 {\hat m}. \nonumber
\end{eqnarray}
If the two heavy masses are equal, $M_P=M_Q$, the functions ${\bar J}_{P P}$
and the constants $k_{P P}$ simplify to
\begin{eqnarray}
{\bar J}_{P P}(t) &=& {t \over 96 \pi^2 M_P^2} + {t^2 \over 960 \pi^2 M_P^4} +
O(t^3), \nonumber \\
\nonumber \\
k_{P P} &=& {1 \over 32 \pi^2} \left( \ln {M_P^2 \over \mu^2} +1 \right) .\nonumber
\end{eqnarray}
Note that $K$--$K$- and $\eta$--$\eta$-loops can only contribute in the $t$-channel.
If both amplitudes are developed in powers of $t$, $\nu$ and
${\hat m}$, an expansion of the $\chi SU(2)$
constants in powers of the strange-quark mass is obtained (see appendix \ref{su3estim}).
\section{Threshold parameters}\label{LET}
The formulae (\ref{t32.eqn}) specify the scattering amplitude to fourth chiral
order. The expression involves 12 combinations of low-energy constants which
can be related to 12 threshold parameters defined by
equation (\ref{scatlen.def}).
These observables are $a^{I}_0$, $b^{I}_0$, $c^{I}_0$, $a^{I}_1$, $b^{I}_1,$
and $a^{I}_2$ for $I \in \{1/2, 3/2 \}$.
For $x \in \{a,b,c\}$ and $l \in \{0,1,2\}$ we define
\begin{eqnarray}\label{x+x-}
x^+_l &=& x^{1/2}_l+2 x^{3/2}_l \nonumber \\
x^-_l &=& x^{1/2}_l- x^{3/2}_l,
\end{eqnarray}
after which it is straightforward to extract these threshold parameters from
the amplitude.
\subsection{Corrections to current algebra and a low-energy
theorem}\label{lep.subsec}
The threshold parameters $x^-_l$ can also be extracted from the amplitude
$T^-_l (\nu,t)$ which is odd in $\nu$ and which therefore contains the
polynomials of first and third chiral order ($\nu$, $\nu
t$, $\nu M_{\pi}^2$, $\nu^3$). Chiral invariance ensures that the lowest-order
polynomial proportional to $\nu$ is parameter-free, as are the
leading terms of some of the threshold parameters $x^-_l$. Here only $a_0^-$,
$a_2^-$ and $b_1^-$ are listed and the parameters $a_1^-$, $b_0^-$ and
$c_0^-$ are given in appendix \ref{lepar}.
{\small \begin{eqnarray}\label{xmin.lec}
a^-_0 \hspace{-1mm}&=&\hspace{-1mm} {M_{\pi} \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}}
{3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 8 \pi F_{\pi}^2} \nonumber \\
&&+ {M_{\pi}^3 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( {M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 16 \pi^3 F_{\pi}^4}-
{3 B_1^r M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 4 \pi F_{\pi}^2} - {3 B_2^r M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 \over 4 \pi
F_{\pi}^2} + {3 B_3^r M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over \pi F_{\pi}^2} - {3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} {\bar \mu}_{\pi} \over
2 \pi F_{\pi}^4} \right), \nonumber \\
&& \nonumber \\
b^-_1 \hspace{-1mm}&=&\hspace{-1mm} {1 \over M_{\pi} (M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi})} \left(-{1 \over 32 \pi F_{\pi}^2
M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}} -{37 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 11520 \pi^3 F_{\pi}^4} + {B_1^r M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 8 \pi F_{\pi}^2}
- {M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} {\bar \mu}_{\pi} \over 24 \pi F_{\pi}^4} \right) \nonumber \\
&&+ {1 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( -{85 \over 2304 \pi^3 F_{\pi}^4}
- {B_1^r \over 8 \pi F_{\pi}^2} - {3 B_2^r M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 \over 8 \pi F_{\pi}^2} -
{17 {\bar \mu}_{\pi} \over 24 \pi F_{\pi}^4} \right), \\
&& \nonumber\\
a^-_2 \hspace{-1mm}&=&\hspace{-1mm} {1 \over M_{\pi}(M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi})} {M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 4800 \pi^3 F_{\pi}^4} +
{1 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( -{1 \over 1152 \pi^3 F_{\pi}^4} +
{B_1^r \over 20 \pi F_{\pi}^2} - {{\bar \mu}_{\pi} \over 60 \pi F_{\pi}^4}
\right). \nonumber
\end{eqnarray}}
It is interesting that loop corrections (factor $\pi^3$ in the denominator) to
$b_1^-$ have the same chiral order as the tree contribution of lowest order,
and the $d$-wave scattering length $a_2^-$ is even dominated by a loop
contribution. This happens because in the definition of these observables (see
equation (\ref{scatlen.def})) a factor $q^l$ has to be extracted, which in
general leads to factors of $M_{\pi}$ in the denominator. For the same reason,
tree contributions from the third-order Lagrangian are less suppressed in
$b_1^-$ and $a_2^-$ and these observables are therefore the natural candidates
to replace the unknown coupling constants $B_1^r$ and $B_2^r$ in the expressions
for the four other observables. The remaining
low-energy constant $B_3^r$ corresponds to a chiral-symmetry breaking term and
therefore cannot be extracted from the angular and energy distributions of the
scattering cross section. However, the theoretical estimate from
appendix \ref{su3estim},
\[
B_3^r(\mu=770) = 7.9 \cdot 10^{-7} \hspace{1mm}{\rm MeV}^{-2},
\]
gives parameterless corrections to the `current algebra' result of the
observables $a_0^-$, $b_0^-$, $a_1^-$ and $c_0^-$. Here only $a_0^-$ is listed.
The parameters $a_1^-$ and $b_0^-$ are given in appendix \ref{lepar}.
{\small \begin{eqnarray}\label{aba}
a^-_0 \hspace{-2mm}&=&\hspace{-2mm} {M_{\pi} \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left({3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 8 \pi F_{\pi}^2} +
{M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 \over 960 \pi^3 F_{\pi}^4} \right) + {M_{\pi}^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left({1 \over 16 \pi F_{\pi}^2} +
{M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 \over 240 \pi^3 F_{\pi}^4} -5 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 a_2^- \right) \nonumber \\
&&+ {M_{\pi}^3 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( {3 B^r_3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over \pi
F_{\pi}^2} + {49 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 384 \pi^3 F_{\pi}^4}- {M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} {\bar \mu}_{\pi} \over 4 \pi F_{\pi}^4} - 15 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 a_2^- + 2
M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 b_1^- \right) \nonumber \\
&&+ {M_{\pi}^4 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( -10 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} a_2^- + 2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} b_1^- \right).
\end{eqnarray}}
\unskip The only input from $\chi SU(3)$ is the constant $B^r_3$, which appears
in the term of $O(p^3)$.
Another interesting combination between scattering lengths is
\[
A^-=a_0^- -6 M_{\pi} M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} a_1^- + 30 M_{\pi}^2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 a_2^- ,
\]
which is discussed (in the $\chi SU(3)$ current algebra (CA) approach) by Lang
\cite{Lang}. Without taking into account loop corrections
\[
A^-_{CA} =0.
\]
From the results (\ref{xmin.lec}) and (\ref{xmin.lec2}) one finds
{\small\begin{eqnarray}
A^- \hspace{-1mm}&=&\hspace{-1mm} {M_{\pi} \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} {M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 \over 160 \pi^3 F_{\pi}^4}
+ {M_{\pi}^3 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( {3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over 32 \pi^3 F_{\pi}^4}
+ {3 B_2^r M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 \over 2 \pi F_{\pi}^2} + {3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} {\bar \mu}_{\pi} \over \pi
F_{\pi}^4}\right),\nonumber
\end{eqnarray}}
\unskip valid to fourth chiral order.
The chiral-symmetry breaking terms (proportional to $B_3^r$) cancel in this
sum.
The first- and the third-order loop corrections are $0.009 M_{\pi}^{-1}$ and
$0.011 M_{\pi}^{-1}$ respectively. The correction from the polynomial $\nu^3$ -- which is of
$\chi SU(3)$ order $p^6$ -- can be estimated with the $\chi SU(3)$ value for
$B_2^r$ and is about 10 \% of the loop corrections. This leads to
\[
A^- = 0.022 M_{\pi}^{-1}.
\]
Alternatively, the unknown constant $B_2^r$ can be expressed in terms of the
observables $a_2^-$ and $b_1^-$,
{\small \begin{eqnarray}\label{exact}
A^- \hspace{-1mm}&=&\hspace{-1mm} {M_{\pi} \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} +
M_{\pi}} {M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 \over 240 \pi^3 F_{\pi}^4} + {M_{\pi}^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}}
\left( -{1 \over 8 \pi F_{\pi}^2} + 10
M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^3 a_2^- -{M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2 \over 480 \pi^3 F_{\pi}^4} \right) \nonumber \\
&&- {M_{\pi}^3 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( 4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} b_1^- +
{ 1 \over 16 \pi^3 F_{\pi}^4} \right)- {M_{\pi}^4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left(
10 a_2^- + 4 b_1^- \right),
\end{eqnarray}}
\unskip giving a result which is exact to fourth $\chi SU(2)$ order and which
involves no low-energy constants.
Table \ref{threshpar.min} shows the numerical values obtained by using
$\chi SU(3)$ estimates for $B_1^r$, $B_2^r$ and $B_3^r$ in
(\ref{xmin.lec}) and (\ref{xmin.lec2}), given in the appendix.
Some experimental values for $a_0^-$ are given
in table \ref{thresh.exp}. More accurate data on threshold parameters of higher
order would be needed to test relations (\ref{aba}), (\ref{exact}) and
(\ref{aba2}).
\begin{table}[!]
\caption{\footnotesize {\bf Threshold parameters associated with the amplitude $T^-$.}
There are tree ({\rm CA}) and loop contributions (loops) from
${\cal L}_{\pi {\mbox{\tiny\hspace{-0.4mm}$K$}}}^{(1)}$. The total result ($\chi SU(2)+\chi SU(3)$) also
contains polynomial contributions from ${\cal L}_{\pi {\mbox{\tiny\hspace{-0.4mm}$K$}}}^{(3)}$, with
low-energy constants estimated from $\chi SU(3)$ at a renormalisation scale of
$\mu=770 {\rm MeV}$. }\label{threshpar.min}
\begin{center} \begin{tabular}{|c|c|c|c|} \hline
\hspace{1mm} thresh. param. \hspace{1mm} & \hspace{7mm} CA \hspace{7mm} &
\hspace{7mm}loops \hspace{3mm} &\hspace{6mm} $\chi SU(2) + \chi SU(3)$
\hspace{6mm} \\ \hline \hline
$a_0^- M_{\pi}$ & 0.20 & 0.01 &$0.23 \pm 0.01$ \\ \hline
$b_0^- M_{\pi}^{3}$& 0.10 & 0.01 &$0.10 \pm 0.01$ \\ \hline
$c_0^- M_{\pi}^{5}$& -0.03 & -0.01 &$-0.052 \pm 0.003$ \\ \hline
$a_1^- M_{\pi}^{3}$& 0.009 & -0.001 &$0.013 \pm 0.002$ \\ \hline
$b_1^- M_{\pi}^{5}$&-0.001 & -0.001 &$-0.002 \pm 0.001$ \\ \hline
$a_2^- M_{\pi}^{5}$& 0 &$2 \cdot 10^{-5}$ & $(20 \pm 7) \cdot 10^{-5}$ \\ \hline
\end{tabular} \end{center}
\end{table}
The values for the scattering lengths are consistent with the result in
\cite{BKMthresh}, where the slope parameters $b_0^{1/2}$ and $b_0^{3/2}$,
however, were obtained as
\begin{equation}\label{bmbk}
(b_0^{1/2})_{\scriptsize \cite{BKMthresh}} = (0.14 \pm 0.02) M_{\pi}^{-3},
\mbox{ } \quad (b_0^{3/2})_{\scriptsize \cite{BKMthresh}} = (-0.011 \pm 0.005)
M_{\pi}^{-3},
\end{equation}
which is inconsistent with our result. In the $\chi SU(3)$ case
$M_{\pi}/M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}$ is not a chiral expansion parameter and
the analytical expressions for the threshold parameters are therefore more
voluminous then in the $\chi SU(2)$ case. Thus only numerical values are
given in \cite{BKMthresh}.
\begin{table}[h]
\caption{\footnotesize {\bf The scattering length $a_0^-$ as determined in various analyses.}
The values are in chronological order. The interval in the second column is chosen
so as to cover the experimental results for $a^{(1/2)}_0$ \cite{Bingham} and
$a^{(3/2)}_0$ \cite{Bakker}. References
\cite{Karabarbounis} and \cite{Krivoruchenko} are dispersion relation analyses
based on more recent experiments, \cite{Estabrooks} and \cite{Aston,
Estabrooks} respectively, with much more data. The error in the second column
is chosen so as to cover the results for $a^{(1/2)}_0$ and
$a^{(3/2)}_0$ given in \cite{Karabarbounis}.}\label{thresh.exp}
\begin{center} \begin{tabular}{|c|c|c|c|} \hline
thresh. param. & before 1977 & \hspace{0.5mm}Karabarbounis
\cite{Karabarbounis}\hspace{0.5mm} &\hspace{1mm} Krivoruchenko \cite{Krivoruchenko}
\hspace{1mm} \\ \hline \hline
$a_0^- M_{\pi}$ &$0.2 \dots 0.4$& $0.26 \pm 0.06$ &$0.22 \pm 0.02$ \\ \hline
\end{tabular} \end{center}
\end{table}
\subsection{Other parameter-free relations between threshold parameters}
The amplitude $T^+_l$ is even in $\nu$ and it therefore involves the
polynomials of second and fourth chiral order. In this case, chiral invariance
does not restrict the associated threshold parameters, which are listed in
appendix \ref{lepar}.
The quantity $A^+$, defined by
{\small \begin{eqnarray}
A^+ \hspace{-2mm}&=&\hspace{-2mm} a_0^+ - M_{\pi}^2 b_0^+ + {3 M_{\pi}^3 \over 2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}} \left( a_1^+ +
{b_0^+ \over 4} \right) \nonumber \\
&&\hspace{-2mm}+ M_{\pi}^4 \left[ c_0^+ - 10 a_2^+
+ \left( {1 \over 288 \pi^2 F_{\pi}^2} - { 5
\over 16 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2} \right) b_0^+ - \left( {1 \over 48 \pi^2 F_{\pi}^2} +
{ 9 \over 8 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2} \right) a_1^+ \right], \nonumber
\end{eqnarray}}
\unskip preserves chiral-power
counting in the sense that no threshold parameter is needed to an unnaturally
high precision. This quantity contains only (combinations of) coupling constants
associated with chiral-symmetry breaking terms:
{\small \begin{eqnarray}\label{A+}
A^+ \hspace{-3mm}&=&\hspace{-2mm} -{M_{\pi}^2 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} {3 \sigma_0 \over 4 \pi F_{\pi}^2} -
{M_{\pi}^3 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} {15 \sigma_0 \over 32 \pi F_{\pi}^2 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}}
\nonumber \\
&&\hspace{-2mm}+ {M_{\pi}^4 \over M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}} + M_{\pi}} \left( {3 \sigma_0 \over 64 \pi F_{\pi}^2
M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}^2} + {3 \sigma_0 (C_5^r + 2 C_6^r) \over 2 \pi F_{\pi}^2 } + { 3
\lambda_{(t m^2)} \over 4 \pi} + { 3 \lambda_{(m^4)} \over 8 \pi} +
{ \sigma_0 \over 128 \pi^3 F_{\pi}^4 } \right) \nonumber \\
\hspace{-3mm}&=&\hspace{-2mm} (0.03 \pm 0.02) M_{\pi}^{-1} + (0.006 \pm 0.004) M_{\pi}^{-1}
+ (0.004 \pm 0.006) M_{\pi}^{-1} \nonumber \\
\hspace{-3mm}&=&\hspace{-2mm} (0.04 \pm 0.03) M_{\pi}^{-1}.
\end{eqnarray}}
\unskip The numerical values again result from theoretical $\chi SU(3)$ estimations of the
$\chi SU(2)$ constants.
The same estimations can be used to obtain numerical values for the threshold
parameters from expressions (\ref{ahoch+}). However, due to the absence of
parameter-free lowest-order predictions these values, shown in table
\ref{threshpar.plu}, are less precise than
those discussed in subsection \ref{lep.subsec}.
The parameter
\[a_1^{(1/2)}=(a_1^+ + 2 a_1^-)/3=(0.018 \pm 0.002 )M_{\pi}^{-3}
\]
is in very good agreement with the dispersion relation
analysis of Karabarbounis and Shaw \cite{Karabarbounis}:
\[ 0.0178 M_{\pi}^{-3} \leq (a_1^{(1/2)})_{\scriptsize \cite{Karabarbounis}}
\leq 0.0185 M_{\pi}^{-3}. \]
Mean experimental values for $a_0^{1/2}$ and $a_0^{3/2}$ are given in
\cite{MBK}:
\[ a_0^{1/2}=0.13 \dots 0.24 M_{\pi}^{-1},\quad a_0^{3/2}=-0.13 \dots -0.05
M_{\pi}^{-1}. \]
These were obtained long before the high-statistics experiment of \cite{Aston} in
1988 which is an important reference for $\pi$--$K$ resonance data collected by the PDG
\cite{PDG}. The absence of precise experimental data for higher-order threshold
parameters precludes the verification of relation (\ref{A+}).
\begin{table}[!]
\caption{\footnotesize {\bf Threshold parameters associated with the amplitude $T^+$.}
Current algebra gives no contributions and loops must be considered form
${\cal L}_{\pi {\mbox{\tiny\hspace{-0.4mm}$K$}}}^{(1)}$ {\it and} ${\cal L}_{\pi {\mbox{\tiny\hspace{-0.4mm}$K$}}}^{(2)}$. The total result
($\chi SU(2)+\chi SU(3)$) contains in addition polynomial contributions from
${\cal L}_{\pi {\mbox{\tiny\hspace{-0.4mm}$K$}}}^{(2)}$ and ${\cal L}_{\pi {\mbox{\tiny\hspace{-0.4mm}$K$}}}^{(4)}$, with low-energy constants
estimated from $\chi SU(3)$
. }\label{threshpar.plu}
\begin{center} \begin{tabular}{|c|c|c|c|} \hline
thresh. param. &\hspace{6mm} $\chi SU(2)+\chi SU(3)$ \hspace{6mm} \\ \hline \hline
$a_0^+ M_{\pi}$ & $0.08 \pm 0.04$ \\ \hline
$b_0^+ M_{\pi}^{3}$& $0.06 \pm 0.03$ \\ \hline
$c_0^+ M_{\pi}^{5}$& $0.015 \pm 0.004$ \\ \hline
$a_1^+ M_{\pi}^{3}$& $0.023 \pm 0.008$ \\ \hline
$b_1^+ M_{\pi}^{5}$& $0.003 \pm 0.002$ \\ \hline
$a_2^+ M_{\pi}^{5}$& $(4 \pm 3) \cdot 10^{-4}$ \\ \hline
\end{tabular} \end{center}
\end{table}
As in the case of $T_{\pi K}^-$, the scattering lengths are consistent with
the result of \cite{BKMthresh}, whereas, as can be seen from equation
(\ref{bmbk}) and definition (\ref{x+x-}), the effective range parameter $b_0^+$
is half the size.
This cannot be due to the fact that the authors of
\cite{BKMthresh} use older values for the $\chi SU(3)$ constants, since the effect of this
would be too small. In the $\chi SU(3)$ case, our calculation of the
$T_{\pi K}^{3/2}$ amplitude coincides exactly with the one in \cite{MBK}.
\section{Summary and Conclusion}\label{summary}
To the best of our knowledge, this is the first analysis of the threshold
structure of pion-kaon scattering using only $\chi SU(2)$ symmetry
of the strong interactions. Chiral $SU(2)$ symmetry is less strongly broken in
nature than $\chi SU(3)$ symmetry and the chiral expansion is therefore expected to
converge more rapidly. Chiral $SU(2)$ symmetry only allows for one interaction
term of $O(p^1)$,
\[
{\cal L}^{(1)}_{\pi,K} = D_{\mu} K^+ D^{\mu} K - M^2 K^+ K,
\]
where $D_{\mu} K= \partial_{\mu} K + \Gamma_{\mu} K$ is the covariant derivative
containing the pion field (definition \ref{Gamma}).
The lowest-order contribution to the isospin $3/2$ scattering amplitude is
therefore parameter-free:
\[
T^{3 \over 2}_1 = -{\nu \over 4 F_\pi^2 },
\]
where $\nu=s-u$ contains indeed a term proportional to the pion mass and
is therefore of first chiral order\footnote{In the standard convention the variable $\nu$ is defined by
$\nu =(s-u)/4 M_{{\mbox{\tiny\hspace{-0.4mm}$K$}}}$.}. The lowest-order contributions to the threshold
parameters do not differ significantly from those given by the $\chi SU(3)$
current algebra result \cite{Lang}.
The isospin $3/2$ amplitude was calculated to fourth chiral order in the
isospin limit and without including elecromagnetic interactions. The result is
given by equation (\ref{t32.eqn}).
The loop correction to the current algebra result for the scattering length
$a_0^-$ amounts to $5\%$ at a renormalisation scale of
$\mu=770{\rm MeV}$. The loop contributions to
the parameters $b_0^-$,
$c_0^-$, $a_1^-$, $b_1^-$ and $a_2^-$ are given in table \ref{threshpar.min}.
The exact result for theses observables -- valid to fourth chiral order --
involves three unknown low-energy constants, $B_1^r$, $B_2^r$ and $B_3^r$.
The observables $b^-_1$ and $a^-_2$ measure the first two constants so that a
theoretical estimate (from chiral $SU(3)$ symmetry) of the ``symmetry-breaking
constant" $B_3^r$ produces parameter-free predictions for $a^-_0$, $b^-_0$ and
$a^-_1$. These relations are given by formulae (\ref{aba}) and (\ref{aba2}).
There is also a pure
$\chi SU(2)$ low-energy theorem -- given by formula (\ref{exact}) --
involving the above observables.
The $\nu$-even amplitude $T_{\pi K}^+$ starts at second chiral order and the associated
threshold parameters $a^+_0$, $b^+_0$, $c^+_0$, $a^+_1$, $b^+_1$ and $a^+_2$
are not restricted by chiral $SU(2)$ symmetry. However, there are relations between
these observables involving only $\chi SU(2)$ constants associated with a
chiral-symmetry-breaking term. Equation (\ref{A+}) shows the result when all these
constants are estimated from $\chi SU(3)$.
The relations between threshold parameters (\ref{aba}, \ref{exact}, \ref{A+}
and \ref{aba2}), valid to fourth
chiral ($SU(2)$) order, are the main result of the present analysis and could
not have been obtained within $\chi SU(3)$. In relations (\ref{aba}),
(\ref{A+}) and (\ref{aba2}) a theoretical $\chi SU(3)$ input
is used (only for $\chi SU(2)$ symmetry-breaking terms) whereas
relation (\ref{exact}) is completely free of any $\chi SU(3)$ approximation.
The present phenomenological data for the threshold parameters do not
allow for a test of the obtained relations between threshold parameters.
An analysis based on dispersion relations, such as was done for example
in 1978 by Johannesson and Nilsson \cite{Johannesson2} but including all
recent data, might improve this situation \cite{MBK}.
Numerical estimates for our low-energy constants were obtained by
matching the scattering amplitude with the corresponding {$\chi SU(3)$} result.
The numerical results for {\it individual} threshold parameters, where these
estimates are used, should therefore be compatible with the
{$\chi SU(3)$} result. This is the case for all scattering lengths. There is
however an incompatibility between our results for the effective ranges and
those given in \cite{BKMthresh}. In that reference no {\it analytic}
expressions are given for the effective ranges.
Another possibility to obtain estimates for the low-energy constants
in the context of chiral $SU(2)$ perturbation theory
would be to use a {\it resonance
saturation} approach, similar to the one performed by Bernard and Kaiser
\cite{BerKai} in the $\chi SU(3)$ case.
I would like to thank Heinrich Leutwyler, who suggested the subject of this
work, for his continual help as well as Gerhard Ecker and G\'{e}rard Wanders
for useful discussions and stimulating comments.
|
\section{\@startsection {section}{1}{\z@}{+3.0ex plus +1ex minus
+.2ex}{2.3ex plus .2ex}{\normalsize\bf\boldmath}}
\def\subsection{\@startsection{subsection}{2}{\z@}{+2.5ex plus +1ex
minus +.2ex}{1.5ex plus .2ex}{\normalsize\bf\boldmath}}
\def\subsubsection{\@startsection{subsubsection}{3}{\z@}{+3.25ex plus
+1ex minus +.2ex}{1.5ex plus .2ex}{\normalsize\it}}
\expandafter\ifx\csname mathrm\endcsname\relax\def\mathrm#1{{\rm #1}}\fi
\renewcommand{\theequation}{\thesection.\arabic{equation}}
\newcounter{saveeqn}
\newcommand{\alpheqn}{\setcounter{saveeqn}{\value{equation}}%
\addtocounter{saveeqn}{1} \setcounter{equation}{0}%
\renewcommand{\theequation}%
{{\thesection.\arabic{saveeqn}\alph{equation}}}}
\newcommand{\reseteqn}{\setcounter{equation}{\value{saveeqn}}%
\renewcommand{\theequation}{\thesection.\arabic{equation}}}
\@addtoreset{equation}{section}
\newcount\@tempcntc
\def\@citex[#1]#2{\if@filesw\immediate\write\@auxout{\string\citation{#2}}\fi
\@tempcnta\z@\@tempcntb\m@ne\def\@citea{}\@cite{\@for\@citeb:=#2\do
{\@ifundefined
{b@\@citeb}{\@citeo\@tempcntb\m@ne\@citea
\def\@citea{,\penalty\@m\ }{\bf ?}\@warning
{Citation `\@citeb' on page \thepage \space undefined}}%
{\setbox\z@\hbox{\global\@tempcntc0\csname
b@\@citeb\endcsname\relax}%
\ifnum\@tempcntc=\z@ \@citeo\@tempcntb\m@ne
\@citea\def\@citea{,\penalty\@m}
\hbox{\csname b@\@citeb\endcsname}%
\else
\advance\@tempcntb\@ne
\ifnum\@tempcntb=\@tempcntc
\else\advance\@tempcntb\m@ne\@citeo
\@tempcnta\@tempcntc\@tempcntb\@tempcntc\fi\fi}}\@citeo}{#1}}
\def\@citeo{\ifnum\@tempcnta>\@tempcntb\else\@citea
\def\@citea{,\penalty\@m}%
\ifnum\@tempcnta=\@tempcntb\the\@tempcnta\else
{\advance\@tempcnta\@ne\ifnum\@tempcnta=\@tempcntb \else
\def\@citea{--}\fi
\advance\@tempcnta\m@ne\the\@tempcnta\@citea\the\@tempcntb}\fi\fi}
\def,{\relax}
\def,{,}
\def\nonumber\\{\nonumber\\}
\def\co\nonumber\\{,\nonumber\\}
\def\\*[-1ex]{\\*[-1ex]}
\def\vspace{-\abovedisplayskip{\vspace{-\abovedisplayskip}
\vspace{\abovedisplayshortskip}}
\newcommand{\lsim}
{\mathrel{\raisebox{-.3em}{$\stackrel{\displaystyle <}{\sim}$}}}
\newcommand{\gsim}
{\mathrel{\raisebox{-.3em}{$\stackrel{\displaystyle >}{\sim}$}}}
\def\asymp#1%
{\mathrel{\raisebox{-.4em}{$\widetilde{\scriptstyle #1}$}}}
\def\Nlim#1{\mathrel{\raisebox{-.4em}
{$\stackrel{\displaystyle\longrightarrow}{\scriptstyle#1}$}}}
\def\Nequal#1%
{\mathrel{\raisebox{-.5em}{$\stackrel{=}{\scriptstyle\rm#1}$}}}
\newcommand{\mathpalette\make@slash}[1]{\not \hspace{-0.7mm}#1}
\def\mathpalette\make@slash{\mathpalette\make@slash}
\def\make@slash#1#2{\setbox\z@\hbox{$#1#2$}%
\hbox to 0pt{\hss$#1/$\hss\kern-\wd0}\box0}
\def\begin{equation}{\begin{equation}}
\def\end{equation}{\end{equation}}
\def\begin{eqnarray}{\begin{eqnarray}}
\def\end{eqnarray}{\end{eqnarray}}
\def\barr#1{\begin{array}{#1}}
\def\end{array}{\end{array}}
\def\begin{figure}{\begin{figure}}
\def\end{figure}{\end{figure}}
\def\begin{table}{\begin{table}}
\def\end{table}{\end{table}}
\def\begin{center}{\begin{center}}
\def\end{center}{\end{center}}
\def\nonumber{\nonumber}
\def\displaystyle{\displaystyle}
\def\textstyle{\textstyle}
\def\footnotesize{\footnotesize}
\def1.2{1.2}
\def\alpha{\alpha}
\def\beta{\beta}
\def\Gamma{\Gamma}
\def\gamma{\gamma}
\def\delta{\delta}
\def\Delta{\Delta}
\def\epsilon{\epsilon}
\def\varepsilon{\varepsilon}
\def\lambda{\lambda}
\def\omega{\omega}
\def\Omega{\Omega}
\def\sigma{\sigma}
\def\Sigma{\Sigma}
\def\vartheta{\vartheta}
\def\ri\epsilon{{\mathrm{i}}\epsilon}
\def\refeq#1{\mbox{(\ref{#1})}}
\def\refeqs#1{\mbox{(\ref{#1})}}
\def\refeqf#1{\mbox{(\ref{#1})}}
\def\reffi#1{\mbox{Figure~\ref{#1}}}
\def\reffis#1{\mbox{Figures~\ref{#1}}}
\def\refta#1{\mbox{Table~\ref{#1}}}
\def\reftas#1{\mbox{Tables~\ref{#1}}}
\def\refse#1{\mbox{Section~\ref{#1}}}
\def\refses#1{\mbox{Sections~\ref{#1}}}
\def\refapp#1{\mbox{App.~\ref{#1}}}
\def\citere#1{\mbox{Ref.~\cite{#1}}}
\def\citeres#1{\mbox{Refs.~\cite{#1}}}
\newcommand{\unskip\,\mathrm{TeV}}{\unskip\,\mathrm{TeV}}
\newcommand{\unskip\,\mathrm{GeV}}{\unskip\,\mathrm{GeV}}
\newcommand{\unskip\,\mathrm{MeV}}{\unskip\,\mathrm{MeV}}
\newcommand{\unskip\,\mathrm{pb}}{\unskip\,\mathrm{pb}}
\newcommand{\unskip\,\mathrm{fb}}{\unskip\,\mathrm{fb}}
\newcommand{{\mathrm{i}}}{{\mathrm{i}}}
\newcommand{{\mathrm{d}}}{{\mathrm{d}}}
\newcommand{{\mathrm{U}}}{{\mathrm{U}}}
\newcommand{{\mathrm{L}}}{{\mathrm{L}}}
\newcommand{{\mathrm{T}}}{{\mathrm{T}}}
\newcommand{\mathswitch{{\cal{O}}(\alpha)}}{\mathswitch{{\cal{O}}(\alpha)}}
\newcommand{\mathswitch{{\cal{O}}(\alpha^2)}}{\mathswitch{{\cal{O}}(\alpha^2)}}
\renewcommand{\L}{{\cal{L}}}
\newcommand{{\cal{M}}}{{\cal{M}}}
\def\mathswitchr#1{\relax\ifmmode{\mathrm{#1}}\else$\mathrm{#1}$\fi}
\newcommand{\mathswitch f}{\mathswitch f}
\newcommand{\mathswitch{\bar f}}{\mathswitch{\bar f}}
\newcommand{\mathswitchr V}{\mathswitchr V}
\newcommand{\mathswitchr W}{\mathswitchr W}
\newcommand{\mathswitchr w}{\mathswitchr w}
\newcommand{\mathswitchr Z}{\mathswitchr Z}
\newcommand{\mathswitchr A}{\mathswitchr A}
\newcommand{\mathswitchr g}{\mathswitchr g}
\newcommand{\mathswitchr H}{\mathswitchr H}
\newcommand{\mathswitchr e}{\mathswitchr e}
\newcommand{\mathswitch \nu_{\mathrm{e}}}{\mathswitch \nu_{\mathrm{e}}}
\newcommand{\mathswitch \bar\nu_{\mathrm{e}}}{\mathswitch \bar\nu_{\mathrm{e}}}
\newcommand{\mathswitchr d}{\mathswitchr d}
\newcommand{\bar{\mathswitchr d}}{\bar{\mathswitchr d}}
\newcommand{\mathswitchr u}{\mathswitchr u}
\newcommand{\bar{\mathswitchr u}}{\bar{\mathswitchr u}}
\newcommand{\mathswitchr s}{\mathswitchr s}
\newcommand{\bar{\mathswitchr s}}{\bar{\mathswitchr s}}
\newcommand{\mathswitchr c}{\mathswitchr c}
\newcommand{\bar{\mathswitchr c}}{\bar{\mathswitchr c}}
\newcommand{\mathswitchr b}{\mathswitchr b}
\newcommand{\mathswitchr{\bar b}}{\mathswitchr{\bar b}}
\newcommand{\mathswitchr t}{\mathswitchr t}
\newcommand{\mathswitchr{\bar t}}{\mathswitchr{\bar t}}
\newcommand{\mathswitchr {e^+}}{\mathswitchr {e^+}}
\newcommand{\mathswitchr {e^-}}{\mathswitchr {e^-}}
\newcommand{\mathswitchr {W^+}}{\mathswitchr {W^+}}
\newcommand{\mathswitchr {W^-}}{\mathswitchr {W^-}}
\newcommand{\mathswitchr {W^\pm}}{\mathswitchr {W^\pm}}
\def\mathswitch#1{\relax\ifmmode#1\else$#1$\fi}
\newcommand{\mathswitch {m_\Pf}}{\mathswitch {m_\mathswitch f}}
\newcommand{\mathswitch {m_{\Pf'}}}{\mathswitch {m_{\mathswitch f'}}}
\newcommand{\mathswitch {m_\Pl}}{\mathswitch {m_\Pl}}
\newcommand{\mathswitch {m_\Pq}}{\mathswitch {m_\Pq}}
\newcommand{\mathswitch {M_\PV}}{\mathswitch {M_\mathswitchr V}}
\newcommand{\mathswitch {M_\PW}}{\mathswitch {M_\mathswitchr W}}
\newcommand{\mathswitch {\lambda}}{\mathswitch {\lambda}}
\newcommand{\mathswitch {M_\PZ}}{\mathswitch {M_\mathswitchr Z}}
\newcommand{\mathswitch {M_\PH}}{\mathswitch {M_\mathswitchr H}}
\newcommand{\mathswitch {m_\Pe}}{\mathswitch {m_\mathswitchr e}}
\newcommand{\mathswitch {m_\mu}}{\mathswitch {m_\mu}}
\newcommand{\mathswitch {m_\tau}}{\mathswitch {m_\tau}}
\newcommand{\mathswitch {m_\Pd}}{\mathswitch {m_\mathswitchr d}}
\newcommand{\mathswitch {m_\Pu}}{\mathswitch {m_\mathswitchr u}}
\newcommand{\mathswitch {m_\Ps}}{\mathswitch {m_\mathswitchr s}}
\newcommand{\mathswitch {m_\Pc}}{\mathswitch {m_\mathswitchr c}}
\newcommand{\mathswitch {m_\Pb}}{\mathswitch {m_\mathswitchr b}}
\newcommand{\mathswitch {m_\Pt}}{\mathswitch {m_\mathswitchr t}}
\newcommand{\Gamma_{\PW}}{\Gamma_{\mathswitchr W}}
\newcommand{\Gamma_{\PZ}}{\Gamma_{\mathswitchr Z}}
\newcommand{\Gamma_{\PV}}{\Gamma_{\mathswitchr V}}
\newcommand{\mathswitch {s_\Pw}}{\mathswitch {s_\mathswitchr w}}
\newcommand{\mathswitch {c_\Pw}}{\mathswitch {c_\mathswitchr w}}
\newcommand{\mathswitch {\bar s_\Pw}}{\mathswitch {\bar s_\mathswitchr w}}
\newcommand{\mathswitch {\bar c_\Pw}}{\mathswitch {\bar c_\mathswitchr w}}
\newcommand{\mathswitch {Q_\Pf}}{\mathswitch {Q_\mathswitch f}}
\newcommand{\mathswitch {Q_\Pl}}{\mathswitch {Q_\Pl}}
\newcommand{\mathswitch {Q_\Pq}}{\mathswitch {Q_\Pq}}
\newcommand{\mathswitch {v_\Pf}}{\mathswitch {v_\mathswitch f}}
\newcommand{\mathswitch {a_\Pf}}{\mathswitch {a_\mathswitch f}}
\newcommand{\mathswitch {g_\Pe}^{\sigma}}{\mathswitch {g_\mathswitchr e}^{\sigma}}
\newcommand{\mathswitch {g_\Pe}^-}{\mathswitch {g_\mathswitchr e}^-}
\newcommand{\mathswitch {g_\Pe}^+}{\mathswitch {g_\mathswitchr e}^+}
\newcommand{\mathswitch {G_\mu}}{\mathswitch {G_\mu}}
\newcommand{\mathrm{\{V,A\}}}{\mathrm{\{V,A\}}}
\newcommand{\mathrm{V}}{\mathrm{V}}
\def\raise.9mm\hbox{\protect\rule{1.1cm}{.2mm}}{\raise.9mm\hbox{\protect\rule{1.1cm}{.2mm}}}
\def\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}{\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}}
\def\rlap{$\cdot$}\hspace*{2mm}{\rlap{$\cdot$}\hspace*{2mm}}
\def\dash\dash\dash\dash{\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\dash\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\dash}
\def\dash\hspace*{-.5mm}\dash\hspace*{1mm}\dash\hspace*{-.5mm}\dash{\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\hspace*{-.5mm}\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\hspace*{1mm}\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\hspace*{-.5mm}\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}}
\def\dot\dot\dot\dot\dot\dot{\rlap{$\cdot$}\hspace*{2mm}\dot\rlap{$\cdot$}\hspace*{2mm}\dot\rlap{$\cdot$}\hspace*{2mm}\dot}
\def\dot\dash\dot\dash\dot{\rlap{$\cdot$}\hspace*{2mm}\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\rlap{$\cdot$}\hspace*{2mm}\raise.9mm\hbox{\protect\rule{2mm}{.2mm}}\hspace*{1mm}\rlap{$\cdot$}\hspace*{2mm}}
\defi.e.\ {i.e.\ }
\defe.g.\ {e.g.\ }
\defcf.\ {cf.\ }
\newcommand{self-energy}{self-energy}
\newcommand{self-energies}{self-energies}
\newcommand{counterterm}{counterterm}
\newcommand{counterterms}{counterterms}
\newcommand{cross section}{cross section}
\newcommand{cross sections}{cross sections}
\newcommand{{\mathrm{br}}}{{\mathrm{br}}}
\newcommand{{\mathrm{QCD}}}{{\mathrm{QCD}}}
\newcommand{{\mathrm{QED}}}{{\mathrm{QED}}}
\newcommand{{\mathrm{LEP}}}{{\mathrm{LEP}}}
\newcommand{{\mathrm{SLD}}}{{\mathrm{SLD}}}
\newcommand{{\mathrm{SM}}}{{\mathrm{SM}}}
\newcommand{{\mathrm{Born}}}{{\mathrm{Born}}}
\newcommand{{\mathrm{Born}}}{{\mathrm{Born}}}
\newcommand{{\mathrm{corr}}}{{\mathrm{corr}}}
\newcommand{{\mathrm{1-loop}}}{{\mathrm{1-loop}}}
\newcommand{{\mathrm{weak}}}{{\mathrm{weak}}}
\newcommand{{\mathrm{cut}}}{{\mathrm{cut}}}
\newcommand{{\mathrm{SB}}}{{\mathrm{SB}}}
\newcommand{\mathrm{unpol}}{\mathrm{unpol}}
\newcommand{{\mathrm{CMS}}}{{\mathrm{CMS}}}
\newcommand{{\mathrm{CC}}}{{\mathrm{CC}}}
\newcommand{{\mathrm{NC}}}{{\mathrm{NC}}}
\newcommand{{\mathrm{virt}}}{{\mathrm{virt}}}
\newcommand{{\mathrm{soft}}}{{\mathrm{soft}}}
\newcommand{{\mathrm{coll}}}{{\mathrm{coll}}}
\newcommand{{\mathrm{sub}}}{{\mathrm{sub}}}
\renewcommand{\min}{{\mathrm{min}}}
\renewcommand{\max}{{\mathrm{max}}}
\newcommand{N_f^{\mathrm{c}}}{N_f^{\mathrm{c}}}
\newcommand{\mathrm{U}}{\mathrm{U}}
\newcommand{\mathrm{SU}}{\mathrm{SU}}
\def\mathop{\mathrm{arctan}}\nolimits{\mathop{\mathrm{arctan}}\nolimits}
\def\mathop{\mathrm{Li}_2}\nolimits{\mathop{\mathrm{Li}_2}\nolimits}
\def\mathop{\overline{\mathrm{Li}_2}}\nolimits{\mathop{\overline{\mathrm{Li}_2}}\nolimits}
\def\mathop{{\cal L}i_2}\nolimits{\mathop{{\cal L}i_2}\nolimits}
\def\mathop{\mathrm{Re}}\nolimits{\mathop{\mathrm{Re}}\nolimits}
\def\mathop{\mathrm{Im}}\nolimits{\mathop{\mathrm{Im}}\nolimits}
\def\mathop{\mathrm{sgn}}\nolimits{\mathop{\mathrm{sgn}}\nolimits}
\hyphenation{brems-strah-lung}
\marginparwidth 1.2cm
\marginparsep 0.2cm
\renewcommand{\topfraction}{1.0}
\renewcommand{\bottomfraction}{1.0}
\renewcommand{\textfraction}{0.2}
\renewcommand{\floatpagefraction}{0.7}
\newcommand{\Pep\Pem\to 4f}{\mathswitchr {e^+}\mathswitchr {e^-}\to 4f}
\newcommand{\eeffff\gamma}{\Pep\Pem\to 4f\gamma}
\renewcommand{\O}{{\cal O}}
\newcommand{\langle p_a p_b \rangle}{\langle p_a p_b \rangle}
\newcommand{\langle p_a p_c \rangle}{\langle p_a p_c \rangle}
\newcommand{\langle p_a p_d \rangle}{\langle p_a p_d \rangle}
\newcommand{\langle p_a p_e \rangle}{\langle p_a p_e \rangle}
\newcommand{\langle p_a p_f \rangle}{\langle p_a p_f \rangle}
\newcommand{\langle p_b p_c \rangle}{\langle p_b p_c \rangle}
\newcommand{\langle p_b p_d \rangle}{\langle p_b p_d \rangle}
\newcommand{\langle p_b p_e \rangle}{\langle p_b p_e \rangle}
\newcommand{\langle p_b p_f \rangle}{\langle p_b p_f \rangle}
\newcommand{\langle p_c p_d \rangle}{\langle p_c p_d \rangle}
\newcommand{\langle p_c p_e \rangle}{\langle p_c p_e \rangle}
\newcommand{\langle p_c p_f \rangle}{\langle p_c p_f \rangle}
\newcommand{\langle p_d p_e \rangle}{\langle p_d p_e \rangle}
\newcommand{\langle p_d p_f \rangle}{\langle p_d p_f \rangle}
\newcommand{\langle p_e p_f \rangle}{\langle p_e p_f \rangle}
\newcommand{\langle p_a k \rangle}{\langle p_a k \rangle}
\newcommand{\langle p_b k \rangle}{\langle p_b k \rangle}
\newcommand{\langle p_c k \rangle}{\langle p_c k \rangle}
\newcommand{\langle p_d k \rangle}{\langle p_d k \rangle}
\newcommand{\langle p_e k \rangle}{\langle p_e k \rangle}
\newcommand{\langle p_f k \rangle}{\langle p_f k \rangle}
\newcommand{\spab^*}{\langle p_a p_b \rangle^*}
\newcommand{\spac^*}{\langle p_a p_c \rangle^*}
\newcommand{\spad^*}{\langle p_a p_d \rangle^*}
\newcommand{\spae^*}{\langle p_a p_e \rangle^*}
\newcommand{\spaf^*}{\langle p_a p_f \rangle^*}
\newcommand{\spbc^*}{\langle p_b p_c \rangle^*}
\newcommand{\spbd^*}{\langle p_b p_d \rangle^*}
\newcommand{\spbe^*}{\langle p_b p_e \rangle^*}
\newcommand{\spbf^*}{\langle p_b p_f \rangle^*}
\newcommand{\spcd^*}{\langle p_c p_d \rangle^*}
\newcommand{\spce^*}{\langle p_c p_e \rangle^*}
\newcommand{\spcf^*}{\langle p_c p_f \rangle^*}
\newcommand{\spde^*}{\langle p_d p_e \rangle^*}
\newcommand{\spdf^*}{\langle p_d p_f \rangle^*}
\newcommand{\spef^*}{\langle p_e p_f \rangle^*}
\newcommand{\spak^*}{\langle p_a k \rangle^*}
\newcommand{\spbk^*}{\langle p_b k \rangle^*}
\newcommand{\spck^*}{\langle p_c k \rangle^*}
\newcommand{\spdk^*}{\langle p_d k \rangle^*}
\newcommand{\spek^*}{\langle p_e k \rangle^*}
\newcommand{\spfk^*}{\langle p_f k \rangle^*}
\newcommand{\mpar}[1]{{\marginpar{\hbadness10000%
\sloppy\hfuzz10pt\boldmath\bf#1}}%
\typeout{marginpar: #1}\ignorespaces}
\marginparwidth 1.2cm
\marginparsep 0.2cm
\def\today{\relax}
\def\relax{\relax}
\def\relax{\relax}
\def\relax{\relax}
\def\draft{
\def******************************{******************************}
\def\thtystars\thtystars{******************************\thtystars}
\typeout{}
\typeout{\thtystars\thtystars**}
\typeout{* Draft mode!
For final version remove \protect\draft\space in source file *}
\typeout{\thtystars\thtystars**}
\typeout{}
\def\today{\today}
\def\relax{\marginpar[\boldmath\hfil$\uparrow$]%
{\boldmath$\uparrow$\hfil}%
\typeout{marginpar: $\uparrow$}\ignorespaces}
\def\relax{\marginpar[\boldmath\hfil$\downarrow$]%
{\boldmath$\downarrow$\hfil}%
\typeout{marginpar: $\downarrow$}\ignorespaces}
\def\relax{\marginpar[\boldmath\hfil$\rightarrow$]%
{\boldmath$\leftarrow $\hfil}%
\typeout{marginpar: $\leftrightarrow$}\ignorespaces}
\def\Mua{\marginpar[\boldmath\hfil$\Uparrow$]%
{\boldmath$\Uparrow$\hfil}%
\typeout{marginpar: $\uparrow$}\ignorespaces}
\def\Mda{\marginpar[\boldmath\hfil$\Downarrow$]%
{\boldmath$\Downarrow$\hfil}%
\typeout{marginpar: $\downarrow$}\ignorespaces}
\def\Mla{\marginpar[\boldmath\hfil$\Rightarrow$]%
{\boldmath$\Leftarrow $\hfil}%
\typeout{marginpar: $\leftrightarrow$}\ignorespaces}
\overfullrule 5pt
\oddsidemargin -15mm
\marginparwidth 29mm
}
\def\strut\leaders\hbox{*}\hfill\strut{\strut\leaders\hbox{*}\hfill\strut}
\def\hfil\strut\hfil\hbox to \textwidth {\stars}\hfil{\hfil\strut\hfil\hbox to \textwidth {\strut\leaders\hbox{*}\hfill\strut}\hfil}
\begin{document}
\thispagestyle{empty}
\def\arabic{footnote}{\fnsymbol{footnote}}
\setcounter{footnote}{1}
\null
\today\hfill BI-TP 99/10 \\
\strut\hfill PSI-PR-99-12\\
\strut\hfill hep-ph/9904472
\vfill
\begin{center}
{\Large \bf\boldmath
Predictions for all processes $\mathswitchr {e^+}\mathswitchr {e^-}\to4 \mbox{ fermions} + \gamma$%
\par} \vskip 2.5em
\vspace{1cm}
{\large
{\sc A.\ Denner$^1$, S.\ Dittmaier$^2$, M. Roth$^{1,3}$ and
D.\ Wackeroth$^1$} } \\[1cm]
$^1$ {\it Paul-Scherrer-Institut, W\"urenlingen und Villigen\\
CH--5232 Villigen PSI, Switzerland} \\[0.5cm]
$^2$ {\it Theoretische Physik, Universit\"at Bielefeld \\
D-33615 Bielefeld, Germany}
\\[0.5cm]
$^3$ {\it Institut f\"ur Theoretische Physik, ETH-H\"onggerberg\\
CH--8093 Z\"urich, Switzerland}
\par \vskip 1em
\end{center}\par
\vskip 2cm
{\bf Abstract:} \par
The complete matrix elements for $\Pep\Pem\to 4f$ and $\eeffff\gamma$ are
calculated in the Electroweak Standard Model for
polarized massless fermions. The matrix elements for all final states are
reduced to a few compact generic functions. Monte Carlo generators for
$\Pep\Pem\to 4f$ and $\eeffff\gamma$ are constructed.
We compare different treatments of
the finite widths of the electroweak gauge bosons; in particular, we
include a scheme with a complex gauge-boson
mass that obeys all Ward identities.
The detailed discussion of numerical results comprises integrated cross
sections as well as photon-energy distributions for all different final
states.
\par
\vskip 1cm
\noindent
April 1999
\null
\setcounter{page}{0}
\clearpage
\def\arabic{footnote}{\arabic{footnote}}
\setcounter{footnote}{0}
\section{Introduction}
\label{se:intro}
When exceeding the \mathswitchr W-pair production threshold, the LEP collider
started a new era in the verification of the Electroweak Standard
Model (SM): the study of the properties of the \mathswitchr W~boson and of its
interactions.
While the most important process at
LEP2 in this respect is certainly $\mathswitchr {e^+}\mathswitchr {e^-}\to\mathswitchr {W^+}\mathswitchr {W^-}\to4f$, many
other reactions have now become accessible. Besides the
4-fermion-production processes, including single \mathswitchr W-boson production,
single \mathswitchr Z-boson production, or \mathswitchr Z-boson-pair production, LEP2
and especially a future linear collider allow us
to investigate another class of processes, namely
$\eeffff\gamma$.
The physical interest in the processes $\eeffff\gamma$ is
twofold. First of all, they are an important building block for the
radiative corrections to $\mathswitchr {e^+}\mathswitchr {e^-}\to4f$, and their effect must be
taken into account in order to get precise predictions for the
observables that are used for the measurement of the \mathswitchr W-boson mass
and the
triple-gauge-boson couplings.
On the other hand, those processes themselves involve interesting
physics. They include, in particular, triple-gauge-boson-production
processes
such as $\mathswitchr {W^+}\mathswitchr {W^-}\gamma$, $\mathswitchr Z\PZ\gamma$, or $\mathswitchr Z\gamma\ga$ production
and can therefore be used to obtain information on the quartic
gauge-boson couplings $\gamma\ga WW$, $\gamma ZWW$, and $\gamma\ga ZZ$.
While only a few events of this kind are expected at LEP2,
these studies can be performed in more detail at future linear
$\mathswitchr {e^+}\mathswitchr {e^-}$ colliders \cite{Be92}.
Important contributions to $\Pep\Pem\to 4f$ and $\eeffff\gamma$ arise from
subprocesses with two resonant gauge bosons. These
subprocesses do not only contain interesting physics, such as the
quartic gauge-boson self-interactions, but also dominate the cross
sections in those regions of phase space where the invariant masses of
certain combinations of final-state particles are close to the mass of
the corresponding nearly resonant
gauge bosons.
Therefore, it is sometimes a reasonable approach to consider only
the resonant contributions.
In the
most naive approximation, the produced gauge bosons are treated as
stable particles. Most of the existing calculations for
triple-gauge-boson production have been performed in this way, such as
many calculations for $\mathswitchr {e^+}\mathswitchr {e^-}\to\mathswitchr {W^+}\mathswitchr {W^-}\gamma$ \cite{WWg}.
In an improved approach, the so-called pole expansion
\cite{polescheme,Ae94}, the resonances are treated exactly, i.e.\ the
decay of the produced gauge bosons is taken into account, but the
matrix elements are expanded about the poles of the resonances. Once
the non-resonant diagrams have been left out, the expansion is in fact
mandatory in order to retain gauge invariance. If only the leading
terms in the expansion are kept, this approach is known as the pole
approximation, or double-pole approximation in the presence of two
resonances.
The accuracy of the (double-)pole approximation is, at best, of the
order of $\Gamma_{\PV}/\mathswitch {M_\PV}$, where $\mathswitch {M_\PV}$ and $\Gamma_{\PV}$ are the mass and the width
of the relevant gauge bosons, and thus typically at the level of
several per cent. Consequently, this approach is only reasonable if the
experimental accuracy is correspondingly low.
The error estimate of $\Gamma_{\PV}/\mathswitch {M_\PV}$ is too optimistic in situations in which
scales of order $\Gamma_{\PV}$, or smaller, are involved. This is, in particular,
the case in threshold regions or if photons with energies of the order
of $\Gamma_{\PV}$ are emitted from the resonant gauge bosons.
On the other hand, the quality of the pole approximation can be
improved by applying appropriate invariant-mass cuts.
Note that the double-pole approximation is particularly well-suited
for the calculation of the radiative corrections to
gauge-boson pair production, since the error resulting from
the double-pole approximation is suppressed by an additional factor
$\alpha/\pi$ in this case. In this approximation the
corrections can be classified into factorizable and non-factorizable
corrections \cite{Ae94,wwrev}. The virtual factorizable corrections
can be composed of the known corrections to
on-shell \mathswitchr W-pair
production \cite{rcwprod} and on-shell \mathswitchr W~decay \cite{rcwdecay}, and the
non-factorizable corrections have recently been calculated \cite{nfc}.
In contrast to the virtual factorizable corrections,
the real factorizable corrections cannot be simply taken over from the
on-shell processes. In the case of real photon
emission the definition of the double-pole approximation is
non-trivial if the energy of the final-state photon is of the order of
$\Gamma_{\PV}$, since the resonances before and after photon emission are not
well separated in phase space.
A possible double-pole approximation of the $\mathswitch{{\cal{O}}(\alpha)}$
corrections to four-lepton production has been discussed in \citere{be98b}.
In our calculation of $\eeffff\gamma$ we
do not use the double-pole approximation since the full calculation is
feasable with reasonable effort. Unlike the double-pole
approximation, the full calculation is valid for arbitrary processes
in the set $\eeffff\gamma$. Moreover, when using the exact results for the
real corrections in an $\mathswitch{{\cal{O}}(\alpha)}$ calculation of 4 fermion-production
processes, the reliability of possible approximations can be tested.
Some results for $\eeffff\gamma$ with an observable photon already exist in
the literature. In \citeres{Ae91,Ae91a} the contributions to the
matrix elements involving two resonant \mathswitchr W~bosons have been calculated
and implemented into a Monte Carlo generator. This generator has been
extended to include collinear bremsstrahlung \cite{vO94} and used to
discuss the effect of hard photons at LEP2 \cite{vO96}. The complete
cross section for the process $\mathswitchr {e^+}\mathswitchr {e^-}\to\mathswitchr u\,\bar{\mathswitchr d}\,\mathswitchr {e^-}\mathswitch \bar\nu_{\mathrm{e}}\gamma$ has
been discussed in \citere{Fu94}. In \citere{Ca97}, the complete
matrix elements for the processes $\eeffff\gamma$ have been calculated
using an iterative numerical algorithm without referring to Feynman
diagrams. We are, however, interested in explicit analytical results
on the amplitudes for various reasons. In particular, we want to have
full control over the implementation of the finite width of the
virtual vector bosons and to select single diagrams, such as the
doubly-resonant ones.
So far no results for $\eeffff\gamma$ with $\mathswitchr {e^+}\mathswitchr {e^-}$ pairs in the final
state have been published.
In order to perform the calculation as efficient as possible we have
reduced all processes to a small number of generic contributions. For
$\Pep\Pem\to 4f$, the calculation is similar to the one in \citere{Be94},
and the generic contributions correspond to individual Feynman diagrams.
In the case of $\eeffff\gamma$ we have combined groups of diagrams
in such a way that the
resulting generic contributions can be classified in the same way as
those for $\Pep\Pem\to 4f$. As a consequence, the generic contributions are
individually gauge-invariant with respect to the external photon. The
number and the complexity of diagrams in the generic contributions for
$\eeffff\gamma$
has been reduced by using a non-linear gauge-fixing condition
for the W-boson field
\cite{nlgauge}. In this way, many cancellations between diagrams are
avoided, without any
further algebraic manipulations. Finally, for the helicity amplitudes
corresponding to the generic contributions concise results
have been obtained by using the
Weyl--van~der~Waerden formalism (see
\citere{wvdw} and references therein).
After the matrix elements have been calculated, the finite widths
of the resonant particles have to be introduced. We have done this in
different
ways and compared the different treatments for $\Pep\Pem\to 4f$ and
$\eeffff\gamma$. In particular, we have discussed a ``complex-mass
scheme'', which preserves all Ward identities and is still rather
simple to apply.
The matrix elements to $\Pep\Pem\to 4f$ and $\eeffff\gamma$
exhibit a complex peaking behaviour owing to propagators
of massless particles and Breit--Wigner resonances,
so that the integration over the
8- and 11-dimensional phase
spaces, respectively, is not straightforward.
In order to obtain numerically stable results,
we adopt the multi-channel integration
method \cite{Be94,Multichannel} and
reduce the Monte Carlo error by the adaptive weight optimization
procedure described in \citere{Kl94}.
In the multi-channel approach,
we define a suitable mapping of random numbers into phase space variables
for each arising propagator structure.
These variables are generated according to distributions that
approximate this specific peaking behavior of the integrand.
For $\Pep\Pem\to 4f$ and $\eeffff\gamma$ we identify up to 128 and 928 channels,
respectively, which necessitates an efficient and generic procedure
for the phase-space generation. We wrote two independent Monte Carlo
programs following the general strategy outlined above.
They differ in the realization of a generic procedure for the construction
of the phase-space generators.
The paper is organized as follows: in \refse{se:anres} we describe the
calculation of the helicity amplitudes for
$\Pep\Pem\to 4f$ and $\eeffff\gamma$ and list the complete results. The Monte Carlo
programs are
described in \refse{se:MC}. In \refse{se:numres} the numerical results
are discussed, and \refse{se:sum}
contains a summary and an outlook.
\section{Analytical results}
\label{se:anres}
\subsection{Notation and conventions}
\label{se:not&con}
We consider reactions of the types
\begin{eqnarray}\label{eq:eeffff}
\mathswitchr {e^+}(p_+,\sigma_+)+\mathswitchr {e^-}(p_-,\sigma_-) &\to&
f_1(k_1,\sigma_1)+\bar f_2(k_2,\sigma_2)+f_3(k_3,\sigma_3)+\bar f_4(k_4,\sigma_4),
\qquad\\
\mathswitchr {e^+}(p_+,\sigma_+)+\mathswitchr {e^-}(p_-,\sigma_-) &\to&
f_1(k_1,\sigma_1)+\bar f_2(k_2,\sigma_2)+f_3(k_3,\sigma_3)+\bar f_4(k_4,\sigma_4)
+\gamma(k_5,\lambda).\qquad
\label{eq:eeffffg}
\end{eqnarray}
The arguments label the momenta $p_\pm$, $k_i$ and helicities
$\sigma_i=\pm1/2$, $\lambda=\pm1$ of the corresponding particles. We often
use only the signs to denote the helicities. The fermion masses are
neglected everywhere.
For the Feynman rules we use the conventions of \citere{sm}. In
particular, all fields and momenta are incoming. It is
convenient to use a non-linear gauge-fixing term \cite{nlgauge}
of the form
\begin{eqnarray}\label{eq:nlgauge}
{\cal L}_{\mathrm{fix}} &=& -\left| \partial^\mu W^+_\mu + {\mathrm{i}} e
(A^\mu - \frac{\mathswitch {c_\Pw}}{\mathswitch {s_\Pw}} Z^\mu) W^+_\mu
-{\mathrm{i}} \mathswitch {M_\PW} \phi^+ \right|^2 \nonumber\\*
& &{}- \frac{1}{2} (\partial^\mu Z_\mu - \mathswitch {M_\PZ} \chi)^2
- \frac{1}{2} (\partial^\mu A_\mu)^2 \;,
\end{eqnarray}
where $\phi^\pm$ and $\chi$ are the would-be Goldstone bosons of the
$W^\pm$ and $Z$ fields, respectively.
With this choice, the $\phi^\pm W^\mp A$ vertices vanish, and the
bosonic couplings that are relevant for $\eeffff\gamma$ read
\begin{eqnarray}
\setlength{\unitlength}{1pt}
\label{fr:vvv}
\barr{l}
\begin{picture}(90,80)(-50,-38)
\Text(-45,3)[lb]{$V_{\mu},k_{V}$}
\Text(35,27)[rb]{$W^+_{\nu},k_{+}$}
\Text(35,-27)[rt] {$W^-_{\rho},k_{-}$}
\Vertex(0,0){2}
\Photon(0,0)(35,25){2}{3.5}
\Photon(0,0)(35,-25){2}{3.5}
\Photon(0,0)(-45,0){2}{3.5}
\end{picture}
\end{array}
&&\barr{l}
= -{\mathrm{i}} e g_{VWW}\left[
g_{\nu\rho}(k_- -k_+)_\mu-2g_{\mu\nu}k_{V,\rho}+2g_{\mu\rho}k_{V,\nu} \right],
\nonumber\\
\end{array}\\%\nlndia
\label{fr:vvvv}
\barr{l}
\begin{picture}(90,80)(-50,-36)
\Text(-35,27)[lb]{$A_{\mu}$}
\Text(-35,-27)[lt]{$V_{\nu}$}
\Text(35,27)[rb]{$W^+_{\rho}$}
\Text(35,-27)[rt]{$W^-_{\sigma}$}
\Vertex(0,0){2}
\Photon(0,0)(35,25){2}{3.5}
\Photon(0,0)(35,-25){2}{3.5}
\Photon(0,0)(-35,25){2}{3.5}
\Photon(0,0)(-35,-25){2}{3.5}
\end{picture}
\end{array}
&&\barr{l}
= -2{\mathrm{i}} e g_{VWW} \, g_{\mu\nu}g_{\rho\sigma},
\end{array}
\end{eqnarray}
with $V=A,Z$, and the coupling factors
\begin{equation}
g_{AWW} = 1, \qquad g_{ZWW} = -\frac{\mathswitch {c_\Pw}}{\mathswitch {s_\Pw}}.
\end{equation}
Note that the gauge-boson propagators have the same simple form as in
the 't~Hooft--Feynman gauge, i.e.\ they are proportional to the metric
tensor $g_{\mu\nu}$.
This gauge choice eliminates some diagrams and simplifies others owing
to the simpler structure of the photon--gauge-boson couplings.
The vector-boson--fermion--fermion couplings have the usual form
\begin{equation}
\label{fr:Vff}
\barr{l}
\begin{picture}(90,80)(-50,-36)
\Text(-45,5)[lb]{$V_{\mu}$}
\Text(35,27)[rb]{$\bar{f}_{i}$}
\Text(35,-27)[rt]{$f_{j}$}
\Vertex(0,0){2}
\ArrowLine(0,0)(35,25)
\ArrowLine(35,-25)(0,0)
\Photon(0,0)(-45,0){2}{3.5}
\end{picture}
\end{array}
\barr{l}
= \displaystyle{\mathrm{i}} e \gamma_\mu \sum_\sigma g^\sigma_{V\bar f_i f_j}\omega_\sigma,
\end{array}
\end{equation}
where
$\omega_\pm=(1\pm\gamma_5)/2$.
The corresponding coupling factors read
\begin{equation}
g^\sigma_{A\bar f_i f_i} = -Q_i, \qquad
g^\sigma_{Z\bar f_i f_i} =
-\frac{\mathswitch {s_\Pw}}{\mathswitch {c_\Pw}}Q_i+\frac{I^3_{{\mathrm{w}},i}}{\mathswitch {c_\Pw}\mathswitch {s_\Pw}}\delta_{\sigma-},
\qquad
g^\sigma_{W\bar f_i f'_i} = \frac{1}{\sqrt{2}\mathswitch {s_\Pw}}\delta_{\sigma-},
\end{equation}
where $Q_i$ and $I^3_{{\mathrm{w}},i}=\pm{1/2}$ denote the
relative charge and the weak isospin of the fermion $f_i$,
respectively, and $f'_i$ is the weak-isospin partner of $f_i$.
The colour factor of a fermion $f_i$ is denoted by
$N^{\mathrm{c}}_{f_i}$, i.e.\
$N^{\mathrm{c}}_{\mathrm{lepton}}=1$ and
$N^{\mathrm{c}}_{\mathrm{quark}}=3$.
\subsection{Classification of final states for $\Pep\Pem\to 4f$}
\label{se:pclass}
The final states for $\Pep\Pem\to 4f$ have
already been classified in
\citeres{Be94,Ba94,CERN9601mcgen}. We introduce a classification that
is very close to the one of \citeres{Ba94,CERN9601mcgen}. It is based
on the production
mechanism, i.e.\ whether the
reactions proceed via charged-current (CC), or
neutral-current (NC) interactions, or via both interaction types. The
classification can be performed by considering the quantum numbers of
the final-state fermion pairs. In the following, $f$ and $F$ denote
different fermions ($f\ne F$) that are neither electrons nor electron
neutrinos
($f,F \ne \mathswitchr {e^-},\nu_\mathswitchr e$), and their weak-isospin partners are denoted
by $f'$ and $F'$, respectively. We find the following 11 classes of
processes (in parenthesis the corresponding classification of
\citere{CERN9601mcgen} is given):
\renewcommand{\labelenumi}{(\roman{enumi})}
\renewcommand{\labelenumii}{(\alph{enumii})}
\newcommand{\ientry}[2]{\mbox{\rlap{#1}\hspace*{5cm}#2}}
\begin{enumerate}
{\samepage
\item CC reactions:
\begin{enumerate}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to f \bar f' F \bar F'$,}%
{({\em CC11} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to \nu_\mathswitchr e \mathswitchr {e^+} f \bar f'$,}%
{({\em CC20} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to f \bar f' \mathswitchr {e^-} \bar\nu_\mathswitchr e$,}%
{({\em CC20} family),}
\end{enumerate}
}
\item NC reactions:
\begin{enumerate}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to f \bar f F \bar F$,}%
{({\em NC32} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to f \bar f f \bar f$,}%
{({\em NC4$\cdot$16} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to \mathswitchr {e^-} \mathswitchr {e^+} f \bar f$,}%
{({\em NC48} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to \mathswitchr {e^-} \mathswitchr {e^+} \mathswitchr {e^-} \mathswitchr {e^+}$,}%
{({\em NC4$\cdot$36} family),}
\end{enumerate}
\item Mixed CC/NC reactions:
\begin{enumerate}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to f \bar f f' \bar f'$,}%
{({\em mix43} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to \nu_\mathswitchr e \bar\nu_\mathswitchr e f \bar f$,}%
{({\em NC21} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to \nu_\mathswitchr e \bar\nu_\mathswitchr e \nu_\mathswitchr e \bar\nu_\mathswitchr e$,}%
{({\em NC4$\cdot$9} family),}
\item \ientry{$\mathswitchr {e^+}\mathswitchr {e^-} \to \nu_\mathswitchr e \bar\nu_\mathswitchr e \mathswitchr {e^-} \mathswitchr {e^+}$,}%
{({\em mix56} family).}
\end{enumerate}
\end{enumerate}
The radiation of an additional photon does not change this
classification.
\subsection{Generic diagrams and amplitudes for \boldmath{$\mathswitchr {e^+}\mathswitchr {e^-}\to 4f$} }
\label{se:ee4f}
In order to explain and to illustrate our generic approach we first
list the results for $\Pep\Pem\to 4f$. All these processes can be composed
from only two generic diagrams, the abelian and non-abelian diagrams
shown in \reffi{ee4fdiags}. All external fermions $f_{a,\dots,f}$ are
assumed to be incoming, and the momenta and helicities are denoted by
$p_{a,\dots,f}$ and $\sigma_{a,\dots,f}$, respectively.
\begin{figure}
\begin{center}
\setlength{\unitlength}{1pt}
\begin{picture}(420,150)(0,-20)
\Text(0,120)[lb]{a) abelian graph}
\Text(210,120)[lb]{b) non-abelian graph}
\put(20,-8){
\begin{picture}(150,100)(0,0)
\ArrowLine(35,70)( 5, 95)
\ArrowLine( 5, 5)(35, 30)
\ArrowLine(35,30)(35,70)
\Photon(35,30)(90,20){2}{6}
\Photon(35,70)(90,80){-2}{6}
\Vertex(35,70){2.0}
\Vertex(35,30){2.0}
\Vertex(90,80){2.0}
\Vertex(90,20){2.0}
\ArrowLine(90,80)(120, 95)
\ArrowLine(120,65)(90,80)
\ArrowLine(120, 5)( 90,20)
\ArrowLine( 90,20)(120,35)
\put(55,82){$V_1$}
\put(55,10){$V_2$}
\put( 20,50){$f_g$}
\put(-15,110){$\bar f_a(p_a,\sigma_a)$}
\put(-15,-10){$f_b(p_b,\sigma_b)$}
\put(125,90){$\bar f_c(p_c,\sigma_c)$}
\put(125,65){$f_d(p_d,\sigma_d)$}
\put(125,30){$\bar f_e(p_e,\sigma_e)$}
\put(125, 5){$f_f(p_f,\sigma_f)$}
\end{picture}
}
\put(245,-8){
\begin{picture}(150,100)(0,0)
\ArrowLine( 15,50)(-15, 65)
\ArrowLine(-15,35)( 15, 50)
\Photon(15,50)(60,50){2}{5}
\Photon(60,50)(90,20){-2}{5}
\Photon(60,50)(90,80){2}{5}
\Vertex(15,50){2.0}
\Vertex(60,50){2.0}
\Vertex(90,80){2.0}
\Vertex(90,20){2.0}
\ArrowLine(90,80)(120, 95)
\ArrowLine(120,65)(90,80)
\ArrowLine(120, 5)( 90,20)
\ArrowLine( 90,20)(120,35)
\put(30,58){$V$}
\put(62,70){$W$}
\put(62,18){$W$}
\put(-35,75){$\bar f_a(p_a,\sigma_a)$}
\put(-35,20){$f_b(p_b,\sigma_a)$}
\put(125,90){$\bar f_c(p_c,\sigma_c)$}
\put(125,65){$f_d(p_d,\sigma_a)$}
\put(125,30){$\bar f_e(p_e,\sigma_e)$}
\put(125, 5){$f_f(p_f,\sigma_f)$}
\end{picture}
}
\end{picture}
\end{center}
\caption{Generic diagrams for $\mathswitchr {e^+}\mathswitchr {e^-}\to 4f$}
\label{ee4fdiags}
\end{figure}
The helicity amplitudes of these diagrams are calculated within the
Weyl--van~der~Waerden (WvdW) formalism following the conventions of
\citere{wvdw} (see also references therein).
\subsubsection{Leptonic and semi-leptonic final states}
We first treat purely leptonic and semi-leptonic final states. In this
case, none of the gauge bosons in the generic graphs of \reffi{ee4fdiags}
can be a gluon, and the colour structure trivially leads to a global
factor $N^{\mathrm{c}}_{f_1}N^{\mathrm{c}}_{f_3}$, which is equal to 1
or 3, after summing the squared amplitude over the colour degrees of
freedom. The results for the generic amplitudes are
\begin{eqnarray}\label{genfun4fNC}
&& \hspace*{-3em}
{\cal M}^{\sigma_a,\sigma_b,\sigma_c,\sigma_d,\sigma_e,\sigma_f}_{V_1 V_2}
(p_a,p_b,p_c,p_d,p_e,p_f)
\nonumber\\*
\hspace*{2em}
&=& -4e^4 \delta_{\sigma_a,-\sigma_b} \delta_{\sigma_c,-\sigma_d} \delta_{\sigma_e,-\sigma_f} \,
g^{\sigma_b}_{V_1\bar f_a f_g} g^{\sigma_b}_{V_2\bar f_g f_b}
g^{\sigma_d}_{V_1\bar f_c f_d} g^{\sigma_f}_{V_2\bar f_e f_f}
\nonumber\\ && {} \times
\frac{P_{V_1}(p_c+p_d) P_{V_2}(p_e+p_f)}{(p_b+p_e+p_f)^2} \,
A^{\sigma_a,\sigma_c,\sigma_e}_2(p_a,p_b,p_c,p_d,p_e,p_f),
\hspace*{3em}
\\[1em]
&& \hspace*{-3em}
{\cal M}^{\sigma_a,\sigma_b,\sigma_c,\sigma_d,\sigma_e,\sigma_f}_{VWW}
(p_a,p_b,p_c,p_d,p_e,p_f)
\nonumber\\*
&=& -4e^4 \delta_{\sigma_a,-\sigma_b} \delta_{\sigma_c,+} \delta_{\sigma_d,-}
\delta_{\sigma_e,+} \delta_{\sigma_f,-} \,
(Q_c-Q_d)g_{VWW} g^{\sigma_b}_{V\bar f_a f_b}
g^{-}_{W\bar f_c f_d} g^{-}_{W\bar f_e f_f}
\nonumber\\ && {} \times
P_V(p_a+p_b) P_W(p_c+p_d) P_W(p_e+p_f) \,
A^{\sigma_a}_3(p_a,p_b,p_c,p_d,p_e,p_f),
\label{genfun4fCC}
\end{eqnarray}
where
the vector-boson propagators are abbreviated by
\begin{equation}
P_V(p) = \frac{1}{p^2-\mathswitch {M_\PV}^2}, \qquad V=A,Z,W,g, \qquad M_A = M_g = 0.
\end{equation}
(The case of the gluon is included for later convenience.)
The auxiliary functions $A^{\sigma_a,\sigma_c,\sigma_e}_2$ and $A^{\sigma_a}_3$ are
expressed in terms of WvdW spinor products,
\begin{eqnarray}
A^{{+}{+}{+}}_2(p_a,p_b,p_c,p_d,p_e,p_f) &=&
\langle p_a p_c \rangle\spbf^*(\spbd^*\langle p_b p_e \rangle+\spdf^*\langle p_e p_f \rangle),
\nonumber\\
A^{{+}{+}{-}}_2(p_a,p_b,p_c,p_d,p_e,p_f) &=&
A^{{+}{+}{+}}_2(p_a,p_b,p_c,p_d,p_f,p_e),
\nonumber\\
A^{{+}{-}{+}}_2(p_a,p_b,p_c,p_d,p_e,p_f) &=&
A^{{+}{+}{+}}_2(p_a,p_b,p_d,p_c,p_e,p_f),
\nonumber\\
A^{{+}{-}{-}}_2(p_a,p_b,p_c,p_d,p_e,p_f) &=&
A^{{+}{+}{+}}_2(p_a,p_b,p_d,p_c,p_f,p_e),
\nonumber\\
A^{{-},\sigma_c,\sigma_d}_2(p_a,p_b,p_c,p_d,p_e,p_f) &=&
\left(A^{{+},-\sigma_c,-\sigma_d}_2(p_a,p_b,p_c,p_d,p_e,p_f)\right)^*,
\\[1em]
A^{+}_3(p_a,p_b,p_c,p_d,p_e,p_f) &=& \phantom{{}+{}}
\spbd^*\spbf^*\langle p_a p_b \rangle\langle p_c p_e \rangle + \spbd^*\spdf^*\langle p_a p_e \rangle\langle p_c p_d \rangle
\nonumber\\
&& {}
+ \spbf^*\spdf^*\langle p_a p_c \rangle\langle p_e p_f \rangle,
\nonumber\\
A^{-}_3(p_a,p_b,p_c,p_d,p_e,p_f) &=&
A^{+}_3(p_b,p_a,p_c,p_d,p_e,p_f).
\end{eqnarray}
The spinor products are defined by
\begin{equation}
\langle pq\rangle=\epsilon^{AB}p_A q_B
=2\sqrt{p_0 q_0} \,\Biggl[
{\mathrm{e}}^{-{\mathrm{i}}\phi_p}\cos\frac{\theta_p}{2}\sin\frac{\theta_q}{2}
-{\mathrm{e}}^{-{\mathrm{i}}\phi_q}\cos\frac{\theta_q}{2}\sin\frac{\theta_p}{2}
\Biggr],
\end{equation}
where $p_A$, $q_A$ are the associated momentum spinors for the momenta
\begin{eqnarray}
p^\mu&=&p_0(1,\sin\theta_p\cos\phi_p,\sin\theta_p\sin\phi_p,\cos\theta_p),\nonumber\\
q^\mu&=&p_0(1,\sin\theta_q\cos\phi_q,\sin\theta_q\sin\phi_q,\cos\theta_q).
\end{eqnarray}
Incoming fermions are turned into outgoing ones by crossing, which
is performed by inverting the corresponding fermion momenta and
helicities.
If the generic functions are called with negative momenta $-p$, $-q$, it
is understood that only the complex conjugate spinor products get the
corresponding sign change. We illustrate this by
simple examples:
\begin{equation}
\begin{array}[b]{rllll}
A(p,q) &= \langle pq \rangle
&=\phantom{-}A(p,-q)
&=\phantom{-}A(-p,q)
&=A(-p,-q), \\
B(p,q) &= \langle pq \rangle^*
&=-B(p,-q)
&=-B(-p,q)
&=B(-p,-q).
\end{array}
\end{equation}
We have checked
the results for the generic diagrams against those of \citere{Be94}
and found
agreement.
Using the results for the generic diagrams of \reffi{ee4fdiags}, the
helicity amplitudes for all possible processes involving
six external fermions can be built up.
It is convenient to construct first
the amplitudes for the process types
CC(a) and NC(a) (see \refse{se:pclass}) in terms of the generic functions
\refeq{genfun4fNC} and \refeq{genfun4fCC}, because these amplitudes
are the basic subamplitudes of
the other channels. The full amplitude for each
process type can be built up from those subamplitudes
by appropriate substitutions and linear combinations.
We first list the helicity amplitudes for the CC processes:
\begin{eqnarray}
\label{eq:ee4f_cca}
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=&
{\cal M}^{\sigma_+,\sigma_-,-\sigma_1,-\sigma_2,-\sigma_3,-\sigma_4}_{WW}
(p_+,p_-,-k_1,-k_2,-k_3,-k_4)
\nonumber\\ && {}
+ \sum_{V=\gamma,Z} \Bigl[ \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,-\sigma_1,-\sigma_2,-\sigma_3,-\sigma_4}_{VWW}
(p_+,p_-,-k_1,-k_2,-k_3,-k_4)
\nonumber\\ && \hphantom{{} + \sum_{V=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_1,-\sigma_2,\sigma_+,\sigma_-,-\sigma_3,-\sigma_4}_{VW}
(-k_1,-k_2,p_+,p_-,-k_3,-k_4)
\nonumber\\ && \hphantom{{} + \sum_{V=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_3,-\sigma_4,\sigma_+,\sigma_-,-\sigma_1,-\sigma_2}_{VW}
(-k_3,-k_4,p_+,p_-,-k_1,-k_2)
\nonumber\\ && \hphantom{{} + \sum_{V=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_1,-\sigma_2,-\sigma_3,-\sigma_4,\sigma_+,\sigma_-}_{WV}
(-k_1,-k_2,-k_3,-k_4,p_+,p_-)
\nonumber\\ && \hphantom{{} + \sum_{V=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_3,-\sigma_4,-\sigma_1,-\sigma_2,\sigma_+,\sigma_-}_{WV}
(-k_3,-k_4,-k_1,-k_2,p_+,p_-) \Bigr],
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CCb}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{\sigma_+,-\sigma_2,\sigma_1,-\sigma_-,\sigma_3,\sigma_4}_{\mathrm{CCa}}
(p_+,-k_2,k_1,-p_-,k_3,k_4),
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CCc}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_3,\sigma_-,\sigma_1,\sigma_2,-\sigma_+,\sigma_4}_{\mathrm{CCa}}
(-k_3,p_-,k_1,k_2,-p_+,k_4).
\end{eqnarray}
The ones for the NC processes are given by
\begin{eqnarray}
\label{eq:ee4f_nca}
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \sum_{V_1,V_2=\gamma,Z} \Bigl[ \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,-\sigma_1,-\sigma_2,-\sigma_3,-\sigma_4}_{V_1 V_2}
(p_+,p_-,-k_1,-k_2,-k_3,-k_4)
\nonumber\\ && \hphantom{\sum_{V_1,V_2=\gamma,Z} \Bigl[}
+ {\cal M}^{\sigma_+,\sigma_-,-\sigma_3,-\sigma_4,-\sigma_1,-\sigma_2}_{V_1 V_2}
(p_+,p_-,-k_3,-k_4,-k_1,-k_2)
\nonumber\\ && \hphantom{\sum_{V_1,V_2=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_1,-\sigma_2,\sigma_+,\sigma_-,-\sigma_3,-\sigma_4}_{V_1 V_2}
(-k_1,-k_2,p_+,p_-,-k_3,-k_4)
\nonumber\\ && \hphantom{\sum_{V_1,V_2=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_3,-\sigma_4,\sigma_+,\sigma_-,-\sigma_1,-\sigma_2}_{V_1 V_2}
(-k_3,-k_4,p_+,p_-,-k_1,-k_2)
\nonumber\\ && \hphantom{\sum_{V_1,V_2=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_1,-\sigma_2,-\sigma_3,-\sigma_4,\sigma_+,\sigma_-}_{V_1 V_2}
(-k_1,-k_2,-k_3,-k_4,p_+,p_-)
\nonumber\\ && \hphantom{\sum_{V_1,V_2=\gamma,Z} \Bigl[}
+ {\cal M}^{-\sigma_3,-\sigma_4,-\sigma_1,-\sigma_2,\sigma_+,\sigma_-}_{V_1 V_2}
(-k_3,-k_4,-k_1,-k_2,p_+,p_-) \Bigl],
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCb}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\* &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_3,\sigma_2,\sigma_1,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_3,k_2,k_1,k_4),
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCc}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\* &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_1,\sigma_-,-\sigma_+,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(-k_1,p_-,-p_+,k_2,k_3,k_4),
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCd}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\* &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_3,\sigma_2,\sigma_1,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_3,k_2,k_1,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_1,\sigma_-,-\sigma_+,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(-k_1,p_-,-p_+,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_3,\sigma_-,\sigma_1,\sigma_2,-\sigma_+,\sigma_4}_{\mathrm{NCa}}
(-k_3,p_-,k_1,k_2,-p_+,k_4)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_1,\sigma_-,\sigma_3,\sigma_2,-\sigma_+,\sigma_4}_{\mathrm{NCa}}
(-k_1,p_-,k_3,k_2,-p_+,k_4)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_3,\sigma_-,-\sigma_+,\sigma_2,\sigma_1,\sigma_4}_{\mathrm{NCa}}
(-k_3,p_-,-p_+,k_2,k_1,k_4).
\end{eqnarray}
Finally, the helicity amplitudes for reactions of mixed CC/NC type read
\begin{eqnarray}
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CC/NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_4,\sigma_3,\sigma_2}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_4,k_3,k_2),
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CC/NCb}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_3,-\sigma_4,\sigma_1,-\sigma_-,-\sigma_+,\sigma_2}_{\mathrm{CCa}}
(-k_3,-k_4,k_1,-p_-,-p_+,k_2),
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CC/NCc}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_3,\sigma_2,\sigma_1,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_3,k_2,k_1,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_1,-\sigma_2,-\sigma_+,\sigma_4,\sigma_3,-\sigma_-}_{\mathrm{CCa}}
(-k_1,-k_2,k_3,-p_-,-p_+,k_4)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_1,-\sigma_4,-\sigma_+,\sigma_2,\sigma_3,-\sigma_-}_{\mathrm{CCa}}
(-k_1,-k_4,k_3,-p_-,-p_+,k_2)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_3,-\sigma_2,-\sigma_+,\sigma_4,\sigma_1,-\sigma_-}_{\mathrm{CCa}}
(-k_3,-k_2,k_1,-p_-,-p_+,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_3,-\sigma_4,-\sigma_+,\sigma_2,\sigma_1,-\sigma_-}_{\mathrm{CCa}}
(-k_3,-k_4,k_1,-p_-,-p_+,k_2),
\\[1em]
&& \hspace*{-4em}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CC/NCd}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ &=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\nonumber\\ && {}
- {\cal M}^{-\sigma_3,\sigma_-,\sigma_1,\sigma_2,-\sigma_+,\sigma_4}_{\mathrm{NCa}}
(-k_3,p_-,k_1,k_2,-p_+,k_4)
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_4,\sigma_3,\sigma_2}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_4,k_3,k_2)
\nonumber\\ && {}
+ {\cal M}^{\sigma_+,-\sigma_4,\sigma_1,-\sigma_-,\sigma_3,\sigma_2}_{\mathrm{CCa}}
(p_+,-k_4,k_1,-p_-,k_3,k_2)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_3,\sigma_-,\sigma_1,\sigma_4,-\sigma_+,\sigma_2}_{\mathrm{CCa}}
(-k_3,p_-,k_1,k_4,-p_+,k_2)
\nonumber\\ && {}
- {\cal M}^{-\sigma_3,-\sigma_4,\sigma_1,-\sigma_-,-\sigma_+,\sigma_2}_{\mathrm{CCa}}
(-k_3,-k_4,k_1,-p_-,-p_+,k_2).
\end{eqnarray}
The relative signs between contributions of the basic subamplitudes
${\cal M}_{\mathrm{CCa}}$ and ${\cal M}_{\mathrm{NCa}}$ to the full matrix
elements account for the sign changes resulting from interchanging
external fermion lines.
For the CC reactions, the amplitudes ${\cal{M}}_{\mathrm{CCa}}$ are the
smallest gauge-invariant subset of diagrams \cite{Bo99}. In the
case of NC reactions, the amplitudes ${\cal{M}}_{\mathrm{NCa}}$ are composed of
three separately gauge-invariant subamplitudes consisting of the first
two lines, the two lines in the middle, and the last two lines of
\refeq{eq:ee4f_nca}.
\subsubsection{Hadronic final states}
\label{se:hadfinstat}
\begin{sloppypar}
Next we inspect purely hadronic final states, i.e.\ the cases where
all final-state fermions $f_i$ are quarks. This concerns only the
channels CC(a), NC(a), NC(b), and CC/NC(a) given in
\refse{se:pclass}. The colour structure of the quarks leads to two
kinds of modifications. Firstly, the summation of the squared
amplitudes over the colour degrees of freedom can become
non-trivial, and secondly, the possibility of virtual-gluon exchange
between the quarks has to be taken into account. More precisely,
there are diagrams of type (a) in \reffi{ee4fdiags} in which one of
the gauge bosons $V_{1,2}$ is a gluon. The other gauge boson of
$V_{1,2}$ can only be a photon or Z~boson, since this boson has to
couple to the incoming $\mathswitchr {e^+}\mathswitchr {e^-}$ pair. Consequently, there is an
impact of gluon-exchange diagrams only for the channels NC(a),
NC(b), and CC/NC(a), but not for CC(a). This can be easily seen by
inspecting the generic diagrams in \reffi{ee4fdiags}: the presence
of a gluon exchange requires two quark--antiquark pairs $q\bar q$ in
the final state.
\end{sloppypar}
We first inspect the colour structure of the purely electroweak
diagrams. Since the colour structure of each diagram contributing to the
basic channels CC(a) and NC(a) is the same, the corresponding amplitudes
factorize into a simple colour part and the ``colour-singlet
amplitudes'' ${\cal M}_{\mathrm{CCa}}$ and ${\cal M}_{\mathrm{NCa}}$,
given in \refeq{eq:ee4f_cca} and \refeq{eq:ee4f_nca}, respectively.
The amplitudes for NC(b) and CC/NC(a) are composed from the ones of CC(a)
and NC(a) in a way that is analogous to the singlet case, but now the
colour indices $c_i$ of the quarks $f_i$ have to be taken into account.
Indicating the electroweak amplitudes for fully hadronic final states
by ``$\mathrm{had,ew}$'', and writing colour indices explicitly, we get
\begin{eqnarray}
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{CCa,had,ew,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_2,k_3,k_4) \, \delta_{c_1 c_2} \delta_{c_3 c_4},
\\[1em]
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{NCa,had,ew,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4) \, \delta_{c_1 c_2} \delta_{c_3 c_4},
\\[1em]
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{NCb,had,ew,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\*
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4) \, \delta_{c_1 c_2} \delta_{c_3 c_4}
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_3,\sigma_2,\sigma_1,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_3,k_2,k_1,k_4) \, \delta_{c_3 c_2} \delta_{c_1 c_4},
\label{eq:ee4f_ncbhadew}
\\[1em]
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{CC/NCa,had,ew,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\*
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{NCa}}
(p_+,p_-,k_1,k_2,k_3,k_4) \, \delta_{c_1 c_2} \delta_{c_3 c_4}
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_4,\sigma_3,\sigma_2}_{\mathrm{CCa}}
(p_+,p_-,k_1,k_4,k_3,k_2) \, \delta_{c_1 c_4} \delta_{c_3 c_2}.
\label{eq:ee4f_ccncahadew}
\end{eqnarray}
In the calculation of the gluon-exchange diagrams we can also make use
of the ``colour-singlet'' result \refeq{genfun4fNC} for the generic diagram
(a) of \reffi{ee4fdiags}, after splitting off the colour structure
appropriately. Since each of these diagrams
involves exactly one internal gluon,
exchanged by the two quark lines, the corresponding matrix elements can
be deduced in a simple way from the diagrams in which the gluon is
replaced by a photon. The
gluon-exchange contributions to the channels NC(b)
and CC/NC(a) can again be composed from the ones for NC(a). Making use of
the auxiliary function
\begin{eqnarray}
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{g}}
(p_+,p_-,k_1,k_2,k_3,k_4)
} \hspace*{3em} &&
\nonumber\\*
& = \displaystyle\frac{g_{\mathrm{s}}^2}{Q_1 Q_3 e^2} \,
\sum_{V=\gamma,Z} \Bigl[ & \phantom{{}+{}}
{\cal M}^{-\sigma_1,-\sigma_2,\sigma_+,\sigma_-,-\sigma_3,-\sigma_4}_{V\gamma}
(-k_1,-k_2,p_+,p_-,-k_3,-k_4)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_3,-\sigma_4,\sigma_+,\sigma_-,-\sigma_1,-\sigma_2}_{V\gamma}
(-k_3,-k_4,p_+,p_-,-k_1,-k_2)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_1,-\sigma_2,-\sigma_3,-\sigma_4,\sigma_+,\sigma_-}_{\gamma V}
(-k_1,-k_2,-k_3,-k_4,p_+,p_-)
\nonumber\\ && {}
+ {\cal M}^{-\sigma_3,-\sigma_4,-\sigma_1,-\sigma_2,\sigma_+,\sigma_-}_{\gamma V}
(-k_3,-k_4,-k_1,-k_2,p_+,p_-) \Bigl],
\hspace*{2em}
\end{eqnarray}
where $g_{\mathrm{s}}=\sqrt{4\pi\alpha_{\mathrm{s}}}$ is the strong
gauge coupling, the matrix elements involving gluon exchange
explicitly read
\begin{eqnarray}
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{NCa,had,gluon,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\*
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{g}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\; \textstyle\frac{1}{4} \, \lambda^a_{c_1 c_2} \lambda^a_{c_3 c_4},
\\[1em]
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{NCb,had,gluon,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\*
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{g}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\; \textstyle\frac{1}{4} \, \lambda^a_{c_1 c_2} \lambda^a_{c_3 c_4}
\nonumber\\ && {}
- {\cal M}^{\sigma_+,\sigma_-,\sigma_3,\sigma_2,\sigma_1,\sigma_4}_{\mathrm{g}}
(p_+,p_-,k_3,k_2,k_1,k_4)
\; \textstyle\frac{1}{4} \, \lambda^a_{c_3 c_2} \lambda^a_{c_1 c_4},
\label{eq:ee4f_ncbhadqcd}
\\[1em]
\lefteqn{
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\mathrm{CC/NCa,had,gluon,}}c_1,c_2,c_3,c_4}
(p_+,p_-,k_1,k_2,k_3,k_4) } \hspace*{7em} &&
\nonumber\\*
&=& \phantom{{}+{}}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}_{\mathrm{g}}
(p_+,p_-,k_1,k_2,k_3,k_4)
\; \textstyle\frac{1}{4} \, \lambda^a_{c_1 c_2} \lambda^a_{c_3 c_4}.
\end{eqnarray}
The colour structure is easily evaluated by making use of the
completeness relation
$\lambda^a_{ij} \lambda^a_{kl} = -\textstyle\frac{2}{3}\delta_{ij}\delta_{kl}
+2\delta_{il}\delta_{jk}$ for the Gell-Mann matrices $\lambda^a_{ij}$.
\begin{sloppypar}
The complete matrix elements for the fully hadronic channels result
from the sum of the
purely electroweak and the gluon-exchange contributions,
\begin{equation}
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\ldots\mathrm{,had,}}c_1,c_2,c_3,c_4}
= {\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\ldots\mathrm{,had,ew,}}c_1,c_2,c_3,c_4}+
{\cal M}^{\sigma_+,\sigma_-,\sigma_1,\sigma_2,\sigma_3,\sigma_4}
_{{\ldots\mathrm{,had,gluon,}}c_1,c_2,c_3,c_4}.
\end{equation}
The gluon-exchange contributions are separately gauge-invariant.
For clarity, we explicitly write down the colour-summed squared matrix
elements for the fully hadronic channels. Abbreviating
${\cal M}_{\dots}
^{\sigma_+,\sigma_-,\sigma_a,\sigma_b,\sigma_c,\sigma_d}(p_+,p_-,k_a,k_b,k_c,k_d)$
by ${\cal M}_{\dots}(a,b,c,d)$ we obtain
\end{sloppypar}
\begin{eqnarray}
\lefteqn{
\sum_{\mathrm{colour}}|{\cal M}_{\mathrm{CCa,had}}(1,2,3,4)|^2
= 9 |{\cal M}_{\mathrm{CCa}}(1,2,3,4)|^2,
} \hspace*{0em} &&
\\[1em]
\lefteqn{
\sum_{\mathrm{colour}}|{\cal M}_{\mathrm{NCa,had}}(1,2,3,4)|^2
= 9 |{\cal M}_{\mathrm{NCa}}(1,2,3,4)|^2
+ 2 |{\cal M}_{\mathrm{g}}(1,2,3,4)|^2,
} \hspace*{0em} &&
\\[1em]
\lefteqn{
\sum_{\mathrm{colour}}|{\cal M}_{\mathrm{NCb,had}}(1,2,3,4)|^2
} \hspace*{0em} &&
\nonumber\\*
&=& 9 |{\cal M}_{\mathrm{NCa}}(1,2,3,4)|^2
+ 9 |{\cal M}_{\mathrm{NCa}}(3,2,1,4)|^2
- 6\mathop{\mathrm{Re}}\nolimits\big\{ {\cal M}_{\mathrm{NCa}}(1,2,3,4)
{\cal M}^*_{\mathrm{NCa}}(3,2,1,4) \big\}
\nonumber\\ && {}
+ 2 |{\cal M}_{\mathrm{g}}(1,2,3,4)|^2
+ 2 |{\cal M}_{\mathrm{g}}(3,2,1,4)|^2
+ \textstyle\frac{4}{3}\mathop{\mathrm{Re}}\nolimits\big\{ {\cal M}_{\mathrm{g}}(1,2,3,4)
{\cal M}^*_{\mathrm{g}}(3,2,1,4) \big\}
\nonumber\\ && {}
- 8\mathop{\mathrm{Re}}\nolimits\big\{ {\cal M}_{\mathrm{NCa}}(1,2,3,4)
{\cal M}^*_{\mathrm{g}}(3,2,1,4) \big\}
- 8\mathop{\mathrm{Re}}\nolimits\big\{ {\cal M}_{\mathrm{NCa}}(3,2,1,4)
{\cal M}^*_{\mathrm{g}}(1,2,3,4) \big\},
\nonumber\\*
\\[1em]
\lefteqn{
\sum_{\mathrm{colour}}|{\cal M}_{\mathrm{CC/NCa,had}}(1,2,3,4)|^2
} \hspace*{0em} &&
\nonumber\\
&=& 9 |{\cal M}_{\mathrm{NCa}}(1,2,3,4)|^2
+ 9 |{\cal M}_{\mathrm{CCa}}(1,4,3,2)|^2
- 6\mathop{\mathrm{Re}}\nolimits\big\{ {\cal M}_{\mathrm{NCa}}(1,2,3,4)
{\cal M}^*_{\mathrm{CCa}}(1,4,3,2) \big\}
\nonumber\\ && {}
+ 2 |{\cal M}_{\mathrm{g}}(1,2,3,4)|^2
- 8\mathop{\mathrm{Re}}\nolimits\big\{ {\cal M}_{\mathrm{CCa}}(1,4,3,2)
{\cal M}^*_{\mathrm{g}}(1,2,3,4) \big\}.
\end{eqnarray}
Owing to the colour structure of the diagrams, a non-zero interference
between purely electroweak and gluon-exchange contributions is only
possible if the four final-state fermions can be combined into one
single closed fermion line in the squared diagram. This implies that
fermion pairs must couple to different resonances in the
electroweak and the gluon-exchange diagrams,
leading to a global suppression of such interference effects in the
phase-space integration (see \refse{se:QCD}).
\subsection{Generic functions and amplitudes for
\boldmath{$\eeffff\gamma$} }
\label{se:genfunction}
The generic functions for $\eeffff\gamma$ can be constructed in a similar
way. The idea is to combine the contributions of all those graphs to
one generic function that reduce to the same graph after removing the
radiated photon. These combined contributions to $\eeffff\gamma$ are
classified in the same way as the diagrams for the corresponding
process $\mathswitchr {e^+}\mathswitchr {e^-}\to 4f$, i.e.\ the graphs of \reffi{ee4fdiags} also
represent the generic functions for $\eeffff\gamma$. Finally, all
amplitudes for $\eeffff\gamma$ can again be constructed from only two
generic functions. Note that the number of individual Feynman
diagrams ranges between 14 and 1008 for the various processes. We
note that the generic functions can in fact be used to construct the
amplitudes for all processes involving exactly six external fermions
and one external photon, such as $\mathswitchr {e^-}\Pem\to4f\gamma$ and $\mathswitchr e\gamma\to5f$.
As a virtue of this approach, the so-defined generic functions fulfill
the QED Ward identity for the external photon, i.e.\ replacing the
photon polarization vector by the photon momentum yields zero for each
generic function. This is simply a consequence of electromagnetic
charge conservation. Consequently, in the actual calculation in the
WvdW formalism the gauge spinor of the photon drops out in each
contribution separately.
Assuming the external fermions as incoming and the photon as outgoing,
the generic functions read
\begin{eqnarray}
&& \hspace*{-3em}
{\cal M}^{\sigma_a,\sigma_b,\sigma_c,\sigma_d,\sigma_e,\sigma_f,\lambda}_{V_1 V_2}
(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
\hspace*{2em}
&=& -4\sqrt{2}e^5 \delta_{\sigma_a,-\sigma_b} \delta_{\sigma_c,-\sigma_d} \delta_{\sigma_e,-\sigma_f} \,
g^{\sigma_b}_{V_1\bar f_a f_g} g^{\sigma_b}_{V_2\bar f_g f_b}
g^{\sigma_d}_{V_1\bar f_c f_d} g^{\sigma_f}_{V_2\bar f_e f_f}
\nonumber\\ && {} \times
A^{\sigma_a,\sigma_c,\sigma_e,\lambda}_2
(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k),
\hspace*{3em}
\\[1em]
&& \hspace*{-3em}
{\cal M}^{\sigma_a,\sigma_b,\sigma_c,\sigma_d,\sigma_e,\sigma_f,\lambda}_{VWW}
(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& -4\sqrt{2}e^5 \delta_{\sigma_a,-\sigma_b} \delta_{\sigma_c,+} \delta_{\sigma_d,-}
\delta_{\sigma_e,+} \delta_{\sigma_f,-} \,
(Q_c-Q_d)g_{VWW} g^{\sigma_b}_{V\bar f_a f_b}
g^{-}_{W\bar f_c f_d} g^{-}_{W\bar f_e f_f}
\nonumber\\ && {} \times
A^{\sigma_a,\lambda}_3
(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k),
\label{eq:Mgenee4fA}
\end{eqnarray}
with the auxiliary functions
\begin{eqnarray}
&& \hspace*{-3em}
A^{{+}{+}{+}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
= -\langle p_a p_c \rangle \biggl\{
\nonumber\\
\hspace*{2em} &&
P_{V_1}(p_c+p_d) P_{V_2}(p_e+p_f)
\nonumber\\
&& \quad
{} \times \biggl[ \frac{\spbf^*}{\langle p_a k \rangle} \biggl(
\frac{Q_c-Q_d}{(p_b+p_e+p_f)^2} \frac{\langle p_a p_c \rangle}{\langle p_c k \rangle}
(\spbd^* \langle p_b p_e \rangle+\spdf^* \langle p_e p_f \rangle)
\nonumber\\
&& \quad
\hphantom{{} \times \biggl[ \frac{\spbf^*}{\langle p_a k \rangle} \biggl(} {}
+\frac{Q_f-Q_e}{(p_a+p_c+p_d)^2} \frac{\langle p_a p_e \rangle}{\langle p_e k \rangle}
(\spad^* \langle p_a p_e \rangle+\spcd^* \langle p_c p_e \rangle) \biggr)
\nonumber\\
&& \quad \hphantom{\times \biggl[}
+\frac{Q_b (\spad^*\langle p_a p_e \rangle+\spcd^*\langle p_c p_e \rangle) (\spbf^*\langle p_a p_b \rangle-\spfk^*\langle p_a k \rangle) }
{(p_a+p_c+p_d)^2\langle p_a k \rangle\langle p_b k \rangle}
\nonumber\\
&& \quad \hphantom{\times \biggl[}
+\frac{(Q_a+Q_c-Q_d) \spbf^*\spcd^*\langle p_a p_c \rangle (\spbk^*\langle p_b p_e \rangle-\spfk^*\langle p_e p_f \rangle) }
{(p_a+p_c+p_d)^2 (p_b+p_e+p_f)^2 \langle p_a k \rangle} \biggr]
\nonumber\\
&& {}
-\frac{Q_d-(Q_c-Q_d) 2(p_d\cdot k) P_{V_1}(p_c+p_d)}{(p_b+p_e+p_f)^2}
P_{V_1}(p_c+p_d-k) P_{V_2}(p_e+p_f) \spbf^*
\nonumber\\
&& {} \hphantom{{}+{}} \;
\times \frac{ \langle p_c p_d \rangle (\spbd^* \langle p_b p_e \rangle+\spdf^* \langle p_e p_f \rangle)
+\langle p_c k \rangle (\spbk^* \langle p_b p_e \rangle-\spfk^* \langle p_e p_f \rangle) } {\langle p_c k \rangle\langle p_d k \rangle}
\nonumber\\
&& {}
+\frac{Q_f-(Q_e-Q_f) 2(p_f\cdot k) P_{V_2}(p_e+p_f)}{(p_a+p_c+p_d)^2}
P_{V_1}(p_c+p_d) P_{V_2}(p_e+p_f-k)
\nonumber\\
&& {} \hphantom{{}+{}} \;
\times \frac{ (\spbf^* \langle p_e p_f \rangle+\spbk^* \langle p_e k \rangle) (\spad^* \langle p_a p_e \rangle+\spcd^* \langle p_c p_e \rangle)}
{\langle p_e k \rangle\langle p_f k \rangle}
\,\biggr\},
\nonumber\\
&& \hspace*{-3em}
A^{{+}{+}{-}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_a,Q_b,Q_c,Q_d,-Q_f,-Q_e,p_a,p_b,p_c,p_d,p_f,p_e,k),
\nonumber\\
&& \hspace*{-3em}
A^{{+}{-}{+}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_a,Q_b,-Q_d,-Q_c,Q_e,Q_f,p_a,p_b,p_d,p_c,p_e,p_f,k),
\nonumber\\
&& \hspace*{-3em}
A^{{+}{-}{-}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_a,Q_b,-Q_d,-Q_c,-Q_f,-Q_e,p_a,p_b,p_d,p_c,p_f,p_e,k),
\nonumber\\
&& \hspace*{-3em}
A^{{-}{+}{+}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_b,Q_a,-Q_e,-Q_f,-Q_c,-Q_d,p_b,p_a,p_e,p_f,p_c,p_d,k),
\nonumber\\
&& \hspace*{-3em}
A^{{-}{+}{-}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_b,Q_a,Q_f,Q_e,-Q_c,-Q_d,p_b,p_a,p_f,p_e,p_c,p_d,k),
\nonumber\\
&& \hspace*{-3em}
A^{{-}{-}{+}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_b,Q_a,-Q_e,-Q_f,Q_d,Q_c,p_b,p_a,p_e,p_f,p_d,p_c,k),
\nonumber\\
&& \hspace*{-3em}
A^{{-}{-}{-}{+}}_2(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}{+}{+}}_2(Q_b,Q_a,Q_f,Q_e,Q_d,Q_c,p_b,p_a,p_f,p_e,p_d,p_c,k),
\nonumber\\
&& \hspace*{-3em}
A^{\sigma_a,\sigma_c,\sigma_d,{-}}_2
(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\\*
&=& \Big( A^{-\sigma_a,-\sigma_c,-\sigma_d,{+}}_2
(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k) \Bigr)^*
\Big|_{P_{V_{1,2}}(p)\to P_{V_{1,2}}^*(p)},
\n
\label{eq:A2}
\end{eqnarray}
and
\begin{eqnarray}
&& \hspace*{-3em}
A^{{+}{+}}_3(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
\hspace*{2em}
&=& P_V(p_a+p_b) P_W(p_c+p_d) P_W(p_e+p_f)
\frac{(Q_c-Q_d)\langle p_c p_e \rangle}{\langle p_c k \rangle\langle p_e k \rangle}
\nonumber\\
&& \quad
{} \times ( \hphantom{{}+{}} \spbd^*\spbf^*\langle p_a p_b \rangle\langle p_c p_e \rangle +\spbd^*\spdf^*\langle p_a p_e \rangle\langle p_c p_d \rangle
\nonumber\\
&& \quad \hphantom{{} \times (}
{} + \spbf^*\spdf^*\langle p_a p_c \rangle\langle p_e p_f \rangle )
\nonumber\\
&& {} +P_V(p_a+p_b-k) P_W(p_c+p_d) P_W(p_e+p_f) \frac{Q_b}{\langle p_a k \rangle\langle p_b k \rangle}
\nonumber\\
&& \quad
{} \times \{ \hphantom{{}+{}} \spdf^* [ \hphantom{{}+{}}
\langle p_a p_e \rangle\langle p_c p_d \rangle (\spbd^*\langle p_a p_b \rangle-\spdk^*\langle p_a k \rangle)
\nonumber\\
&& \quad \hphantom{{} \times \{ {}+ \spdf^* [}
{} +\langle p_a p_c \rangle\langle p_e p_f \rangle (\spbf^*\langle p_a p_b \rangle-\spfk^*\langle p_a k \rangle) ]
\nonumber\\
&& \quad \hphantom{{} \times \{ }
{} + \langle p_c p_e \rangle (\spbd^*\langle p_a p_b \rangle-\spdk^*\langle p_a k \rangle) (\spbf^*\langle p_a p_b \rangle-\spfk^*\langle p_a k \rangle) \}
\nonumber\\
&& {} +P_V(p_a+p_b) P_W(p_c+p_d-k) P_W(p_e+p_f)
\frac{Q_d-(Q_c-Q_d)2(p_d\cdot k) P_W(p_c+p_d)}{\langle p_c k \rangle\langle p_d k \rangle}
\nonumber\\
&& \quad
{} \times \{ \hphantom{{}+{}}
\spbf^* [ \hphantom{{}+{}} \langle p_a p_c \rangle\langle p_e p_f \rangle (\spdf^*\langle p_c p_d \rangle-\spfk^*\langle p_c k \rangle)
\nonumber\\
&& \quad \hphantom{{} \times \{ \hphantom{{}+{}} \spbf^* [ }
{} +\langle p_c p_e \rangle\langle p_a p_b \rangle (\spbd^*\langle p_c p_d \rangle+\spbk^*\langle p_c k \rangle) ]
\nonumber\\
&& \quad \hphantom{{} \times \{ }
{} + \langle p_a p_e \rangle (\spdf^*\langle p_c p_d \rangle-\spfk^*\langle p_c k \rangle) (\spbd^*\langle p_c p_d \rangle+\spbk^*\langle p_c k \rangle) \}
\nonumber\\
&& {} +P_V(p_a+p_b) P_W(p_c+p_d) P_W(p_e+p_f-k)
\frac{Q_f+(Q_f-Q_e)2(p_f\cdot k) P_W(p_e+p_f)}{\langle p_e k \rangle\langle p_f k \rangle}
\nonumber\\
&& \quad
{} \times \{ \hphantom{{}+{}}
\spbd^* [ \hphantom{{}+{}} \langle p_c p_e \rangle\langle p_a p_b \rangle (\spbf^*\langle p_e p_f \rangle+\spbk^*\langle p_e k \rangle)
\nonumber\\
&& \quad \phantom{ {} \times \{ \hphantom{{}+{}} \spbd^* [ }
{} + \langle p_a p_e \rangle\langle p_c p_d \rangle (\spdf^*\langle p_e p_f \rangle+\spdk^*\langle p_e k \rangle) ]
\nonumber\\
&& \quad \phantom{ {} \times \{ }
{}+\langle p_a p_c \rangle (\spbf^*\langle p_e p_f \rangle+\spbk^*\langle p_e k \rangle) (\spdf^*\langle p_e p_f \rangle+\spdk^*\langle p_e k \rangle) \},
\nonumber\\
&& \hspace*{-3em}
A^{{-}{+}}_3(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\nonumber\\*
&=& A^{{+}{+}}_3(-Q_b,-Q_a,Q_c,Q_d,Q_e,Q_f,p_b,p_a,p_c,p_d,p_e,p_f,k),
\nonumber\\
&& \hspace*{-3em}
A^{\sigma_a,{-}}_3(Q_a,Q_b,Q_c,Q_d,Q_e,Q_f,p_a,p_b,p_c,p_d,p_e,p_f,k)
\\*
&=& \Bigl(
A^{-\sigma_a,{+}}_3(Q_a,Q_b,-Q_d,-Q_c,-Q_f,-Q_e,p_a,p_b,p_d,p_c,p_f,p_e,k)
\Bigr)^* \Big|_{P_{V,W}(p)\to P_{V,W}^*(p)}.
\n
\label{eq:A3}
\end{eqnarray}
The replacements $P_V\to P_V^*$ after the complex conjugation
in the last lines of \refeq{eq:A2} and \refeq{eq:A3} ensure that the
vector-boson propagators remain unaffected.
Note that the vector-boson masses do not enter explicitly in the above
results, but only via $P_V$.
In gauges such as the 't~Hooft--Feynman or the unitary gauge this feature
is obtained only after combining different Feynman graphs for $\eeffff\gamma$;
in the
non-linear gauge \refeq{eq:nlgauge} this is the case diagram by diagram.
The helicity amplitudes for $\eeffff\gamma$ follow from the
generic functions ${\cal M}_{V_1 V_2}$ and ${\cal M}_{VWW}$ of
\refeq{eq:Mgenee4fA} in exactly the same way as described in
\refse{se:ee4f}
for $\mathswitchr {e^+}\mathswitchr {e^-}\to 4f$.
This holds also for the gluon-exchange matrix elements and for the
colour factors.
Moreover, the classification of gauge-invariant sets of diagrams
for $\Pep\Pem\to 4f$ immediately yields such sets for $\eeffff\gamma$, if the additional
photon is attached to all graphs of a set in all possible ways.
We have checked analytically that the electromagnetic Ward identity
for the external photon is fulfilled for each generic contribution
separately. In addition, we have numerically compared the amplitudes
for all processes with amplitudes generated by {\sl Madgraph} \cite{St94}
for zero width of the vector bosons and found complete agreement.
We could not compare our results with {\sl Madgraph} for finite width,
because {\sl Madgraph} uses the unitary gauge for massive vector-boson
propagators and the `t~Hooft--Feynman gauge for the photon propagators,
while we are using the non-linear gauge \refeq{eq:nlgauge}. Therefore,
the matrix elements differ after introduction of finite vector-boson widths.
While the calculation with {\sl Madgraph} is fully automized,
in our calculation we have full control over the matrix
element and can, in particular, investigate various implementations of
the finite width.
A comparison of our results with those of \citeres{Ae91,Ae91a}, which include
only the matrix elements that involve two resonant \mathswitchr W~bosons,
immediately reveals the virtues of our generic approach.
\subsection{Implementation of finite gauge-boson widths}
\label{se:finwidth}
We have implemented the finite widths of the $\mathswitchr W$ and $\mathswitchr Z$ bosons in
different ways:
\begin{itemize}
\item{\it fixed width} in all propagators:
$P_V(p) = [p^2-\mathswitch {M_\PV}^2+{\mathrm{i}}\mathswitch {M_\PV}\Gamma_{\PV}]^{-1}$,
\item{\it running width} in time-like propagators:
$P_V(p) = [p^2-\mathswitch {M_\PV}^2+{\mathrm{i}} p^2(\Gamma_{\PV}/\mathswitch {M_\PV})\theta(p^2)]^{-1}$,
\item{\it complex-mass scheme:} complex gauge-boson masses everywhere,
i.e.\ $\sqrt{M_V^2-{\mathrm{i}} M_V\Gamma_V}$ instead of $M_V$ in the propagators
and in the couplings. This results, in particular, in a constant
width in all propagators,
\begin{equation}\label{Vprop}
P_V(p) = [p^2-\mathswitch {M_\PV}^2+{\mathrm{i}}\mathswitch {M_\PV}\Gamma_V]^{-1},
\end{equation}
and in a complex weak mixing angle:
\begin{equation}\label{complangle}
\mathswitch {c_\Pw}^2=1-\mathswitch {s_\Pw}^2= \frac{\mathswitch {M_\PW}^2-{\mathrm{i}}\mathswitch {M_\PW}\Gamma_{\PW}}{\mathswitch {M_\PZ}^2-{\mathrm{i}}\mathswitch {M_\PZ}\Gamma_{\PZ}}.
\end{equation}
\end{itemize}
The virtues and drawbacks of the first two schemes
have been discussed in \citere{bhf2}.
Both violate $\mathrm{SU}(2)$ gauge invariance, the running width
also $\mathrm{U}(1)$ gauge invariance.
The complex-mass scheme obeys
all Ward identities%
\footnote{In the context of electromagnetic gauge invariance, the
introduction of a complex mass has been proposed in \citere{Lo91}}
and thus gives a consistent description of the
finite-width effects in any tree-level calculation. While the
complex-mass scheme
works in general, it is particularly simple for $\eeffff\gamma$ in the
non-linear gauge \refeq{eq:nlgauge}. In this case, no couplings
involving explicit gauge-boson masses appear,
and it is sufficient to
introduce the finite gauge-boson widths in the propagators [cf.\
\refeq{Vprop}] and to introduce the complex weak mixing angle
\refeq{complangle} in the couplings. We note that a generalization of
this scheme to higher orders requires to introduce complex mass
counterterms in order to compensate for the complex masses in the
propagators \cite{St90}. We did not consider the
fermion-loop scheme \cite{bhf2,FLscheme}, which is also fully
consistent for lowest-order predictions, since it requires the
calculation of fermionic one-loop corrections to $\eeffff\gamma$
which is beyond the scope of this work.
\section{The Monte Carlo programs}
\label{se:MC}
The cross section for $\mathswitchr {e^+}\mathswitchr {e^-}\to 4f(\gamma)$ is given by
\begin{eqnarray}\label{eq:crosssection}
{\mathrm{d}} \sigma &=& \frac{(2 \pi)^{4-3 n}}{2 s}
\left[\prod\limits_{i=1}^n {\mathrm{d}}^4 k_i \,
\delta\left(k_i^2\right) \theta(k_i^0)\right]
\delta^{(4)} \left( p_+ +p_- -\sum_{i=1}^n k_i \right)
\nonumber \\
&& {}
\times |{\cal M}(p_+,p_-,k_1,\ldots ,k_n)|^2,
\end{eqnarray}
where
$n=4,5$ is the number of outgoing particles.
The helicity amplitudes ${\cal M}$ for $\mathswitchr {e^+}\mathswitchr {e^-}\to 4f(\gamma)$
have been calculated in \refses{se:ee4f} and \ref{se:genfunction},
respectively.
The phase-space integration is performed
with the help of a Monte Carlo technique, since
the Monte Carlo method allows
us to calculate a variety of observables
simultaneously and to
easily implement cuts in order to
account for the experimental situation.
The helicity amplitudes in \refeq{eq:crosssection}
exhibit a complicated peaking behaviour in different
regions of the integration domain. In order to obtain a numerically stable
result and to reduce the Monte Carlo integration error
we use a multi-channel Monte Carlo method \cite{Be94,Multichannel}, which
is briefly outlined in the following.
Before turning to the multi-channel method, we consider the treatment
of a single channel. We choose a suitable set $\vec{\Phi}$ of $3n-4$
phase-space variables to describe a point in phase space, and
determine the corresponding physical region $V$ and the relation
$k_i(\vec{\Phi})$ between the phase-space variables $\vec{\Phi}$ and
the momenta $k_1, \ldots k_n$. The phase-space integration of
\refeq{eq:crosssection} reads
\begin{eqnarray}
I_n &=& \int {\mathrm{d}} \sigma=
\int_V {\mathrm{d}} \vec{\Phi} \, \rho\Big(k_i(\vec{\Phi})\Big) \,
f\Big(k_i(\vec{\Phi})\Big),
\\\nonumber
f\Big(k_i(\vec{\Phi})\Big)&=& \frac{(2 \pi)^{4-3 n}}{2 s}
\left|{\cal M}\left(p_+,p_-,k_1(\vec{\Phi}),\ldots ,
k_n(\vec{\Phi})\right)\right|^2,
\end{eqnarray}
where $\rho$ is the phase-space density.
For the random generation of the events,
we further transform
the integration variables $\vec{\Phi}$
to $3n-4$ new variables $\vec{r}=(r_i)$
with a hypercube as integration domain:
$\vec{\Phi}=\vec{h}(\vec{r}\,)$ with $0\le r_i \le 1$.
We obtain
\begin{equation}\label{eq:phint}
I_n = \int_V {\mathrm{d}} \vec{\Phi} \, \rho\Big(k_i(\vec{\Phi})\Big) \,
f\Big(k_i(\vec{\Phi})\Big)
= \int_0^1 {\mathrm{d}} \vec{r} \,
\frac{f\Big(k_i(\vec{h}(\vec{r\,}))\Big)}
{g\Big(k_i(\vec{h}(\vec{r\,}))\Big)} \; ,
\end{equation}
where $g$ is the probability density of events generated in phase
space, defined by
\begin{equation} \label{eq:dens}
\frac{1}{g\Big(k_i(\vec{\Phi})\Big)} = \rho\Big(k_i(\vec{\Phi})\Big)
\left| \frac{\partial\vec{h}(\vec{r\,})}
{\partial\vec{r}} \right|_{\vec{r}=\vec{h}^{-1}(\vec{\Phi})} \; .
\end{equation}
If $f$ varies strongly, the efficiency of the Monte Carlo method
can be considerably enhanced by choosing the
mapping of random numbers $\vec{r}$ into $\vec{\Phi}$ in such a way that
the resulting density $g$ mimics the behaviour of $|f|$.
For this {\it importance sampling}, the
choice of $\vec{\Phi}$ is guided by the peaking structure of $f$,
which is determined by the propagators in a characteristic Feynman diagram.
We choose the variables $\vec{\Phi}$ in such a way that the invariants
corresponding to the propagators are included. Accordingly, we
decompose the $n$-particle final state into $2 \to 2$ scattering
processes with subsequent $1 \to 2$ decays. The variables
$\vec{\Phi}$ consist of Lorentz invariants $s_i, t_i$, defined as the
squares of time- and space-like momenta, respectively, and of polar
and azimuthal angles $\theta_i,\phi_i$, defined in appropriate frames.
A detailed description of the parameterization of an $n$-particle
phase space in terms of invariants and angles can be found in
\citere{By73}. The parameterization of the invariants $s_i,t_i$ in
$\vec{\Phi}=\vec{h}(\vec{r}\,)$ is chosen in such a way that the
propagator structure of the function $f$ is compensated by a similar
behaviour in the density $g$. More precisely, if $f$ contains
Breit--Wigner resonances or distributions like $s_i^{-\nu}$, which are
relevant for massless propagators, appropriate parameterizations of
$s_i$ are given by:
\begin{itemize}
\item Breit--Wigner resonances:
\begin{eqnarray}\label{eq:maps1}
s_i &=& \mathswitch {M_\PV}^2+\mathswitch {M_\PV}\Gamma_{\PV}\tan[y_1+(y_2-y_1) r_i]
\\ && \mbox{with } \;
y_{1,2}=\arctan\left(\frac{s_{\min,\max}-\mathswitch {M_\PV}^2}{\mathswitch {M_\PV}\Gamma_{\PV}}\right); \nonumber
\end{eqnarray}
\item propagators of massless particles:
\begin{eqnarray}\label{eq:maps2}
\nu\ne 1: \qquad s_i &=& \left[s_{\max}^{1-\nu} r_i
+s_{\min}^{1-\nu}(1-r_i)\right]^{1/(1-\nu)} ,
\nonumber\\[.5em]
\nu=1: \qquad
s_i &=&\exp\left[\ln(s_{\max}) r_i + \ln(s_{\min})(1-r_i)\right].
\end{eqnarray}
\end{itemize}
The parameter $\nu $ can be tuned to optimize the
Monte Carlo integration and should be chosen
$\gsim 1$. The naive expectation $\nu =2$ is not necessarily the best
choice, because the propagator poles of the differential cross section
are partly cancelled in the collinear limit. The remaining variables
in $\vec{\Phi}=\vec{h}(\vec{r}\,)$, i.e.\ those for which $f$ is
expected not to exhibit a peaking behaviour, are generated as follows:
\begin{eqnarray}\label{eq:nomaps}
s_i &=& s_{\max} r_i+s_{\min} (1-r_i), \qquad
\phi_i = 2 \pi r_i, \qquad \cos\theta_i =2 r_i-1.
\end{eqnarray}
The absolute values of the
invariants $t_i$ are generated in the same way as $s_i$.
The resulting density $g$ of events in phase space
is obtained as the product of the corresponding Jacobians,
as given in \refeq{eq:dens}.
In the appendix, we provide an explicit example for an event generation
with a specific choice of mappings $k_i(\vec{\Phi})$ and $\vec{h}(\vec{r}\,)$,
and for the calculation of the corresponding density $g$.
The differential cross sections of the processes
$\mathswitchr {e^+}\mathswitchr {e^-}\to 4f$ and especially $\eeffff\gamma$ possess very
complex peaking structures so that the peaks in the integrand
$f(\vec{\Phi})$ in \refeq{eq:phint} cannot be described properly
by only one single density $g(\vec{\Phi})$.
The {\it multi-channel approach} \cite{Be94,Multichannel}
suggests a solution to this problem. For each peaking structure we choose
a suitable set $\vec{\Phi}_k$, and accordingly a
mapping of random numbers $r_i$ into $\vec{\Phi}_k$:
$\vec{\Phi}_k=\vec{h}_k(\vec{r}\,)$ with $0 \le r_i \le 1$,
so that the resulting density
$g_k$ describes this particular peaking behaviour of $f$.
All densities $g_k$ are combined
into one density $g_{\mathrm {tot}}$
that is expected to smoothen the integrand over the whole phase-space
integration region. The phase-space integral of \refeq{eq:phint} reads
\begin{eqnarray}
I_n &=& \sum_{k=1}^M \int_V {\mathrm{d}} \vec{\Phi}_k \,
\rho_k\Big(k_i(\vec{\Phi}_k)\Big) \,
g_k\Big(k_i(\vec{\Phi}_k)\Big) \,
\frac{f\Big(k_i(\vec{\Phi}_k)\Big)}{g_{\mathrm{tot}}\Big(k_i(\vec{\Phi}_k)\Big)}
=\sum_{k=1}^M \int_0^1 {\mathrm{d}} \vec{r} \,
\frac{f\Big(k_i(\vec{h}_k(\vec{r}\,))\Big)}
{g_{\mathrm {tot}}\Big(k_i(\vec{h}_k(\vec{r}\,))\Big)},
\hspace*{2em}
\end{eqnarray}
with
\begin{eqnarray}\label{eq:mdens}
g_{\mathrm {tot}}\Big(k_i(\vec{\Phi}_k)\Big)&=&
\sum_{l=1}^M g_l\Big(k_i(\vec{\Phi}_k)\Big)
,\qquad
\frac{1}{g_l\Big(k_i(\vec{\Phi}_k)\Big)}=\rho_l\Big(k_i(\vec{\Phi}_k)\Big) \,
\left| \frac{\partial\vec{h}_l(\vec{r}\,)}
{\partial \vec{r}}\right|_{\vec{r}=\vec{h}_k^{-1}(\vec{\Phi}_k)}.
\end{eqnarray}
The different mappings $\vec{h}_k(\vec{r}\,)$ are called channels,
and $M$ is the number of all channels.
In order to reduce the Monte Carlo error further, we adopt the method
of weight optimization of \citere{Kl94} and
introduce {\em a-priori weights} $\alpha_k, k=1,\dots, M$ ($\alpha_k \ge 0$
and $\sum_{k=1}^M \alpha_k=1$).
The channel $k$ that is used to generate the event is picked randomly
with probability $\alpha_k$, i.e.\
\begin{eqnarray}
I_n &=& \sum_{k=1}^M \alpha_k \int_V {\mathrm{d}} \vec{\Phi}_k \,
\rho_k\Big(k_i(\vec{\Phi}_k)\Big)
g_k\Big(k_i(\vec{\Phi}_k)\Big)
\, \frac{f\Big(k_i(\vec{\Phi}_k)\Big)}
{g_{\mathrm{tot}}\Big(k_i(\vec{\Phi}_k)\Big)}
\nonumber \\
&=& \int_0^1 {\mathrm{d}} r_0 \, \sum_{k=1}^M
\theta(r_0-\beta_{k-1})\theta(\beta_k-r_0)
\int_V {\mathrm{d}} \vec{\Phi}_k \, \rho_k \Big(k_i(\vec{\Phi}_k)\Big)
g_k\Big(k_i(\vec{\Phi}_k)\Big) \,
\frac{f\Big(k_i(\vec{\Phi}_k)\Big)}
{g_{\mathrm{tot}}\Big(k_i(\vec{\Phi}_k)\Big)} \nonumber\\
&=&\int_0^1 {\mathrm{d}} r_0\,
\sum_{k=1}^M \theta(r_0-\beta_{k-1})\theta(\beta_k-r_0)
\int_0^1 {\mathrm{d}} \vec{r} \,
\frac{f\Big(k_i(\vec{h}_k(\vec{r}\,))\Big)}
{g_{\mathrm{tot}}\Big(k_i(\vec{h}_k(\vec{r}\,))\Big)},
\end{eqnarray}
where $\beta_0=0, \beta_j=\sum_{k=1}^j \alpha_k, j=1,\ldots ,M-1$,
$\beta_M=\sum_{k=1}^M \alpha_k=1$, and
\begin{equation}
g_{\mathrm{tot}}\Big(k_i(\vec{\Phi}_k)\Big)=
\sum_{l=1}^M \alpha_l g_l\Big(k_i(\vec{\Phi}_k)\Big),
\end{equation}
is the total density of the event.
For the processes $\mathswitchr {e^+} \mathswitchr {e^-} \to 4 f$ we have between 6 and 128
different channels, for $\eeffff\gamma$ between 14 and 928 channels.
Each channel smoothens a particular combination of propagators that
results from a characteristic Feynman diagram. We have written
phase-space generators in a generic way for several classes of
channels determined by the chosen set of invariants $s_i, t_i$. The
channels within one class differ in the choice of the mappings
\refeqs{eq:maps1}, \refeqf{eq:maps2}, and \refeqf{eq:nomaps} and the
order of the external particles. We did not include special
channels for interference contributions.
The $\alpha_k$-dependence of the quantity
\begin{equation}
W(\vec{\alpha})= \frac{1}{N} \sum_{j=1}^N [w(r_0^j,{\vec{r}}^{\,j})]^2,
\end{equation}
where $w=f/g_{\mathrm{tot}}$ is the weight assigned to the Monte Carlo point
$(r_0^j, \vec{r}^{\,j})$ of the $j$th event,
can be exploited to minimize the expected Monte Carlo error
\begin{equation}
\delta \bar I_n=\sqrt{\frac{W(\vec{\alpha})-\bar{I}_n^2}{N}},
\end{equation}
with the Monte Carlo estimate of $I_n$
\begin{equation}
\bar I_n= \frac{1}{N} \sum_{j=1}^N w(r_0^j,\vec{r}^{\,j})
\end{equation}
by trying to choose an optimal set of a-priori weights.
We perform the search for an optimal set of $\alpha_k$ by using an
{\it adaptive optimization} method, as described in \citere{Kl94}.
After a certain number of generated events a new set of
a-priori weights $\alpha_k^{\mathrm{new}}$ is calculated according to
\begin{eqnarray}
\alpha_k^{\mathrm{new}} &\propto& \alpha_k
\sqrt{\frac{1}{N}\sum_{j=1}^N \,
\frac{g_k\Big(k_i(\vec{h}_k({\vec{r}}^{\,j}))\Big) \,
[w(r_0^j,{\vec{r}}^{\,j})]^2}
{g_{\mathrm{tot}}\Big(k_i(\vec{h}_k({\vec{r}}^{\,j}))\Big)}}, \qquad
\sum_{k=1}^M \alpha_k^{\mathrm{new}}=1.
\end{eqnarray}
Based on the above approach, we have written two independent Monte
Carlo programs. While the general strategy is similar,
the programs differ in the explicit phase-space generation.
\section{Numerical results}
\label{se:numres}
If not stated otherwise we use the complex-mass scheme and the
following parameters:
\begin{equation}
\begin{array}[b]{rlrl}
\alpha =& 1/128.89, & \qquad \alpha_s =& 0.12, \\
\mathswitch {M_\PW} =& 80.26\unskip\,\mathrm{GeV},& \Gamma_{\PW} =& 2.05\unskip\,\mathrm{GeV}, \\
\mathswitch {M_\PZ} =& 91.1884\unskip\,\mathrm{GeV},& \Gamma_{\PZ} =& 2.46\unskip\,\mathrm{GeV}.
\end{array}
\end{equation}
In the complex-mass scheme, the weak mixing angle is defined in
\refeq{complangle}, in all other schemes it
is fixed by $\mathswitch {c_\Pw}=\mathswitch {M_\PW}/\mathswitch {M_\PZ}$, $\mathswitch {s_\Pw}^2=1-\mathswitch {c_\Pw}^2$.
The energy in the centre-of-mass (CM) system of the incoming electron
and positron
is denoted by $\sqrt{s}$.
Concerning the phase-space integration, we apply
the canonical cuts of the ADLO/TH detector,
\begin{equation}
\begin{array}[b]{rlrlrl}
\theta (l,\mathrm{beam})> & 10^\circ, & \qquad
\theta( l, l^\prime)> & 5^\circ, & \qquad
\theta( l, q)> & 5^\circ, \\
\theta (\gamma,\mathrm{beam})> & 1^\circ, &
\theta( \gamma, l)> & 5^\circ, &
\theta( \gamma, q)> & 5^\circ, \\
E_\gamma> & 0.1\unskip\,\mathrm{GeV}, & E_l> & 1\unskip\,\mathrm{GeV}, & E_q> & 3\unskip\,\mathrm{GeV}, \\
m(q,q')> & 5\unskip\,\mathrm{GeV},
\end{array}
\label{eq:canonicalcuts}
\end{equation}
where $\theta(i,j)$ specifies the angle between the particles $i$ and
$j$ in the CM system,
and $l$, $q$, $\gamma$, and ``beam'' denote charged leptons,
quarks, photons, and the beam electrons or positrons, respectively.
The invariant mass of a quark pair $qq'$ is denoted by
$m(q,q')$.
The cuts coincide with those defined in \citere{CERN9601mcgen},
except for the additional angular cut between charged leptons.
The canonical cuts exclude all collinear and infrared singularities
from phase space for all processes.
Although our helicity amplitudes and Monte Carlo programs
allow for a treatment of arbitrary polarization configurations, we
consider only unpolarized quantities in this paper.
All results are produced with $10^7$ events.
The calculation of the cross section for
$\mathswitchr {e^+}\mathswitchr {e^-}\to\mathswitchr {e^+}\mathswitchr {e^-}\mu^+\mu^-$ requires
about 50 minutes on a DEC ALPHA workstation with 500 MHz,
the calculation of the cross section for
$\mathswitchr {e^+}\mathswitchr {e^-}\to\mathswitchr {e^+}\mathswitchr {e^-}\mu^+\mu^-\gamma$ takes about 5 hours.
The results of our two Monte Carlo programs agree very well.
The numbers in
parentheses in the following tables
correspond to the statistical errors
of the results of the Monte Carlo integrations.
\subsection{Comparison with existing results}
In order to compare our results for $\Pep\Pem\to 4f$ with Tables 6--8
of \citere{CERN9601table},
we use the corresponding set of phase-space cuts and input parameters,
i.e.\ the canonical cuts defined in \refeq{eq:canonicalcuts},
a CM energy of $\sqrt{s}=190\unskip\,\mathrm{GeV}$, and the parameters
$\alpha=\alpha (2 \mathswitch {M_\PW})=1/128.07$, $\alpha_s=0.12$,
$\mathswitch {M_\PW}=80.23\unskip\,\mathrm{GeV}$, $\Gamma_{\PW}=2.0337\unskip\,\mathrm{GeV}$, $\mathswitch {M_\PZ}=91.1888\unskip\,\mathrm{GeV}$, and $\Gamma_{\PZ}=2.4974\unskip\,\mathrm{GeV}$.
The value of $\mathswitch {s_\Pw}$,
which enters the couplings, is calculated from
$\alpha (2 \mathswitch {M_\PW})/(2 \mathswitch {s_\Pw}^2)=\mathswitch {G_\mu} \mathswitch {M_\PW}^2/(\pi \sqrt{2})$
with $\mathswitch {G_\mu}=1.16639\times 10^{-5} \unskip\,\mathrm{GeV}^{-2}$.
\begin{table}
\renewcommand{1.2}{1.1}
\newdimen\digitwidth
\setbox0=\hbox{0}
\digitwidth=\wd0
\catcode`!=\active
\def!{\kern\digitwidth}
\newdimen\dotwidth
\setbox0=\hbox{$.$}
\dotwidth=\wd0
\catcode`?=\active
\def?{\kern\dotwidth}
\begin{center}
{\begin{tabular}{|c||r@{}l|r@{}l|r@{}l|}
\hline
$\sigma/\unskip\,\mathrm{fb}$ &
\multicolumn{2}{c|}{$\begin{array}{c}
\mathswitchr {e^+}\mathswitchr {e^-} \to 4 f \\
\mbox{running width}
\end{array}$} &
\multicolumn{2}{c|}{$\begin{array}{c}
\mathswitchr {e^+}\mathswitchr {e^-} \to 4 f \\
\mbox{constant width}
\end{array}$} &
\multicolumn{2}{c|}{$\begin{array}{c}
\eeffff\gamma \\
\mbox{constant width}
\end{array}$}
\\\hline\hline
$\mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}} \mathswitchr {e^-} \mathswitchr {e^+}$
&\hspace{1.15cm}$ 256.7 $&$( 3)$
&\hspace{1.15cm}$ 257.1 $&$( 7)$
&\hspace{0.85cm}$ 89.4 $&$( 2)$
\\\hline
$\nu_\mu \mu^+ \mathswitchr {e^-} \mathswitch \bar\nu_{\mathrm{e}}$
&$ 227.4 $&$( 1)$
&$ 227.5 $&$( 1)$
&$ 79.1 $&$( 1)$
\\\hline
$\nu_\mu \bar{\nu}_\mu \mu^- \mu^+$
&$ 228.7 $&$( 1)$
&$ 228.8 $&$( 1)$
&$ 81.0 $&$( 2)$
\\\hline
$\nu_\mu \mu^+ \tau^- \bar{\nu}_\tau$
&$ 218.55 $&$( 9)$
&$ 218.57 $&$( 9)$
&$ 76.7 $&$( 1)$
\\\hline
$\mathswitchr {e^-} \mathswitchr {e^+} \mathswitchr {e^-} \mathswitchr {e^+}$
&$ 109.1 $&$( 3)$
&$ 109.4 $&$( 3)$
&$ 38.8 $&$( 4)$
\\\hline
$\mathswitchr {e^-} \mathswitchr {e^+} \mu^- \mu^+$
&$ 116.6 $&$( 3)$
&$ 116.4 $&$( 3)$
&$ 43.4 $&$( 4)$
\\\hline
$\mu^- \mu^+ \mu^- \mu^+$
&$ 5.478 $&$( 5)$
&$ 5.478 $&$( 5)$
&$ 3.37 $&$( 1)$
\\\hline
$\mu^- \mu^+ \tau^- \tau^+$
&$ 11.02 $&$( 1)$
&$ 11.02 $&$( 1)$
&$ 6.78 $&$( 3)$
\\\hline
$\mathswitchr {e^-} \mathswitchr {e^+} \nu_\mu \bar{\nu}_\mu $
&$ 14.174 $&$( 9)$
&$ 14.150 $&$( 9)$
&$ 5.36 $&$( 1)$
\\\hline
$\mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}} \mu^- \mu^+$
&$ 17.78 $&$( 6)$
&$ 17.73 $&$( 6)$
&$ 6.63 $&$( 2)$
\\\hline
$\nu_\tau \bar{\nu}_\tau\mu^- \mu^+$
&$ 10.108 $&$( 8)$
&$ 10.103 $&$( 8)$
&$ 4.259 $&$( 9)$
\\\hline
$\mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}} \mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}}$
&$ 4.089 $&$( 1)$
&$ 4.082 $&$( 1)$
&$ 0.7278 $&$( 7)$
\\\hline
$\mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}} \nu_\mu \bar{\nu}_\mu$
&$ 8.354 $&$( 2)$
&$ 8.337 $&$( 2)$
&$ 1.512 $&$( 1)$
\\\hline
$\nu_\mu \bar{\nu}_\mu \nu_\mu \bar{\nu}_\mu $
&$ 4.069 $&$( 1)$
&$ 4.057 $&$( 1)$
&$ 0.7434 $&$( 7)$
\\\hline
$\nu_\mu \bar{\nu}_\mu \nu_\tau \bar{\nu}_\tau$
&$ 8.241 $&$( 2)$
&$ 8.218 $&$( 2)$
&$ 1.511 $&$( 1)$
\\\hline
$\mathswitchr u\, \bar{\mathswitchr d}\, \mathswitchr {e^-} \mathswitch \bar\nu_{\mathrm{e}}$
&$ 693.5 $&$( 3)$
&$ 693.6 $&$( 3)$
&$ 220.8 $&$( 4)$
\\\hline
$\mathswitchr u\, \bar{\mathswitchr d}\, \mu^- \bar{\nu}_\mu$
&$ 666.7 $&$( 3)$
&$ 666.7 $&$( 3)$
&$ 214.5 $&$( 4)$
\\\hline
$\mathswitchr {e^-} \mathswitchr {e^+} \mathswitchr u\, \bar{\mathswitchr u}$
&$ 86.87 $&$( 9)$
&$ 86.82 $&$( 9)$
&$ 32.3 $&$( 2)$
\\\hline
$\mathswitchr {e^-} \mathswitchr {e^+} \mathswitchr d\, \bar{\mathswitchr d}$
&$ 43.02 $&$( 4)$
&$ 42.95 $&$( 4)$
&$ 16.17 $&$( 8)$
\\\hline
$\mathswitchr u\, \bar{\mathswitchr u}\, \mu^- \mu^+$
&$ 24.69 $&$( 2)$
&$ 24.69 $&$( 2)$
&$ 12.70 $&$( 4)$
\\\hline
$\mathswitchr d\, \bar{\mathswitchr d}\, \mu^- \mu^+$
&$ 23.73 $&$( 1)$
&$ 23.73 $&$( 1)$
&$ 10.43 $&$( 2)$
\\\hline
$\mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}} \mathswitchr u\, \bar{\mathswitchr u}$
&$ 24.00 $&$( 2)$
&$ 23.95 $&$( 2)$
&$ 6.84 $&$( 1)$
\\\hline
$\mathswitch \nu_{\mathrm{e}} \mathswitch \bar\nu_{\mathrm{e}} \mathswitchr d\, \bar{\mathswitchr d}$
&$ 20.657 $&$( 8)$
&$ 20.62 $&$( 1)$
&$ 4.319 $&$( 6)$
\\\hline
$\mathswitchr u\, \bar{\mathswitchr u}\, \nu_\mu \bar{\nu}_\mu$
&$ 21.080 $&$( 5)$
&$ 21.050 $&$( 5)$
&$ 6.018 $&$( 9)$
\\\hline
$\mathswitchr d\, \bar{\mathswitchr d}\, \nu_\mu \bar{\nu}_\mu$
&$ 19.863 $&$( 5)!!!$
&$ 19.817 $&$( 5)!!!$
&$ 4.156 $&$( 5)!!$
\\\hline
$\mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr d\, \bar{\mathswitchr d}$
&\multicolumn{2}{c|}{$ 2064.1 ( 9), !2140.8 ( 9)$}
&\multicolumn{2}{c|}{$ 2064.3 ( 9), !!?2141 ( 1)$}
&\multicolumn{2}{c|}{$ !?615 ( 1), !!?672 ( 1)$}
\\\hline
$\mathswitchr u\, \bar{\mathswitchr d}\, \mathswitchr s\, \bar{\mathswitchr c}$
&$ 2015.2 $&$( 8)$
&$ 2015.3 $&$( 8)$
&$ 598 $&$( 1)$
\\\hline
$\mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr u\, \bar{\mathswitchr u}$
&\multicolumn{2}{c|}{$ 25.738 ( 7), !!71.28 ( 4)$}
&\multicolumn{2}{c|}{$ 25.721 ( 7), !!71.30 ( 4)$}
&\multicolumn{2}{c|}{$ !9.78 ( 2), !! 42.1 ( 1)$}
\\\hline
$\mathswitchr d\, \bar{\mathswitchr d}\, \mathswitchr d\, \bar{\mathswitchr d}$
&\multicolumn{2}{c|}{$ 23.494 ( 6), !!51.35 ( 3)$}
&\multicolumn{2}{c|}{$ 23.448 ( 6), !!51.32 ( 3)$}
&\multicolumn{2}{c|}{$ 5.527 ( 7), !28.68 ( 4)$}
\\\hline
$\mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr c\, \bar{\mathswitchr c}$
&\multicolumn{2}{c|}{$ !51.61 ( 1), !144.72 ( 9)$}
&\multicolumn{2}{c|}{$ !51.57 ( 1), !144.75 ( 9)$}
&\multicolumn{2}{c|}{$ 19.61 ( 4), !! 86.1 ( 2)$}
\\\hline
$\mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr s\, \bar{\mathswitchr s}$
&\multicolumn{2}{c|}{$ !49.68 ( 1), !126.52 ( 8)$}
&\multicolumn{2}{c|}{$ !49.62 ( 1), !126.52 ( 8)$}
&\multicolumn{2}{c|}{$ 15.17 ( 2), !! 75.1 ( 2)$}
\\\hline
$\mathswitchr d\, \bar{\mathswitchr d}\, \mathswitchr s\, \bar{\mathswitchr s}$
&\multicolumn{2}{c|}{$ !47.13 ( 1), !104.79 ( 6)$}
&\multicolumn{2}{c|}{$ !47.02 ( 1), !104.74 ( 6)$}
&\multicolumn{2}{c|}{$ 11.10 ( 2), !! 59.2 ( 1)$}
\\\hline
\end{tabular}}
\end{center}
\caption[]{Integrated cross sections for all representative processes
$\Pep\Pem\to 4f$ with running widths and constant widths and for the
corresponding processes $\eeffff\gamma$ with constant widths.
If two numbers are given, the first results from
pure electroweak diagrams and the second involves in addition
gluon-exchange contributions.}
\label{ta:yellowreport}
\end{table}
In \refta{ta:yellowreport}, we list the integrated cross sections for
various processes $\Pep\Pem\to 4f$ with running widths and
constant widths,
and for the corresponding processes $\eeffff\gamma$
with constant widths. For processes involving gluon-exchange
diagrams we give the cross sections resulting from the purely
electroweak diagrams and those including the gluon-exchange
contributions. The latter results include also the interference terms
between purely electroweak and gluon-exchange diagrams.
In \refta{ta:yellowreport} we provide a complete list of processes
for vanishing fermion masses. All processes $\Pep\Pem\to 4f(\gamma)$ not explicitly
listed are equivalent to one of the given processes.
For NC processes $\Pep\Pem\to 4f$
with four neutrinos or four quarks in the final state
we find small deviations of roughly $0.2\%$ between the results with
constant and running widths. Assuming that a running width has been
used in \citere{CERN9601table}, we find very good agreement.
Unfortunately we cannot compare with most of the publications
\cite{vO94,vO96,Fu94,Ca97} for the bremsstrahlung processes
$\eeffff\gamma$. In those papers, either the cuts are not (completely) specified,
or collinear photon emission is not excluded, and the corresponding
fermion-mass effects are taken into account. Note that the
contributions of collinear photons dominate the results given there.
We have compared our results with the ones given in \citeres{Ae91,Ae91a},
where the total cross sections for $\eeffff\gamma$ have been calculated for
the purely leptonic and the semi-leptonic final states. As done in
\citeres{Ae91,Ae91a} only diagrams involving two resonant \mathswitchr W~bosons have
been taken into account
for this comparison. Table~\ref{ta:aepplitable} contains our
results corresponding to \refta{ta:aepplitable} of \citere{Ae91}.
Based on \citeres{Ae91,Ae91a},
we have chosen $\sqrt{s}=200\unskip\,\mathrm{GeV}$ and the input parameters
$\alpha=1/137.03599$, $\mathswitch {M_\PW}=80.9\unskip\,\mathrm{GeV}$, $\Gamma_{\PW}=2.14\unskip\,\mathrm{GeV}$, $\mathswitch {M_\PZ}=91.16\unskip\,\mathrm{GeV}$, $\Gamma_{\PZ}=2.46\unskip\,\mathrm{GeV}$,
$\mathswitch {s_\Pw}$
obtained from $\alpha/(2 \mathswitch {s_\Pw}^2)=\mathswitch {G_\mu} \mathswitch {M_\PW}^2/(\pi \sqrt{2})$ with
$\mathswitch {G_\mu}=1.16637\times 10^{-5} \unskip\,\mathrm{GeV}^{-2}$, and constant gauge-boson
widths.
The energy of the photon is
required to be larger than $E_{\gamma,\min}$,
and the angle between the
photon and any charged fermion must be larger than
$\theta_{\gamma,\min}$. A maximal photon energy is required,
$E_\gamma<60\unskip\,\mathrm{GeV}$, in order to exclude contributions from the $\mathswitchr Z$
resonance. Our results
are consistent with those of \citeres{Ae91,Ae91a} within
the statistical error of 1\% given there. In some cases we find
deviations of 2\%.%
\footnote{Note that the input specified in \citeres{Ae91,Ae91a} is not
completely clear even if the information of both publications is
combined.}
\begin{table}
\newdimen\digitwidth
\setbox0=\hbox{0}
\digitwidth=\wd0
\catcode`!=\active
\def!{\kern\digitwidth}
\begin{center}
{\begin{tabular}{|l|c|r@{}l|r@{}l|r@{}l|r@{}l|}
\hline
\multicolumn{2}{|r|}{$E_{\gamma,\min}=$} &
\multicolumn{2}{c|}{$1\unskip\,\mathrm{GeV}$} &
\multicolumn{2}{c|}{$5\unskip\,\mathrm{GeV}$} &
\multicolumn{2}{c|}{$10\unskip\,\mathrm{GeV}$} &
\multicolumn{2}{c|}{$15\unskip\,\mathrm{GeV}$} \\
\hline
&$\theta_{\gamma,\min}$&
\multicolumn{8}{c|}{$\sigma/\mathrm{fb}$} \\\hline\hline
&$ !1^\circ $
&$ 53.54 $&$( 8)$
&$ 27.57 $&$( 3)$
&$ 16.96 $&$( 2)$
&$ 11.22 $&$( 2)$
\\\cline{2-10}
leptonic
&$!5^\circ $
&$ 32.65 $&$( 4)$
&$ 16.98 $&$( 3)$
&$ 10.48 $&$( 2)$
&$ 6.94 $&$( 1)$
\\\cline{2-10}
process
&$10^\circ $
&$ 23.48 $&$( 3)$
&$ 12.30 $&$( 2)$
&$ 7.61 $&$( 2)$
&$ 5.04 $&$( 1)$
\\\cline{2-10}
&$15^\circ $
&$ 18.03 $&$( 2)$
&$ 9.51 $&$( 2)$
&$ 5.90 $&$( 1)$
&$ 3.90 $&$( 1)$
\\\hline\hline
&$!1^\circ $
&$ 141.9 $&$( 2)$
&$ 71.90 $&$( 8)$
&$ 43.56 $&$( 5)$
&$ 28.26 $&$( 4)$
\\\cline{2-10}
semi-leptonic
&$!5^\circ $
&$ 86.8 $&$( 1)$
&$ 44.25 $&$( 6)$
&$ 26.78 $&$( 4)$
&$ 17.40 $&$( 3)$
\\\cline{2-10}
process
&$10^\circ $
&$ 62.29 $&$( 7)$
&$ 31.92 $&$( 5)$
&$ 19.40 $&$( 4)$
&$ 12.61 $&$( 3)$
\\\cline{2-10}
&$15^\circ $
&$ 47.42 $&$( 5)$
&$ 24.50 $&$( 4)$
&$ 14.97 $&$( 3)$
&$ 9.77 $&$( 2)$
\\\hline
\end{tabular}}
\end{center}
\caption[]{Comparison with Table 2 of \citere{Ae91}:
Cross sections resulting from diagrams involving two resonant W~bosons
for purely
leptonic and semi-leptonic final states and several photon separation
cuts
}
\label{ta:aepplitable}
\end{table}
\subsection{Comparison of finite-width schemes}
{}As discussed in \citeres{bhf2,FLscheme},
particular care has to be
taken when implementing the finite gauge-boson widths. Differences
between results obtained with running or constant widths can already
be seen in \refta{ta:yellowreport}, where a typical LEP2 energy is
considered.
\begin{table}
\begin{center}
{\begin{tabular}{|c|c|r@{}l|r@{}l|r@{}l|r@{}l|}
\hline
\multicolumn{1}{|c|}{$\sigma/\unskip\,\mathrm{fb}$} &
\multicolumn{1}{r|}{$\sqrt{s}=$} &
\multicolumn{2}{c|}{$189\unskip\,\mathrm{GeV}$} &
\multicolumn{2}{c|}{$500\unskip\,\mathrm{GeV}$} &
\multicolumn{2}{c|}{$2\unskip\,\mathrm{TeV}$} &
\multicolumn{2}{c|}{$10\unskip\,\mathrm{TeV}$}
\\\hline\hline
& constant width
&$ 703.5 $&$( 3)$
&$ 237.4 $&$( 1)$
&$ 13.99 $&$( 2)$
&$ 0.624 $&$( 3)$
\\\cline{2-10}
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr d}\, \mu^- \bar{\nu}_\mu $
& running width
&$ 703.4 $&$( 3)$
&$ 238.9 $&$( 1)$
&$ 34.39 $&$( 3)$
&$ 498.8 $&$( 1)$
\\\cline{2-10}
& complex-mass scheme
&$ 703.1 $&$( 3)$
&$ 237.3 $&$( 1)$
&$ 13.98 $&$( 2)$
&$ 0.624 $&$( 3)$
\\\hline\hline
& constant width
&$ 224.0 $&$( 4)$
&$ 83.4 $&$( 3)$
&$ 6.98 $&$( 5)$
&$ 0.457 $&$( 6)$
\\\cline{2-10}
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr d}\, \mu^- \bar{\nu}_\mu \,\gamma$
& running width
&$ 224.6 $&$( 4)$
&$ 84.2 $&$( 3)$
&$ 19.2 $&$( 1)$
&$ 368 $&$( 6)$
\\\cline{2-10}
& complex-mass scheme
&$ 223.9 $&$( 4)$
&$ 83.3 $&$( 3)$
&$ 6.98 $&$( 5)$
&$ 0.460 $&$( 6)$
\\\hline\hline
& constant width
&$ 730.2 $&$( 3)$
&$ 395.3 $&$( 2)$
&$ 211.0 $&$( 2)$
&$ 32.38 $&$( 6)$
\\\cline{2-10}
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr d}\, \mathswitchr {e^-} \mathswitch \bar\nu_{\mathrm{e}}$
& running width
&$ 729.8 $&$( 3)$
&$ 396.9 $&$( 2)$
&$ 231.5 $&$( 2)$
&$ 530.2 $&$( 6)$
\\\cline{2-10}
& complex-mass scheme
&$ 729.8 $&$( 3)$
&$ 395.1 $&$( 2)$
&$ 210.9 $&$( 2)$
&$ 32.37 $&$( 6)$
\\\hline\hline
& constant width
&$ 230.0 $&$( 4)$
&$ 136.5 $&$( 5)$
&$ 84.0 $&$( 7)$
&$ 16.8 $&$( 5)$
\\\cline{2-10}
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr d}\, \mathswitchr e^- \mathswitch \bar\nu_{\mathrm{e}} \,\gamma $
& running width
&$ 230.6 $&$( 4)$
&$ 137.3 $&$( 5)$
&$ 95.7 $&$( 7)$
&$ 379 $&$( 6)$
\\\cline{2-10}
& complex-mass scheme
&$ 229.9 $&$( 4)$
&$ 136.4 $&$( 5)$
&$ 84.1 $&$( 6)$
&$ 16.8 $&$( 5)$
\\\hline
\end{tabular}}
\end{center}
\caption[]{Comparison of different width schemes for several processes
and energies}
\label{ta:width}
\end{table}
In \refta{ta:width} we compare predictions for integrated cross
sections obtained
by using a constant width, a running width, or the
complex-mass scheme for several energies. We consider
two semi-leptonic final states for $\Pep\Pem\to 4f(\gamma)$.
The numbers show that the constant width and the complex-mass scheme
yield the same results within the statistical accuracy for
$\Pep\Pem\to 4f$ and
$\eeffff\gamma$. In contrast, the results with the running width produce
totally wrong results for high energies. The difference of the
running width with respect to the other implementations of the finite
width is up to $1\%$ already for $500\unskip\,\mathrm{GeV}$. Thus, the running width
should not be used for linear-collider energies. As already stated
above, our default treatment of the finite width is the complex-mass
scheme in the following.
\subsection{Survey of photon-energy spectra}
\label{se:photonspectra}
In \reffi{fi:photonspectra} we show the photon-energy spectra of several
processes for the typical LEP2 energy of $189\unskip\,\mathrm{GeV}$ and a possible
linear-collider energy of $500\unskip\,\mathrm{GeV}$.
The upper plots contain CC and CC/NC processes, the plots in the
middle and the lower plots contain NC processes.
\begin{figure}
\centerline{
\setlength{\unitlength}{1cm}
\begin{picture}(7.2,7)
\put(0,0){\special{psfile=1.189.eps hscale=100 vscale=100
angle=0 hoffset=-85 voffset=-440}}
\put(-0.3,4.9){\makebox(1,1)[c]{$\frac{{\mathrm{d}} \sigma}{{\mathrm{d}} E_\gamma}/
\frac{\unskip\,\mathrm{fb}}\unskip\,\mathrm{GeV}$}}
\end{picture}
\begin{picture}(7.2,7)
\put(0,0){\special{psfile=1.500.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\end{picture}
}
\centerline{
\setlength{\unitlength}{1cm}
\begin{picture}(7.2,7)
\put(0,0){\special{psfile=2.189.eps hscale=100 vscale=100
angle=0 hoffset=-85 voffset=-440}}
\put(-0.3,5.15){\makebox(1,1)[c]{$\frac{{\mathrm{d}} \sigma}{{\mathrm{d}} E_\gamma}/
\frac{\unskip\,\mathrm{fb}}\unskip\,\mathrm{GeV}$}}
\end{picture}
\begin{picture}(7.2,7)
\put(0,0){\special{psfile=2.500.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\end{picture}
}
\centerline{
\setlength{\unitlength}{1cm}
\begin{picture}(7.2,7)
\put(0,0){\special{psfile=3.189.eps hscale=100 vscale=100
angle=0 hoffset=-85 voffset=-440}}
\put(-0.3,5.5){\makebox(1,1)[c]{$\frac{{\mathrm{d}} \sigma}{{\mathrm{d}} E_\gamma}/
\frac{\unskip\,\mathrm{fb}}\unskip\,\mathrm{GeV}$}}
\put(4.5,-0.3){\makebox(1,1)[cc]{{$E_\gamma/\unskip\,\mathrm{GeV}$}}}
\end{picture}
\begin{picture}(7.2,7)
\put(0,0){\special{psfile=3.500.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(3.5,-0.3){\makebox(1,1)[cc]{{$E_\gamma/\unskip\,\mathrm{GeV}$}}}
\end{picture}
}
\caption[]{Photon-energy spectra for several processes and
for $\sqrt{s}=189\unskip\,\mathrm{GeV}$ and $500\unskip\,\mathrm{GeV}$}
\label{fi:photonspectra}
\end{figure}
Several spectra show threshold or peaking structures. These structures
are caused by diagrams in which the
photon is emitted from the initial state. The two important classes of
diagrams are shown in \reffi{fi:resonance}.
\begin{figure}
{
\begin{center}
\begin{picture}(200,120)(-5,0)
\ArrowLine(0,10)(40,50)
\ArrowLine(40,70)(0,110)
\Photon(50,60)(150,60){2}{11}
\Photon(50,63)(120,90){2}{8}
\Photon(50,57)(120,30){2}{8}
\Vertex(120,90){2}
\Vertex(120,30){2}
\ArrowLine(120,90)(150,110)
\ArrowLine(150,70)(120,90)
\ArrowLine(120,30)(150,50)
\ArrowLine(150,10)(120,30)
\GCirc(50,60){20}{0.8}
\put(95,24){\makebox(1,1)[c]{$V_2$}}
\put(95,96){\makebox(1,1)[c]{$V_1$}}
\put(158,60){\makebox(1,1)[c]{$\gamma$}}
\Text(-10,120)[rt]{a)}
\end{picture}
\begin{picture}(200,120)
\ArrowLine(0,10)(40,50)
\ArrowLine(40,70)(0,110)
\Photon(50,63)(150,100){2}{12}
\Photon(50,57)(110,30){2}{8}
\Vertex(110,30){2}
\ArrowLine(110,30)(140,50)\ArrowLine(140,50)(170,70)
\ArrowLine(150,10)(110,30)
\Vertex(140,50){2}
\Photon(140,50)(170,30){2}{4}
\Vertex(170,30){2}
\ArrowLine(170,30)(200,50)
\ArrowLine(200,10)(170,30)
\GCirc(50,60){20}{0.8}
\put(85,26){\makebox(1,1)[c]{$Z$}}
\put(153,28){\makebox(1,1)[c]{$V_3$}}
\put(158,99){\makebox(1,1)[c]{$\gamma$}}
\Text(-10,120)[rt]{b)}
\end{picture}
\end{center}
}
\caption[]{Diagrams for important subprocesses, where
$V_1,V_2=\mathswitchr W,\mathswitchr Z,\gamma$, and $V_3=\gamma,\mathswitchr g$ }
\label{fi:resonance}
\end{figure}
The first class, shown in \reffi{fi:resonance}a, corresponds
to triple-gauge-boson-production subprocesses which yield dominant
contributions as long as the two virtual gauge bosons $V_1$ and $V_2$
can become simultaneously resonant. If the real photon takes the
energy $E_\gamma$, defined
in the CM system, only the energy $\sqrt{s'}$, with
\begin{equation}
s' = s-2 \sqrt{s}\, E_\gamma,
\end{equation}
is available for the production of the gauge-boson pair $V_1V_2$.
If at least one of the gauge bosons is massive,
and if the photon becomes too hard,
the two gauge bosons cannot be produced
on shell anymore, so that the spectrum falls off for $E_\gamma$ above
the corresponding threshold $E_\gamma^{V_1 V_2}$.
Using the threshold condition for the on-shell production of the
$V_1V_2$ pair,
\begin{equation}
\sqrt{s'}>M_{V_1}+M_{V_2},
\end{equation}
the value of $E_\gamma^{V_1 V_2}$ is determined by
\begin{equation}
E_\gamma^{V_1 V_2} = \frac{s-(M_{V_1}+M_{V_2})^2}{2\sqrt{s}}.
\end{equation}
The values of the photon energies
that cause such thresholds
can be found in \refta{ta:resonances}.
\begin{table}
\renewcommand{1.2}{1.1}
\begin{center}
{\begin{tabular}{|c|c|c|c|c|c|c|c|c|}
\hline
$\sqrt{s}/\unskip\,\mathrm{GeV}$ &
\multicolumn{4}{c|}{$189$} &
\multicolumn{4}{c|}{$500$}
\\ \hline \hline
$V_1 V_2$ &
$\mathswitchr W \mathswitchr W$ & $\mathswitchr Z \mathswitchr Z$ & $\gamma \mathswitchr Z$ & $\gamma \gamma$ &
$\mathswitchr W \mathswitchr W$ & $\mathswitchr Z \mathswitchr Z$ & $\gamma \mathswitchr Z$ & $\gamma \gamma$
\\ \hline
$E_\gamma^{V_1 V_2}/\unskip\,\mathrm{GeV}$ &
$26.3$ & $6.5$ & $72.5$& $94.5$ &
$224$ & $217$ & $242$ & $250$
\\ \hline
\end{tabular}}
\end{center}
\caption[]{Photon energies $E^{V_1 V_2}_\gamma$ corresponding to thresholds}
\label{ta:resonances}
\end{table}
The value $E^{\gamma\gamma}_\gamma$ corresponds to the
upper endpoint of the photon-energy spectrum, which is given by the beam
energy $\sqrt{s}/2$.
Since $\sqrt{s'}$ is fully determined by $s$ and
$E_\gamma$, the contribution of the $V_1 V_2$-production subprocess to the
$E_\gamma$ spectrum qualitatively follows the energy
dependence of the total cross section for $V_1 V_2$ production (cf.\
\citere{CERN9601table}, Fig.~1) above the corresponding thresholds.
The cross sections for $\gamma\ga$ and $\gamma\mathswitchr Z$ production strongly
increase with decreasing energy, while the ones for $\mathswitchr Z\PZ$ and
$\mathswitchr W\PW$ production are comparably flat.
Thus, the $\gamma\ga$ and $\gamma\mathswitchr Z$-production subprocesses introduce contributions
in the photon-energy spectra with resonance-like structures,
whereas the ones with $\mathswitchr Z\PZ$ or $\mathswitchr W\PW$ pairs yield
edges.
The second class of important diagrams, shown in
\reffi{fi:resonance}b corresponds to the production of a photon and a
resonant \mathswitchr Z~boson that decays into four fermions. These diagrams are
important
if the gauge boson $V_3$
is also resonant, i.e.\ a photon or a gluon with small
invariant mass. In this case, the kinematics fixes the energy of the
real photon to
\begin{equation}
E_\gamma = E_{\gamma}^{\gamma\mathswitchr Z}=\frac{s-\mathswitch {M_\PZ}^2}{2\sqrt{s}},
\end{equation}
which corresponds to the $\gamma\mathswitchr Z$ threshold in
\refta{ta:resonances}. This subprocess gives rise to resonance
structures at $E_{\gamma}^{\gamma\mathswitchr Z}$, which are even enhanced by
$\alpha_{\mathrm{s}}/\alpha$ in the presence of gluon exchange.
In the photon-energy spectra of \reffi{fi:photonspectra} all these
threshold and resonance effects are visible. The effect of the $\gamma
\mathswitchr Z$ peak can be nicely seen in different photon-energy spectra, in
particular in those where gluon-exchange diagrams contribute (cf.\ also
\reffi{fi:QCD}). The effect of the WW threshold is present in the
upper two plots of \reffi{fi:photonspectra}. In the plot for
$\sqrt{s}=189\unskip\,\mathrm{GeV}$ the threshold for single W production causes the
steep drop of the spectrum for the pure CC processes above $70\unskip\,\mathrm{GeV}$.
Note that the CC cross sections are an order of magnitude larger than
the NC cross sections if the WW channel is open. The ZZ threshold is
visible in the middle and lower plots for $\sqrt{s}=500\unskip\,\mathrm{GeV}$. The
$\gamma$Z threshold (resulting from the graphs of \reffi{fi:resonance}a)
is superimposed on the $\gamma \mathswitchr Z$ peak (resulting from the graphs of
\reffi{fi:resonance}b) and therefore best recognizable in those
channels where the $\gamma \mathswitchr Z$ peak is absent or suppressed, i.e.\ where a
neutrino pair is present in the final state or where at least no
gluon-exchange diagrams contribute. Processes with four neutrinos in
the final state do not involve photonic diagrams and are therefore
small above the ZZ threshold. The effects of the
triple-photon-production subprocess appear as a tendency of some
photon-energy spectra to increase near the maximal value of $E_\gamma$
for two charged fermion--antifermion pairs in the final state.
\subsection{Triple-gauge-boson-production subprocesses}
In \reffi{fi:signal} we compare predictions that are based on the full
set of diagrams with those that include only the graphs associated
with the triple-gauge-boson-production subprocesses, i.e.\ the graphs
in \reffi{fi:resonance}a. In addition we consider the contributions
of the $\mathswitchr Z\PZ\gamma$-production subprocess alone.
\begin{figure}
\centerline{
\setlength{\unitlength}{1.1cm}
\begin{picture}(14.5,6.3)
\put(0.8,0){\special{psfile=CC.189.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(5.1,0){\special{psfile=CC.500.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(9.4,0){\special{psfile=CC.2000.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(0,5.5){\makebox(1,1)[c]{$\frac{{\mathrm{d}} \sigma}{{\mathrm{d}} E_\gamma}/
\frac{\unskip\,\mathrm{fb}}\unskip\,\mathrm{GeV}$}}
\end{picture}
}
\centerline{
\setlength{\unitlength}{1.1cm}
\begin{picture}(14.5,6.3)
\put(0.8,0){\special{psfile=NC.189.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(5.1,0){\special{psfile=NC.500.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(9.4,0){\special{psfile=NC.2000.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(0,5.95){\makebox(1,1)[c]{$\frac{{\mathrm{d}} \sigma}{{\mathrm{d}} E_\gamma}/
\frac{\unskip\,\mathrm{fb}}\unskip\,\mathrm{GeV}$}}
\put(7.5,-0.3){\makebox(1,1)[cc]{{$E_\gamma/\unskip\,\mathrm{GeV}$}}}
\end{picture}
}
\caption[]{Photon-energy spectra resulting from the
triple-gauge-boson-production subprocesses compared to those
resulting from all diagrams ($V_1V_2$ includes ZZ, $\gamma$Z, and
$\gamma\ga$)}
\label{fi:signal}
\end{figure}
For CC processes, the photon-energy spectra resulting from the
$\mathswitchr {W^+}\mathswitchr {W^-}\gamma$-production subprocess are close to those resulting from
all diagrams at LEP2 energies, but large differences are found for
higher energies and $\mathswitchr e^\pm$ in the final state. Note that the
spectra are shown on a logarithmic scale. Even at LEP2 energies the
differences between the predictions for different final states may be
important, as can be seen, for instance, in \refta{ta:yellowreport} by
comparing the cross sections of $\mathswitchr {e^+}\mathswitchr {e^-}\to
\mathswitchr u\,\bar{\mathswitchr d}\,\mu^-\bar\nu_\mu\gamma$ and $\mathswitchr {e^+}\mathswitchr {e^-}\to
\mathswitchr u\,\bar{\mathswitchr d}\,\mathswitchr {e^-}\mathswitch \bar\nu_{\mathrm{e}}\gamma$. In the case of NC processes, already
for $189\unskip\,\mathrm{GeV}$ the contributions from $\mathswitchr Z\PZ\gamma$, $\mathswitchr Z\gamma\ga$, and
$\gamma\ga\gamma$ production are not sufficient: in the vicinity of the
$\gamma\mathswitchr Z$ peak sizeable contributions result from the
$\gamma\mathswitchr Z$-production subprocess (\reffi{fi:resonance}b) even for the
$\mu^+\mu^-\mathswitchr u\,\bar{\mathswitchr u}\,\gamma$ final state. For
$\mathswitchr {e^+}\mathswitchr {e^-}\to\mathswitchr {e^-}\mathswitchr {e^+}\mathswitchr u\,\bar{\mathswitchr u}\,\gamma$ other diagrams become dominating
everywhere. The contribution of $\mathswitchr Z\PZ\gamma$ production is always
small and could only be enhanced by invariant-mass cuts. Note that
the triple-gauge-boson-production diagrams form a gauge-invariant
subset for NC processes, while this is not the case for CC processes.
\subsection{Relevance of gluon-exchange contributions}
\label{se:QCD}
In the analytical calculation of the matrix elements for
$\Pep\Pem\to 4f(\gamma)$ in \refse{se:anres} we have seen that NC processes
with four quarks in the final state involve, besides purely electroweak,
also gluon-exchange diagrams. Table \ref{ta:QCD}
illustrates the impact of these diagrams on the integrated cross
sections for a CM energy of $500\unskip\,\mathrm{GeV}$.
\begin{table}
\begin{center}
{\begin{tabular}{|l|r@{}l|r@{}l|r@{}l|r@{}l|}
\hline
\multicolumn{1}{|c|}{$\sigma/\unskip\,\mathrm{fb}$} &
\multicolumn{2}{c|}{ew and gluon} &
\multicolumn{2}{c|}{purely ew} &
\multicolumn{2}{c|}{gluon} &
\multicolumn{2}{c|}{interference}
\\\hline\hline
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr c\, \bar{\mathswitchr c}$
&\hspace{0.5cm}$ 52.98 $&$( 4)$
&$ 21.560 $&$( 6)$
&\hspace{0.2cm}$ 31.38 $&$( 3)$
&\hspace{0.35cm}$ 0.04 $&$( 5)$
\\\hline
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr c\, \bar{\mathswitchr c}\, \gamma$
&$ 29.8 $&$( 1)$
&$ 10.38 $&$( 4)$
&$ 19.6 $&$( 1)$
&$ -0.1 $&$( 1)$
\\\hline
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr u} \,\mathswitchr u\, \bar{\mathswitchr u}$
&$ 26.25 $&$( 2)$
&$ 10.765 $&$( 3)$
&$ 15.34 $&$( 1)$
&$ 0.14 $&$( 2)$
\\\hline
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\, \bar{\mathswitchr u}\, \mathswitchr u\, \bar{\mathswitchr u}\, \gamma$
&$ 14.83 $&$( 7)$
&$ 5.16 $&$( 2)$
&$ 9.52 $&$( 5)$
&$ 0.15 $&$( 9)$
\\\hline
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr d\, \bar{\mathswitchr d}\, \mathswitchr u\, \bar{\mathswitchr u}$
&$ 901.2 $&$( 6)$
&$ 876.4 $&$( 5)$
&$ 24.24 $&$( 2)$
&$ 0.6 $&$( 8)$
\\\hline
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr d\, \bar{\mathswitchr d}\, \mathswitchr u\, \bar{\mathswitchr u}\, \gamma$
&$ 290 $&$( 1)$
&$ 275 $&$( 1)$
&$ 14.82 $&$( 8)$
&$ 0 $&$( 1)$
\\\hline
\end{tabular}}
\end{center}
\caption[]{Full lowest order cross section (ew and gluon) and
contributions of purely electroweak diagrams (ew), of
gluon-exchange diagrams (gluon), and their interference for $500\unskip\,\mathrm{GeV}$}
\label{ta:QCD}
\end{table}
The results for the interference are obtained by subtracting the
purely electroweak and the gluon contribution from the total cross
section. For pure NC processes the contributions of gluon-exchange
diagrams dominate over the purely electroweak graphs. This can be
understood from the fact that the gluon-exchange diagrams are enhanced
by the strong coupling constant, and, as discussed in
\refse{se:photonspectra}, that the diagrams with gluons replaced by
photons yield a sizeable contribution to the cross section. For the
mixed CC/NC processes the purely electroweak diagrams dominate the
cross section. Here, the contributions from the
$\mathswitchr {W^+}\mathswitchr {W^-}\gamma$-production subprocess are large compared to all other
diagrams, even if the latter are enhanced by the strong coupling. At
$500\unskip\,\mathrm{GeV}$ the gluon-exchange diagrams contribute to the cross section
at the level of several per cent. The interference contributions are
relatively small. As discussed at the end of \refse{se:hadfinstat},
this is due to the fact that interfering electroweak and
gluon-exchange diagrams involve different resonances. Note that the
interference vanishes for $\mathswitchr {e^+} \mathswitchr {e^-} \to
\mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr c\,\bar{\mathswitchr c}\,\gamma$, and the corresponding numbers in
\refta{ta:QCD} are only due to the Monte Carlo integration error.
In \reffi{fi:QCD} we show the photon-energy spectra for the processes
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr d\,\bar{\mathswitchr d}\,\gamma$ and $\mathswitchr {e^+} \mathswitchr {e^-} \to
\mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr u\,\bar{\mathswitchr u}\,\gamma$ together with the separate
contributions from purely electroweak and gluon-exchange diagrams.
The pure electroweak contributions are similar to the ones for
$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\,\bar{\mathswitchr d}\,\mu^-\bar\nu_\mu \,\gamma$ and $\mathswitchr {e^+} \mathswitchr {e^-} \to
\mathswitchr {e^-} \mathswitchr {e^+} \mathswitchr u\, \bar{\mathswitchr u} \,\gamma$ in \reffi{fi:photonspectra}. For the NC
process $\mathswitchr {e^+}\mathswitchr {e^-} \to \mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr u\,\bar{\mathswitchr u}\,\gamma$, the
photon-energy spectrum is dominated by the gluon-exchange
contribution, which shows a strong peak at $72.5\unskip\,\mathrm{GeV}$ owing to the
$\gamma\mathswitchr Z$-production subprocess. For the CC/NC process $\mathswitchr {e^+} \mathswitchr {e^-} \to
\mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr d\,\bar{\mathswitchr d}\,\gamma$, the electroweak diagrams dominate
below the WW threshold, whereas the gluon-exchange diagrams dominate
at the $\gamma\mathswitchr Z$ peak and above. The interference between purely
electroweak and gluon-exchange diagrams is generally small.
\begin{figure}
\centerline{
\setlength{\unitlength}{1cm}
\begin{picture}(7.2,8)
\put(0,0){\special{psfile=MC.189.qcd.eps hscale=100 vscale=100
angle=0 hoffset=-85 voffset=-440}}
\put(-0.3,5.2){\makebox(1,1)[c]{$\frac{{\mathrm{d}} \sigma}{{\mathrm{d}} E_\gamma}/
\frac{\unskip\,\mathrm{fb}}\unskip\,\mathrm{GeV}$}}
\put(4.5,-0.3){\makebox(1,1)[cc]{{$E_\gamma/\unskip\,\mathrm{GeV}$}}}
\put(5.2,6.7){\makebox(1,1)[cc]{\small
{$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr d\,\bar{\mathswitchr d}\,\gamma$}}}
\end{picture}
\begin{picture}(7.2,8)
\put(0,0){\special{psfile=NC.189.qcd.eps hscale=100 vscale=100
angle=0 hoffset=-105 voffset=-440}}
\put(3.5,-0.3){\makebox(1,1)[cc]{{$E_\gamma/\unskip\,\mathrm{GeV}$}}}
\put(4.5,6.7){\makebox(1,1)[cc]{
{\small$\mathswitchr {e^+} \mathswitchr {e^-} \to \mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr u\,\bar{\mathswitchr u}\,\gamma$}}}
\end{picture}
}
\caption[]{Electroweak and gluon-exchange contributions to the
photon-energy spectra for $\mathswitchr {e^+} \mathswitchr {e^-} \to
\mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr d\,\bar{\mathswitchr d}\,\gamma$ and $\mathswitchr {e^+} \mathswitchr {e^-} \to
\mathswitchr u\,\bar{\mathswitchr u}\,\mathswitchr u\,\bar{\mathswitchr u}\,\gamma$ at $\sqrt{s}=189\unskip\,\mathrm{GeV}$}
\label{fi:QCD}
\end{figure}
\section{Summary and outlook}
\label{se:sum}
The class of processes $\mathswitchr {e^+}\mathswitchr {e^-}\to 4\,\mbox{fermions} + \gamma$ has been
discussed in detail. After classifying the different final states
according to their production mechanism, the sets of all Feynman
graphs are reduced to two generic subsets that are related to the two
basic graphs of the non-radiative processes $\mathswitchr {e^+}\mathswitchr {e^-}\to
4\,\mbox{fermions}$. In this way, all helicity matrix elements are
expressed in terms of two generic functions. Using the
Weyl--van~der~Waerden spinor formalism, we have given compact
expressions for these functions. We wrote two independent Monte Carlo
programs, both using the multi-channel integration technique and an
adaptive weight optimization procedure to reduce the Monte Carlo
error. The results obtained with the two Monte Carlo programs agree
well within the integration error.
The detailed discussion of numerical results comprises a survey of
integrated cross sections and photon-energy spectra for all different
final states. Moreover, we have numerically compared different ways to
introduce finite decay widths of the massive gauge bosons. Similar to
the known results for $\mathswitchr {e^+}\mathswitchr {e^-}\to 4\,\mbox{fermions}$, we find that
the application of running gauge-boson widths leads to totally wrong
results for the radiative processes. Using constant widths
consistently, leads to meaningful predictions. For the considered
observables, the results for constant widths practically coincide with
those obtained in a complex-mass scheme that fully respects gauge
invariance. In the latter scheme gauge-boson masses are treated as
complex parameters everywhere, in particular, leading to complex
couplings. Similar to the situation for the well-known non-radiative
processes, we find that for precise predictions the diagrams
corresponding to triple-gauge-boson-production subprocesses are not
sufficient and the inclusion of the other graphs is mandatory.
Finally, we have investigated the relevance of gluon-exchange
contributions. In general, both the purely electroweak contributions
and the gluon-exchange contributions are relevant and either one can be
dominant depending on the process and the considered observable. The
interference of both contributions is at most at the per-cent level.
In this paper, we have assumed that the radiated photon appears as a
detectable particle in the final state, i.e.\ soft and collinear photons
are excluded by cuts. The inclusion of soft and collinear photons is,
however, necessary if $\mathswitchr {e^+}\mathswitchr {e^-}\to 4\,\mbox{fermions}+\gamma$ is considered
as a correction to four-fermion production. The analytical results of
this paper and the constructed Monte Carlo programs
will be used as a building block in the
evaluation of four-fermion production in $\mathswitchr {e^+}\mathswitchr {e^-}$ collisions
including ${\cal O}(\alpha)$ corrections.
|
\section{Introduction}
The last few years were marked by an outburst of research devoted to
the problem of reconstruction of quantum states for various physical
systems (see, e.g., Ref.~\cite{BM99} for an extensive list of the
literature on the subject). The problem, as stated already in the
fifties by Fano \cite{Fano57} and Pauli \cite{Pauli58}, is to
determine the density matrix $\rho$ from information obtained by a
set of measurements performed on an ensemble of identically prepared
systems. Significant theoretical and experimental progress has been
achieved during the last decade in the reconstruction of quantum
states of the light field \cite{Leon:book}. Also, numerous works
were devoted to reconstruction methods for other physical systems.
Most recently, a general theory of quantum-state reconstruction for
physical systems with Lie-group symmetries was developed \cite{BM99}.
In the present work we consider state-reconstruction methods
for some quantum systems possessing SU(2) symmetry. The principal
procedure for the reconstruction of spin states was recently
presented by Agarwal \cite{Agar98}. A similar approach was also
proposed by Dodonov and Man'ko \cite{DoMa97}, while the basic idea
underlying this method goes back to the pioneering work
by Royer \cite{Royer}. In brief, one applies a phase-space
displacement [specifically, a rotation in the SU(2) case] to
the initial quantum state and then measures the probability to
find the displaced system in a specific state (the so-called
``quantum ruler'' state). Repeating this procedure with identically
prepared systems for many phase-space points [many rotation angles
in the SU(2) case], one determines a function on the phase space
(the so-called operational phase-space probability distribution
\cite{Wod84,BKK95,Ban98}).
In particular, by measuring the population of the ground state,
one obtains the so-called $Q$ function. The information contained
in the operational phase-space probability distribution is
sufficient to completely reconstruct the unknown density matrix of
the initial quantum state. A general group-theoretical description
of this method and some examples, including SU(2), are presented
in Ref.~\cite{BM99}.
The aim of the present paper is to study how the general
state-reconstruction procedure outlined above can be implemented
in practice for a number of specific physical systems with SU(2)
symmetry. Three systems are considered: a collection of two-level
atoms, a two-mode quantized radiation field with a fixed total
number of photons, and a single laser-cooled ion in a
two-dimensional harmonic trap with a fixed total number of
vibrational quanta. We show that a simple rearrangement of
conventional spectroscopic and interferometric schemes enables one
to measure unknown quantum states of these systems.
\section{Reconstruction of quantum states for systems with SU(2)
symmetry}
We start with some basic properties of SU(2) which is the dynamical
symmetry group for the angular momentum or spin and for many other
systems (e.g., a collection of two-level atoms, the Stokes
operators describing the polarization of the quantized light field,
two light modes with a fixed total photon number, etc.). The su(2)
simple Lie algebra consists of the three operators
$\{J_{x},J_{y},J_{z}\}$,
\begin{equation}
\label{su2alg}
[J_{p},J_{r}] = i \epsilon_{p r t} J_{t} .
\end{equation}
The Casimir operator is a constant times the unit operator,
${\mathbf{J}}^2 = j(j+1) I$, for any unitary irreducible
representation of the SU(2) group; so the representations are
labeled by the single index $j$ that takes the values
$j = 0,1/2,1,3/2,\ldots$. The representation Hilbert space
${\cal H}_{j}$ is spanned by the complete orthonormal basis
$\{ |j,\mu\rangle \}$ (where $\mu=j,j-1,\ldots,-j$):
\[
{\mathbf{J}}^2 |j,\mu\rangle = j(j+1) |j,\mu\rangle , \hspace{8mm}
J_z |j,\mu\rangle = \mu |j,\mu\rangle .
\]
In the following we assume that the state $|\psi\rangle$ of the
system belongs to ${\cal H}_{j}$ (or, for mixed states, that the
density matrix $\rho$ is an operator on ${\cal H}_{j}$).
Group elements can be
parametrized using the Euler angles $\alpha,\beta,\gamma$:
\begin{equation}
\label{eq:gEuler}
g=g(\alpha,\beta,\gamma) = e^{ i \alpha J_{z} }
e^{ i \beta J_{y} } e^{ i \gamma J_{z} } .
\end{equation}
We will employ two very useful concepts: the phase space (which
is the group coset space of maximum symmetry) and the coherent
states (each point of the phase space corresponds to a coherent
state). For SU(2), the phase space is the unit sphere
${\Bbb{S}}^2 = {\rm SU}(2) / {\rm U}(1)$, and each coherent
state is characterized by a unit vector \cite{ACGT72,Per}
\begin{equation}
{\bf n} = (\sin\theta \cos\phi, \sin\theta \sin\phi,
\cos\theta) .
\end{equation}
Specifically, the coherent states $|j;{\bf n}\rangle$ are given by
the action of the group element
\begin{equation}
\label{eq:oSU2}
g({\bf n}) = e^{- i \phi J_{z} } e^{- i \theta J_{y} }
\end{equation}
on the highest-weight state $|j,j\rangle$:
\begin{eqnarray}
|j;{\bf n}\rangle = g({\bf n}) |j,j\rangle
& = & \sum_{\mu=-j}^{j} {2j \choose j+\mu}^{1/2}
\cos^{j+\mu}(\theta/2) \nonumber \\
& & \times \sin^{j-\mu}(\theta/2) e^{- i \mu \phi} |j,\mu\rangle .
\label{jcohstates}
\end{eqnarray}
An important property of the coherent states is the resolution of
the identity:
\begin{equation}
\frac{2j+1}{4\pi} \int_{{\Bbb{S}}^2} d {\bf n}\, |j;{\bf n}\rangle
\langle j;{\bf n}| = I ,
\end{equation}
where $d {\bf n} = \sin\theta\, d \theta\, d \phi$.
A possible procedure for the quantum-state reconstruction is as
follows \cite{BM99,Agar98,DoMa97}. First, the system, whose initial
state is described by the density matrix $\rho$, is displaced in
the phase space:
\begin{equation}
\label{eq:displ}
\rho \rightarrow \rho({\bf n}) = g^{-1}({\bf n}) \rho g({\bf n}) ,
\hspace{8mm} {\bf n} \in {\Bbb{S}}^2 .
\end{equation}
Then one measures the probability to find the displaced system in
one of the states $|j,\mu\rangle$ (e.g., in the highest state
$|j,j\rangle$). This probability
\begin{equation}
\label{eq:pmu-def}
p_{\mu} ({\bf n}) = \langle j,\mu| \rho({\bf n}) |j,\mu\rangle
\end{equation}
(which is sometimes called the operational phase-space probability
distribution) can be formally considered as the expectation value
\begin{equation}
p_{\mu} ({\bf n}) = {\rm Tr}\, [\rho \Gamma_{\mu} ({\bf n})]
\end{equation}
of the so-called displaced projector
\begin{equation}
\Gamma_{\mu} ({\bf n}) = g({\bf n}) |j,\mu\rangle
\langle j,\mu| g^{-1}({\bf n}) .
\end{equation}
Repeating this procedure (with a large number of identically
prepared systems) for a large number of phase-space points ${\bf n}$,
one can determine the function $p_{\mu} ({\bf n})$.
Knowledge of the function $p_{\mu} ({\bf n})$ is sufficient for the
reconstruction of the initial density matrix $\rho$.
We can use the following expansion for the density matrix (such an
expansion exists for any operator on ${\cal H}_j$):
\begin{equation}
\rho = \sum_{l=0}^{2j} \sum_{m=-l}^{l} {\cal R}_{l m} D_{l m} ,
\hspace{6mm} {\cal R}_{l m} = {\rm Tr}\, (\rho D^{\dagger}_{l m}) .
\end{equation}
Here, $D_{l m}$ are the so-called tensor operators (also known in
the context of angular momentum as the Fano multipole operators
\cite{Fano53}),
\begin{equation}
D_{l m} = \sqrt{\frac{2l+1}{2j+1}} \sum_{k,q=-j}^{j}
\langle j,k;l,m|j,q \rangle |j,q \rangle \langle j,k| ,
\end{equation}
where $\langle j_{1},m_{1};j_{2},m_{2}|j,m \rangle$
are the Clebsch-Gordan coefficients.
Now, one can reconstruct the density matrix by using the relation
\cite{BM99,Agar98}
\begin{equation}
{\cal R}_{l m} = \frac{ \sqrt{(2j+1)/4\pi} }{
\langle j,\mu;l,0|j,\mu \rangle } \int_{{\Bbb{S}}^2} d {\bf n}\,
p_{\mu} ({\bf n}) Y_{l m}^{\ast}({\bf n}) ,
\end{equation}
where $Y_{l m}({\bf n})$ are the spherical harmonics.
Other ways to deduce the density matrix from the measured
probabilities $p_{\mu} ({\bf n})$ were also proposed
\cite{AmWe99}.
Let us also consider the useful concept of phase-space
quasiprobability distributions (QPDs). In the SU(2) case, one can
introduce an $s$-parametrized family of the QPDs
\cite{BM99,Agar81,VaGB89}
\begin{equation}
P({\bf n};s) = \sum_{l=0}^{2j} \sum_{m=-l}^{l}
\frac{ \sqrt{4\pi/(2j+1)} }{ \langle j,j;l,0|j,j \rangle^{s} }
{\cal R}_{l m} Y_{l m}({\bf n}) .
\end{equation}
For $s=0$, we have the SU(2) equivalent of the Wigner function,
\begin{equation}
W({\bf n}) = \sqrt{ \frac{4\pi}{2j+1} }
\sum_{l=0}^{2j} \sum_{m=-l}^{l} {\cal R}_{l m} Y_{l m}({\bf n}) .
\end{equation}
For $s=1$, we obtain the SU(2) equivalent of the Glauber-Sudarshan
function (also known as Berezin's contravariant symbol),
$P({\bf n})$, whose defining property is
\begin{equation}
\rho = \frac{2j+1}{4\pi} \int_{{\Bbb{S}}^2} d {\bf n}\,
P({\bf n}) |j;{\bf n}\rangle \langle j;{\bf n}| .
\end{equation}
The function which is probably the most important for the
reconstruction problem is the SU(2) equivalent of the Husimi
function (also known as Berezin's covariant symbol),
\begin{equation}
Q({\bf n}) = \langle j;{\bf n}| \rho |j;{\bf n}\rangle ,
\end{equation}
obtained for $s=-1$. As is seen from Eq.~(\ref{eq:pmu-def}),
the function $Q({\bf n})$ gives the probability to find the
displaced system in the highest spin state $|j,j\rangle$,
\begin{equation}
\label{eq:Q-p}
Q({\bf n}) = p_j ({\bf n}) .
\end{equation}
Also, one can see that the probability $p_{-j}(\theta,\phi)$
to find the displaced system in the lowest spin state
$|j,-j\rangle$ is equal to $Q(\theta+\pi,\phi)$.
More generally, any one of the QPDs can be reconstructed using
the relation \cite{BM99}
\begin{eqnarray}
&& P({\bf n};s) = \frac{2j+1}{4\pi} \int_{{\Bbb{S}}^2} d {\bf n}'\,
K_{\mu,s}^{-}({\bf n}, {\bf n}')\, p_{\mu} ({\bf n}') , \\
&& K_{\mu,s}^{-}({\bf n}, {\bf n}') = \sum_{l=0}^{2j}
\frac{2l+1}{2j+1} \frac{ \langle j,j;l,0|j,j \rangle^{-s} }{
\langle j,\mu;l,0|j,\mu \rangle } P_l ( {\bf n} \cdot {\bf n}' ) ,
\end{eqnarray}
where $P_l (x)$ are the Legendre polynomials.
For $s=-1$ and $\mu=j$ we recover the relation (\ref{eq:Q-p}).
\section{General description of experimental schemes}
\subsection{Spectroscopy and interferometry}
Quantum transformations which constitute the basic operations
in spectroscopic and interferometric measurements can be
conveniently described as rotations in an abstract 3-dimensional
space. In this description, the system is characterized by the
vector ${\mathbf{J}} = (J_x , J_y , J_z)^T$, where the three
operators $J_x$, $J_y$, and $J_z$ satisfy the su(2) algebra
(\ref{su2alg}).
A spectroscopic or interferometric process is usually described
in the Heisenberg picture as a unitary transformation
\begin{equation}
{\mathbf{J}}_{\mathrm{out}} =
U(\vartheta_1,\vartheta_2,\varphi) {\mathbf{J}}
U^{\dagger}(\vartheta_1,\vartheta_2,\varphi)
= {\mathsf{U}}(\vartheta_1,\vartheta_2,\varphi) {\mathbf{J}} ,
\end{equation}
where ${\mathsf{U}}(\vartheta_1,\vartheta_2,\varphi)$ is a $3 \times 3$
transformation (rotation) matrix, and $\vartheta_1$, $\vartheta_2$,
$\varphi$ are transformation parameters (rotation angles).
A standard transformation consists of three steps:
\begin{enumerate}
\item[(i)] rotation around the $\hat{\mathbf{y}}$ axis by $\vartheta_1$,
with the transformation matrix ${\mathsf{R}}_y (\vartheta_1)$,
\item[(ii)] rotation around the $\hat{\mathbf{z}}$ axis by $\varphi$,
with the transformation matrix ${\mathsf{R}}_z (\varphi)$,
\item[(iii)] rotation around the $\hat{\mathbf{y}}$ axis by
$\vartheta_2$, with the transformation matrix
${\mathsf{R}}_y (\vartheta_2)$.
\end{enumerate}
The overall transformation performed on ${\mathbf{J}}$ is
\begin{equation}
{\mathsf{U}}(\theta,\phi) = {\mathsf{R}}_y (\vartheta_2)
{\mathsf{R}}_z (\varphi) {\mathsf{R}}_y (\vartheta_1) .
\end{equation}
This transformation is slightly more general than those routinely
made in spectroscopy and interferometry. The usual choice is
$\vartheta_2 = -\vartheta_1 = \pm \pi/2$, so
${\mathsf{U}} = {\mathsf{R}}_x (\pm \varphi)$, respectively,
while $\varphi$ is the parameter to be estimated in the experiment.
In the Schr\"{o}dinger picture, the density matrix of the system
transforms as
\begin{equation}
\label{eq:Srot}
\rho_{\mathrm{out}} = U^{\dagger}(\vartheta_1,\vartheta_2,\varphi)
\rho U(\vartheta_1,\vartheta_2,\varphi) ,
\end{equation}
where the transformation operator is
\begin{equation}
\label{eq:Uoperator}
U(\vartheta_1,\vartheta_2,\varphi) = e^{ i \vartheta_1 J_y }
e^{ i \varphi J_z } e^{ i \vartheta_2 J_y } .
\end{equation}
Now, the aim is to measure the value of $\varphi$ which is
proportional to the transition frequency in a spectroscopic
experiment or to the optical path difference between the two arms of
an interferometer.
The information on $\varphi$ is inferred from the measurement of
the observable $J_z$ at the output. The quantum uncertainty
in the estimation of $\varphi$ is
\begin{equation}
\label{eq:uncert}
\Delta \varphi = \frac{\Delta J_{z \mathrm{out}} }{| \partial
\langle J_{z \mathrm{out}} \rangle/ \partial \varphi |} ,
\end{equation}
where the expectation values are taken over the initial quantum
state of the system. This state is assumed to be known, so one
can estimate the value of $\varphi$ and the corresponding
uncertainty.
\subsection{Reconstruction of the initial state}
In this paper we consider how to use the spectroscopic or
interferometric arrangement for the inverse purpose, i.e., for the
measurement of an unknown initial quantum state by means of a large
number of transformations with known parameters.
As discussed in Sec.~II, the first part of the reconstruction
procedure is the phase-space displacement of Eq.~(\ref{eq:displ}).
With the phase space being the sphere, this displacement is just
a rotation produced by the operator $g({\bf n})$ of
Eq.~(\ref{eq:oSU2}). Now, compare this rotation with the one
made during a spectroscopic or interferometric experiment, as
given by Eqs.~(\ref{eq:Srot}) and (\ref{eq:Uoperator}).
One can immediately conclude that the phase-space displacement
needed for the SU(2) state-reconstruction procedure can be neatly
implemented by means of the spectroscopic and interferometric
techniques. One only needs to omit the first rotation (i.e., take
$\vartheta_1 = 0$), and recognize the two spherical angles as:
\begin{equation}
\theta = -\vartheta_2 , \hspace{8mm}
\phi = -\varphi .
\end{equation}
After the rotation $g({\bf n})$ is made, one should
measure the probability $p_{\mu}({\bf n})$ to find the displaced
system in the state $|j,\mu\rangle$. Perhaps the most convenient
choice is to measure the population of the lowest state
$|j,-j\rangle$, which is usually the ground state of the system
(e.g., this state corresponds to the case where all the atoms are
unexcited; in the atomic case such a measurement can be made by
monitoring the resonant fluorescence for an auxiliary dipole
transition).
This procedure should be repeated for many phase-space points
${\bf n}$ with a large number of identically prepared systems,
thereby determining the function $p_{\mu}({\bf n})$ (e.g., for
$\mu = j$ or $\mu = -j$). According to the formalism presented
in Sec.~II, this information is sufficient to reconstruct the
initial quantum state.
\section{Collections of two-level atoms}
In Ramsey spectroscopy \cite{Ramsey} one deals with a collection
of $N$ two-level systems (usually atoms or ions) interacting with
classical light fields. One can equivalently describe this physical
situation as the interaction of $N$ spin-$\frac{1}{2}$ particles
with classical magnetic fields. Denoting by ${\mathbf{S}}_i$ the
spin of $i$th particle, one can use the collective spin operators:
\begin{equation}
{\mathbf{J}} = \sum_{i=1}^{N} {\mathbf{S}}_i .
\end{equation}
The orthonormal basis $\{ |j,\mu\rangle \}$ consists of the symmetric
Dicke states \cite{Dicke54}:
\begin{equation}
|j,\mu\rangle = {N \choose p}^{-1/2} \sum \prod_{k=1}^{p}
|+\rangle_{l_k} \prod_{l \neq l_k} |-\rangle_{l} ,
\end{equation}
where $|+\rangle_l$ and $|-\rangle_l$ are the upper and lower states,
respectively, of the $l$th atom, and the summation is over all
possible permutations of $N$ atoms. If only symmetric states
are considered, then the ``cooperative number'' $j$ is equal to
$N/2$ and $p = \mu+j$ is just the number of excited atoms.
Usually the atoms (ions) are far enough apart so their wave functions
do not overlap and the direct dipole-dipole coupling or other direct
interactions between the atoms may be neglected.
In the spin formulation (see, e.g., Ref.~\cite{Wine94} for a very good
description), the magnetic moment $\bbox{\mu} = \mu_0 {\mathbf{S}}$
is associated with each particle. If a uniform external magnetic
field ${\mathbf{B}}_0 = B_0 \hat{\mathbf{z}}$
is applied, the Hamiltonian for each particle is given by
\begin{equation}
H_0 = - \bbox{\mu} \cdot {\mathbf{B}}_0 = \hbar \omega_0 S_z ,
\end{equation}
where $\hbar \omega_0 = -\mu_0 B_0$ is the separation in energy
between the two levels. The corresponding Heisenberg equation for
the collective spin operator is
\begin{equation}
\partial {\mathbf{J}}/ \partial t = \bbox{\omega}_0 \times
{\mathbf{J}} ,
\end{equation}
where $\bbox{\omega}_0 = \omega_0 \hat{\mathbf{z}}$. Then one
applies the so-called clock radiation which is a classical field
of the form
\begin{equation}
{\mathbf{B}}_{\perp} = B_{\perp} \left( \hat{\mathbf{y}} \cos \omega t
- \hat{\mathbf{x}} \sin \omega t \right) ,
\end{equation}
where $\omega \simeq \omega_0$ and we assume $\omega_0 > 0$.
In the reference frame that rotates at frequency $\omega$, the
collective spin ${\mathbf{J}}$ interacts with the effective field
\begin{equation}
{\mathbf{B}} = B_r \hat{\mathbf{z}} + B_{\perp} \hat{\mathbf{y}} ,
\end{equation}
where $B_r = B_0 (\omega_0 -\omega)/\omega_0$. In the rotating
frame, the Hamiltonian is
$H = - \mu_0 {\mathbf{J}} \cdot {\mathbf{B}}$, and the Heisenberg
equation for $\mathbf{J}$ is
\begin{equation}
\label{eq:Jrot}
\partial {\mathbf{J}}/ \partial t = \bbox{\omega}' \times
{\mathbf{J}} ,
\end{equation}
where $\bbox{\omega}' = (\omega_0 - \omega) \hat{\mathbf{z}}
+ \omega_{\perp} \hat{\mathbf{y}}$ and
$\omega_{\perp} = -\mu_0 B_{\perp}/\hbar$.
The Ramsey method breaks the evolution time of the system into three
parts. In the first part $B_{\perp}$ is nonzero and constant with
value $B_{1}$ during the time interval $0 \leq t \leq t_{\vartheta}$.
During this period (the first Ramsey pulse),
${\mathbf{B}} = B_r \hat{\mathbf{z}} + B_{1} \hat{\mathbf{y}}
\simeq B_{1} \hat{\mathbf{y}}$, where we made the assumption
$|B_{1}| \gg |B_{r}|$, i.e., $|\omega_{1}| \gg |\omega_0 - \omega|$,
with $\omega_{1} = -\mu_0 B_{1}/\hbar$. Therefore, in the rotating
frame of Eq.~(\ref{eq:Jrot}), $\mathbf{J}$ rotates around the
$\hat{\mathbf{y}}$ axis by the angle
$\vartheta_1 = \omega_{1} t_{\vartheta}$.
During the second period, of duration $T$,
(usually $T \gg t_{\vartheta}$), $B_{\perp}$ is zero, so
${\mathbf{B}} = B_r \hat{\mathbf{z}}$, and $\mathbf{J}$ rotates around
the $\hat{\mathbf{z}}$ axis by the angle
$\varphi = (\omega_0 - \omega) T$. The third period is exactly as the
first one but with the field $B_{\perp} = B_{2}$ and the corresponding
angular frequency $\omega_{2} = -\mu_0 B_{2}/\hbar$.
This gives a rotation around the $\hat{\mathbf{y}}$ axis by the
angle $\vartheta_2 = \omega_{2} t_{\vartheta}$. These three Ramsey
pulses provide the rotations we described in Sec.~III~A (usually,
$\vartheta_1 = \vartheta_2 = \pi/2$).
The aim of spectroscopic experiments is to measure the transition
frequency $\omega_0$ (which is equivalent to the measurement of
$\varphi$, as $\omega$ and $T$ are determined by the experimenter).
Usually, one measures the number of atoms in the upper state
$|+\rangle$,
\begin{equation}
N_{+ {\rm out}} = J_{z {\rm out}} + N/2 ,
\end{equation}
and thus obtains the information about the angle $\varphi$ or,
equivalently, about the frequency $\omega_0$. Of course, in order to
infer this information one should know the initial quantum state of
the system. The measurement sensitivity, as seen from
Eq.~(\ref{eq:uncert}), also depends on the initial quantum state.
In the state-reconstruction procedure, the first Ramsey pulse should
be omitted, while the second and third pulses produce the desired
phase-space displacement $g^{\dagger}({\bf n})$. After the displacement
is completed, one should measure the probability to find the system in
one of the states $|j,\mu\rangle$, for example, measure the population
of the ground state $|j,-j\rangle$ or of the most excited state
$|j,j\rangle$.
This measurement can be made by driving a dipole transition to an
auxiliary atomic level and then observing the resonance fluorescence.
The phase space is scanned by repeating the measurement with many
identically prepared systems for various durations of the
Ramsey pulses, $T$ and $t_{\vartheta}$. Of course, the apparatus
should be first calibrated by measuring the transition frequency
$\omega_0$.
\section{Two-mode light fields}
The basic device employed in a passive optical interferometer is
a beam splitter (a partially transparent mirror). A Mach-Zehnder
interferometer consists of two beam splitters and its operation
is as follows. Two light modes (with boson annihilation operators
$a_1$ and $a_2$) are mixed by the first beam splitter,
accumulate phase shifts $\varphi_1$ and $\varphi_2$, respectively,
and then they are once again mixed by the second beam splitter.
Photons in the output modes are counted by two photodetectors.
In fact, a Michelson interferometer works in the same way, but
due to its geometric layout the two beam splitters may coincide.
Each beam splitter has two input and two output ports.
Let ${\bf a} = (a_1 , a_2)^T$ and ${\bf b} = (b_1 , b_2)^T$ be the
column-vectors of the boson operators of the input and output modes,
respectively. Then, in the Heisenberg picture, the action of
the beam splitter is described by the transformation
\begin{equation}
{\bf b} = {\sf B} {\bf a} ,
\end{equation}
where ${\sf B}$ is a $2 \times 2$ matrix. For a lossless beam
splitter ${\sf B}$ must be unitary, thereby assuring the energy
(photon number) conservation. A possible form of ${\sf B}$ is
\begin{equation}
\label{eq:BSmatrix}
{\sf B}(\vartheta) = \left( \begin{array}{cc}
\cos (\vartheta/2) & -\sin (\vartheta/2) \\
\sin (\vartheta/2) & \cos (\vartheta/2) \end{array} \right) ,
\end{equation}
with $T = \cos^2 (\vartheta/2)$ and $R = \sin^2 (\vartheta/2)$
being the transmittance and reflectivity, respectively.
When the two light modes accumulate phase shifts $\varphi_1$ and
$\varphi_2$, respectively, the corresponding transformation is
\begin{equation}
\label{eq:PSmatrix}
{\bf b} = {\sf P} {\bf a} , \hspace{8mm}
{\sf P} = \left( \begin{array}{cc}
e^{ i \varphi_1 } & 0 \\ 0 & e^{ i \varphi_2 } \end{array} \right) .
\end{equation}
The group-theoretic description of the interferometric process
\cite{YMK86} is based on the Schwinger realization of the su(2)
algebra:
\begin{eqnarray}
\label{eq:Schwinger}
& & J_x =
( a_1^{\dagger} a_2 + a_2^{\dagger} a_1 )/2 , \nonumber \\
& & J_y =
- i ( a_1^{\dagger} a_2 - a_2^{\dagger} a_1 )/2 , \\
& & J_z =
( a_1^{\dagger} a_1 - a_2^{\dagger} a_2 )/2 . \nonumber
\end{eqnarray}
Actions of the interferometer elements (mixing by the beam splitters
and phase shifts) can be represented as rotations of the column-vector
${\bf J} = ( J_x , J_y , J_z )^T$.
The beam-splitter transformation of Eq.~(\ref{eq:BSmatrix}) is
represented by rotation ${\sf R}_y (\vartheta)$ around the
$\hat{\mathbf{y}}$ axis by the angle $\vartheta$, and the phase shift
of Eq.~(\ref{eq:PSmatrix}) is represented by rotation
${\sf R}_z (\varphi)$ around the $\hat{\mathbf{z}}$ axis by the angle
$\varphi = \varphi_2 - \varphi_1$.
Now, if the transmittances of the two beam splitters are
$T_1 = \cos^2 (\vartheta_1 /2)$ and $T_2 = \cos^2 (\vartheta_2 /2)$,
respectively, then the interferometer action is given by the
three rotations described in Sec.~III~A. (Usually, one uses 50-50
beam splitters, so $\vartheta_1 = -\vartheta_2 = \pi/2$.)
Interferometers are constructed to measure the relative phase
shift $\varphi$, which is proportional to the optical path difference
between the two arms. Usually, one measures the difference between
the photocurrents due to the two output light beams. This quantity is
proportional to the photon-number difference at the output,
$q_{\mathrm{out}} = 2 J_{z \mathrm{out}}$. If the input state of
light is known, then the measurement of $q_{\mathrm{out}}$ can
be used to infer the phase shift $\varphi$ and estimate the
measurement error due to the quantum fluctuations of the light
field.
A simple calculation gives ${\bf J}^2 = (N/2)(1+N/2)$, where
$N = a_1^{\dagger} a_1 + a_2^{\dagger} a_2$ is the total number
of photons in the two modes. If $N$ has a fixed value for the input
state of the two-mode light field, then this state belongs to the
Hilbert space ${\cal H}_j$ of a specific SU(2) representation with
$j = N/2$. Because $N$ is the SU(2) invariant, this state will remain
in ${\cal H}_j$ during the interferometric process.
Such input states of the two-mode light field can be reconstructed
using a rearrangement of the interferometric scheme, according to the
general procedure described in Secs.~II and III~B.
The phase-space displacement $g^{\dagger}({\bf n})$ needed for the
state-reconstruction procedure can be implemented by using an
interferometer without the first beam splitter. Then one should
measure the probability $p_{\mu}({\bf n})$ to find the output light
in one of the states $|j,\mu\rangle$. Note that these states are
given by
\begin{equation}
\label{eq:staterel}
|j,\mu\rangle = |j+\mu \rangle_1 \otimes |j-\mu \rangle_2
\end{equation}
in the terms of the Fock states of the two light modes.
So, $\mu$ is just one half of the photon-number difference
measured at the output. Averaging over many measurements, one
obtains the probabilities $p_{\mu}({\bf n})$.
For example, $p_{j}({\bf n})$ is the probability that all photons
exit in the first output beam while the number of photons in the
second output beam is zero.
The measurement should be repeated with identically prepared input
light beams for many phase-space displacements. This means that one
needs a well-calibrated apparatus which can be tuned for various
values of the relative phase shift $\varphi$. These phase shifts
can be conveniently produced by moving a mirror with a precise
electro-mechanical system. Various values of the angle
$\vartheta_2$ can be realized using a collection of partially
transparent mirrors with different reflectivities for the second
beam splitter. An alternative possibility is to use the dependence
of the reflectivity on the angle of incidence for light polarized
in the plane of incidence.
In general, the state reconstruction for two-mode light fields is
a tedious task, because the corresponding Hilbert space is very
large \cite{KWV95,RMAL96,OWV97,Richter97,PTKJ97}.
Obviously, this task can be greatly simplified for the subclass of
two-mode states with a fixed total number of photons, by means of
the reconstruction method presented here. However, this method is
in principle suitable also for other two-mode states as well.
In general, the whole Hilbert space of the two-mode system can
be decomposed as
\begin{equation}
\label{eq:decomp}
{\cal H} = \bigoplus_{j} {\cal H}_j .
\end{equation}
The method of inverted interferometry enables one to reconstruct
the part of the density matrix corresponding to each irreducible
subspace ${\cal H}_j$.
One case for which our method is applicable is the subclass of
states, whose density matrices are block-diagonal in terms of the
decomposition (\ref{eq:decomp}).
This means that the corresponding operator can be written as
\begin{equation}
\rho = \sum_{j} \rho_j ,
\end{equation}
where $\rho_j$ is an operator on ${\cal H}_j$. Each component
$\rho_j$ evolves independently during the phase-space displacement;
hence the state of the whole system can be measured by
reconstructing all invariant components $\rho_j$.
The other case for which our method works is the subclass of pure
states,
\begin{equation}
|\psi\rangle = \sum_{j} |\psi_j\rangle , \hspace{8mm}
|\psi_j\rangle = \sum_{\mu = -j}^{j} c_{j\mu} |j,\mu\rangle .
\end{equation}
Then the density matrix can be written as
\begin{equation}
\label{eq:pure-decomp}
\rho = \sum_{j} | \psi_j \rangle \langle \psi_j | +
\sum_{j \neq j'} |\psi_j \rangle \langle \psi_{j'} | .
\end{equation}
The populations of the states $|j,\mu\rangle$ are unaffected by
the second term in (\ref{eq:pure-decomp}), and one can reconstruct
all invariant components
$\rho_j = | \psi_j \rangle \langle \psi_j |$.
This gives information about the state $|\psi\rangle$ of the whole
system, except for relative phases between different
$| \psi_j \rangle$. From the technical point of view,
each measurement of the photon-number difference $2 \mu$, needed
to determine the probabilities $p_{\mu}({\bf n})$, should be
accompanied by a measurement of the photon-number sum $N = 2j$,
in order to determine to which invariant subspace ${\cal H}_j$
does the detected value of $\mu$ correspond.
Consequently, one needs to make many more measurements, in order
to accumulate enough data for each value of $j$.
A technical problem is that quantum efficiencies of realistic
photodetectors are always less then unity.
While this problem is not too serious for the measurement of the
photon-number difference (as long as both detectors have the
same efficiency), it puts a serious limitation on the accuracy
of the measurement of the total number of photons.
\section{Two-dimensional vibrations of a trapped ion}
As was recently demonstrated by Wineland \emph{et al.} \cite{Wine98},
a single laser-cooled ion in a harmonic trap can be used to simulate
various interactions governing many well-known optical processes.
In particular, one can simulate transformations produced by elements
of a Mach-Zehnder optical interferometer.
Consider a single ion confined in a two-dimensional harmonic trap,
with angular frequencies of oscillations in two orthogonal
directions $\Omega_1$ and $\Omega_2$. Two internal states
of the ion, $|+\rangle$ and $|-\rangle$, are separated in energy
by $\hbar \omega_0$. The internal and motional degrees of freedom
can be coupled by applying classical laser beams, with
electric fields of the form
\[ {\bf E}({\bf x},t) = {\bf E}_0 \cos ({\bf k} \cdot {\bf x}
- \omega t + \Phi) . \]
For example, one can apply two laser beams to produce stimulated
Raman transitions. We denote by $\omega = \omega_1 - \omega_2$,
${\bf k} = {\bf k}_1 - {\bf k}_2$, and $\Phi = \Phi_1 - \Phi_2$
the differences between the angular frequencies, the wave
vectors, and the phases, respectively, of the two applied fields.
Then, in the rotating-wave approximation, the interaction
Hamiltonian reads
\begin{equation}
H_I = \hbar \kappa \exp[ i ({\bf k} \cdot {\bf x} - \delta t +
\Phi) ] + {\rm H.c.} ,
\end{equation}
where $\delta = \omega-\omega_0$ is the frequency detuning,
${\bf x}$ is the ion's position relative to its equilibrium, and
$\kappa$ is the coupling constant (the Rabi frequency). Each of
the two modes of the ion's motion can be modelled by a quantum
harmonic oscillator:
\begin{equation}
x_r = x_{0 r} (a_r + a_r^{\dagger}), \hspace{6mm}
x_{0 r} = \sqrt{ \hbar/(2 M \Omega_r) } ,
\end{equation}
where $r=1,2$ and $M$ is the ion's mass. Also, let
$\eta_r = k_r x_{0 r}$ ($r=1,2$) be the Lamb-Dicke parameters for the
two oscillatory modes.
It is convenient to use the interaction picture for the ion's motion:
\begin{eqnarray}
\tilde{H}_I & = & \exp( i H_0 t/\hbar) H_I \exp(- i H_0 t/\hbar)
\nonumber \\
& = & \hbar \kappa e^{ i (\Phi - \delta t)} \prod_{r=1,2}
\exp[ i \eta_r (\tilde{a}_r + \tilde{a}_r^{\dagger}) ] + {\rm H.c.},
\label{eq:Ham2}
\end{eqnarray}
where $H_0$ is the free Hamiltonian for the ion's motion,
\begin{equation}
H_0 = \hbar \Omega_1 \left( a_1^{\dagger} a_1 + \mbox{$\frac{1}{2}$}
\right) + \hbar \Omega_2 \left( a_2^{\dagger} a_2 +
\mbox{$\frac{1}{2}$} \right) ,
\end{equation}
and $\tilde{a}_r = a_r \exp(- i \Omega_r t)$, $r=1,2$.
If the coupling constant $\kappa$ is small enough and $\Omega_1$ and
$\Omega_2$ are incommensurate, one can resonantly excite only one
spectral component of the possible transitions. For a particular
resonance condition $\delta = \Omega_2 - \Omega_1$ (and in the
Lamb-Dicke limit of small $\eta_1$ and $\eta_2$), the product in
Eq.~(\ref{eq:Ham2}) will be dominated by the single term
$( i \eta_1 a_1 )( i \eta_2 a_2^{\dagger} )$. Therefore, one
obtains
\begin{equation}
\label{eq:H-bs}
\tilde{H}_I \approx -\hbar \kappa \eta_1 \eta_2 \left( e^{ i \Phi}
a_1 a_2^{\dagger} + e^{- i \Phi} a_1^{\dagger} a_2 \right).
\end{equation}
Returning to the Schr\"{o}dinger picture, the total evolution
operator reads:
\begin{eqnarray}
U(t) & = & \exp(- i H_0 t/\hbar) \exp(- i \tilde{H}_I t/\hbar)
\nonumber \\
& = & \exp[- i (\Omega_1 + \Omega_2)(N+1) t/2 ]
\exp[ i (\Omega_2 - \Omega_1) J_z t ] \nonumber \\
& & \times \exp( 2 i \kappa \eta_1 \eta_2 J_{\Phi} t ) .
\label{eq:evolution}
\end{eqnarray}
Here, $N = a_1^{\dagger} a_1 + a_2^{\dagger} a_2$ is the total number
of vibrational quanta in the two modes,
$J_{\Phi} = J_x \cos\Phi + J_y \sin\Phi$, and we used the Schwinger
realization (\ref{eq:Schwinger}) for the SU(2) generators.
Now, let us consider only such motional states of the ion for which
$N$ has a fixed value, i.e., which belong to the irreducible Hilbert
space ${\cal H}_j$ (with $j = N/2$). For these states,
the first exponent in (\ref{eq:evolution}) will just produce an
unimportant phase factor and can be omitted. Clearly, the evolution
operator (\ref{eq:evolution}) can be used to simulate the action of
an optical interferometer, with two vibrational modes of a trapped
ion employed instead of two light beams. In order to simulate the
action of a beam splitter, one should apply the interaction
(\ref{eq:H-bs}) during time $t_{\theta}$ and ensure that
$|2 \kappa \eta_1 \eta_2| \gg |\Omega_2 - \Omega_1|$, so the effect
of the free evolution can be neglected. Then, for $\Phi = \pi/2$,
the evolution operator reads
\begin{equation}
\label{eq:Utheta}
U_y (\theta) = \exp( i \theta J_y ) , \hspace{8mm}
\theta = 2 \kappa \eta_1 \eta_2 t_{\theta} .
\end{equation}
A relative phase shift between the two modes can be produced
just by using the free evolution, i.e., with no external laser
fields applied. Letting the system evolve freely during time $T$,
one obtains
\begin{equation}
\label{eq:Uphi}
U_z (\phi) = \exp( i \phi J_z ) , \hspace{8mm}
\phi = (\Omega_2 - \Omega_1) T .
\end{equation}
It is obvious that applying consequently the transformations
(\ref{eq:Uphi}) and (\ref{eq:Utheta}) one will produce the
phase-space displacement $g^{\dagger}({\bf n})$, employed in the
state-reconstruction procedure. The whole phase space can be
scanned by repeating the procedure with identically prepared
systems for various durations $T$ and $t_{\theta}$. Each
phase-space displacement should be followed by the measurement of
the probability $p_{\mu}({\bf n})$ to find the system in one of
the states $|j,\mu\rangle$. For example, $p_{j}({\bf n})$ is
the probability that the first oscillatory mode is excited to the
$N$th level ($N = 2j$) while the second mode is in the ground
state. Such a measurement can be made with the method used
recently by the NIST group \cite{Leibfr} to reconstruct the
one-dimensional motional state of a trapped ion.
The principle of this method is as follows. One of the
oscillatory modes is coupled to the internal transition
$|+\rangle \leftrightarrow |-\rangle$. This is done by applying
one classical laser field, so single-photon transitions are
excited. This results in an interaction of the Jaynes-Cummings
type \cite{JC63} between the oscillatory mode and the internal
transition. Then the population $P_{-}(t)$ of the lower internal
state $|-\rangle$ is measured for various values of the
interaction time $t$ (as we already mentioned, this measurement
can be made by monitoring the resonant fluorescence produced in
an auxiliary dipole transition). If $|-\rangle$ is the internal
state at $t=0$, then the signal averaged over many measurements
is
\[
P_{-}(t) = \frac{1}{2} \left[ 1 +
\sum_{n=0}^{\infty} P_{n} \cos(2 \Omega_{n,n+1} t)
e^{-\gamma_{n} t} \right] ,
\]
where $\Omega_{n,n+1}$ are the Rabi frequencies and $\gamma_{n}$
are the experimentally determined decay constants. This
relation allows one to determine the populations $P_{n}$
of the motional eigenstates $|n\rangle$. By virtue of
Eq.~(\ref{eq:staterel}), this gives the populations $p_{\mu}$ of
the SU(2) states $|j,\mu\rangle$ (with $\mu = n-j$ for the first
mode and $\mu = j-n$ for the second mode). For example, $p_{-j}$
and $p_{j}$ are given by $P_0$ for the first and second modes,
respectively.
\section{Conclusions}
In this paper we presented practical methods for the reconstruction
of quantum states for a number of physical systems with SU(2)
symmetry. All these methods employ the same basic idea---the
measurement of displaced projectors---which in principle is
applicable to any system possessing a Lie-group symmetry. Practical
realizations, of course, vary for different physical systems.
In our approach, we exploited the fact that transformations applied
in conventional spectroscopic and interferometric schemes are, from
the mathematical point of view, just rotations. In the context of
the SU(2) group, these rotations constitute phase-space
displacements needed to implement a part of the reconstruction
procedure. Therefore, the spectroscopic and interferometric
measurements can be easily rearranged in order to enable one to
determine unknown quantum states for an ensemble of identically
prepared systems. As the spectroscopic and interferometric
measurements are known for their high accuracy, we hope that the
corresponding rearrangements will allow accurate reconstructions
of unknown quantum states.
\acknowledgements
This work was supported by the Fund for Promotion of Research
at the Technion and by the Technion VPR Fund.
|
Subsets and Splits